WO2025008191A1

WO2025008191A1 - Communication devices and methods

Info

Publication number: WO2025008191A1
Application number: PCT/EP2024/067213
Authority: WO
Inventors: Thomas Handte; Dana CIOCHINA-KAR; Ken Tanaka; Salma OTHMAN
Original assignee: Sony Europe Bv; Sony Group Corp
Current assignee: Sony Europe Bv; Sony Group Corp
Priority date: 2023-07-04
Filing date: 2024-06-20
Publication date: 2025-01-09
Anticipated expiration: 2026-01-04

Abstract

A first communication device configured to communicate with a second communication device, the first communication device comprising circuitry configured to receive a data unit transmitted by the second communication device or a third communication device; determine channel state information (CSI) based on the received data unit; encode the determined CSI using an encoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data; and transmit CSI feedback comprising the encoded CSI to the second communication device.

Description

Applicants:

Sony Group Corporation 20.06.2024 1-7-1 Konan 4727P365WO - SK

Minato-Ku

Tokyo 108-0075

JAPAN

Sony Europe B.V.

The Heights, Brooklands

Weybridge, Surrey

KT13 OXW

UNITED KINGDOM

COMMUNICATION DEVICES AND METHODS

BACKGROUND

FIELD OF THE DISCLOSURE

[0001] The present disclosure relates to first and second communication devices that are configured to communicate with each other. The present disclosure relates further to corresponding first and second communication methods.

DESCRIPTION OF RELATED ART

[0002] In communication systems there is often the need to transfer channel state information (CSI) wirelessly to a peer communication device. For example, in beamforming, CSI is required by the peer communication device to compute beamforming weights for im- proved communication. Similarly in sensing applications, CSI may be used to detect movements or changes in the environment. As of today, CSI is compressed by quantization and subsampling (in time or frequency) before transmission, but a compression beyond quantization and down sampling is rarely done.

[0003] The “background” description provided herein is for the purpose of generally presenting the context of the disclosure. Work of the presently named inventor(s), to the extent it is described in this background section, as well as aspects of the description which may not otherwise qualify as prior art at the time of filing, are neither expressly nor impliedly admitted as prior art against the present disclosure.

SUMMARY

[0004] It is an object to enable a more effective compression of CSI and to provide corresponding communication devices and methods. It is a further object to provide a corresponding method as well as a corresponding computer program and a non-transitory computer- readable recording medium that stores therein a computer program product for implementing said method.

[0005] According to an aspect there is provided a first communication device configured to communicate with a second communication device, the first communication device comprising circuitry configured to: receive a data unit transmitted by the second communication device or a third communication device; determine channel state information (CSI) based on the received data unit; encode the determined CSI using an encoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data; and transmit CSI feedback comprising the encoded CSI to the second communication device. [0006] According to a further aspect there is provided a second communication device configured to communicate with a first communication device, the second communication device comprising circuitry configured to: transmit a data unit to the first communication device; receive channel state information (CSI) feedback comprising encoded CSI from the first communication device; and decode the encoded CSI using a decoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data, the decoder corresponding to an encoder of the trained algorithm or computing system that has been used for encoding the CSI.

[0007] According to still further aspects a computer program comprising program means for causing a computer to carry out the steps of the methods disclosed herein, when said computer program is carried out on a computer, as well as a non-transitory computer- readable recording medium that stores therein a computer program product, which, when executed by a processor, causes the methods disclosed herein to be performed are provided.

[0008] Embodiments are defined in the dependent claims. It shall be understood that the disclosed communication methods, the disclosed computer program and the disclosed computer-readable recording medium have similar and/or identical further embodiments as the claimed communication devices and as defined in the dependent claims and/or disclosed herein.

