WO2025231714A1

WO2025231714A1 - Method and apparatus for communication

Info

Publication number: WO2025231714A1
Application number: PCT/CN2024/091940
Authority: WO
Inventors: Yiqun Ge; Adam Christian CAVATASSI; Nghia DOAN
Original assignee: Huawei Technologies Co Ltd
Current assignee: Huawei Technologies Co Ltd
Priority date: 2024-05-09
Filing date: 2024-05-09
Publication date: 2025-11-13
Anticipated expiration: 2026-11-09

Abstract

Embodiments of the present application provide a method and an apparatus for communication. The method includes: receiving a first reference signal through a set of communication channels; generating first information based on the first reference signal, where a dimension of the first information is smaller than a dimension of the set of communication channels; and inputting the first information to a first model, to obtain second information, wherein the second information is used to predict a channel state of the set of communication channels. The signal process on the first reference signal through the set of communication channels can reduce the dimension of input of the first model, which reduce computation complexity of the first model. An application of the first model in large-scale channels becomes feasible.

Description

METHOD AND APPARATUS FOR COMMUNICATION

TECHNICAL FIELD

Embodiments of the present application relate to the field of wireless technologies, and more specifically, to a method and apparatus for communication.

BACKGROUND

Artificial intelligence (AI) technology and the machine learning (ML) technology can help to manage or optimize operations of communication network. For example, an AI model, can be used to perform channel estimation, to improve the reliability of communication.

However, the communications system becomes more complex as the number of antenna ports and the bandwidth increases. This greatly increases the scale and computation of the model, making it difficult to apply a model in large-scale communication network.

Therefore, an urgent technical problem that needs to be solved is how to make an application of a model in large-scale communications system feasible.

SUMMARY

Embodiments of the present application provide a method and an apparatus for communication, which can make an application of a model in large-scale communications system feasible.

According to a first aspect, an embodiment of the present application provides a communication method, and the method could be performed by a communication apparatus, the communication apparatus may be a communication device (for example, an electronic device (ED) ) , or a chip, a circuit, or a processing system configured in the communication device. The method includes: receiving a first reference signal through a set of communication channels; generating first information based on the first reference signal, where a dimension of the first information is smaller than a dimension of the set of communication channels; and inputting the first information to a first model, to obtain second information, where the second information is used to predict a channel state of the set of communication channels.

According to the above technical solution, a first model can be used to predict the channel state of the set of communication channels between the base station and the ED. Moreover, taking the low-dimensional first information as input of the first model can reduce the size of the first model, which reduce computation complexity of the first model. An application of the first model in large-scale channels becomes feasible.

With reference to the first aspect, in some embodiments, the method further includes: receiving compression information, where the first information is generated by performing a signal process on the first reference signal, and the compression information is for assisting the signal process.

According to the above technical solution, a signal process on the first reference signal through the set of communication channels can reduce the dimension of the input of the first model, the compression information may indicate one or more parameters related to the signal process, and the ED may perform the signal process based on the compression information.

With reference to the first aspect, in some embodiments, the signal process is used to compress the dimension of the set of communication channels.

According to the above technical solution, the signal process may include one or more dimension-related actions, and these actions may be used for reducing the size and complexity of the first model.

With reference to the first aspect, in some embodiments, the first information includes a low-dimensional channel coefficient vector of the set of communication channels.

According to the above technical solution, the ED may estimate the channel coefficients on the transmitted first reference signal into a channel coefficient vector, and generate a low-dimensional channel coefficient vector from the set of communication channels.

With reference to the first aspect, in some embodiments, the first reference signal is transmitted in a first time interval, the second information is used to predict a channel state of the set of communication channels in a second time interval, and the method further includes: receiving a second reference signal through the set of communication channels in the second time interval; and retraining the first model based on the second information and the second reference signal.

According to the above technical solution, during the channel estimation process, the ED can refine or optimize the first model. The first model may maintain accuracy in the changing communication environment.

With reference to the first aspect, in some embodiments, the method further includes: transmitting information that indicates a retrained first model; and receiving information that indicates a second model, where the second model is generated based on M models that include the retrained first model, M is a positive integer.

With reference to the first aspect, in some embodiments, the M models are from M EDs.

According to the above technical solution, the second model may contain characteristics of M EDs, and each of the M EDs can predict the channel states using the high-performance first model.

With reference to the first aspect, in some embodiments, the inputting the first information to a first model to obtain second information, includes: inputting the first information and third information to the first model to obtain the second information, where the third information indicates one or more of: measurement parameter (s) of the first reference signal; location parameter (s) of an electronic device that receives the first reference signal through the set of communication channels; movement parameter (s) of an electronic device that receives the first reference signal through the set of communication channels; and service parameter (s) corresponding to an electronic device that receives the first reference signal through the set of communication channels.

According to the above technical solution, with the input of multiple information into the first model, the first model can predict the channel state from multiple aspects.

With reference to the first aspect, in some embodiments, the compression information indicates one or more of: a dimension of the first information generated by the signal process; a common basis (U) of the set of communication channels involved in the signal process; a low-dimension matrix (P) of the U, where the P indicates a pattern of the first reference signal involved in the signal process; and a unit scheme, where the unit scheme indicates the signal process involves dividing the set of communication channel based on one or more of: frequency domain, time domain and space domain.

According to the above technical solution, the compression indicates one or more parameters related to the signal process, and the ED can perform the signal process based on the one or more parameters.

With reference to the first aspect, in some embodiments, the first model is generated based on N initial models, and the N initial models are from N EDs, N is a positive integer.

According to the above technical solution, the first model may contain characteristics of N sets of communication channels of the N EDs, each of the N EDs can receive information that indicates the first model from the BS, and each of the N EDs can predict the channel states using the high-performance first model.

With reference to the first aspect, in some embodiments, the method further includes: receiving K third reference signals through the set of communication channels, K is a positive integer; performing the signal process on the K third reference signals, to generate training information; establishing an initial model based on the training information; transmitting information that indicates the initial model; and receiving information that indicates the first model, where the first model is generated based on N initial models from N electronic devices, N is a positive integer.

With reference to the first aspect, in some embodiments, the information that indicates the initial model includes the training information.

According to the above technical solution, N EDs and the base station may collaborate to generate the first model. The first model man contain characteristics of N sets of communication channels of the N EDs, making the first model high performance.

With reference to the first aspect, in some embodiments, the receiving a first reference signal through communication channels, includes: receiving, in a period, reference signals that include the first reference signal through the set of communication channels, the first reference signal is transmitted in a first time interval, the second information is used to predict a channel state of the set of communication channels in a second time interval, and the first time interval and the second time interval are two adjacent time intervals of the period.

According to the above technical solution, the first model can predict future channel state based on current state, reducing the need for real-time data transmission and measuring overhead.

With reference to the first aspect, in some embodiments, the method further includes: transmitting, in part or all of time intervals of the period, information that indicates a channel state of the set of communication channels.

According to the above technical solution, this allows that the ED may not transmit information that indicates the channel state to the BS for each transmission of the reference signal, reducing the transmission consumption.

With reference to the first aspect, in some embodiments, the first model is retrained in part or all of time intervals of the period.

According to the above technical solution, for the variability of the transmission environment between the ED and base station, the ED and/or base station also optimize or retrain the first model during channel estimation process. Therefore, the predication accuracy of the first model may maintain accuracy.

According to a second aspect, an embodiment of the present application provides a communication method, and the method could be performed by a communication apparatus, the communication apparatus may be a communication device (for example, a base station (BS) ) , or a chip, a circuit, or a processing system configured in the communication device. The method includes: transmitting a first reference signal through a set of communication channels, where first information is generated based on the first reference signal, an input of a first model includes the first information, and second information obtained by the first model is used to predict a channel state of the set of communication channels.

Various implementations of the second aspect are corresponding to the various implementations of the first aspect. For the beneficial technical effects of the various implementations of the second aspect, reference may be made to the descriptions of the relevant implementations of the second aspect, which will not be repeated here.

With reference to the second aspect, in some embodiments, the method further includes: transmitting compression information, where the first information is generated by performing a signal process on the first reference signal, the compression information is for assisting in the signal process.

With reference to the second aspect, in some embodiments, the signal process is used to compress the dimension of the set of communication channels.

With reference to the second aspect, in some embodiments, the first information includes low-dimensional channel coefficient vector of the set of communication channels.

With reference to the second aspect, in some embodiments, the first reference signal is transmitted in a first time interval, the second information is used to predict a channel state of the set of communication channels in a second time interval, and the method further includes: transmitting a second reference signal through the set of communication channels in the second time interval, where the second reference signal and the second information are used to retrain the first model.

With reference to the second aspect, in some embodiments, the method further includes: receiving information that indicates M models that include a retrained first models, M is a positive integer; generating a second model based on the M models; and receiving information that indicates the second model.

With reference to the second aspect, in some embodiments, the M models are from M electronic devices.

With reference to the second aspect, in some embodiments, the input of the first model further includes third information, where the third information indicates one or more of: measurement parameter (s) of the first reference signal; location parameter (s) of an electronic device that receives the first reference signal through the set of communication channels; movement parameter (s) of an electronic device that receives the first reference signal through the set of communication channels; and service parameter (s) corresponding to an electronic device that receives the first reference signal through the set of communication channels.

With reference to the second aspect, in some embodiments, the compression information indicates one or more of:a dimension of the first information generated by the signal process; a common basis (U) of the set of communication channels involved in the signal process; a low-dimension matrix (P) of the U, where the P indicates a pattern of the first reference signal involved in the signal process; and a unit scheme, where the unit scheme indicates the signal process involves dividing the set of communication channel based on one or more of: frequency domain, time domain and space domain.

With reference to the second aspect, in some embodiments, the first model is generated based on N initial models, and the N initial models are from N electronic devices, N is a positive integer.

With reference to the second aspect, in some embodiments, the method further includes: transmitting K third reference signals through N sets of communication channels, where the N sets of communication channels correspond to N electronic devices, K and N are positive integers; receiving information that indicates N initial models from the N electronic devices, where an initial model of the N initial models is generated by training information, and the training information is generated by the K third reference signals; generating the first model based on the N initial models; and transmitting information that indicates the first model.

With reference to the second aspect, in some embodiments, the information that indicates the N initial models includes the training information.

With reference to the second aspect, in some embodiments, the transmitting a first reference signal through communication channels, includes: transmitting, in a period, reference signals that include the first reference signal through the set of communication channels, the first reference signal is transmitted in a first time interval, the second information is used to predict a channel state of the set of communication channels in a second time interval, and the first time interval and the second time interval are two adjacent time intervals of the period.

With reference to the second aspect, in some embodiments, the method further includes: receiving, in part or all of time intervals of the period, information that indicates a channel state of the set of communication channels.

With reference to the second aspect, in some embodiments, the first model is retrained in part or all of time intervals of the period.

With reference to the second aspect, in some embodiments, the method further includes: obtaining the first model and using the first model to predict channel states of the set of communication channels.

According to a third aspect, an embodiment of the present application provides a communication method, and the method could be performed by a communication apparatus, the communication apparatus may be a communication device (for example, a base station (BS) ) , or a chip, a circuit, or a processing system configured in the communication device. The method includes: transmitting K reference signals through N sets of communication channels, where the N sets of communication channels correspond to N electronic devices, K and N are positive integers; receiving information that indicates N initial models, where the N initial models are generated by the K reference signals; and generating a first model based on the N initial models, where a dimension of an input of the first model is smaller than or equal to a dimension of each of the N sets of communication channels, and the first model is used to predict a channel state of the N sets of communication channels.

According to the above technical solution, a base station and at N EDs collaborate to generate a first model, and a dimension of an input of the first model is smaller than or equal to a dimension of each of the N sets of communication channels. The size of the first model based on N sets of communication channels, making the first model in large-scale channels feasible.

With reference to the third aspect, in some embodiments, the method further includes: transmitting (990) information that indicates the first model.

According to the above technical solution, each ED can obtain the first model to predict the channel states, the ED can gain valuable learning outcomes from others. This is particularly valuable in mobile environments, as EDs can use these shared learning experiences to predict the state of their channels in new locations, rather than relying solely on their known state information.

With reference to the third aspect, in some embodiments, an initial model of the N initial models is established based on training information of a corresponding set of communication channels, the training information includes first training information that is generated by performing a signal process on the K sets of reference signal.

According to the above technical solution, the size of the initial model can be reduced by the signal process, and the transmission consumption can be reduced.

With reference to the third aspect, in some embodiments, the method further includes: transmitting compression information, where the compression information is for assisting the signal process.

According to the above technical solution, the compression information may indicate one or more parameters related to the signal process, and the ED may perform the signal process based on the compression information.

With reference to the third aspect, in some embodiments, the signal process is used to compress the dimension of the corresponding set of communication channels.

With reference to the third aspect, in some embodiments, the first training information includes low-dimensional channel coefficient vector of the corresponding set of communication channels.

With reference to the third aspect, in some embodiments, the training information further includes second training information that indicates one or more of: measurement parameter (s) of the K reference signals through the corresponding set of communication channels; location parameter (s) of a corresponding electronic device; movement parameter (s) of a corresponding electronic device; and service parameter (s) of a corresponding electronic device.

With reference to the third aspect, in some embodiments, the information that indicates the N initial models includes part or all of the training information.

With reference to the third aspect, in some embodiments, the compression information indicates one or more of: a dimension of the first training information generated by the signal process; a common basis (U) of each of the N sets of communication channels involved in the signal process; a low-dimension matrix (P) of the U, where the P indicates a pattern of the K reference signal involved in the signal process; and a unit scheme, where the unit scheme indicates the signal process involves dividing each of the N sets of communication channel based on one or more of: frequency domain, time domain and space domain.

According to a fourth aspect, an embodiment of the present application provides a communication method, and the method could be performed by a communication apparatus, the communication apparatus may be a communication device (for example, an electronic device (ED) ) , or a chip, a circuit, or a processing system configured in the communication device. The method includes: receiving K reference signals through a set of communication channels, K is a positive integer; establishing an initial model based on the K reference signals; and transmitting information that indicates the initial model, where the initial model is used to generate a first model, a dimension of an input of the first model is smaller than or equal to a dimension of the set of communication channels, and the first model is used to predict a channel state of the set of communication channels.

With reference to the fourth aspect, in some embodiments, receiving information that indicates the first model.

With reference to the fourth aspect, in some embodiments, the establishing an initial model based on the K reference signals includes: performing a signal process on the K reference signals, to generate first training information; and establishing the initial model based on training information that includes the first training information.

With reference to the fourth aspect, in some embodiments, the method further includes: receiving compression information, where the compression information is for assisting the signal process.

With reference to the fourth aspect, in some embodiments, the signal process is used to compress the dimension of the set of communication channels.

With reference to the fourth aspect, in some embodiments, the first training information includes low-dimensional channel coefficient vector of the set of communication channels.

With reference to the fourth aspect, in some embodiments, the training information further includes second training information, where the second training information that indicates one or more of: measurement parameter (s) of the K reference signals of the set of communication signals; location parameter (s) of an electronic device that receives the K reference signals from the set of communication signals; movement parameter (s) of an electronic device that receives the K reference signals from the set of communication signals; and service parameter (s) corresponding to an electronic device that receives the K reference signals through the set of communication channels.

With reference to the fourth aspect, in some embodiments, the information that indicates the initial model includes the training information.

With reference to the fourth aspect, in some embodiments, the compression information indicates one or more of:a dimension of the first training information generated by the signal process; a common basis (U) of each of the N sets of communication channels involved in the signal process; a low-dimension matrix (P) of the U, where the P indicates a pattern of the K reference signal involved in the signal process; and a unit scheme, where the unit scheme indicates the signal process involves dividing each of the N sets of communication channel based on one or more of: frequency domain, time domain and space domain.

According to a fifth aspect, an ED is provided. The ED includes a function or unit configured to perform the method according to the first aspect or any one of the possible embodiments of the first aspect.

According to a sixth aspect, a BS is provided. The BS includes a function or unit configured to perform the method according to the second aspect or any one of the possible embodiments of the second aspect.

According to a seventh aspect, a system is provided. The system includes: the ED according to the fifth aspect and the BS according to the sixth aspect.

According to an eighth aspect, an BS is provided. The BS includes a function or unit configured to perform the method according to the third aspect or any one of the possible embodiments of the third aspect.

According to a ninth aspect, an ED is provided. The ED includes a function or unit configured to perform the method according to the fourth aspect or any one of the possible embodiments of the fourth aspect.

According to a tenth aspect, a system is provided. The system includes: the BS according to the eighth aspect and the ED according to the ninth aspect.

According to an eleventh aspect, a communication apparatus is provided. The communication apparatus includes at least one processor, and the at least one processor is coupled to at least one memory. The at least one memory is configured to store a computer program or one or more instructions. The at least one processor is configured to: invoke the computer program or the one or more instructions from the at least one memory and run the computer program or the one or more instructions, so that the communication apparatus performs the method in any one of the first aspect or the possible implementations of the first aspect, or the communication apparatus performs the method in any one of the second aspect or the possible implementations of the second aspect, or the communication apparatus performs the method in any one of the third aspect or the possible implementations of the third aspect, or the communication apparatus performs the method in any one of the fourth aspect or the possible implementations of the fourth aspect.