[0009] One of the aspects of the disclosure is to use an encoder of a trained algorithm or computing system, e.g. an autoencoder (AE) or a neural network (NN), to compress CSI and to provide a framework for two interacting communication devices for it to be applied. Furthermore, requirements and mechanisms are presented in embodiments to determine in advance if CSI compression with an encoder is appropriate. Thus, according to the present disclosure, a “true” and more effective compression of CSI, i.e., a compression beyond quantization and down sampling (e.g. by exploiting correlations or by applying simplifications or by non-linear compression or by extracting the essence or by applying a transformation), is provided. [0010] The foregoing paragraphs have been provided by way of general introduction and are not intended to limit the scope of the following claims. The described embodiments, together with further advantages, will be best understood by reference to the following detailed description taken in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWING

[0011] A more complete appreciation of the disclosure and many of the attendant advantages thereof will be readily obtained as the same becomes better understood by reference to the following detailed description when considered in connection with the accompanying drawings, wherein:

Fig. 1 shows a diagram of a conventional communication scheme using CSI feedback.

Fig. 2 shows a diagram of a conventional communication scheme using a neural network.

Fig. 3 shows a flowchart of a first communication method carried out by a first communication device according to the present disclosure.

Fig. 4 shows a flowchart of a second communication method carried out by a second communication device according to the present disclosure.

Fig. 5 shows an embodiment of a trained algorithm or computer system that may be used in embodiments of the present disclosure.

Fig. 6 shows a first embodiment of communication devices according to the present disclosure.

Fig. 7 shows a second embodiment of communication devices according to the present disclosure. Fig. 8 shows a third embodiment of communication devices according to the present disclosure.

Fig. 9 shows an embodiment of an encoder built up by various encoders of a basic configuration.

Fig. 10 shows an embodiment of CSI feedback comprising multiple encoded CSI partitions.

Fig. 11A shows an embodiment of first and second communication devices using a decision mechanism for deciding upon the CSI feedback type.

Fig. 11 B shows a flowchart of a corresponding method of the decision mechanism illustrated in Fig. 11A.

Fig. 12A shows an embodiment of first and second communication devices using a decision mechanism for deciding upon update training.

Fig. 12B shows a flowchart of a corresponding method of the decision mechanism illustrated in Fig. 12A.

Fig. 13 shows a fourth embodiment of communication devices according to the present disclosure.

DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] In 802.11 wireless LAN, CSI feedback is implemented as follows and as illustrated in Fig. 1 showing a diagram of a conventional communication scheme using CSI feedback. In an initial step 10 a second communication device 2 transmits a packet (e.g. a PPDU) to a first communication device 1. This PPDU can be a null data packet (NDP) and may generally be referred to as channel sounding data unit, although it may be a conventional data unit like a PPDU as well. The NDP is preceded by an NDP announcement (NDPA), generally referred to as channel sounding announcement herein (not shown in Fig. 1), which configures the feedback parameters. Based on this packet, the first communication device 1 derives channel state information (CSI) in subsequent step 11 , e.g., by evaluating measured signals on an OFDM subcarrier of a training field and comparing it with its reference value. Subsequently, the first communication device 1 applies quantization to the measured CSI values which includes (i) mapping a CSI value to discrete levels, (ii) down sampling the CSI values by considering only every N-th subcarrier (N = 1 , 2, 4). The compressed CSI indicated by Q(CSI) is transmitted in step 12 to the second communication device 2, where it is evaluated and further processed, e.g., for beamforming or sensing purposes.

[0013] In the prior art the use of artificial intelligence (Al) is considered to provide a better compression with less reconstruction loss compared the IEEE 802.11 baseline. It is considered not to use CSI but beamforming matrix information that can be represented by two angles ip and (p; hence, the feedback can only be used for the purpose for which the computed beamforming was derived. In an option, a clustering of all possible CSI is performed, and the feedback consists just of a cluster index number.

[0014] In another option, an autoencoder (AE) for CSI compression is proposed with online training for CSI as shown in Fig. 2 depicting a diagram of a conventional communication scheme using a neural network (NN) 30, representing an AE and comprising an encoder (E) 31 (sometimes also called encoder part) and a decoder (D) 32 (sometimes also called decoder part). In a first step 20 the second communication device 2, e.g. an access point (AP), transmits the encoder (E) to the first communication device 1 , e.g. a station (STA) 1. In a second step 21 an NDP is transmitted to the STA 1 , which derives CSI in step 22 and computes the two angles ip and (p in step 23. Subsequently, in step 24 the computed angles ip and (p are encoded, and the encoded angles E(ip, <p) are then transmitted to the AP 2, which decodes the encoded angles E(ip, <p) in step 25, denoted as D(E(ip, <p)) resulting in angles ip and (p.

[0015] The known solutions have a number of shortcomings as follows. A first shortcoming is that the known solutions represent AP centric solutions. The STAs must support the model provided by the AP which is difficult for a STA with limited capabilities in terms of power consumption and memory limits. Further, the AP requires regular CSI feedback to train its model (NN 30 and/or E 31 and/or D 32) and/or to figure out if a renewed training is needed. A second shortcoming is that the known solutions require online training of the AE. The online training requires significant resources at the AP in terms of memory and processing power. This may often not be available, especially in consumer grade or battery powered products. Further, the compression cannot be readily used since a significant amount of training data needs to be captured before the training and model sharing can be done. A third shortcoming is that the known solutions have a lack of universality. Feedback of the angles i , (p cannot reconstruct CSI; hence, the feedback can only be used for the underlying beamforming design criterion. Further, predistortion of i , (p may be needed to fulfill the requirements by AE training. This predistortion may not be bijective and reconstruction may be ambiguous. Still further, just one PHY (physical layer) configuration (e.g. number of carriers, number of receive antennas) is supported, i.e. , an AE needs to be newly trained for each configuration. Also, serval types of STAs with different capabilities need to be supported; hence, each STA type set requires its own AE.

[0016] Fig. 3 shows a flowchart of a first communication method 100 carried out by a first communication device (e.g. STA 1) according to the present disclosure. In a first step 101 a data unit transmitted by the second communication device (e.g. AP 2) (or transmitted by a third communication device, e.g., a communication device of a different basic service set (BSS) such as an overlapping BSS (OBSS); this may be another STA that is associated with another AP in an OBSS) is received. In a second step 102 CSI is determined based on the received data unit. In a third step 103 the determined CSI is encoded using an encoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data. In a fourth step 104 CSI feedback comprising the encoded CSI is transmitted to the AP 2.

[0017] Fig. 4 shows a flowchart of a second communication method 200 carried out by a second communication device (e.g. AP 2) according to the present disclosure. In a first step 201 a data unit is transmitted to the STA 1. In a second step 202 CSI feedback comprising encoded CSI is received from the STA 1. In a third step 203 the encoded CSI is decoded using a decoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data, the decoder corresponding to an encoder of the trained algorithm or computing system that has been used for encoding the CSI.

[0018] In the following, details of the present disclosure will be explained by referring to a neural network (NN) as an exemplary embodiment of a trained algorithm or computer system. It shall be understood that in the following embodiments, generally other trained algorithms or computer systems may be used instead of the NN, such as an autoencoder, a learning system, a convolutional neural network or a U-net-like network. In an exemplary implementation, a neural network of an autoencoder may comprise at least two layers, wherein each layer may include affine layers, convolutional layers, or different layers as well as an activation function such as ReLLI (rectifier linear unit).

[0019] The NN of the AE may be predefined or pretrained and resides at the STA side. The shortcomings explained above may be avoided or overcome as will become apparent from the following description. In an embodiment the first communication device 1 features a NN that can be split into an encoder and a decoder, wherein the NN performs CSI compression (by the encoder) and decompression (by the decoder). The NN is pretrained. In this embodiment, the second communication device 2 features a memory to store a received NN and a processing unit to execute the NN.

[0020] Fig. 5 shows an embodiment of a trained algorithm or computer system that may be used in embodiments of the present disclosure, in this exemplary embodiment in the form of a neural network (NN) 30 representing an autoencoder (AE) with an encoder (E) 31 and a corresponding decoder (D) 32. At the input of the encoder 31 , CSI 33, which is estimated by the receiver, has a size of N_ltl (e.g. 64) real values. It is encoded to N_etlc (e.g. 32) real values denoted as E(CSI) 34. The decoder 32 decodes E(CSI) 34 to D(E(CSI)) 35 including N_ltl real values. In an embodiment it is the objective of the training of the NN 30 to minimize the reconstruction error which is e =

- CSI |. In this application, E(CSI) 34 is quantized and transmitted wirelessly as CSI feedback. Thereby, N_etlc < N_ltl

N / holds such that a compression of ^enc _N ■ 100% is achieved (e.g. 50%).