With reference to the eleventh aspect, in some implementations of the eleventh aspect, the communication apparatus may be an ED or a component (for example, a chip or an integrated circuit) installed in the ED. For another example, the communication apparatus may be a BS or a component (for example, a chip or an integrated circuit) installed in the BS.

According to a twelfth aspect, a communication apparatus is provided. The communication apparatus includes a processor and a communications interface. The processor is connected to the communications interface. The processor is configured to execute one or more instructions, and the communications interface is configured to communicate with other network elements under the control of the processor. The processor is enabled to perform the method according to the first aspect, any one of the possible embodiments of the first aspect, the second aspect, the third aspect, the fourth aspect, or any one of the possible embodiments of the above aspects.

According to a thirteenth aspect, a computer storage medium is provided. The computer storage medium stores program code, and the program code is used to execute one or more instructions for the method according to the first aspect, any one of the possible embodiments of the first aspect, the second aspect, the third aspect, the fourth aspect, or any one of the possible embodiments of the above aspects.

According to a fourteenth aspect, this application provides a computer program product including one or more instructions, where when the computer program product runs on a computer, the computer performs the method according to the first aspect, any one of the possible embodiments of the first aspect, the second aspect, the third aspect, the fourth aspect, or any one of the possible embodiments of the above aspects.

According to a fifteenth aspect, this application provides a non-transitory computer-readable medium storing instruction the instructions causing a processor in a device to implement the method according to the first aspect or any one of the possible embodiments of the first aspect, or the second aspect or any one of the possible embodiments of the second aspect, or the third aspect or any one of the possible embodiments of the third aspect, or the fourth aspect or any one of the possible embodiments of the fourth aspect.

According to a sixteenth aspect, this application provides a device configured to perform the method according to the first aspect or any one of the possible embodiments of the first aspect, or the second aspect or any one of the possible embodiments of the second aspect, or the third aspect or any one of the possible embodiments of the third aspect, or the fourth aspect or any one of the possible embodiments of the fourth aspect.

According to a seventeenth aspect, this application provides a processor, configured to execute instructions to cause a device to perform the method according to the first aspect or any one of the possible embodiments of the first aspect, or the second aspect or any one of the possible embodiments of the second aspect, or the third aspect or any one of the possible embodiments of the third aspect, or the fourth aspect or any one of the possible embodiments of the fourth aspect.

According to an eighteenth aspect, this application provides an integrated circuit configure to perform the method according to the first aspect or any one of the possible embodiments of the first aspect, or the second aspect or any one of the possible embodiments of the second aspect, or the third aspect or any one of the possible embodiments of the third aspect, or the fourth aspect or any one of the possible embodiments of the fourth aspect.

According to a nineteenth aspect, this application provides a communication apparatus, comprising a transceiver unit, configured to perform the receiving step according to the first aspect or any one of the possible embodiments of the first aspect, and a processing unit, configured to perform the processing step according to the first aspect or any one of the possible embodiments of the first aspect.

According to a twentieth aspect, this application provides a communication apparatus, comprising a transceiver unit, configured to perform the transmitting step according to the second aspect or any one of the possible embodiments of the second aspect.

According to a twenty-first aspect, this application provides a communication apparatus, comprising a transceiver unit, configured to perform the receiving step according to the third aspect or any one of the possible embodiments of the third aspect, and a processing unit, configured to perform the processing step according to the third aspect or any one of the possible embodiments of the third aspect.

According to a twenty-second aspect, this application provides a communication apparatus, comprising a transceiver unit, configured to perform the receiving step according to the fourth aspect or any one of the possible embodiments of the fourth aspect, and a processing unit, configured to perform the processing step according to the fourth aspect or any one of the possible embodiments of the fourth aspect.

DESCRIPTION OF DRAWINGS

One or more embodiments are exemplarily described by corresponding accompanying drawings, and these exemplary illustrations and accompanying drawings constitute no limitation on the embodiments. Elements with the same reference numerals in the accompanying drawings are illustrated as similar elements, and the drawings are not limited to scale, in which:

FIG. 1 is a schematic diagram of an application scenario according to an embodiment of the present application.

FIG. 2 illustrates an example of a communication system.

FIG. 3 illustrates another example of an electronic device (ED) and a base station.

FIG. 4 is an example of a channel model of a MIMO system.

FIG. 5 is an example of 6G system conceptual structure.

FIG. 6 illustrates an example of T-MIMO channel space.

FIG. 7 illustrates sizes of deep neural networks of different systems.

FIG. 8 is a schematic flowchart of a communication method 800 according to an embodiment of this application.

FIG. 9 is a schematic flowchart of a communication method 900 according to an embodiment of this application.

FIG. 10 is a schematic flowchart of a communication method 1000 according to an embodiment of this application.

FIG. 11 is a schematic diagram of a signal process and a first model according to an embodiment of this application.

FIG. 12 illustrates an example of feature extraction with common basis U.

FIG. 13 illustrates an example of reference-signal selection with placement scheme P.

FIG. 14 illustrates an example of DMD.

FIG. 15 illustrates an example of obtaining ultra low-dimensional space G by an ED.

FIG. 16 illustrates an example of obtaining ultra low-dimensional space with DMD.

FIG. 17 illustrates an example of transmission of compression information.

FIG. 18 illustrates an example of an ED on the move.

FIG. 19 illustrates an example of delivering a retrained model.

FIG. 20 illustrates an example of predicting a channel state corresponding a single ED by a base station.

FIG. 21 illustrates an example of generating a general DNN corresponding to multiple EDs by a base station.

FIG. 22 illustrates an example of interaction between a base station and an ED.

FIG. 23 illustrates an example of transmitting reference signals periodically.

FIG. 24 illustrates an example of receiving reference signals periodically.

FIG. 25 illustrates an example of training a deep neural network by a ED.

FIG. 26 illustrates an example of multiple EDs reporting model parameters.

FIG. 27 illustrates an example of obtaining an averaged deep neural network corresponding to multiple EDs by a base station.

FIGs. 28 and 29 are schematic block diagrams of possible devices according to embodiments of this application.

DESCRIPTION OF EMBODIMENTS

In order to understand features and technical contents of embodiments of the present application in detail, implementations of the embodiments of the present application will be described in detail below with reference to the accompanying drawings, and the attached drawings are only for reference and illustration purposes, and are not intended to limit the embodiments of the present applications. In the following technical descriptions, for ease of explanation, numerous details are set forth to provide a thorough understanding of the disclosed embodiments.

The technical solutions in embodiments of this application may be applied to various communications systems, such as a Global System for Mobile Communications (GSM) , a Code Division Multiple Access (CDMA) system, a Wideband Code Division Multiple Access (WCDMA) system, a general packet radio service (GPRS) system, a Long Term Evolution (LTE) system, an LTE frequency division duplex (FDD) system, an LTE time division duplex (TDD) system, a Universal Mobile Telecommunications System (UMTS) , a Worldwide Interoperability for Microwave Access (WiMAX) communications system, a wireless local area network (WLAN) , a fifth generation (5G) wireless communications system, a new ratio (NR) wireless communications system, a sixth generation (6G) wireless communications system, or other evolving communications systems.

For ease of understanding the embodiments of this application, a communications system shown in FIGs. 1-3 is first used as an example to describe in detail a communications system to which the embodiments of this application are applicable.

FIG. 1 is a schematic diagram of an application scenario according to this application. Referring to FIG. 1, as an illustrative example without limitation, a simplified schematic illustration of a communication system is provided. The communication system 100 comprises a radio access network 120. The radio access network 120 may be a next generation (e.g. sixth generation (6G) or later) radio access network, or a legacy (e.g. 5G, 4G, 3G or 2G) radio access network. One or more communication electronic device (ED) 110a-110j (generically referred to as 110) may be interconnected to one another or connected to one or more network nodes (170a, 170b, generically referred to as 170) in the radio access network 120. A core network 130 may be a part of the communication system and may be dependent or independent of the radio access technology used in the communication system 100. Also the communication system 100 comprises a public switched telephone network (PSTN) 140, the internet 150, and other networks 160.

FIG. 2 illustrates an example communication system 100. In general, the communication system 100 enables multiple wireless or wired elements to communicate data and other content. The purpose of the communication system 100 may be to provide content, such as voice, data, video, and/or text, via broadcast, multicast and unicast, etc. The communication system 100 may operate by sharing resources, such as carrier spectrum bandwidth, between its constituent elements. The communication system 100 may include a terrestrial communication system and/or a non-terrestrial communication system. The communication system 100 may provide a wide range of communication services and applications (such as earth monitoring, remote sensing, passive sensing and positioning, navigation and tracking, autonomous delivery and mobility, etc. ) . The communication system 100 may provide a high degree of availability and robustness through a joint operation of the terrestrial communication system and the non-terrestrial communication system. For example, integrating a non-terrestrial communication system (or components thereof) into a terrestrial communication system can result in what may be considered a heterogeneous network comprising multiple layers. Compared to conventional communication networks, the heterogeneous network may achieve better overall performance through efficient multi-link joint operation, more flexible functionality sharing, and faster physical layer link switching between terrestrial networks and non-terrestrial networks.

The terrestrial communication system and the non-terrestrial communication system could be considered sub-systems of the communication system. In the example shown, the communication system 100 includes electronic devices (ED) 110a-110d (generically referred to as ED 110) , radio access networks (RANs) 120a-120b, non-terrestrial communication network 120c, a core network 130, a public switched telephone network (PSTN) 140, the internet 150, and other networks 160. The RANs 120a-120b include respective base stations (BSs) 170a-170b, which may be generically referred to as terrestrial transmit and receive points (T-TRPs) 170a-170b. The non-terrestrial communication network 120c includes an access node 120c, which may be generically referred to as a non-terrestrial transmit and receive point (NT-TRP) 172.

Any ED 110 may be alternatively or additionally configured to interface, access, or communicate with any other T-TRP 170a-170b and NT-TRP 172, the internet 150, the core network 130, the PSTN 140, the other networks 160, or any combination of the preceding. In some examples, ED 110a may communicate an uplink and/or downlink transmission over an interface 190a with T-TRP 170a. In some examples, the EDs 110a, 110b and 110d may also communicate directly with one another via one or more sidelink air interfaces 190b. In some examples, ED 110d may communicate an uplink and/or downlink transmission over an interface 190c with NT-TRP 172.

The air interfaces 190a and 190b may use similar communication technology, such as any suitable radio access technology. For example, the communication system 100 may implement one or more channel access methods, such as code division multiple access (CDMA) , time division multiple access (TDMA) , frequency division multiple access (FDMA) , orthogonal FDMA (OFDMA) , or single-carrier FDMA (SC-FDMA) in the air interfaces 190a and 190b. The air interfaces 190a and 190b may utilize other higher dimension signal spaces, which may involve a combination of orthogonal and/or non- orthogonal dimensions.

The air interface 190c can enable communication between the ED 110d and one or multiple NT-TRPs 172 via a wireless link or simply a link. For some examples, the link is a dedicated connection for unicast transmission, a connection for broadcast transmission, or a connection between a group of EDs and one or multiple NT-TRPs for multicast transmission.

The RANs 120a and 120b are in communication with the core network 130 to provide the EDs 110a 110b, and 110c with various services such as voice, data, and other services. The RANs 120a and 120b and/or the core network 130 may be in direct or indirect communication with one or more other RANs (not shown) , which may or may not be directly served by core network 130, and may or may not employ the same radio access technology as RAN 120a, RAN 120b or both. The core network 130 may also serve as a gateway access between (i) the RANs 120a and 120b or EDs 110a 110b, and 110c or both, and (ii) other networks (such as the PSTN 140, the internet 150, and the other networks 160) . In addition, some or all of the EDs 110a 110b, and 110c may include functionality for communicating with different wireless networks over different wireless links using different wireless technologies and/or protocols. Instead of wireless communication (or in addition thereto) , the EDs 110a 110b, and 110c may communicate via wired communication channels to a service provider or switch (not shown) , and to the internet 150. PSTN 140 may include circuit switched telephone networks for providing plain old telephone service (POTS) . Internet 150 may include a network of computers and subnets (intranets) or both, and incorporate protocols, such as Internet Protocol (IP) , Transmission Control Protocol (TCP) , User Datagram Protocol (UDP) . EDs 110a 110b, and 110c may be multimode devices capable of operation according to multiple radio access technologies, and incorporate multiple transceivers necessary to support such.

FIG. 3 illustrates another example of an ED 110 and a base station 170a, 170b and/or 170c. The ED 110 is used to connect persons, objects, machines, etc. The ED 110 may be widely used in various scenarios, for example, cellular communications, device-to-device (D2D) , vehicle to everything (V2X) , peer-to-peer (P2P) , machine-to-machine (M2M) , machine-type communications (MTC) , internet of things (IOT) , virtual reality (VR) , augmented reality (AR) , industrial control, self-driving, remote medical, smart grid, smart furniture, smart office, smart wearable, smart transportation, smart city, drones, robots, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery and mobility, etc.

Each ED 110 represents any suitable end user device for wireless operation and may include such devices (or may be referred to) as a user equipment/device (UE) , a wireless transmit/receive unit (WTRU) , a mobile station, a fixed or mobile subscriber unit, a cellular telephone, a station (STA) , a machine type communication (MTC) device, a personal digital assistant (PDA) , a smartphone, a laptop, a computer, a tablet, a wireless sensor, a consumer electronics device, a smart book, a vehicle, a car, a truck, a bus, a train, or an IoT device, an industrial device, or apparatus (e.g. communication module, modem, or chip) in the forgoing devices, among other possibilities. Future generation EDs 110 may be referred to using other terms. The base station 170a and 170b is a T-TRP and will hereafter be referred to as T-TRP 170. Also shown in FIG. 3, a NT-TRP will hereafter be referred to as NT-TRP 172. Each ED 110 connected to T-TRP 170 and/or NT-TRP 172 can be dynamically or semi-statically turned-on (i.e., established, activated, or enabled) , turned-off (i.e., released, deactivated, or disabled) and/or configured in response to one of more of: connection availability and connection necessity.

The ED 110 includes a transmitter 201 and a receiver 203 coupled to one or more antennas 204. Only one antenna 204 is illustrated. One, some, or all of the antennas may alternatively be panels. The transmitter 201 and the receiver 203 may be integrated, e.g. as a transceiver. The transceiver is configured to modulate data or other content for transmission by at least one antenna 204 or network interface controller (NIC) . The transceiver is also configured to demodulate data or other content received by the at least one antenna 204. Each transceiver includes any suitable structure for generating signals for wireless or wired transmission and/or processing signals received wirelessly or by wire. Each antenna 204 includes any suitable structure for transmitting and/or receiving wireless or wired signals.

The ED 110 includes at least one memory 208. The memory 208 stores instructions and data used, generated, or collected by the ED 110. For example, the memory 208 could store software instructions or modules configured to implement some or all of the functionality and/or embodiments described herein and that are executed by the processing unit (s) 210. Each memory 208 includes any suitable volatile and/or non-volatile storage and retrieval device (s) . Any suitable type of memory may be used, such as random access memory (RAM) , read only memory (ROM) , hard disk, optical disc, subscriber identity module (SIM) card, memory stick, secure digital (SD) memory card, on-processor cache, and the like.

The ED 110 may further include one or more input/output devices (not shown) or interfaces (such as a wired interface to the internet 150 in FIG. 1) . The input/output devices permit interaction with a user or other devices in the network. Each input/output device includes any suitable structure for providing information to or receiving information from a user, such as a speaker, microphone, keypad, keyboard, display, or touch screen, including network interface communications.

The ED 110 further includes a processor 210 for performing operations including those related to preparing a transmission for uplink transmission to the NT-TRP 172 and/or T-TRP 170, those related to processing downlink transmissions received from the NT-TRP 172 and/or T-TRP 170, and those related to processing sidelink transmission to and from another ED 110. Processing operations related to preparing a transmission for uplink transmission may include operations such as encoding, modulating, transmit beamforming, and generating symbols for transmission. Processing operations related to processing downlink transmissions may include operations such as receive beamforming, demodulating and decoding received symbols. Depending upon the embodiment, a downlink transmission may be received by the receiver 203, possibly using receive beamforming, and the processor 210 may extract signaling from the downlink transmission (e.g. by detecting and/or decoding the signaling) . An example of signaling may be a reference signal transmitted by NT-TRP 172 and/or T-TRP 170. In some embodiments, the processor 276 implements the transmit beamforming and/or receive beamforming based on the indication of beam direction, e.g. beam angle information (BAI) , received from T-TRP 170. In some embodiments, the processor 210 may perform operations relating to network access (e.g. initial access) and/or downlink synchronization, such as operations relating to detecting a synchronization sequence, decoding and obtaining the system information, etc. In some embodiments, the processor 210 may perform channel estimation, e.g. using a reference signal received from the NT-TRP 172 and/or T-TRP 170.