' * ’ in [0021] It is noted that CSI is typically complex-valued; hence, the CSI is decomposed into a real and an imaginary part, e.g., before it is processed by the encoder 31. The decomposition is inversed in a bijective way at decoder output. Similarly, the CSI 33 could also be decomposed in magnitude and phase, but the phase may be difficult to process by the NN 30 due to the mod 2TT ambiguity and non-Gaussian probability density function. Furthermore, the CSI 33 may have a matrix shape in MIMO communication systems, i.e. , multiple channels between transmitter (e.g. AP 2) and receiver (e.g. STA 1) may exist. In this context, a transform into scalar values may be used before the CSI is processed by the encoder 31. For example, if the CSI matrix of a 2x2 MIMO system is

transform into scalar values may be c₁₁, c₂i, c₁₂, c₂₂. Decomposition of real and imaginary parts and transform to scalar values are independent of each other, and both may be done in arbitrary order, as long as both are inversed in a bijective way at decoder output.

[0022] Fig. 6 shows a first embodiment of communication devices according to the present disclosure. In this exemplary embodiment the first communication device 1 is a non-AP STA (briefly referred to as STA 1 in the following), which holds a pretrained NN 30 having an encoder 31 and a decoder 32, and the second communication device 2 is an AP STA (briefly referred to as AP 2 in the following). In an initial step 40 the STA 1 can send its decoder 32 to the AP 2. Once the AP 2 received the decoder 32, it stores it in its memory for later processing. Alternatively, the decoder 32 (preferably the complete NN 30) may be agreed in advance and may be obtained by the AP 2 from a server or database. For instance, a system indication indicating the encoder used for encoding by the STA 1 or indicating a corresponding decoder to be used by the AP 2 for decoding or indicating the NN from which the encoder has been used, may be transmitted from the STA 1 to the AP 2 so that the AP 2 can obtain the correct decoder. It shall be noted that step 40 is preferably done rarely in comparison to other steps, i.e., the other steps 41-44 are performed consecutively more often. Since the NN 30 is pretrained and may not change, step 40 may even be just done once during setup of the communication link between the STA 1 and the AP 2, for example during association of the non-AP STA to the AP STA.

[0023] In step 41 the AP 2 transmits a PPDll (generally referred to as data unit), e.g. a channel sounding data unit such as an NDP, to the STA 1. It shall be understood that the PPDll 41 can be any sort of PPDll; hence, a dedicated sounding PPDll (null data packet NDP) or a regular PPDll (e.g. a PPDU holding MAC layer data units, control data unit, or management data units) may be used for this purpose. For simplicity, all these PPDU types are referred to as NDP, i.e., the transmitted data unit will be referred to as NDP, but other data units may be used for the same purpose.

[0024] The STA 1 retrieves CSI from the NPD in step 42. The CSI is encoded by the STA 1 using the encoder 31 of the pretrained NN 30 and transmitted to the AP 2 in step 43 as encoded CSI feedback E(CSI). The AP 2 receives the encoded CSI feedback and decodes it with the decoder 32 (received in step 40 or otherwise obtained) of the NN 30 to retrieve the CSI in step 44. Optionally, the STA 1 may update the decoder D and transmit the updated decoder to the AP 2 (not shown in Fig. 6, but it substantially corresponds to step 40), and the AP 2 may store the updated decoder (e.g. overwrite the previously stored decoder D) for later use.

[0025] The pretrained NN 30 should have one or more of the following requirements that should be fulfilled in order to compress and decompress properly. Essentially, there are two types of requirements: i) The NN should see similar CSI which it has been seen during training. During training assumptions on the training data are imposed from which the measured CSI data should not deviate too much to achieve a reasonable reconstruction loss. ii) The NN features fixed input and output dimensions N_in as well as the latent dimension N_enc. These dimensions should be identical during training and during inference. As shown in Fig. 5, N_in depends on properties of the CSI, i.e., on the number of subcarriers, the number of receive antennas of the first communication device, and the number of transmit antennas of the second communication device. The number of subcarriers is a function of the bandwidth, guard bands, and number of pilots that is used by a PPDU sent from second communication device. For example, if 242 out of 256 subcarriers should be compressed for a SISO system, a N_in of N_in = 2 • 242 = 484 results, because real and imaginary part are separated at the encoder input for each of the 242 subcarriers. If a similar exemplary system would be a 2x2 MIMO scheme, N_in = 4 - 2 • 242 = 1936 results. [0026] In an embodiment the STA 1 indicates allowed settings for the AP 2 to respect during transmission of a channel sounding packet (e.g. NDP). Those channel sounding settings only apply for channel sounding packets and not for regular data communication and may include one or more of the allowed spatial mapping matrix during transmission of an NDP, the cyclic shift delay (CSD) values during transmission of an NDP, the number of transmit antennas that can be active during transmission of a NDP, the allowed bandwidth of the NDP, and the allowed PPDll format (e.g. a PPDll according to IEEE 802.11ac or IEEE 802.11ax). The STA 1 the receives a NDP from the AP 2 with the allowed channel sounding settings and determines CSI out of this NDP as shown in Fig. 6.

[0027] Further, in an embodiment the STA 1 indicates allowed settings for the AP 2 to respect during transmission of a channel sounding announcement (e.g. NDPA). Those announcement setting may include one or more of the supported feedback bandwidths (which can be less than the allowed bandwidth of the NDP), supported compression ratios, and a recommended quantization of compressed CSI feedback (which can be recommended because quantization can be done afterwards without significant impact to compression performance).

[0028] It shall be noted that the allowed channel sounding settings and/or the allowed announcement settings are determined during training of the pretrained NN and may depend on the vendor or user of the device.

[0029] One or more of the configurations (i.e. configurations in accordance with the allowed settings) can be considered during the training of the NN. This means that non-unique configurations may be possible if the training of the NN considered those and if those configurations do not change N_in and N_enc. For example, support of multiple spatial mapping matrices and/or support of a range of transmit antennas is feasible. Further, configurations may depend on each other. For example, the supported spatial mapping matrices may depend on the number of transmit antennas.

[0030] If settings are provided by the STA 1 , the AP 2 should respect the allowed channel sounding settings when transmitting an NDP and the allowed announcement settings when transmitting an NDPA, i.e., the AP 2 transmits an NDPA and/or NDP with appropriate configurations. If multiple decoders supporting different configurations have been received by the AP 2 from the STA 1, then the NDPA or NDP may contain an identifier of the encoder or decoder or NN, e.g., a system indication mentioned above that indicates the encoder, the decoder and/or the NN.

[0031] It shall be noted that a configuration is generally related to the settings information (also called requirements list), i.e., the settings information carries information of the configuration, except for the number of receive antennas of the STA 1 which has no impact to channel sounding settings. All transmitter related parameters, e.g. the number of transmit antennas have an impact to channel sounding settings. For instance, if there are two configurations (i) for one or two transmit antennas, and (ii) for three or four transmit antennas, then the setting should be followed.

[0032] Fig. 7 shows a second embodiment of communication devices according to the present disclosure that illustrates the behavior described above in which the STA 1 has two NNs (NNA, NNB) 30A, 30B available. Each NN 30A, 30B comes with a requirement list 36A, 36B including allowed channel sounding settings for NDP and/or allowed announcement settings for NDPA, as described above. The STA 1 decides to initially transmit the decoder DA 32A of the NNA 30A to the AP1 in step 50. The requirements list 36A, that the NDP and/or NDPA needs to fulfill, is signaled as part of it or in a separate signaling beforehand or afterwards. If the AP 2 transmits a NDP or NDPA (step 51) to the STA 1, it respects the requirements described by the received requirement list 36A.

[0033] In general multiple decoders may be transmitted by the STA 1 to the AP 2, which increases the degrees of freedom that a NDPA or NDP needs to respect. In such a case, the STA 1 selects the appropriate encoder based on the NDPA settings (the announcement settings) or NDP transmission parameters (the channel sounding settings), respectively. The encoded CSI information (step 43 in Fig. 6) transmitted by the STA 1 may hold an identifier (e.g., a system indication) of the encoder or of the sounding phase such that the STA 2 can readily apply the appropriate decoder. [0034] The STA 1 may hold multiple pretrained NNs, which support different configuration settings. For example, a NN for IEEE 802.11ax and IEEE 802.11ac carrier spacing can be pretrained or multiple compression ratios can be supported. Therefore, the configurations may depend on the signaled decoder of the NN. Fig. 8 shows a corresponding embodiment of communication devices according to the present disclosure depicting two options for selecting an appropriate NN from a number of NNs (NNA 30A and NNB 30B, each having its requirements list 36A and 36B, respectively).