Although not illustrated, the processor 210 may form part of the transmitter 201 and/or receiver 203. Although not illustrated, the memory 208 may form part of the processor 210.

The processor 210, and the processing components of the transmitter 201 and receiver 203 may each be implemented by the same or different one or more processors that are configured to execute instructions stored in a memory (e.g. in memory 208) . Alternatively, some or all of the processor 210, and the processing components of the transmitter 201 and receiver 203 may be implemented using dedicated circuitry, such as a programmed field-programmable gate array (FPGA) , a graphical processing unit (GPU) , or an application-specific integrated circuit (ASIC) .

The T-TRP 170 may be known by other names in some embodiments, such as a base station, a base transceiver station (BTS) , a radio base station, a network node, a network device, a device on the network side, a transmit/receive node, a Node B, an evolved NodeB (eNodeB or eNB) , a Home eNodeB, a next Generation NodeB (gNB) , a transmission point (TP) ) , a site controller, an access point (AP) , or a wireless router, a relay station, a remote radio head, a terrestrial node, a terrestrial network device, or a terrestrial base station, base band unit (BBU) , remote radio unit (RRU) , radio unit (RU) , active antenna unit (AAU) , remote radio head (RRH) , central unit (CU) , distribute unit (DU) , positioning node, among other possibilities. The T-TRP 170 may be macro BSs, pico BSs, relay node, donor node, or the like, or combinations thereof. The T-TRP 170 may refer to the forging devices or apparatus (e.g. communication module, modem, or chip) in the forgoing devices.

The CU (or CU-control plane (CP) and CU-user plane (UP) ) , DU or RU may be known by other names in some embodiments. For example, in open RAN (ORAN) system, the CU may also be referred to as open CU (O-CU) , DU may also be referred to as open DU (O-DU) , CU-CP may also be referred to open CU-CP (O-CU-CP) , CU-UP may also be referred to as open CU-UP (O-CU-CP) , and RU may also be referred to open RU (O-RU) . Any one of the CU (or CU-CP, CU-UP) , DU, or RU could be implemented through a software module, a hardware module, or a combination of software and hardware modules.

In some embodiments, the parts of the T-TRP 170 may be distributed. For example, some of the modules of the T-TRP 170 may be located remote from the equipment housing the antennas of the T-TRP 170, and may be coupled to the equipment housing the antennas over a communication link (not shown) sometimes known as front haul, such as common public radio interface (CPRI) . Therefore, in some embodiments, the term T-TRP 170 may also refer to modules on the network side that perform processing operations, such as determining the location of the ED 110, resource allocation (scheduling) , message generation, and encoding/decoding, and that are not necessarily part of the equipment housing the antennas of the T-TRP 170. The modules may also be coupled to other T-TRPs. In some embodiments, the T-TRP 170 may actually be a plurality of T-TRPs that are operating together to serve the ED 110, e.g. through coordinated multipoint transmissions.

The T-TRP 170 includes at least one transmitter 252 and at least one receiver 254 coupled to one or more antennas 256. Only one antenna 256 is illustrated. One, some, or all of the antennas may alternatively be panels. The transmitter 252 and the receiver 254 may be integrated as a transceiver. The T-TRP 170 further includes a processor 260 for performing operations including those related to: preparing a transmission for downlink transmission to the ED 110, processing an uplink transmission received from the ED 110, preparing a transmission for backhaul transmission to NT-TRP 172, and processing a transmission received over backhaul from the NT-TRP 172. Processing operations related to preparing a transmission for downlink or backhaul transmission may include operations such as encoding, modulating, precoding (e.g. MIMO precoding) , transmit beamforming, and generating symbols for transmission. Processing operations related to processing received transmissions in the uplink or over backhaul may include operations such as receive beamforming, and demodulating and decoding received symbols. The processor 260 may also perform operations relating to network access (e.g. initial access) and/or downlink synchronization, such as generating the content of synchronization signal blocks (SSBs) , generating the system information, etc. In some embodiments, the processor 260 also generates the indication of beam direction, e.g. BAI, which may be scheduled for transmission by scheduler 253. The processor 260 performs other network-side processing operations described herein, such as determining the location of the ED 110, determining where to deploy NT-TRP 172, etc. In some embodiments, the processor 260 may generate signaling, e.g. to configure one or more parameters of the ED 110 and/or one or more parameters of the NT-TRP 172. Any signaling generated by the processor 260 is sent by the transmitter 252. Note that “signaling” , as used herein, may alternatively be called control signaling. Dynamic signaling may be transmitted in a control channel, e.g. a physical downlink control channel (PDCCH) , and static or semi-static higher layer signaling may be included in a packet transmitted in a data channel, e.g. in a physical downlink shared channel (PDSCH) .

A scheduler 253 may be coupled to the processor 260. The scheduler 253 may be included within or operated separately from the T-TRP 170, which may schedule uplink, downlink, and/or backhaul transmissions, including issuing scheduling grants and/or configuring scheduling-free ( “configured grant” ) resources. The T-TRP 170 further includes a memory 258 for storing information and data. The memory 258 stores instructions and data used, generated, or collected by the T-TRP 170. For example, the memory 258 could store software instructions or modules configured to implement some or all of the functionality and/or embodiments described herein and that are executed by the processor 260.

Although not illustrated, the processor 260 may form part of the transmitter 252 and/or receiver 254. Also, although not illustrated, the processor 260 may implement the scheduler 253. Although not illustrated, the memory 258 may form part of the processor 260.

The processor 260, the scheduler 253, and the processing components of the transmitter 252 and receiver 254 may each be implemented by the same or different one or more processors that are configured to execute instructions stored in a memory, e.g. in memory 258. Alternatively, some or all of the processor 260, the scheduler 253, and the processing components of the transmitter 252 and receiver 254 may be implemented using dedicated circuitry, such as a FPGA, a GPU, or an ASIC.

Although the NT-TRP 172 is illustrated as a drone only as an example, the NT-TRP 172 may be implemented in any suitable non-terrestrial form. Also, the NT-TRP 172 may be known by other names in some embodiments, such as a non-terrestrial node, a non-terrestrial network device, or a non-terrestrial base station. The NT-TRP 172 includes a transmitter 272 and a receiver 274 coupled to one or more antennas 280. Only one antenna 280 is illustrated. One, some, or all of the antennas may alternatively be panels. The transmitter 272 and the receiver 274 may be integrated as a transceiver. The NT-TRP 172 further includes a processor 276 for performing operations including those related to: preparing a transmission for downlink transmission to the ED 110, processing an uplink transmission received from the ED 110, preparing a transmission for backhaul transmission to T-TRP 170, and processing a transmission received over backhaul from the T-TRP 170. Processing operations related to preparing a transmission for downlink or backhaul transmission may include operations such as encoding, modulating, precoding (e.g. MIMO precoding) , transmit beamforming, and generating symbols for transmission. Processing operations related to processing received transmissions in the uplink or over backhaul may include operations such as receive beamforming, and demodulating and decoding received symbols. In some embodiments, the processor 276 implements the transmit beamforming and/or receive beamforming based on beam direction information (e.g. BAI) received from T-TRP 170. In some embodiments, the processor 276 may generate signaling, e.g. to configure one or more parameters of the ED 110. In some embodiments, the NT-TRP 172 implements physical layer processing, but does not implement higher layer functions such as functions at the medium access control (MAC) or radio link control (RLC) layer. As this is only an example, more generally, the NT-TRP 172 may implement higher layer functions in addition to physical layer processing.

The NT-TRP 172 further includes a memory 278 for storing information and data. Although not illustrated, the processor 276 may form part of the transmitter 272 and/or receiver 274. Although not illustrated, the memory 278 may form part of the processor 276.

The processor 276 and the processing components of the transmitter 272 and receiver 274 may each be implemented by the same or different one or more processors that are configured to execute instructions stored in a memory, e.g. in memory 278. Alternatively, some or all of the processor 276 and the processing components of the transmitter 272 and receiver 274 may be implemented using dedicated circuitry, such as a programmed FPGA, a GPU, or an ASIC. In some embodiments, the NT-TRP 172 may actually be a plurality of NT-TRPs that are operating together to serve the ED 110, e.g. through coordinated multipoint transmissions.

The T-TRP 170, the NT-TRP 172, and/or the ED 110 may include other components, but these have been omitted for the sake of clarity.

For ease of understanding the embodiments of this application, the following briefly describes a process of transmitting reference signals and measuring channels based on the reference signals.

Multiple input multiple-output (MIMO) technology allows an antenna array of multiple antennas to perform signal transmissions and receptions to meet high transmission rate requirements. The above ED110 and T-TRP 170, and/or NT-TRP use MIMO to communicate over the wireless resource blocks. MIMO utilizes multiple antennas at the transmitter and/or receiver to transmit wireless resource blocks over parallel wireless signals. MIMO may beamform parallel wireless signals for reliable multipath transmission of a wireless resource block. MIMO may bond parallel wireless signals that transport different data to increase the data rate of the wireless resource block.

In recent years, a MIMO (large-scale MIMO) wireless communication system with the above T-TRP 170, and/or NT-TRP 172 configured with a large number of antennas has gained wide attentions from the academia and the industry. In the large-scale MIMO system, the T-TRP 170, and/or NT-TRP 172 is generally configured with more than ten antenna units (such as 128 or 256) , and serves dozens of the ED 110 (such as 40) . A large number of antenna units of the T-TRP 170, and NT-TRP 172 can greatly increase the degree of spatial freedom of wireless communication, greatly improve the transmission rate, spectrum efficiency and power efficiency, and eliminate the interference between cells to a large extent. The increased number of antennas allows each antenna unit to be smaller in size with a lower cost. Using the degree of spatial freedom provided by the large-scale antenna units, the T-TRP 170, and NT-TRP 172 of each cell can communicate with many ED 110 in the cell on the same time-frequency resource at the same time, thus greatly increasing the spectrum efficiency. A large number of antenna units of the T-TRP 170, and/or NT-TRP 172 also enable each user to have better spatial directivity for uplink and downlink transmission, so that the transmitting power of the T-TRP 170, and/or NT-TRP 172 and an ED 110 is reduced, and the power efficiency is increased. When the antenna number of the T-TRP 170, and/or NT-TRP 172 is sufficiently large, random channels between each ED 110 and the T-TRP 170, and/or NT-TRP 172 can approach orthogonal, and the interference between the cell and the users and the effect of noises can be eliminated. The plurality of advantages described above enable large-scale MIMO systems to have good prospects for application.

A MIMO system may include a receiver connected to a receive (Rx) antenna, a transmitter connected to transmit (Tx) antenna, and a signal processor connected to the transmitter and the receiver. Each of the Rx antenna and the Tx antenna may include a plurality of antennas. For instance, the Rx antenna may have an ULA antenna array in which the plurality of antennas are arranged in line at even intervals. When a radio frequency (RF) signal is transmitted through the Tx antenna, the Rx antenna may receive a signal reflected and returned from a forward target.

FIG. 4 illustrates units or modules in a device. One or more steps of the embodiment methods provided herein may be performed by corresponding units or modules, according to FIG. 4. FIG. 4 illustrates units or modules in a device, such as in ED 110, in T-TRP 170, or in NT-TRP 172. For example, a signal may be transmitted by a transmitting unit or a transmitting module. For example, a signal may be transmitted by a transmitting unit or a transmitting module. A signal may be received by a receiving unit or a receiving module. A signal may be processed by a processing unit or a processing module. Other steps may be performed by an artificial intelligence (AI) or machine learning (ML) module. The respective units or modules may be implemented using hardware, one or more components or devices that execute software, or a combination thereof. For instance, one or more of the units or modules may be an integrated circuit, such as a programmed FPGA, a GPU, or an ASIC. It will be appreciated that where the modules are implemented using software for execution by a processor for example, they may be retrieved by a processor, in whole or part as needed, individually or together for processing, in single or multiple instances, and that the modules themselves may include instructions for further deployment and instantiation.

Additional details regarding the EDs 110, T-TRP 170, and NT-TRP 172 are known to those of skill in the art. As such, these details are omitted here.

Artificial Intelligence (AI) technologies can be applied in communication, including artificial intelligence or machine learning (AI/ML) based communication in the physical layer and/or AI/ML based communication in the higher layer, e.g., medium access control (MAC) layer. For example, in the physical layer, the AI/ML based communication may aim to optimize component design and/or improve the algorithm performance. For the MAC layer, the AI/ML based communication may aim to utilize the AI/ML capability for learning, prediction, and/or making a decision to solve a complicated optimization problem with possible better strategy and/or optimal solution, e.g. to optimize the functionality in the MAC layer, e.g. intelligent TRP management, intelligent beam management, intelligent channel resource allocation, intelligent power control, intelligent spectrum utilization, intelligent modulation and coding scheme (MCS) , intelligent hybrid automatic repeat request (HARQ) strategy, intelligent transmit/receive (Tx/Rx) mode adaption, etc.

An AI architecture may involve multiple nodes, where the multiple nodes may possibly be organized in one of two modes, i.e., centralized and distributed, both of which may be deployed in an access network, a core network, or an edge computing system or third party network. A centralized training and computing architecture is restricted by possibly large communication overhead and strict user data privacy. A distributed training and computing architecture may comprise several frameworks, e.g., distributed machine learning and federated learning. Distribution learning is a machine learning technique that focuses on the modeling and analysis of probability distributions. The goal of distribution learning is to develop algorithms and techniques that can accurately estimate the underlying probability distribution of a given dataset. This is an important task in many areas of machine learning, including density estimation, generative modeling, and anomaly detection. Distribution learning techniques are used in a wide range of applications, including image and speech recognition, natural language processing, and finance. Some of the most commonly used distribution learning algorithms include Gaussian mixture models, kernel density estimation, and variational autoencoders. Distribution learning is a data-driven approach, where the algorithms learn from the available data to model the underlying distributions. While stochastic gradient descent is a widely used optimization technique in many machine learning algorithms, it is not the only method employed in distribution learning. Several other techniques, such as expectation-maximization (EM) algorithms, Markov Chain Monte Carlo (MCMC) methods, and variational inference, are also commonly used in distribution learning algorithms. These methods can be particularly useful when the objective function is non-convex or when the data has a complex structure. Additionally, some distribution learning algorithms may employ non-parametric methods, which do not assume a specific parametric form for the underlying distribution. In this application, learning (training) or AI can be generalized to describe a data-driven and adaptive method.

In some embodiments, an AI architecture may comprise an intelligent controller which can perform as a single agent or a multi-agent, based on joint optimization or individual optimization. New protocols and signaling mechanisms are desired so that the corresponding interface link can be personalized with customized parameters to meet particular requirements while minimizing signaling overhead and maximizing the whole system spectrum efficiency by personalized AI technologies.

New protocols and signaling mechanisms are provided for operating within and switching between different modes of operation, including between AI and non-AI modes, and for measurement and feedback to accommodate the different possible measurements and information that may need to be fed back, depending upon the implementation.

An air interface that uses AI as part of the implementation, e.g. to optimize one or more components of the air interface, will be referred to herein as an “AI enabled air interface” . In some embodiments, there may be two types of AI operation in an AI enabled air interface: both the network and the ED implement learning; or learning is only applied by the network.

The following are some terminologies which are used in AI/ML or data-driven and adaptiveness field:

1) Data collection:

Data is the very important component for AI/ML techniques. Data collection is a process of collecting data by the network nodes, management entity, or ED for the purpose of Model training, data analytics and inference.

2) Model training and validation:

Model training is a process to train a model, or simply model in this application, by learning the input/output relationship in a data driven manner and obtain the trained model for inference.

As a sub-process of training, validation is used to evaluate the quality of a model using a dataset different from the one used for model training. Validation can help selecting model parameters that generalize beyond the dataset used for model training. The model parameter after training can be adjusted further by the validation process.

3) Model inference:

A process of using a trained model to produce a set of outputs based on a set of inputs.

4) Model testing:

Similar with validation, testing is also a sub-process of training, and it is used to evaluate the performance of a final model using a dataset different from the one used for model training and validation. Differently from model validation, testing do not assume subsequent tuning of the model.

5) Online training:

Online training means a training process where the model being used for inference is typically continuously trained in (near) real-time with the arrival of new training samples.

6) Offline training:

A training process where the model is trained based on collected dataset, and where the trained model is later used or delivered for inference.

7) Model delivery/transfer:

A generic term referring to delivery of a model from one entity to another entity in any manner. Delivery of a model over the air interface includes either parameters of a model structure known at the receiving end or a new model with parameters. Delivery may contain a full model or a partial model.

8) Federated learning:

Federated learning (FL) is a machine learning technique that is used to train a model by a central node (e.g., server) and a plurality of decentralized edge nodes (e.g., EDs, next Generation NodeBs, “gNBs” ) .

According to the wireless FL technique, a server may provide, to an edge node, a set of model parameters (e.g., weights, biases, gradients) that describe a global model. The edge node may initialize a local model with the received global model parameters. The edge node may then train the local model using local data samples to, thereby, produce a trained local model. The edge node may then provide, to the serve, a set of model parameters that describe the local model.