[0035] According to a first option the STA 1 may select the appropriate NN by receiving (step 60) information on supported communications standard and/or capabilities of the AP 2 and determining (step 61) the appropriate NN and its requirements list based on the supported communications standard and/or capabilities of the AP 2. According to a second option the STA 1 may select the appropriate NN by advertising (step 62) (i.e. transmitting) all different configuration settings to the AP 2 and receiving (step 63) a selection request from the AP 2.

[0036] The STA 1 should support various PHY configurations, meaning that the NN needs to support various configurations. Various NNs could thus be pretrained, which may increase memory requirements and signaling overhead when transferring the decoder of the NN to the AP 2. To avoid this, in an embodiment multiple NNs are held available to support different configurations. The NN that fits to the current transmission configuration of the AP 2 may then be selected. If there is no NN meeting the supported criterions, the proposed mechanism may not be used, or it may be considered to change the transmission configuration.

[0037] Another embodiment works with one NN that supports a basic configuration. Fig. 9 shows a diagram of an embodiment of an encoder 70 for a basic configuration. A basic configuration can be seen as the greatest common divisor of all or most configurations. In this case, the STA 1 applies the encoder of this NN one or more times independently and such that a configuration is met. A useful basic configuration in WLAN system may e.g. be a NN supporting 20 MHz (e.g. 242 subcarriers) and one receive antenna. If the STA 1 would like to vary the number of receive antennas, for example, it applies the same encoder separately for each receive antenna. The same may be applied for 40 MHz and two receive antennas 71 , 72, where the same encoder 73 is independently applied four times as shown in Fig. 9.

[0038] The embodiment depicted in Fig. 9 results in lower compression as correlations among the basic configurations are not exploited. For example, a strong correlation among the receive antennas cannot be exploited. A combination of both embodiments is possible as well, in which multiple NNs for a few basic configurations are provided. It shall be noted that the number of antennas and the supported bandwidth by the encoders 73 is exemplary; other numbers and bandwidths may be used as well.

[0039] Generally, the CSI feedback may comprise multiple encoded CSI partitions (or parts or blocks) as illustrated in Fig. 10, in particular if the embodiment depicted in Fig. 9 is applied. In this exemplary embodiment the CSI feedback 80 comprises CSI partition information 81 and one or more (in this example four) encoded CSI partitions p; (only three partitions 82, 83, 84 are shown). The CSI partition information contains control information which indicates to the AP 2 how to reassemble the encoded CSI information blocks. Each encoded CSI partition, indicated by E2OMHZ,IRX(CSI pi) with i = 1, ... , 4, contains the encoded CSI of a basic configuration on a CSI partition indicated by CSI pi.

[0040] In case one or more basic configuration NNs are used, it may still happen that CSI data is not available for all input dimensions. This may e.g. happen in the context of pilots, which may change their position depending on the considered bandwidth portion. Further, due to the nature of pilot implementation in WLAN, CSI may not be determined for those positions. In such a case, CSI values may be interpolated (e.g. linearly) to fill the gaps in the input data dimensions.

[0041] Typically, it is sufficient to quantize E(CSI) after training, i.e. , consideration of quantization during NN training is not necessarily required because it does not provide much performance gain. Thus, a recommendation by the STA 1 is sufficient and the AP 2 can select the desired quantization in an NDPA if needed. If for very severe compression the NN considers quantization during pretraining, the quantization parameter can be set as a requirement similar to the number of transmit antennas, for example. [0042] A NN generally comprises one or more layers. Each layer can have a different type and may be followed by a non-linearity of a different type. One layer can be identified by layer type (e.g. affine, convolution), (quantized) weight matrix, non-linearity type, and input and output dimensions. If multiple layers are connected, the order matters. Furthermore, input and output dimensions of the connected layer should match. Each NN provided by the STA 1 has properties in terms of NN type, NN structure, N_ltl, and N_etlc. At the same time, the AP 2 has capabilities for these parameters due to its memory and computation performance. If the properties of the parameters by the STA 1 do not exceed the capabilities of the AP 2, the disclosed concept may be used. Otherwise, regular CSI compression, e.g. as known in the art, may be applied.