Upon receiving, from a plurality of edge nodes, a plurality of sets of model parameters that describe respective local models at the plurality of edge nodes, the server may aggregate the local model parameters reported from the plurality of EDs and, based on such aggregation, update the global Model. A subsequent iteration progresses much like the first iteration. The server may transmit the aggregated global model to a plurality of edge nodes. The above procedures are performed multiple iterations until the global Model is considered to be finalized, e.g., the Model is converged or the training stopping conditions are satisfied.

Notably, the wireless FL technique may not involve exchange of local data samples. Indeed, the local data samples remain at respective edge nodes.

Distributed learning involves frequent transmission of model’s parameters such as neural network neuron’s weights matrix. Each working node, worker, uploads its model’s parameters to a convergence point, which then merges multiple versions of the model’s parameters and redistributes them to the working nodes. For a model to process a channel data of large-scale 6G MIMO, this size of the model’s parameters even larger than the MIMO channel data itself is typically required. The transmission and merging process for model’s parameters of such scale becomes exceedingly time-consuming, bandwidth-intensive, and energy-consuming, which is the crux of the issue.

Regrettably, the industry is still in the early stages of addressing these AI-related challenges. Deep neural networks in the wireless physical layer or link layer are mainly confined to handling smaller-scale problems and are not equipped to tackle larger-scale, more critical issues. This fundamentally limits the spread of the 'A I-for-Net'concept in 6G communications, where one of the main goals is to enhance large-scale throughput and reduce latency for the communications at large. These issues have yet to be satisfactorily resolved.

9) Deep neural network (DNN)

DNN is a type of artificial network. An DNN can be modeled as a set of connected neurons in an acyclic graph. The output of some neurons may become the input of other neurons. The DNN may be organizes into multiple neurons layers. The operations of each layer can be represented by : whererepresents input vector, represents output vector, represents offset vector, W represents weight matrix (also referred to as coefficients or model’s parameters) , and α () represents activation function. In other words, an DNN can be parameterized by the weights of the neurons of the neural network.

With the increasing complexity and scale of networks, AI has become an essential tool to improve efficiency. Particularly, with a significant increase in the scale of antenna ports and the size of operational bandwidth in MIMO system, computational complexity is significantly increase because of a dramatic expansion of the MIMO channel space. AI can be used in the MIMO system to reduce the computational complexity.

Currently, AI still faces challenges in addressing foundational issues in wireless communication networks. A primary concern is that existing deep neural network approaches, when dealing with large-scale antenna arrays, are typically limited to addressing relatively smaller-scale problems such as beamforming, scheduling, narrowband, and low real-time applications. This limitation stems from the scale of the original signal, rendering these methods ineffective for dealing with the vast channel spaces characteristic of 6G communications.

Particularly in processing 6G MIMO channels, representing them as extremely long vectors results in a massive deep neural network. This enormity complicates the storage, training, transmission, and updating of the deep neural network over the air. Moreover, when considering distributed learning, the situation becomes even more complex.

FIG. 5 is an example of a channel model of a 4-by-4 MIMO system. A transmitter is connected to four TX antennas, x1 to x4, a receiver is connected to four RX antennas, y1 to y4, and a transmission channel may be formed between each TX antenna and each RX antenna. For example, an RF signal transmitted through x1 may be received by y2 through channel h21. The RF signal transmitted through x3 may be received by y1 through channel h13.

All the channels h11, h12, etc. can constitute a channel matrix for each frequency unit (e.g., a subcarrier) observed at a time unit (e.g., a symbol) . The channel matrix can be used to describe effect of channels on signals. The channel matrix H, received signals y and transmitted signals x may satisfy: y=Hx+n, where n represents noise interference. It is noted that the transmitted signals y is the signals sent by the transmitter. After the signals are transmitted through the channels, they are received by the receiver and are called received signals y.

Dimension of the channel matrix for a frequency unit is related to the number of TX antenna (s) (represented by n_{Tx_Ant} in this application) and the number of the RX antenna (s) (represented by n_{Rx_Ant} in this application) . For example, when the channel matrix for a subcarrier is vectorized, the dimension of the vectorized vector is equal to: n_{Tx_Ant}×n_{Rx_Ant}. For n_subcarrier subcarriers, the dimension of the vectorized vector is equal to: n_{Tx_Ant}×n_{Rx_Ant}×n_subcarrier. The n_{Tx_Ant}, n_{Rx_Ant} and n_subcarrier are positive integers.

In a MIMO system, to implement functions such as system synchronization, channel information feedback, and data transmission, channel estimation needs to be performed on an uplink channel or a downlink channel. Channel estimation refers to the process of reconstructing or restoring received signals to compensate for signal distortion caused by channel fading and noise. In channel estimation, a reference signal predicted by a transmitter and a receiver may be used to track a change in the time domain and/or frequency domain of a channel, so as to reconstruct or restore a received signal. The reference signal may also be referred to as a pilot signal, a reference sequence or the like, and is described as a reference signal in the following for ease of understanding. The reference signal comprises, for example, a channel state information-reference signal (CSI-RS) , a sounding reference signal (SRS) , a demodulation reference signal (DMRS) , phase track reference signals (PT-RS) , or cell reference signals (CRS) . The reference signals listed above are merely examples, and shall not constitute any limitation on this application. This application does not exclude the possibility that other reference signals are defined in a future protocol to implement the same or similar function.

To facilitate understanding of the embodiments of this application, the CSI-RS is described in detail by example below. The CSI-RS is mainly used for downlink channel estimation corresponding to a physical antenna port. For example, a receiving apparatus (e.g., an ED) may perform channel estimation on each physical antenna port based on a CSI-RS sent by a transmitting apparatus (e.g., a base station) , to feedback channel state information (CSI) based on a channel estimation result. The CSI may include related information such as a channel quality indicator (channel quality indicator, CQI) , a precoding matrix indicator (precoding matrix indicator, PMI) , a layer indicator (layer indicator, LI) , and a rank indicator (rank indicator, RI) . The CSI is used to reconstruct or precode the downlink channel. In some embodiments, a process in which the base station obtains CSI may include: the base station sends a reference signal to the ED; the ED obtains an estimated CSI value according to the received reference signal, selects a precoding vector from a codebook according to the estimated CSI value, and feedback related to the index of the precoding vector to the base station; and the base station determines a CSI reconstruction value with reference to the index of the precoding vector. The CSI reconstruction value can be a CSI closest to the true value of the CSI that can be obtained by the base station.

6G T-MIMO (tera-bps-MIMO) technology represents a significant leap forward in wireless communication, primarily due to its utilization of a substantially increased number of base-station antenna ports. It is an ultra-massive MIMO system. The benefits of this dense antenna array are multifaceted. Firstly, it allows for a more precise directionality of signal transmission, akin to a highly focused beam of light, which enhances the signal quality and reliability for the ED.

A high base-station-to-ED antenna ratio further refines this process. With more antenna ports at the base station relative to the number on an ED, the system can engage in more sophisticated signal processing techniques. This disparity allows for the mitigation of interference and the improvement of spectral efficiency, which translates to faster data rates and more stable connections for users.

The operational frequency bands of 6G T-MIMO, for example, cmWave (i.e., 10GHz to 14GHz and mmWave bands, offer distinct advantages and trade-offs. The 10GHz to 14GHz band provides a balance between coverage and capacity, offering wider channels than those available in lower frequency ranges, which supports higher data throughput. In contrast, the mmWave bands, operating at frequencies above 24GHz, offer even higher capacities and data rates, suitable for extremely bandwidth-intensive applications. However, mmWave signals have a shorter range and are more susceptible to attenuation by obstacles such as buildings and foliage, necessitating a denser network of base stations to ensure coverage.

The deployment of a large number of antenna ports operating over these broad bandwidths introduces challenges in terms of power efficiency and sustainability. The energy requirements for powering such expansive networks are considerable, and there is a pressing need to develop energy-efficient technologies and protocols to mitigate the environmental impact. This includes optimizing the hardware design for lower power consumption and implementing intelligent network management systems that can dynamically adjust resource allocation based on real-time demand.

Additionally, the control overhead, the necessary background communication to manage the network becomes more complex as the number of antenna ports and the bandwidth increases. Efficiently managing this overhead is critical to prevent it from negating the benefits provided by the larger bandwidths and antenna arrays. Strategies such as advanced algorithms for signal processing and network coordination are essential to maintain a lean operational profile, ensuring that the vast capabilities of 6G T-MIMO are delivered with an eye towards sustainability and operational efficiency.

FIG. 6 illustrates an example of T-MIMO channel space. As shown in FIG. 6, a base station is connected to 32 rows and 16 columns of TX antennas, where the number of the TX antennas n_{TX_Ant}=32×16×2=1024 when dual polarization is applied. An ED is connected to 2 rows and 4 columns of RX antennas, where the number of the RX antennas n_{RX_Ant}=2×4 ×2=16. As aforementioned, when 273 RBs with 30KHz subcarrier space on a 100MHz band, the total T-MIMO channel dimension of channel matrix observed at a time unit, e.g., one OFDM symbol, for one ED is massive: 273×12×1024×16=53,673,984.

As aforementioned, a model (for example an DNN) can be applied to channel estimation, and a size of the model is related to the dimension of channels. That is, a learned model used in the T-MIMO channel space may be very huge.

FIG. 7 illustrates sizes of DNNs of different systems. When a model applied to channel estimation in 4G (LTE) system, the deep neutral network may take 1K to 5K tokens. When it applied to channel estimation in 5G system, it may take 10K to 100K tokens. However, when a deep neural network applied to channel estimation in 6G T-MIMO system, the deep neutral network may take 1200K to 10M tokens. Therefore, a direct approach of feeding the original channel data into a deep neural network is not feasible for 6G T-MIMO. For a reference, the state-of-the-art LLMs can only handle up to 10K tokens at most. Moreover, the inference latency is too high.

In addressing the challenges in 6G wireless communication technology, we are confronted with a significant increase in the scale of antenna ports and the size of operational bandwidth. This has led to a dramatic expansion of the MIMO channel space, raising a main technical issue that how to effectively manage vast channel space.

This application is addressed to the issue above related to the MIMO system such as 6G T-MIMO or other ultra-massive MIMO system. This application provides a communication method in which a data-driven and adaptive model, or a first model for simplicity, can be used to predict the channel state of a set of channels between a BS and an ED. Moreover, taking the low-dimensional first information as input of the first model can reduce the size of an input of the first model, which can reduce computation complexity of the first model. The application of a data-driven and adaptive model in ultra-large-scale MIMO channels therefore becomes feasible.

At step 810, a BS transmits a first reference signal to an ED through a set of communication channels. Correspondingly, the ED receives the first reference signal from the BS through the set of communication channels.

The first reference signal may be any kind of a downlink reference signal that is described above (e.g., CSI-RS, DMRS, and etc. ) . Alternatively, the first reference signal may also be a sequence of future-defined reference signal with a corresponding function. This is not limited to this application.

The channels between an ED and the BS can be regarded as a set of communication channels. As aforementioned in FIG. 5, the set of communication channels may be presented as a channel matrix or a vectorized vector. A dimension of the set of communication channels for a single frequency unit is related to antenna ports of the BS (n_{Tx_Ant}) and antenna ports of the ED (n_{Rx_Ant}) , and it may be equal to: n_{Tx_Ant}×n_{Rx_Ant}.

Notably, the first reference signal transmitted through the set of channels carries the channel state coefficients, so the received first reference signal may be represented as channel coefficient vector y.

In some embodiments, the BS can broadcast or multicast the first reference signal. For example, the BS could transmit the first reference signal on broadcast downlink channels (PBCH) . Thereby, one or more EDs in the base station coverage area may receive the first reference signal simultaneously.

Notably, the first reference signal may be received by multiple EDs, that is, the first reference signal may be transmitted through multiple sets of communication channel.

In some embodiments, the BS may transmit, in a period, reference signals that include the first reference signal through the set of communication channels. The first reference signal may be transmitted in a first time interval. The first time interval may any one of time intervals of the period. In other words, the BS may transmit reference signals periodically, and the ED can receive the reference signals in each time interval. The first time interval can be referred to as a time unit of any granularity. For example, the first time interval could be a slot, a symbol, a TTI and etc.

At step 820, the ED generates first information based on the first reference signal.

A dimension of the first information is smaller than a dimension of the set of communication channels. For ease of description, the dimension of the first information may be represented by r and the dimension of the set of communication channels may be represented by N, N and r are positive integers, r < N. As aforementioned, a dimension of the set of communication channels may be equal to: N=n_{Tx_Ant}×n_{Rx_Ant}×n_subcarrier. When n_{Tx_Ant} (the number of TX antenna (s) ) and n_{Rx_Ant} (the number of the RX antenna (s) ) are large, the dimension of the set of communication channels is large correspondingly. If the first reference signal through the set of communication channels is input to the first model without the signal process, the first model needs to process at least N-dimension information, which means that the computation complexity of the first model would be huge. The first information may contain the channel state of the set of communication channels after dimension reduction. For example, the first reference signal through the set of communication channels may be represented by a vectorized vector and the first information may be represented by a corresponding low-dimensional vector. In embodiments of this application, the r-dimension first information as the input of the first model, reducing computation complexity of the first model. An application of a model in large-scale channels becomes feasible.

In some embodiments, the first information may be generated by performing a signal process on the first reference signal. In other words, the ED may perform signal process on the first reference signal to generate the first information.

The signal process may include a variety of actions. In some embodiments, the signal process may be used to compress the dimension of the set of communication channels. The signal process may include a variety of dimension-related actions (or compression actions) . The ED may perform one or more dimension-related actions on the first reference signal, and these actions are important for reducing the size and complexity of the first model.

In some embodiments, the signal process may be further used to estimate the set of communication channels. The signal process may include a variety of estimation-related actions. For example, the ED may estimate the channel coefficients on the transmitted first reference signal into a channel coefficient vector y. The signal process may be used to compress the channel coefficient vector of the set of communication channels. For example, the ED can compress the channel coefficient vector y into a low-dimensional channel coefficient vector c, where the first information may include the low-dimensional channel coefficient vector c. The signal process may be used for generating the low-dimensional channel coefficient vector c.

Notably, the dimension-related action and the estimation-related action can be separate actions or they can be the same action. This is not limited to this application. For example, the signal process may include one or more of: feature extraction, reference-signal selection, dynamic mode decomposition (DMD) and etc. Any one of the above actions can be implemented independently, or their any combinations can be implemented thereof.

Feature extraction is the process (or action) of finding a suitable basis (referred to as common basis U) for representing the MIMO channel state using a set of features that capture the dominant patterns or modes of variation in the data. This action can reduce the dimension of the original channel space. Details of this action will be given in FIG. 12 and its related description hereafter.

Reference signal selection is the process (or action) of choosing a subset of candidate reference signal locations that maximize the information content or variance captured by the reference signals for the ED. This reference signal selection can be represented as a low-dimension matrix (P) of the U. Details of this action will be given in FIG. 13 and its related description hereafter.

DMD is a data-driven technique for extracting spatiotemporal coherent structures from high-dimensional data. DMD is based on the idea of decomposing a data matrix into a low-rank approximation that captures the dominant modes and frequencies of the underlying dynamics. Details of this action will be given in FIG. 14 and its related description hereafter.

Notably, this application does not exclude other possible dimension-related and estimation-related actions. In mathematics, what a model learns is a commonality among all training data samples, the signal process can handle the large channel space to generate low-dimensional information.

At step 830, the ED inputs the first information to the first model, to obtain second information.

The second information is used to predict a channel state of the set of communication channel. The first model can be any types of model with prediction function. For example, the first model may be a DNN. Weights of the neurons of the DNN may be obtained from a data set, where the data set includes channel states of multiple time intervals. The weights of the neurons may be trained to represent changes in channel states over two time intervals. Therefore, the ED inputs the first information to the DNN, and the first information is weighted based on the neurons to obtain the second information for prediction. This application does not exclude other possible models.

The channel state, or referred to as channel state information, may indicate a set of parameters that describe the condition of a wireless channel. For example, the set of parameters may include one or more of: the received signal strength indication (RSSI) , channel delay, timing advance (TA) , Doppler frequency shift, and the like. This is not limited to this application.

In some embodiments, the BS transmits reference signals in a period and the first reference signal is transmitted in a first time interval, and the second information may be used to predict a channel state of the set of communication channels in a second time interval. The first time interval and the second time interval may be two adjacent time intervals of the period. In other words, the first model can predict the transition of the channel state of the set of communication channels from the first time interval to the second time interval.

In some embodiments, the ED may further input third information to the first model . The third information indicates one or more of: measurement parameter (s) of the first reference signal; location parameter (s) of the ED; movement parameter (s) of the ED; and service parameter (s) corresponding to the ED. Therefore, with the input of multiple information (in these embodiments, the first information and the second information) into the first model, the first model can predict the channel state from multiple aspects, and therefore more precise.