[0043] For the training of the trained algorithm or computing system, e.g. the trained NN, a training data set can e.g. be created by an indoor channel model simulation. The CSI samples of the channel model should reflect one or more of a different number of transmit antennas, different channel model types (at least most complex channel model), addition noise with different signal to noise ratios, modification by cyclic shift delays according to WLAN standard, and modification by different (supported) spatial mapping matrices. Different supported configurations of the STA 1 can be considered as separate NN or by a basic configuration as explained above.

[0044] Although parameters of the NDP match the requirements by the NN, it may still happen that compression with AE is not appropriate. This may happen in case the estimated CSI is significantly different to what has been used for training. Therefore, it may make sense for the STA 1 to check, if compression by NN is feasible. Fig. 11 illustrates the envisioned setup of a corresponding embodiment, wherein Fig. 11 A shows an embodiment of first and second communication devices using a decision mechanism for deciding upon the CSI feedback type and Fig. 11B shows a flowchart of a corresponding method of a decision mechanism.

[0045] After transmission of the decoder 32 to the AP 2 in step 90 and reception of a NDP in step 91, the STA 1 estimates CSI in step 92. Subsequently, it computes the following error criterion (step 93) by encoding the CSI as described above and applies the decoder to the encoded CSI: e = \CSI - D(E(CS )\

The error e is a measure of the reconstruction loss by the NN compression. If it is below a predefined and tolerable threshold T (as checked in step 94), NN compression may be applied by transmitting E(CSI) to the AP 2 (step 95). Otherwise, regular compression by transmitting Q(CSI) to the AP 2 may be appropriate (step 96)). The threshold T may be defined as a fixed value or dynamically based on the actual CSI, e.g. as T = CSI - Q(CSi In the latter case, the best possible reconstruction can be achieved.

[0046] Some first communication devices (STAs) may have sufficient computational resources available such that they can perform training of the NN. Similarly to the approach to decide if NN compression is suitable, the error e may be also used to decide if a retraining of the NN may be appropriate or not.

[0047] Fig. 12 illustrates the procedure, wherein Fig. 12A shows an embodiment of first and second communication devices using a decision mechanism for deciding upon update training and Fig. 12B shows a flowchart of a corresponding method of such a decision mechanism. Steps 90-94 are identical as illustrated above with respect to Fig. 11. If e is above a threshold T, a retraining of the NN makes sense. The retraining should include the CSI values which caused e to exceed T. If e is below T, no retraining is needed (step 97). In case of a retraining (step 98), the NN needs to be updated, resulting in E_new and Dnew, and a new decoder D_new is to be transmitted (step 99) to the AP 2.

[0048] Fig. 13 shows a fourth embodiment of communication devices according to the present disclosure. This embodiment is similar to the embodiment shown in Fig. 6, but in step 41 ’ the STA 1 receives an NDP from a third communication device, in this example station STA 3 of an OBSS and then (step 42) determines the CSI of the channel between the oBSS STA 3 and itself based on the received NDP. The settings of the encoder/decoder (system indication) are in this case specifically indicated by the STA 2 as the AP 2 cannot know with which parameters the NDP from the oBSS STA 3 has been sent. Further, it may happen that there is no NN or encoder or decoder available that fits to the NDP sent by the oBSS STA 3. In this case regular feedback may be used. The oBSS STA 3 for which the STA 1 performs the CSI compression and feedback may be indicated by the AP 2.

[0049] Thus, the foregoing discussion discloses and describes merely exemplary embodiments of the present disclosure. As will be understood by those skilled in the art, the present disclosure may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. Accordingly, the disclosure of the present disclosure is intended to be illustrative, but not limiting of the scope of the disclosure, as well as other claims. The disclosure, including any readily discernible variants of the teachings herein, defines, in part, the scope of the foregoing claim terminology such that no inventive subject matter is dedicated to the public.