For example, the measurement parameter (s) of the first reference signal may include one or more of: the RSSI, channel delay, TA, Doppler frequency shift, and etc. The location parameters may include location of the ED, and etc. The movement parameters may include one or more of movement velocity, moving direction, and etc. The service parameters may include one or more of type of service, quality of service (QoS) , and etc. In other words, the input of the first model may include not only the received reference signal, but also various static or dynamic information, improving the accuracy of model prediction.

Notably, the performance of the first model may lie in its multitude of information, including the channel's intrinsic information, delay, power, multipath propagation, Doppler effects and the like. A state channel includes various modal information, including frequency domain, time domain, power, and delay information. This multimodal state changes with time and location. The first model (e.g., a DNN) for learning channel state transitions may utilize a method similar to Markov chains. The method can predict future states for several moments based on the current data transmitted. The advantage of this method lies in its predictive capability, which can estimate future states based on the current state and known state transition probabilities, reducing the need for real-time data transmission and measuring overhead.

The ED can obtain the first model in a variety of ways. In a first embodiment, the ED may generate or establish the first model from a data set related to a channel state of the set of communication channels. The first model may contain the characteristics of the set of communication channels. Optionally, the ED may transmit information that indicates the first model to the base station. Therefore, the BS may also use the first model to predict the channel states. For example, the ED may establish a DNN, and transmit information that indicates the DNN and initial information (e.g., channel state of a time interval#0) . The BS can predict channel states of the future time intervals based on the DNN and the initial information.

In a second embodiment, the base station may transmit information that indicates the first model to the ED. That is, the ED may obtain the first model from the base station. For example, the ED may transmit information that indicates a data set (e.g., channel state information) to the BS, and the BS may generate the first model from the data set, and deliver the first model to the ED. Then the BS and/or the ED can predict the channel states based on the first model.

For another example, the first model is generated based on N initial models, and the N initial models are from N EDs, N is a positive integer. The N EDs may be the EDs in BS coverage area. Each of the N EDs may establish an initial model based on respective data, and transmit information that indicates the initial model to the BS. The BS may receive the N initial models from the N EDs, and generate the first model based on the N initial models. The first model may contain characteristics of N sets of communication channels of the N EDs, each of the N EDs can receive information that indicates the first model from the BS, and each of the N EDs can predict the channel states using the high-performance first model. More details of this example will be given in FIG. 10, FIG. 21 and FIG. 22 and their related descriptions hereafter.

The ED may determine the signal process in a variety of ways. For example, the parameters related to the signal process may be signaled by the base station, or be predefined based on the application scenario, or be determined by the ED as a function of other parameters that are known by the ED, or may be fixed, or a combination thereof. In some embodiments, at least part of the parameters related to the signal process may be indicated by the BS. That is, the BS and the ED may perform step 840 before step 820.

Optionally, at step 840, the BS transmits information compression information to the ED. Correspondingly, the ED receives the compression information from the BS.

The compression information assists in the signal process. The ED may perform the signal process on the first reference signal based on the compression information.

The compression information may indicate a variety of parameters related to the signal process. For example, the compression information may indicate one or more of:

a dimension of the first information generated by the signal process;

a common basis (U) of the set of communication channels involved in the signal process;

a low-dimension matrix (P) of the U, where the P indicates a pattern of the first reference signal involved in the signal process; and

a unit scheme, where the unit scheme indicates the signal process involves dividing the set of communication channel based on one or more of: frequency domain, time domain and space domain.

The above parameter (s) may assist in the signal process. The ED may perform the signal process on the first reference signal based on the indicated parameter (s) . The common basis U may assist in feature extraction. The U and the P may assist in reference signal selection. The unit scheme may assist in DMD. This is not limited to this application.

In some embodiments, the BS may broadcast the compression information. For example, the compression information may be included in downlink control messages. The first model may be employed to multiple EDs. The multiple EDs in the base station coverage area can perform the signal process on the first reference signal consistently. This ensures consistency and coordination of information within the network, enabling EDs to interpret and utilize channel state information based on a unified pattern.

From the examples that obtaining the first model, it can be seen that the first model can be employed to the ED and the BS, and both the ED and the BS may predict the channel state using the first model. Therefore, the ED may not transmit information that indicates the channel state (for example, the CSI) to the BS for each transmission of the reference signal. When the BS transmits reference signals in a period, the method 800 may further include the step 850.

Optionally, at step 850, the ED transmits, in part or all of time intervals of the period, information that indicates a channel state of the set of communication channels to the BS. Correspondingly, the BS receives, in the part or all of time intervals of the period, information that indicates of the channel state of the set of communication channels from the ED.

For ease of description, the information that indicates the channel state is represented by information#1 in embodiments of this application.

In some embodiments, the information#1 may include part or all of channel sate information (CSI) . For example, the CSI may include one or more of: precoding matrix index (PMI) , channel quality indicator (CQI) and rank indicator (RI) .

The part or all of the time intervals (represented by time intervals#1 below) that transmit the information#1 may be determined in a variety of ways. In some embodiments, the ED may receive the reference signals with periodicity#1 and transmit the information#1 with periodicity#2. The periodicity#2 may be longer than the periodicity#1. For example, the periodicity#1 is an integer multiple of the periodicity#2. The ED may receive the reference signals in every time interval, and transmit the information#1 in every few time intervals.

In some embodiments, the time intervals#1may be determined based on the accuracy of the first model. For example, when the accuracy is higher than or equal to an accuracy threshold, the periodicity#2 may be determined as longer than or equal to a periodicity threshold.

The ED may obtain the time intervals#1 in a variety of ways. In some embodiments, the time intervals#1 may be signaled by a network device (e.g. base station) dynamically, e.g. in physical layer control signaling such as DCI, or semi-statically, e.g. in radio resource control (RRC) signaling or in the medium access control (MAC) layer, or be predefined based on the application scenario; or be determined by the ED as a function of other parameters that are known by the ED, or may be fixed, e.g. by a standard. This is not limited to this application.

Notably, the information#1 may indicate the real channel state based on received reference signals, but not the predicated channel state obtained from the first model. The information#1 may also assist in retraining the first model.

Notably, when the ED does not feed back the channel state information to the BS, the BS may use the first model to predict the channels states. That is, the method 800 may include the step 860.

Optionally, at step 860, the BS obtains the first model and uses the first model to predict channel states.

As aforementioned, the BS may obtain the first model in a variety of ways. In a first embodiment, the ED may generate the first model and transmit information that indicates the first model to the BS. This information may further indicate initial channel state corresponding to the first model. Therefore, the BS may use the initial channel state and the first model to predict the channel states of the set of the communication channels between the BS and the ED. A detailed example is given in conjunction with FIG. 10.

In a second embodiments, N EDs may generate N initial models and transmit information indicates the initial model respectively. This information may further indicate initial channel state corresponding to the corresponding initial model. The BS can generate the first model based on the N initial models. Therefore, the BS may use the initial channel states and the first model to predict the channel states of the sets of the communication channels between the BS and the N EDs. A detailed example is given in conjunction with FIG. 21.

Notably, the BS may use the initial channel state as the first input to the model, and each output can be used as the next input. In other words, this allows the ED not feed back the channel states constantly.

Notably, the ED can use measured reference signal information (e.g., first information) as each input of the first model. Alternatively, the ED may use predicted reference signal information (e.g., second information) as next input of the model, where the reference signal transmission can be reduced. This is not limited to this application.

Therefore, by intelligently processing ED reports and combining them with efficient predictive algorithms, the system can maintain communication efficiency while providing accurate state predictions. This method effectively balances communication load and prediction accuracy, offering robust support for efficient and flexible network operation. For instance, in this system, the strategy for ED reports, combined with periodicity, allows EDs to report intermittently rather than in every cycle, significantly reducing uplink communication volume. The BS uses the deep neural network to predict the state changes of each ED for the next few moments, greatly enhancing network efficiency.

As aforementioned, the predication accuracy of the first model (e.g., DNN) is important. If the first model can precisely predict state transitions, the future time period it covers may become longer. In other words, the stronger the predictive capability of the network, the less it relies on real-time data. This not only reduces the communication overhead between EDs and the system but also enhances the network's efficiency and reliability in handling dynamic environmental changes.

In some embodiments, the first model may be retrained in part or all of time intervals of the period. For the variability of the transmission environment between the ED and base station, the ED and/or base station also optimize or retrain the first model during channel estimation process. Therefore, the predication accuracy of the first model may maintain accuracy.

The first model can be retrained in a variety of ways. In some embodiments, the first model may be retrained based on real channel state information and corresponding predicated channel state information at the same interval. This application does not exclude other possible model retraining technology. For ease of description, a second model may denote the retrained model. The term “retrain” and “optimize” may be used inter-exchangeable in embodiments of this application.

The first model may be retrained by the ED, or by the base station, or by the ED and the base station. For example, the ED may retrain the first model to obtain the second model. The ED may further transmit information that indicates the second model to the base station. For another example, the ED may transmit information#1 that indicates the channel state to the base station. The base station retrains the first model to obtain the second model, and transmit information that indicates the second model to the ED. For another example, the ED may retrain the first model to obtain a retrained first model (i.e., an intermediate state model of the first model and the second model) , and transmit information that indicates the retrained first model to the base station. The base station may perform a process on the retrained first model to obtain the second model, and transmit information that indicates the second model to the ED. In this example, the retraining process may involve multiple EDs. The details of this example will be given in conjunction with FIG. 9.

The first model may be retrained in part or all of time intervals (represented by time intervals#2 below) of a period. In some embodiments, the ED may receive the reference signals within periodicity#1, and the first model is retrained within periodicity#3. The periodicity#3 may be longer than the periodicity#1. For example, the periodicity#1 is an integer multiple of the periodicity#3. The ED may receive the reference signals in every time interval, and the first model may be retrained in every few time intervals.

Notably, the time interval#2 and the time interval#1 may be the same or different, this is not limited to this application. The time interval#2 may be determined in a manner similar to the time interval#1 determination described above. For example, the time interval#2 may be determined based on the accuracy of the first model. Details are omitted here for brevity.

From the above technical solution, a first model can be used to predict the channel state of the set of communication channels between the base station and the ED. Moreover, taking the low-dimensional first information as input of the first model can reduce the dimension of input of the first model, which reduce computation complexity of the first model. An application of the first model in large-scale channels becomes feasible.

As aforementioned, the base station and the ED may collaborate to retrain the first model. In some embodiments, the retraining process may involve M EDs, M is a positive integer. The M EDs can be those within the coverage area of the base station. For ease of description, two EDs are taken as an example in FIG. 9, represented as ED#1 and ED#2. The first model deployed in the ED#1 is represented by a first model#1, and the first model deployed in ED#2 is represented by a first model#2. A set of communication channels between the base station and the ED#1 is represented by a set of communication channels#1, and a set of communication channels between the base station and the ED#2 is represented by a set of communication channels#2.

Notably, the retraining process may take place at any time during the channel estimation. The retraining process may use related-information of one or more time intervals, where the related-information may include: information that indicates a channel state obtained from real-time reference signal (s) transmitted in the one or more time intervals, and information indicates predicated channel state obtained from reference signal (s) transmitted earlier.

The retaining process may take place after the step 830, the first reference signal and the second information are used to retrain the first model. In other words, the ED#1 and ED#2 may each have the first model (first model#1 and first model#2) , and obtain second information#1 and second information#2 based on the first model#1 and the first model#2 respectively.

At step 910, the base station transmits a second reference signal to ED#1 and ED#2. Correspondingly, the ED#1 and ED#2 receive the second reference signal from the base station.

For example, the base station broadcasts or multi-casts the second reference signal in a second time interval. The ED#1 may receive the second reference signal trough the set of communication channels#1, and the ED#2 may receive the second reference signal through the set of communication channels#2.

The ED#1 may estimate a channel state#1 of the set of communication channels#1, and the ED#2 may estimate a channel state#2 of the set of communication channels#2. That is, the ED#1 and the ED#2 obtain a real channel state of corresponding set of communication channels.

At step 920, the ED#1 retrains the first model#1 based on the second information#1 and the second reference signal, to obtain retrained first model#1.

The second information#1 is used to predict the channel state of channels#1 in the second time interval. The ED#1 may retrain the first model#1 based on real channel state and predicted channel state in the second time interval. For example, when the first model is a DNN, the ED#1 may update the weights of the neurons of the DNN. This is not limited to this application.

At step 930, the ED#2 retrains the first model#2 based on the second information#2 and the second reference signal, to obtain retrained first model#2.

The second information#2 is used to predict the channel state of channels#2 in the second time interval. The ED#2 may retrain the first model#2 based on real channel state and predicted channel state in the second time interval. For example, when the first model is a DNN, the ED#2 may update the weights of the neurons of the DNN. This is not limited to this application.

Notably, the retrained first model#1 and the retrained first model#2 are intermediate state model of the first model and the second model, and are obtained based on different sets of communication channels.

At step 940, the ED#1 transmits information that indicates the retrained first model#1 to the base station. Correspondingly, the base station receives information that indicates the retrained first model#1 from the ED#1.

The information that indicates the retrained first model#1 may include a full retrained first model#1, or a partial retrained model#1 that contains the changed part. This is not limited to this application.

In some embodiments, the information that indicates the retrained first model#1 may further indicate one or more of parameters of the ED#1: information that indicates the channel state of the set of communicate channels obtained from estimation of the second reference signal (this information may be referred to as information#1 described in step 850) , measurement parameter (s) of the second reference signal; location parameter (s) of the ED#1 in the second time interval; movement parameter (s) of the ED#1 in the second time interval; and service parameter (s) corresponding to the ED#1 in the second time interval. The above parameters may be referred to as real-time state of the ED#1 in the second time interval. That is, the information that indicates the retrained first model#1 may further include related parameters used in retraining.

Notably, what parameters are carried in the information are related to actual application scenarios. This is not limited to this application. In some embodiments, this information may carry parameters that have changed from the first time interval. Therefore, the uplink transmission can be effectively reduced.

At step 950, the ED#2 transmits information that indicates the retrained first model#2 to the base station. Correspondingly, the base station receives information that indicates the retrained first model#2 from the ED#2.

The information that indicates the retrained first model#2 may include a full retrained first model#2, or a partial retrained model#2 that contains the changed part. This is not limited to this application.

In some embodiments, the information that indicates the retrained first model#2 may further include related parameters used in retraining. These parameters are similar to parameters described in step 940. Details are omitted here for brevity.

At step 960, the base station generates a second model based on the retrained first model#1 and retrained first model#2.

For example, the base station may perform a model process on the retrained first model#1 and retrained first model#2. The model process is dependent on the type of the first model. For example, when the first model is a DNN, the base station may perform a weighted average process on the retrained first model#1 and retrained first model#2, to obtain the second model.

The retrained first model#1 contains characteristics (e.g., variation characteristics from the first time interval to the second time interval) of the set of communication channels#1 between the base station and the ED#1. Similarly, the retrained first model#2 contains characteristics of the set of communication channels#2 between the base station and the ED#2. Therefore, the generated second model contain characteristics of multiple sets of communication channels, and it can have a good performance in the variable communication environment.

At step 970, the base station transmits information that indicates the second model to the ED#1 and the ED#2. Correspondingly, the ED#1 and the ED#2 receive the information that indicates the second model from the base station.

In some embodiments, the base station may broadcast or multi-cast the information that indicates the second model. This information may contain an entire second model or part of the second model (e.g., changed part) , this is not limited to this application.

In the subsequent channel estimation, ED#1 and ED#2 can use the updated second model to maintain the accuracy of channel estimation.

From the above technical solution, during the channel estimation process, the ED can refine or optimize the first model. The model can maintain accuracy in the changing communication environment.

The method 800 that an ED using a first model to predict a channel state is described in combination with FIG.

8.And the method that an ED retraining the first model is described in combination with FIG. 9. This application also provides a model training method, which can be implemented in combination with the above method 800 and/or method 900 or can be implemented independently. Details will be given below.

At step 1010, a base station transmits K reference signals to N electronic devices. Correspondingly, the N electronic devices receive the K reference signals.

The K reference signals may be any kind of downlink reference signals that is described above (e.g., CSI-RS, DMRS, and etc. ) . Alternatively, the K reference signals may also be a sequence of future-defined reference signals with a corresponding function. This is not limited to this application. K and N are positive integers. Notably, for ease of illustration, ED#1 and ED#2 (i.e., N=2) are taken as an example in FIG. 10. This is not limited in this application.

The channels between a single ED and the BS can be regarded as a set of communication channels. That is, the BD transmits the K reference signals through N sets of communication channels, an ED receives the K reference signals through its corresponding set of communication channels.

In some embodiments, the BS can broadcast or multicast the K reference signals. For example, the BS could transmit the K reference signals on broadcast downlink channels (PBCH) . Thereby, multiple EDs in the base station coverage area may receive the K reference signals simultaneously.

In some embodiments, the BS may transmit the K reference signals in a period, and the K reference signals may be transmitted in K time intervals of the period respectively. That is, the BS and ED may collect K reference signal transmissions to train a model.

Notably, the value of the K may be determined in a variety of ways. For example, the K may be greater than or equal to a threshold, as the greater the K and the better the accuracy of a model. For another example, the K may be determined based on communication environment. When the communication environment is changing frequently (e.g., urban environment) , the K may be greater than or equal to a threshold#1. When the communication environment is not changing frequently (e.g., rural environment) , the K may be greater than or equal to a threshold#2, where threshold#2 is smaller than the threshold#1. This is not limited in this application.

The value K mat be signaled by the base station, or be predefined based on the application scenario, or be determined by the ED#1 as a function of other parameters that are known by the ED#1, or may be fixed. This is not limited in this application.

At step 1020, ED#1 establishes an initial model#1 based on the K reference signals.

The initial model#1 can be used to generate a first model, where the first model is used to predict a channel state. The first model can be any types of model with prediction function. For example, the first model may be a DNN. This application does not exclude other possible models.

The initial model#1 is an intermediate model of generating a first model for a single set of communication channels#1. In some embodiments, the ED#1 may obtain training information#1, and use the training information#1 to establish the initial model#1.

For example, the training information#1 may include first training information#1, where the first training information#1 is generated by performing a signal process on the K reference signals. In this example, the signal process may be referred to description in FIG. 8. The signal process may be used to compress the dimension of the set of communication channels#1. The signal process may be further used to estimate the set of communication channels#1. Details are omitted here for brevity.

Notably, the first training information may refer to a description of the first information in FIG. 8, as they share a similar signal process.

In some embodiments, the training information#1 may further include second training information#1. The second training information indicates one or more of: measurement parameter (s) of the K reference signals through the set of communication channels#1; location parameter (s) of the ED#1; movement parameter (s) of ED#1; and service parameter (s) of ED#1. The second training information#1 may refer to a description of the third information in FIG. 8. Based on the characteristics of a data-driven and adaptive model, the dimensions of the training information used in training process may be similar to the input information (i.e., the first information and the third information) of the model. Details are omitted here for brevity.

At step 1030, ED#2 establishes an initial model#2 based on the K reference signals.

The ED#2 may establish the initial model#2 in a similar way as ED#1. That is, this step may be deduced from the step 1020, details are omitted here for brevity. Notably, this does not mean that ED#1 and ED#2 act exactly the same, as the set of communication channels#1 and the set of communication channels#2 may be different.

Notably, when the initial model#1 and initial model#2 are established with a signal process, the dimension of the initial model#1 and the dimension of the initial model#2 may be the same.

At step 1040, ED#1 transmits information that indicates the initial model#1 to the base station. Correspondingly, the base station receives the information that indicates the initial model#1 from the ED#1.

The initial model#1 may contain characteristics of the set of communication channels#1 between the ED#1 and the base station.

Notably, when the initial model#1 is generated with a signal process, the dimension of the initial model#1 can be effectively reduced. Therefore, uplink transmission consumption can be reduced.

In some embodiments, the information that indicates the initial model#1 may further include part or all of the training information. The part or all of the training information may assist in generation of the first model.

At step 1050, ED#2 transmits information that indicates the initial model#2 to the base station. Correspondingly, the base station receives the information that indicates the initial model#2 from the ED#2.

This step may be referred to the step 1040, and detail are omitted here for brevity.

At step 1060, the base station generates the first model based on the initial model#1 and the initial model#2.

For example, the base station may perform a process on the initial model#1 and the initial model#2, this process is related to the type of the first model. For example, when the first model is a DNN, the base station may perform a weighted average process on the initial model#1 and initial model#2, to obtain the first model.

Notably, initial model#1 and the initial model#2 are only for illustrative purposes. The base station may generate the first model based on M initial models from M electronic devices. This is not limited to this application. Each initial model contains characteristics of the corresponding set of communication channels. Therefore, the first model generated from the M initial models may contain characteristics of M sets of communication channels. The base station may predict channel states of multiple sets of channels in the coverage area using the first model.

In some embodiments, the base station may deliver the first model to the M electronic devices. That is, the method may further include step 1070 optionally.

At step 1070, the base station transmits information that indicates the first model to the ED#1 and the ED#2. Correspondingly, the ED#1 and the ED#2 receive the information that indicates the first model from the base station.

The base station may broadcast or multi-cast the information that indicates the first model. The training method that the base station and M electronic devices may be referred to as federated learning framework. It brings significant benefits to the ED side. By aggregating information and learning experiences from different EDs at the base station, each ED can gain valuable learning outcomes from others. This is particularly valuable in mobile environments, as EDs can use these shared learning experiences to predict the state of their channels in new locations, rather than relying solely on their known state information.

In some embodiments, the base station may transmit compression information to the M electronic devices. That is, the method may further include step 1080 before step 1020 optionally.

At step 1080, the base station transmits compression information to the ED#1 and the ED#2. Correspondingly, the ED#1 and the ED#2 receive the compression information from the base station.

The compression information may be referred to description in 840 and details are omitted here. Notably, when the method 800 and the method 1000 are implemented in a combination, the step 1080 may be omitted because the EDs may have obtained the compression information in step 840.

From the above technical solution, a base station and N EDs collaborate to generate a first model, and a dimension of an input of the first model is smaller than or equal to a dimension of each of the N sets of communication channels. The size of the first model based on N sets of communication channels, making the first model in large-scale channels feasible.

The main content of the method 800, method 900 and method 1000 are described in conjunction with FIGs. 8-10 above. For ease of understanding embodiments of this application. Some examples will be given in conjunction with FIGs. 11-27. Firstly, the signal process and the first model are illustrated with examples, and a general schematic block diagram is given in FIG. 11.

FIG. 11 is a schematic diagram of a signal process and a first model according to an embodiment of this application. For example, an ED may perform (820) a signal process on the first reference signal and input (830) first information and third information (optional) to a first model. The first information generated from first reference signals with a signal process, and the third information may contain some other current state of the ED (details can be referred back to the description of FIG. 8) .

For example, the signal process may use a common basis (U) as illustrated in FIG. 12. For another example, the signal process may use a low-dimension matrix (P) of the U as illustrated in FIG. 13. For another example, the signal process may use a unit scheme as illustrated in FIG. 14.

FIG. 12 illustrates an example of feature extraction with common basis U. One common way to do this is to apply SVD or proper orthogonal decomposition (POD) to find the eigen vectors of the channel data matrix A contributed by all EDs. The resulting eigen vectors correspond to the principal directions of variation of matrix A and can be ordered by their corresponding singular values or eigen values, which measure the amount of variance explained by each vector. By truncating the eigen vector matrix to retain only the most significant vectors, one can obtain a low-rank approximation of the data matrix A, that preserves most of the information and is called as common basis U.

The base station may determine a common basis U, which can reduce the dimension of original space H. As aforementioned, the original space H may be an n-dimensional vector, where n is a positive integer and is equal to N_subcarriers×n_TXAnt×n_RXAnt, where N_subcarriers represents the number of subcarriers, n_TXAnt represents the number of TX antennas and n_TXAnt represents the number of RX antennas. The base station could constitute a common basis U which is a matrix with n rows and r columns, where n is a positive integer and r (number of the principal components of the common basis) is less than n, r＜n. Therefore, an equivalent low-dimensional space c could be obtained that c=U^HH, where U^H represents conjugate transpose matrix of U. The low-dimensional space c is a r-dimensional vector, which means that the dimension of the c is less than the dimension of the original (which is n) space H.

A common basis U may transform a MIMO channel (H) from the original space to an equivalent low-dimensional space (c) . This is a linear transformation. The method, which involves learning a common basis from a set of training data samples and then projecting the channel into a low-dimensional equivalent space, has been key in addressing these challenges. In this reduced-dimensional space, tasks such as channel analysis, beamforming, and multi-ED matching can be performed more efficiently, significantly reducing computational complexity while maintaining performance.

In some embodiments, the signal process may further include reference-signal selection. The reference-signal selection involves processing the common basis U through pseudorandom or QR-based sampling scheme, significantly reducing its size and achieving compression. This method not only reduces the required transmission bandwidth but also allows for effective signal processing at the receiver end by computing the projection of the original channel into the equivalent low-dimensional space, even without fully restoring the original channel.

FIG. 13 illustrates an example of reference-signal selection with placement scheme P. Reference signal selection is the process of choosing a subset of candidate reference signal locations that maximize the information content or variance captured by the reference signals for all EDs.

As aforementioned, the original space H is an n-dimensional vector. The base station may determine a reference signal placement scheme P and transition the original H to a t-dimensional vector (represented by y in the FIG. 13) based on this placement scheme P, where t is a positive integer and is less than n, t＜n. Then a similar common basis constitution method (feature extraction) as described above can be used, and the size of the common basis (represented by θ in FIG. 13) can be reduced because of the placement scheme. This compact common basis θ is a matrix with t rows and r columns. Therefore, an equivalent low-dimensional space c could be obtained thatwhereis pseudo-inverse of θ.

In this embodiment, it applies the sparse subsampling scheme on the common basis U to get a compact (θ) and transform a MIMO channel (H) from the original space to a low-dimensional space (c) that is equivalent, and this transformation is linear.

From the 6G wireless communication perspective, this type of pseudorandom or QR decomposition-based sampling process (reference-signal selection) can be defined as a method for selecting (or placing) reference signals, i.e. a reference-signal placement scheme. This approach efficiently converts even very sparse channel information directly into an equivalent low-dimensional space, thus reducing the complexity and overhead resource requirements while preserving and transmitting channel information.

This example may initially involve acquiring the common basis for the entire cell area. Both the common basis and the reference signal placement scheme are designed based on data-driven methods, aimed at efficiently representing the channel characteristics of the entire cell area while providing a standardized framework for processing and transmitting channel information.

In some embodiments, the signal process may further include dynamic mode decomposition (DMD) . The DMD allows the channel to be divided into different units or blocks, distributed equidistantly along a specific channel direction (such as subcarriers or transmitting antennas) . Once this equidistant distribution is achieved, the DMD algorithm can be applied effectively in the equivalent low-dimensional space of these units. This method enables the identification of multiple modes in the equivalent low-dimensional space, revealing variations between units. Theoretically, with only the DMD modes and an estimation of one of the equivalent low-dimensional spaces, the base station and/or ED can reconstruct the spatial information of the entire channel. These modes constitute an equivalent ultra-low-dimensional space.

FIG. 14 illustrates an example of DMD. DMD is a data-driven technique for extracting spatiotemporal coherent structures from high-dimensional data. It can be used to analyze complex nonlinear dynamical systems, such as fluid flows, combustion, neuroscience, and epidemiology. DMD is based on the idea of decomposing a data matrix into a low-rank approximation that captures the dominant modes and frequencies of the underlying dynamics. DMD can also provide a linear model that approximates the nonlinear evolution of the system, which can be used for prediction, control, and optimization.

For example, the base station and/or the ED may determine a unit scheme method that how to divide a massive channel dimension (i.e., the original space H) into contiguous or equal-spaced subblocks (or units) . For example, the original space H is divided into k units equidistantly along a subcarrier direction where k is a positive integer, and the k units are presented as (H₀, H₁, …, H_k) . Then the base station could find a common reference placement scheme P that can be shared by all the k units, to transition the (H₀, H₁, …, H_k) to (y₀, y₁, …, y_k) . Furthermore, the base station could find a common basis U and its compact version θ that can be shared by the all k units, to obtain (c₀, c₁, …, c_k) .

As aforementioned, the common basis U, its compact version θ, common reference signal placement scheme (P) for a unit scheme involved in the signal process are given for illustrative purpose. These embodiments offer an effective solution for compressing and reconstructing channels in ultra-large-scale antenna array communications, especially when dealing with vast channel spaces.

When the DMD is involved in the signal process, by applying DMD in the equivalent low-dimensional space of each unit, the base station and/or the ED can effectively analyze and compress the channel information for all units. The result of DMD, the modes (represented by matrix G, or decomposition of G) or high-order derivatives, constitutes the equivalent ultra-low-dimensional space.

FIG. 15 illustrates an example of obtaining ultra low-dimensional space G by an ED. The base station transmits reference signals, and each ED could estimate channels based on the space parameters. For example, the ED take pilots of the reference signals based on unit scheme and placement scheme, to estimate channels on the pilots into a vector y. Then for each y, the ED could convert it to c domain directly based on the common basis (e.g., compact version of common basis θ) : c_j=θy_j, j=1, 2, ..., k. Over all c, the DMD could be employed to obtain matrix G =DMD (c₀, c₁, …, c_k) .

FIG. 16 illustrates an example of obtaining ultra low-dimensional space with DMD. The obtained low dimensional space (c₀, c₁, …, c_k) can form ultra low-dimensional space G and c₀. This ultra low-dimensional space G and c₀ can be projected back to the low dimensional: Then it can be projected back to the original space:

The above examples of the signal process are for illustrative purposes, this application does not exclude other possible dimension-related actions and estimation-related actions. This is not limited to this application.

As aforementioned, in some embodiments, the base station may broadcast compression information, where the compression information assists in the signal process.

FIG. 17 illustrates an example of transmission of compression information (840, 1080) . For example, the base station may broadcast compression information, which may indicate one or more of: unit scheme, common basis (e.g., compact version of common basis θ) , reference signal scheme P. The ED-1, ED-2, …, ED-N could receive the compression information and perform signal process based on the compression information. The compression information may be carried in one or more signals or messages, for example, the above parameters may be included in downlink (DL) controlling message (s) . This ensures consistency and coordination of information within the network, enabling EDs to interpret and utilize channel information based on a unified pattern.

As aforementioned, the communication environment may be constantly changing, and the ED may move constantly. FIG. 18 illustrates an example of an ED on the move. FIG. 18 shows the trajectory from time t1 to time t0 of the ED.A moving ED would be subjected to dynamic channels, channel is strongly related to the surroundings, moving trajectory, both of which may be traceable.

A key challenge in the field of communication is the spatial and temporal variance of the channel. As a system that constantly changes over time and space, influenced by the surrounding environment, the channel's characteristics also undergo alterations. This dynamic nature means that channel information from one location cannot be simply applied to another. Even at the same geographical location, channel characteristics might change due to variations in the surrounding environment, such as the movement of vehicles.

In order to maintain the accuracy of the first model in the changing communication environment, in some embodiments, when the base station transmits reference signals in a period, the first model may be retrained in part or all of time intervals of the period. FIG. 19 illustrates an example of delivering a retrained model.

Referring to FIG. 19, a DNN for predicting a channel state is given for illustrative purposes. For example, an ED may retrain its DNN and transmit information that indicates the retrained DNN to the base station with a periodicity, and this periodicity is longer than the periodicity of transmitting reference signals. The retrained DNN for predicting state transitions is dynamic and continuously updated in real-time under the previously DNN (e.g., the DNN established with federated learning framework) . This means that if significant environmental changes occur within the cell area, these changes will also be reflected in the state transition neural network based on the federated learning framework. Thus, the network can adapt not only to individual ED behavior changes but also to the dynamic changes in the entire cell environment.

In some embodiments, the process between the base station and EDs is meticulously designed to ensure effective communication and data exchange. Specific time slots are set within the system to coordinate data transmission. For example, a first time slot (or may be referred to as a network data transmission slot) may be allocated to deliver the retrained DNN (e.g., retrained DNN#0, retrained DNN#1 and retrained DNN#2 in the FIG. 19) . The ED may to send the neuron data of its retrained DNN to the base station. This step in the federated learning process, may enable that the network can be updated in real-time and adapt to current network conditions.

As aforementioned, information that indicates a channel state based on a real-time reference signal may be transmitted in part or all of time intervals (e.g., the state_T0, the state_T1, the state_T2, the state_T3, the state_T4 in the FIG. 19) . For example, a second time slot (or may be referred to as a state information transmission slot) may be allocated to transmit this information. The second time slot may be used for the ED to send its multimodal state information at a specific moment to the base station. This information might include channel modalities, signal delay, and other data related to the ED's state. This may help the base station understand the current network conditions and requirements of the ED.

The first time slot and the second time slot may occur synchronously or asynchronously, depending on the design and requirements of the network. The base station may explicitly instruct the ED on when to send its DNN neuron data and when to send their state information. Typically, information related to time slots may be conveyed to ED via broadcast channels or downlink control channels.

In some embodiments, different EDs can choose to operate synchronously in the same time slot or be assigned independent time slots.

When EDs feedback their DNN neuron data at the designated transmission slot, it is typically done via the uplink data channel or uplink control channel. EDs may have some degree of freedom to choose when to send their neural network data and state information and inform the base station of their transmission plan through the uplink control channel.

Within each federated learning cycle, EDs may be constantly estimating new channel states. However, to reduce the volume of uplink transmission, EDs don't need to transmit the entire state sequence. Instead, they can selectively transmit only key parts of their state at certain intervals, such as channel information, modal changes, delay, channel power, and even speed and location. This selective transmission method reduces network load while retaining sufficient information for effective state analysis.

When the ED is allowed not to feed back channel states for each reference-signal transmission in each time interval, the base station can predict the channel states for future time interval using DNN. For example, the base station may receive multiple DNNs from multiple EDs, and use respective DNN to predict the respective channel state of set of communication channels (afirst implementation of 860) .

FIG. 20 illustrates an example of predicting a channel state corresponding a single ED by a base station. For example, time intervals of the reference-signal transmission period may be represented as T₀, T₁, T₂……The channel states corresponding to the time intervals may be represented as state_T0, state_T1, state_T2…. The base station may obtain a state_T0 from the ED, where the state_T0 may be referred to as an initial channel state. The base station can input the stat_eT0 into the DNN#1, and obtain a sequence of channel states: state_T1, state_T2.... These may represent a prediction of a time-series of the ED’s state over the time.

Notably, each output of the DNN predicts a channel state corresponding to a time interval, and the output is used as the DNN input for the next time. The lower figure in FIG. 20 depicts the relationship in a time-expanded form.

For another example, the base station may perform a process on the multiple DNNs to generate a general DNN, and this general DNN may be able to predict multiple sets of communication channels (asecond implementation of 860) .

FIG. 21 illustrates an example of generating a general DNN corresponding to multiple EDs by a base station. For example, the base station may obtain multiple DNN from multiple EDs (three DNNs are illustrated in FIG. 21) , and it can obtain multiple sequences of channel states corresponding to the multiple EDs. The base station may generate a data set that may be referred to as re-generated data set, where the data set may assist in generating a general DNN. The general DNN contains characteristics of multiple sets of communication channels, that is, this general DNN can be more powerful.

FIG. 22 illustrates an example of interaction between a base station and EDs for generating or retraining (or updating) first model. This interaction may be implemented in generating process (e.g., FIG. 10) and/or in retraining process (e.g., FIG. 9) . The base station and EDs (represented ass ED#1, ED#2, …, ED#N) may loop the process:

At step 1: each ED trains its local model (e.g., deep neural network, which is represented as DNN-ED#1, DNN-ED#2, …, DNN-ED#N) . For example, each ED uses local data (e.g., the training information#1, 2, …, N) , to obtain model parameters (e.g., the deep neural network weights) .

At step 2: each ED transmits information that indicates the model parameters to the base station. For example, each ED send its deep neural network weights (or part or all of EDs local deep neural networks) to the base station.

At step 3: the base station processes the model parameters from EDs, to obtain BS model parameters, then send them (i.e., DNN) to each ED. For example, the base station averages the weights from EDs and broadcast the averaged weights to each ED.

Notably, the above steps may be in a loop during retraining process. The above “training” can be replaced with “retraining” .

FIG. 23 illustrates an example of transmitting reference signals periodically. The base station could transmit reference signals periodically to each ED, where 810 may be regarded as a transmission of a reference signal (i.e., the first reference signal) . Thereby, the EDs could perform channel estimation dynamically.

As the base station broadcasts reference signals periodically, such as at each transmission time interval (TTI) , EDs will obtain a time series containing multimodal states at various moments. This series reflects the time-varying channel and ED state information.

Additionally, EDs can also obtain other channel-related information, such as signal strength (RSSI) , channel delay (TA, timing advance) , Doppler frequency shift, etc. Moreover, EDs may consider other factors related to their state, like location, movement velocity, moving direction, and even the type and quality of service (QoS) they are currently using. These factors together constitute the ED's current state.

The base station inserts multiple reference signals into the downlink channels, enabling every ED to receive these broadcast reference signals and obtain an equivalent low-dimensional space channel estimation for each unit. This process, including the specific methods by which EDs obtain equivalent space channel estimation, has been detailed in our previous patents.

FIG. 24 illustrates an example of receiving reference signals periodically. As aforementioned, the ED could obtain first information (represented by G_T0, G_T1…) , and third information that includes, RSSI (represented by RSSI_T0, RSSI_T1…) , CFO (represented by CFO_T0, CFO_T1…) , TA (represented by TA_T0, TA_T1…) , position of the ED (represented by Pos_T0, Pos_T1…) , etc.

ED periodically receives the reference signals and compute the dynamic modes at each time interval, together with other measured values, which forms a multi-modality state.

Thus, at each time point, EDs obtain a multimodal state that includes information about the channel's equivalent ultra-low-dimensional space estimation, that is, the channel's modes and an estimation of at least one of its equivalent low-dimensional spaces. Each ED as worker may use its local data over time as epoch to train a state-transition multi-modality model.

FIG. 25 illustrates an example of training a DNN by a ED, which may be regarded as an example of 1020 or 1030 described in FIG. 10. As each ED gradually accumulates a certain amount of multimodal time series data, these data are considered a "epoch" for training their local deep neural networks. Using these time series data, each ED trains a deep neural network, with the deep neural network taking the multimodal state at time T as input and predicting the multimodal state at time T+1 as output. Thus, the deep neural network can effectively predict the transition of the ED's multimodal state, from the current moment to the next.

FIG. 26 illustrates an example of multiple EDs reporting model parameters, which may be regarded as an example of 1040 or 1050 described in FIG. 10. At the end of each epoch, each ED sends their local deep neural network weights and the first state to the BS.

With the base station's periodic broadcasting of reference signals, this federated learning process (steps two and three) is ongoing. In this way, both the base station and EDs continuously obtain the latest deep neural networks, capable of predicting the multimodal state transition of EDs at any spatial point and time point. This dynamic learning process ensures that the network can adapt to environmental changes and continuously enhance the accuracy and efficiency of predictions.

Another aspect worth noting is the flexibility of the deep neural network. While it can use complete state information for learning during the training phase, it doesn't necessarily require all the comprehensive state information for inference. For instance, even without precise ED location and velocity information, effective predictions (inference) can still be made using channel information feedback from EDs, such as ultra-small space modal information, delay, signal strength, and service information. This data could be sufficient to predict the ED's velocity and location in the next moment, allowing the neural network to function effectively even with some missing input data. Optionally, some deep neural network can output the scores (likelihood) about the predictions.

This approach enhances the network's flexibility and adaptability. Even with incomplete information, the base station can still use the existing data to accurately predict ED states through the deep neural network. This capability allows the network to better handle various uncertainties, improving overall communication efficiency and ED experience.

FIG. 27 illustrates an example of obtaining an averaged DNN corresponding to multiple EDs by a base station, which may be regarded as an example of 1060 and 1070 described in FIG. 10.

After completing a training epoch, EDs transmit the neuron data of their neural networks to the base station via the wireless uplink, particularly through the data channel. The base station receives neuron data from multiple EDs and performs weighted averaging on this data. Then, the averaged neuron data is broadcast back to each ED via the wireless downlink, particularly through the data channel too. Upon receiving the updated neuron data, EDs update their deep neural networks and continue the training and updating process with the next round of input data.

In embodiments of this application, in exploring how the base station utilizes the federated learning framework, the base station may be already equipped with a deep neural network capable of predicting state transitions. This network is specifically designed to process EDs'multimodal state information as input, including diverse data such as location, velocity, and channel modalities. The network predicts the possible next changes in ED states through inference (reasoning) based on the current state information received.

In embodiments of this application, a key innovation lies in the "digital twin" concept realized through the model (e.g. deep neural network) at the base station. The digital twin is not just a real-time mapping of ED states and situations. it also has the capability to predict future states and situations. This means that the digital twin can simulate the possible behaviors and network conditions of EDs in the next few moments, providing the base station with a highly accurate decision-support tool.

This concept has revolutionary significance for the scheduling of wireless systems, especially in the 6G era. One of the main challenges faced by 6G networks is to reduce latency. To achieve this, we must shift towards predictive control scheduling based on predicted results rather than just observed results. Traditional feedback-based scheduling methods might not meet the low latency requirements due to the rapid changes in channel states.

In embodiments of this application, with digital twins and predictive scheduling methods, the communication system can adapt to changes in channel conditions in real-time, making adjustments in advance rather than reacting after actual changes occur. This predictive scheduling not only reduces latency but also enhances resource allocation efficiency and network performance.

In some embodiments of this application, a dynamic approach, as opposed to a static one, is required to handle these channel variations. Different areas covered by a single base station might be affected by environmental changes. Such changes necessitate continuous updates in channel information to track its dynamic characteristics.

In some embodiments of this application, for predicting channel states, methods like Kalman filters or continuing with DMD (Dynamic Mode Decomposition) algorithms can be utilized. However, these algorithms typically face the challenge of primarily handling single-modal data for state prediction and analysis, failing to fully exploit the multimodal nature of the channel.

In some embodiments of this application, to more effectively handle the complex multimodal state changes of the channel, methods capable of processing multimodal information are required. This is where the use of deep neural networks for state transitions comes into play. Deep neural networks, due to their suitability for processing multimodal information, become powerful tools for handling complex channel state changes, promising to enhance the performance and reliability of communication systems.

In some embodiments of this application, EDs may just provide their current location, speed (or velocity) , moving direction, and other relevant state information. The deep learning network can then use this information, along with data learned from other EDs, to predict future state changes more accurately. This predictive mechanism not only improves the efficiency of the network but also significantly reduces unnecessary network signaling overhead. In other words, EDs do not need to send extensive channel state information, as the deep learning network can already accurately predict the future state of the channel.

It could be noted that the message in the disclosure could be replaced with information, which may be carried in one single message, or be carried in more than one separate message.

Without special noting, the terms “apparatus” and “device” are used exchangeable, and the terms “identity” and “identifier” are sued exchangeable. The terms "system" and "network" may be used interchangeably in embodiments of this application.

In the disclosure, the word “a” or “an” when used in conjunction with the term “comprising” or “including” in the claims and/or the specification may mean “one” , but it is also consistent with the meaning of “one or more” , “at least one” , and “one or more than one” unless the content clearly dictates otherwise. Similarly, the word “another” may mean at least a second or more unless the content clearly dictates otherwise.

In the disclosure, the words “first” , “second” , etc., when used before a same term (e.g., ED, or an operating step) does not mean an order or a sequence of the term. For example, the “first ED” and the “second ED” , means two different EDs without specially indicated, and similarly, the “first step” and the “second step” means two different operating steps without specially indicated, but does not mean the first step have to happen before the second step. The real order depends on the logic of the two steps.

The terms “coupled” , “coupling” or “connected” as used herein can have several different meanings depending on the context in which these terms are used. For example, as used herein, the terms coupled, coupling, or connected can indicate that two elements or devices are directly connected to one another or connected to one another through one or more intermediate elements or devices via a mechanical element depending on the particular context.

Note that the expression “at least one of A or B” , as used herein, is interchangeable with the expression “A and/or B” . It refers to a list in which you may select A or B or both A and B. Similarly, “at least one of A, B, or C” , as used herein, is interchangeable with “A and/or B and/or C” or “A, B, and/or C” . It refers to a list in which you may select: A or B or C, or both A and B, or both A and C, or both B and C, or all of A, B and C. The same principle applies for longer lists having a same format.

The methods according to embodiments of this application are described above in detail with reference to FIGS. 8-27. The apparatuses provided in embodiments of this application are described below in detail with reference to FIGS. 28-29. The description of apparatus embodiments corresponds to the description of the method embodiments. Therefore, for content that is not described in detail, refer to the foregoing method embodiments. For brevity, details are not described herein again.

Referring to FIG. 28, a schematic block diagram of a communication apparatus according to an embodiment of this application is shown. The communication apparatus 10 includes a transceiver unit 11 and a processing unit 12. The transceiver unit 11 may implement a corresponding communication function, and the processing unit 11 is configured to perform data processing. The transceiver unit 11 may also be referred to as a communication interface or a communication unit.

In some embodiments, the communication apparatus 10 may further include a storage unit. The storage unit may be configured to store instructions and/or data. The processing unit 12 may read instructions and/or data in the storage unit, to enable the communication apparatus to implement the foregoing method embodiments.

The communication apparatus 10 may be configured to perform actions performed by the ED in the foregoing method embodiments. In this case, the communication apparatus 10 may be the ED or a component that can be configured in the ED. The transceiver unit 11 is configured to perform communicating-related (e.g., receiving/transmitting-related) operations on the ED side in the foregoing method embodiments. The processing unit 12 is configured to perform processing-related operations on the ED side in the foregoing method embodiments.

The communication apparatus 10 may implement steps or procedures performed by the ED in FIGS. 8-27 according to embodiments of this application. The communication apparatus 10 may include units configured to perform the method performed by the ED in FIGS. 8-27. In addition, the units in the communication apparatus 10 and the foregoing other operations and/or functions are separately used to implement corresponding procedures in FIGS. 8-27.

Alternatively, the communication apparatus 10 may be configured to perform actions performed by the base station in the foregoing method embodiments. In this case, the communication apparatus 10 may be the base station or a component that can be configured in the base station. The transceiver unit 11 is configured to perform communicating-related (e.g., receiving/transmitting-related) operations on the base station side in the foregoing method embodiments. The processing unit 12 is configured to perform processing-related operations on the base station side in the foregoing method embodiments.

The communication apparatus 10 may implement steps or procedures performed by the base station in FIGS. 8-27 according to embodiments of this application. The communication apparatus 10 may include units configured to perform the method performed by the base station in FIGS. 8-27. In addition, the units in the communication apparatus 10 and the foregoing other operations and/or functions are separately used to implement corresponding procedures in FIGS. 8-27.

A specific process in which the units perform the foregoing corresponding steps is described in detail in the foregoing method embodiments. For brevity, details are not described herein again.

Referring to FIG. 29, a schematic block diagram of another communication apparatus according to an embodiment of this application is shown. The communication apparatus 20 includes a processor 21. The processor 21 is coupled to a memory 22. The memory 22 is configured to store a computer program or instructions and/or data. The processor 21 is configured to execute the computer program or instructions and/or data stored in the memory 22, so that the methods in the foregoing method embodiments are executed.

In some embodiments, the communication apparatus 20 includes one or more processors 21.

In an example, as shown in FIG. 29, the communication apparatus 20 may further include the memory 22.

In some embodiments, the communication apparatus 20 may include one or more memories 22.

In an example, the memory 22 may be integrated with the processor 21, or disposed separately from the processor 21.

In an example, as shown in FIG. 29, the communication apparatus 20 may further include a transceiver 23, where the transceiver 23 is configured to receive and/or transmit a signal. For example, the processor 21 may be configured to control the transceiver 23 to receive and/or transmit a signal.

In some embodiments, the communication apparatus 20 may be a ED or a component (e.g., a chip, a circuit, or a processing system) that can be configured in the ED; or the communication apparatus 20 may be a base station or a component (e.g., a chip, a circuit, or a processing system) that can be configured in the base station.

In a solution, the communication apparatus 20 is configured to perform the operations performed by the ED in the foregoing method embodiments.

For example, the processor 21 may be configured to perform a processing-related operation performed by the ED in the foregoing method embodiments, and the transceiver 23 may be configured to perform a communicating-related (e.g., receiving/transmitting-related) operation performed by the ED in the foregoing method embodiments.

In another solution, the communication apparatus 20 is configured to perform the operations performed by the base station in the foregoing method embodiments.

For example, the processor 21 may be configured to perform a processing-related operation performed by the base station in the foregoing method embodiments, and the transceiver 23 may be configured to perform a communicating-related (e.g., receiving/transmitting-related) operation performed by the base station in the foregoing method embodiments.

An embodiment of this application further provides a computer-readable storage medium. The computer-readable storage medium stores computer instructions used to implement the method performed by the ED or the method performed by the base station in the foregoing method embodiments.

For example, when the computer program is executed by a computer, the computer may be enabled to implement the method performed by the ED or the method performed by the base station in the foregoing method embodiments.

An embodiment of this application further provides a computer program product including instructions. When the instructions are executed by a computer, the computer is enabled to implement the method performed by the ED or the method performed by the base station in the foregoing method embodiments.

An embodiment of this application further provides a communication system. The communication system includes the ED and the base station in the foregoing embodiments.

For explanations and beneficial effects of related content of any communication apparatus provided above, refer to a corresponding method embodiment provided above. Details are not described herein again.

The processor mentioned in embodiments of this application may be a central processing unit (CPU) . The processor may further be another general-purpose processor, a digital signal processor (DSP) , an application-specific integrated circuit (ASIC) , a field programmable gate array (FPGA) , or another programmable logic device, a discrete gate, a transistor logic device, a discrete hardware component, or the like. The general-purpose processor may be a microprocessor, or the processor may be any conventional processor or the like.

The memory mentioned in embodiments of this application may be a volatile memory or a non-volatile memory, or may include a volatile memory and a non-volatile memory. The non-volatile memory may be a read-only memory (ROM) , a programmable read-only memory (programmable ROM, PROM) , an erasable programmable read-only memory (erasable PROM, EPROM) , an electrically erasable programmable read-only memory (electrically EPROM, EEPROM) , or a flash memory. The volatile memory may be a random access memory (RAM) . For example, the RAM may be used as an external cache. By way of example but not limitation, the RAM may include a plurality of forms such as the following: a static random access memory (static RAM, SRAM) , a dynamic random access memory (dynamic RAM, DRAM) , a synchronous dynamic random access memory (synchronous DRAM, SDRAM) , a double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM) , an enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM) , a synchlink dynamic random access memory (synchlink DRAM, SLDRAM) , and a direct rambus random access memory (direct rambus RAM, DR RAM) .

It should be noted that when the processor is a general-purpose processor, a DSP, an ASIC, an FPGA, another programmable logic device, a discrete gate or a transistor logic device, or a discrete hardware component, the memory (storage module) may be integrated into the processor.

It should be further noted that the memory described in this specification is intended to include, but is not limited to, these memories and any other memory of a suitable type.

A person of ordinary skill in the art may be aware that, in combination with the examples described in embodiments disclosed in this specification, units and methods may be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed by hardware or software depends on particular applications and design constraints of the technical solutions. A person skilled in the art may use different methods to implement the described functions for each particular application, but it should not be considered that the implementation goes beyond the protection scope of this application.

It should be noted that the term “receive” or “receiving” used herein may refer to receiving or otherwise obtaining from an element/component in same apparatus or from another device separate from the apparatus. Similarly, the term “transmit” or “transmitting” may refer to outputting or sending to/for an element/component in same apparatus or to/for another device separate from the apparatus. For example, any of the methods/procedures described herein may be performed by a chipset, in which case any sending or receiving steps may occur between elements of the chipset.

It may be clearly understood by a person skilled in the art that, for the purpose of convenient and brief description, for a detailed working process of the foregoing apparatus and unit, refer to a corresponding process in the foregoing method embodiment. Details are not described herein again.

In the several embodiments provided in this application, the disclosed apparatuses and methods may be implemented in other manners. For example, the described apparatus embodiment is merely an example. For example, division into the units is merely logical function division and may be other division in an actual implementation. For example, a plurality of units or components may be combined or integrated into another system, or some features may be ignored or not performed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections may be implemented through some interfaces. The indirect couplings or communication connections between the apparatuses or units may be implemented in electronic forms, mechanical forms, or other forms.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on a plurality of network units. Some or all of the units may be selected based on an actual requirement to implement the solutions provided in this application.

In addition, function units in embodiments of this application may be integrated into one unit, or each of the units may exist alone physically, or two or more units may be integrated into one unit.

All or some of the foregoing embodiments may be implemented by using software, hardware, firmware, or any combination thereof. When the software is used to implement embodiments, all or a part of embodiments may be implemented in a form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the procedures or functions according to embodiments of this application are all or partially generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or another programmable apparatus. For example, the computer may be a personal computer, a server, a network device, or the like. The computer instructions may be stored in a computer-readable storage medium or may be transmitted from a computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website, computer, server, or data center to another website, computer, server, or data center in a wired (for example, a coaxial cable, an optical fiber, or a digital subscriber line (DSL) ) or wireless (for example, infrared, radio, and microwave, or the like) manner. The computer-readable storage medium may be any usable medium accessible by the computer, or a data storage device, for example, a server or a data center, integrating one or more usable media. The usable medium may be a magnetic medium (for example, a floppy disk, a hard disk, or a magnetic tape) , an optical medium (for example, a DVD) , a semiconductor medium (for example, an SSD) , or the like. For example, the usable medium may include but is not limited to any medium that can store program code, such as a USB flash drive, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disc.

The foregoing description is merely a specific implementation of this application, but is not intended to limit the protection scope of this application. Any variation or replacement readily figured out by a person skilled in the art within the technical scope disclosed in this application shall fall within the protection scope of this application. Therefore, the protection scope of this application shall be subject to the protection scope of the claims and the specification.

Claims

A method for communication, comprising:

receiving (810) a first reference signal through a set of communication channels;

generating (820) first information based on the first reference signal, wherein a dimension of the first information is smaller than a dimension of the set of communication channels; and

inputting (830) the first information to a first model, to obtain second information, wherein the second information is used to predict a channel state of the set of communication channels.
The method according to claim 1, wherein the method further comprises:

receiving (840) compression information, wherein the first information is generated by performing a signal process on the first reference signal, and the compression information is for assisting the signal process.
The method according to claim 1 or 2, wherein the signal process is used to compress the dimension of the set of communication channels.
The method according to any one of claims 1 to 3, wherein the first information comprises a low-dimensional channel coefficient vector of the set of communication channels.
The method according to any one of claims 1 to 4, wherein the first reference signal is transmitted in a first time interval, the second information is used to predict a channel state of the set of communication channels in a second time interval,

and the method further comprises:

receiving (910) a second reference signal through the set of communication channels in the second time interval; and

retraining (920 or 930) the first model based on the second information and the second reference signal.
The method according to claim 5, wherein the method further comprises:

transmitting (940 or 950) information that indicates a retrained first model; and

receiving (970) information that indicates a second model, wherein the second model is generated based on M models that comprise the retrained first model, M is a positive integer.
The method according to claim 4, wherein the M models are from M electronic devices.
The method according to any one of claims 1 to 7, wherein the inputting (830) the first information to a first model to obtain second information, comprises:

inputting the first information and third information to the first model to obtain the second information, wherein the third information indicates one or more of:

measurement parameter (s) of the first reference signal;

location parameter (s) of an electronic device that receives the first reference signal through the set of communication channels;

movement parameter (s) of an electronic device that receives the first reference signal through the set of communication channels; and

service parameter (s) corresponding to an electronic device that receives the first reference signal through the set of communication channels.
The method according to any one of claims 2 to 8, wherein the compression information indicates one or more of:

a dimension of the first information generated by the signal process;

a common basis (U) of the set of communication channels involved in the signal process;

a low-dimension matrix (P) of the U, wherein the P indicates a pattern of the first reference signal involved in the signal process; and

a unit scheme, wherein the unit scheme indicates the signal process involves dividing the set of communication channel based on one or more of: frequency domain, time domain and space domain.
The method according to any one of claims 1 to 9, wherein the first model is generated based on N initial models, and the N initial models are from N electronic devices, N is a positive integer.
The method according to any one of claims 1 to 10, wherein the method further comprises:

receiving K third reference signals through the set of communication channels, K is a positive integer;

performing the signal process on the K third reference signals, to generate training information;

establishing an initial model based on the training information;

transmitting information that indicates the initial model; and

receiving information that indicates the first model, wherein the first model is generated based on N initial models from N electronic devices, N is a positive integer.
The method according to claim 11, wherein the information that indicates the initial model comprises the training information.
The method according to any one of claims 1 to 12, wherein the receiving (810) a first reference signal through communication channels, comprises:

receiving, in a period, reference signals that comprise the first reference signal through the set of communication channels, the first reference signal is transmitted in a first time interval, the second information is used to predict a channel state of the set of communication channels in a second time interval, and the first time interval and the second time interval are two adjacent time intervals of the period.
The method according to claim 13, wherein the method further comprises:

transmitting, in part or all of time intervals of the period, information that indicates a channel state of the set of communication channels.
The method according to claim 13 or 14, wherein the first model is retrained in part or all of time intervals of the period.
A method for communication, comprising:

transmitting (810) a first reference signal through a set of communication channels, wherein first information is generated based on the first reference signal, an input of a first model comprises the first information, and second information obtained by the first model is used to predict a channel state of the set of communication channels.
The method according to claim 16, wherein the method further comprises:

transmitting (840) compression information, wherein the first information is generated by performing a signal process on the first reference signal, and the compression information is for assisting in the signal process.
The method according to claim 16 or 17, wherein the signal process is used to compress the dimension of the set of communication channels.
The method according to any one of claims 16 to 18, wherein the first information comprises low-dimensional channel coefficient vector of the set of communication channels.
The method according to any one of claims 16 to 19, wherein the first reference signal is transmitted in a first time interval, the second information is used to predict a channel state of the set of communication channels in a second time interval,

and the method further comprises:

transmitting (910) a second reference signal through the set of communication channels in the second time interval, wherein the second reference signal and the second information are used to retrain the first model.
The method according to claim 20, wherein the method further comprises:

receiving (940 or 950) information that indicates M models that comprise a retrained first models, M is a positive integer;

generating (960) a second model based on the M models; and

receiving (970) information that indicates the second model.
The method according to claim 21, wherein the M models are from M electronic devices.
The method according to any one of claims 16 to 22, wherein the input of the first model further comprises third information, wherein the third information indicates one or more of:

measurement parameter (s) of the first reference signal;

location parameter (s) of an electronic device that receives the first reference signal through the set of communication channels;

movement parameter (s) of an electronic device that receives the first reference signal through the set of communication channels; and

service parameter (s) corresponding to an electronic device that receives the first reference signal through the set of communication channels.
The method according to any one of claims 17 to 23, wherein the compression information indicates one or more of:

a dimension of the first information generated by the signal process;

a common basis (U) of the set of communication channels involved in the signal process;

a low-dimension matrix (P) of the U, wherein the P indicates a pattern of the first reference signal involved in the signal process; and

a unit scheme, wherein the unit scheme indicates the signal process involves dividing the set of communication channel based on one or more of: frequency domain, time domain and space domain.
The method according to any one of claims 16 to 24, wherein the first model is generated based on N initial models, and the N initial models are from N electronic devices, N is a positive integer.
The method according to any one of claims 16 to 25, wherein the method further comprises:

transmitting K third reference signals through N sets of communication channels, wherein the N sets of communication channels correspond to N electronic devices, K and N are positive integers;

receiving information that indicates N initial models from the N electronic devices, wherein an initial model of the N initial models is generated by training information, and the training information is generated by the K third reference signals;

generating the first model based on the N initial models; and

transmitting information that indicates the first model.
The method according to claim 26, wherein the information that indicates the N initial models comprises part or all of the training information.
The method according to any one of claims 16 to 27, wherein the transmitting (810) a first reference signal through communication channels, comprises:

transmitting, in a period, reference signals that comprise the first reference signal through the set of communication channels, the first reference signal is transmitted in a first time interval, the second information is used to predict a channel state of the set of communication channels in a second time interval, and the first time interval and the second time interval are two adjacent time intervals of the period.
The method according to claim 28, wherein the method further comprises:

receiving, in part or all of time intervals of the period, information that indicates a channel state of the set of communication channels.
The method according to claim 28 or 29, wherein the first model is retrained in part or all of time intervals of the period.
The method according to any one of claims 15 to 30, wherein the method further comprises: obtaining the first model and using the first model to predict channel states of the set of communication channels.
A method for communication, comprising:

transmitting (1010) K reference signals through N sets of communication channels, wherein the N sets of communication channels correspond to N electronic devices, K and N are positive integers;

receiving (1040 or 1050) information that indicates N initial models, wherein the N initial models are generated by the K reference signals; and

generating (1060) a first model based on the N initial models, wherein a dimension of an input of the first model is smaller than or equal to a dimension of each of the N sets of communication channels, and the first model is used to predict a channel state of the N sets of communication channels.
The method according to claim 32, wherein the method further comprises:

transmitting (1070) information that indicates the first model.
The method according to claim 32 or 33, wherein an initial model of the N initial models is established based on training information of a corresponding set of communication channels, the training information comprises first training information that is generated by performing a signal process on the K sets of reference signal.
The method according to claim 34, wherein the method further comprises:

transmitting compression information, wherein the compression information is for assisting the signal process.
The method according to claim 34 or 35, wherein the signal process is used to compress a dimension of the corresponding set of communication channels.
The method according to any one of claims 34 to 36, wherein the first training information comprises low-dimensional channel coefficient vector of the corresponding set of communication channels.
The method according to any one of claims 34 to 37, wherein the training information further comprises second training information that indicates one or more of:

measurement parameter (s) of the K reference signals through the corresponding set of communication channels;

location parameter (s) of a corresponding electronic device;

movement parameter (s) of a corresponding electronic device; and

service parameter (s) of a corresponding electronic device.
The method according to any one of claims 34 to 38, wherein the information that indicates the N initial models comprises part or all of the training information.
The method according to any one of claims 35 to 39, wherein the compression information indicates one or more of:

a dimension of the first training information generated by the signal process;

a common basis (U) of each of the N sets of communication channels involved in the signal process;

a low-dimension matrix (P) of the U, wherein the P indicates a pattern of the K reference signal involved in the signal process; and

a unit scheme, wherein the unit scheme indicates the signal process involves dividing each of the N sets of communication channel based on one or more of: frequency domain, time domain and space domain.
A method for communication, comprising:

receiving (1010) K reference signals through a set of communication channels, K is a positive integer;

establishing (1020 or 1030) an initial model based on the K reference signals; and

transmitting (1040 or 1050) information that indicates the initial model, wherein the initial model is used to generate a first model, a dimension of an input of the first model is smaller than or equal to a dimension of the set of communication channels, and the first model is used to predict a channel state of the set of communication channels.
The method according to claim 41, wherein the method further comprises:

receiving (1070) information that indicates the first model.
The method according to claim 41 or 42, wherein the establishing (1020 or 1030) an initial model based on the K reference signals comprises:

performing a signal process on the K reference signals, to generate first training information; and

establishing the initial model based on training information that comprises the first training information.
The method according to claim 43, wherein the method further comprises:

receiving compression information, wherein the compression information is for assisting the signal process.
The method according to claim 43 or 44, wherein the signal process is used to compress the dimension of the set of communication channels.
The method according to any one of claims 43 to 45, wherein the first training information comprises low-dimensional channel coefficient vector of the set of communication channels.
The method according to any one of claims 43 to 46, wherein the training information further comprises second training information, wherein the second training information that indicates one or more of:

measurement parameter (s) of the K reference signals of the set of communication signals;

location parameter (s) of an electronic device that receives the K reference signals from the set of communication signals;

movement parameter (s) of an electronic device that receives the K reference signals from the set of communication signals; and

service parameter (s) corresponding to an electronic device that receives the K reference signals through the set of communication channels.
The method according to any one of claims 43 to 47, wherein the information that indicates the initial model comprises part or all of the training information.
The method according to any one of claims 44 to 48, wherein the compression information indicates one or more of:

a dimension of the first training information generated by the signal process;

a common basis (U) of each of the N sets of communication channels involved in the signal process;

a low-dimension matrix (P) of the U, wherein the P indicates a pattern of the K reference signal involved in the signal process; and

a unit scheme, wherein the unit scheme indicates the signal process involves dividing each of the N sets of communication channel based on one or more of: frequency domain, time domain and space domain.
An apparatus, wherein the apparatus comprises a processor and a memory storing one or more instructions that is capable of being run on the processor, and when the one or more instructions are run, the apparatus is enabled to perform the method according to any one of claims 1 to 15, or perform the method according to any one of claims 16 to 31, or perform the method according to any one of claims 32 to 40, or perform the method according to any one of claims 41 to 49.
An apparatus, wherein the apparatus comprises a function or unit to perform the method according to any one of claims 1 to 15, or perform the method according to any one of claims 16 to 31, or perform the method according to any one of claims 32 to 40, or perform the method according to any one of claims 41 to 49.
A communications system, comprising an electronic device and a network device, wherein the electronic device performs the method according to any one of claims 1 to 15, and the network device performs the method according to any one of claims 16 to 31.
The system according to claim 52, wherein the electronic device further performs the method according to any one of claims 32 to 40, and the network device further performs the method according to any one of claims 41 to 49.
A communications system, comprising an electronic device and a network device, wherein the electronic device performs the method according to any one of claims 32 to 40, and the network device performs the method according to any one of claims 41 to 49.
A computer-readable storage medium, comprising one or more instructions, wherein when the one or more instructions are run on a computer, the computer performs the method according to any one of claims 1 to 15, or the method according to any one of claims 16 to 31, or the method according to any one of claims 32 to 40, or the method according to any one of claims 41 to 49.
A non-transitory computer-readable medium storing instruction the instructions causing a processor in a device to implement the method according to any one of claims 1 to 15, or the method according to any one of claims 16 to 31, or the method according to any one of claims 32 to 40, or the method according to any one of claims 41 to 49.
A device configured to perform the method according to any one of claims 1 to 15, or the method according to any one of claims 16 to 31, or the method according to any one of claims 32 to 40, or the method according to any one of claims 41 to 49.
A processor, configured to execute instructions to cause a device to perform the method according any one of claims 1 to 15, or the method according to any one of claims 16 to 31, or the method according to any one of claims 32 to 40, or the method according to any one of claims 41 to 49.
An integrated circuit configure to perform the method according to any one of claims 1 to 15, or the method according to any one of claims 16 to 31, or the method according to any one of claims 32 to 40, or the method according to any one of claims 41 to 49.
A communication apparatus, comprising:

a transceiver unit, configured to perform the receiving step according to any one of claims 1 to 15;

a processing unit, configured to perform the processing step according to any one of claims 1 to 15.
A communication apparatus, comprising a transceiver unit, configured to perform the transmitting step according to any one of claims 16 to 31.
A communication apparatus, comprising:

a transceiver unit, configured to perform the receiving step according to any one of claims 32 to 40;

a processing unit, configured to perform the processing step according to any one of claims 32 to 40.
A communication apparatus, comprising:

a transceiver unit, configured to perform the receiving step according to any one of claims 41 to 49;

a processing unit, configured to perform the processing step according to any one of claims 41 to 49.