[0050] In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single element or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0051] In so far as embodiments of the disclosure have been described as being implemented, at least in part, by software-controlled data processing apparatus, it will be appreciated that a non-transitory machine-readable medium carrying such software, such as an optical disk, a magnetic disk, semiconductor memory or the like, is also considered to represent an embodiment of the present disclosure. Further, such a software may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

[0052] The elements of the disclosed devices, apparatus and systems may be implemented by corresponding hardware and/or software elements, for instance appropriate circuits or circuitry. A circuit is a structural assemblage of electronic components including conventional circuit elements, integrated circuits including application specific integrated circuits, standard integrated circuits, application specific standard products, and field programmable gate arrays. Further, a circuit includes central processing units, graphics processing units, and microprocessors which are programmed or configured according to software code. A circuit does not include pure software, although a circuit includes the abovedescribed hardware executing software. A circuit or circuitry may be implemented by a single device or unit or multiple devices or units, or chipset(s), or processor(s).

[0053] It follows a list of further embodiments of the disclosed subject matter:

1. First communication device configured to communicate with a second communication device, the first communication device comprising circuitry configured to: receive a data unit transmitted by the second communication device or a third communication device; determine channel state information (CSI) based on the received data unit; encode the determined CSI using an encoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data; and transmit CSI feedback comprising the encoded CSI to the second communication device.

2. First communication device according to embodiment 1 , wherein the trained algorithm or computing system comprises an encoder and decoder and is trained to minimize a reconstruction error between a plurality of CSI training data sets and reconstructed CSI data obtained by subsequent application of an encoder and the corresponding decoder to the plurality of CSI training data sets.

3. First communication device according to any preceding embodiment, wherein the CSI comprises a plurality of the CSI values for one or more channels between transmit antennas of the second or third communication device and receive antennas of the first communication device; and wherein the circuitry is configured to decompose the CSI into scalar values and/or to decompose CSI values into their real and imaginary parts or into their magnitude and phase before encoding them by the encoder.

4. First communication device according to any preceding embodiment, wherein the circuitry is configured to transmit, to the second communication device, a decoder of the trained algorithm or computing system; or a system indication indicating the encoder used for encoding by the first communication device or indicating a corresponding decoder to be used by the second communication device for decoding or indicating the trained algorithm or computing system from which the encoder has been used.

5. First communication device according to any preceding embodiment, wherein the circuitry is configured to transmit, to the second communication device, settings information comprising: channel sounding settings to be observed by the second communication device for transmitting the data unit, the channel sounding settings comprising one or more of an allowed spatial mapping matrix, cyclic shift delay values, a number or a range of numbers of active transmit antennas, an allowed bandwidth, and an allowed format of the data unit; and/or announcement settings to be observed by the second communication device for transmitting a channel sounding announcement, the announcement settings comprising one or more of a supported feedback bandwidth, one or more supported compression ratios, and a recommended quantization of the encoded CSI.

6. First communication device according to any preceding embodiment, wherein the circuitry is configured to select one of a set of two or more trained algorithms or computing systems, each being trained for a different configuration, a configuration comprising one or more of a number of subcarriers, a number of receive antennas of the first communication device, a number or a range of numbers of transmit antennas of the second communication device, bandwidth, guard bands, a number of pilot signals, carrier spacing, and compression ratio; use the encoder of the selected trained algorithm or computing system for encoding the determined CSI, and/or transmit the decoder of the selected trained algorithm or computing system or a system indication indicating the encoder or the decoder of the selected trained algorithm or computing system or indicating the selected trained algorithm or computing system.

7. First communication device according to embodiment 6, wherein the circuitry is configured to select the one trained algorithm or computing system by transmitting configuration information indicating the configuration of the two or more trained algorithms or computing systems, receiving a selection request from the second communication device indicating a desired trained algorithm or computing system, and selecting the trained algorithm or computing system based on the selection request; or determining a communications standard supported by the second communication device and/or capabilities of the second communication device, in particular by receiving corresponding capability information from the second communication device, and selecting the trained algorithm or computing system based on the determined communications standard and/or capabilities.

8. First communication device according to any preceding embodiment, wherein the circuitry is configured to partition the determined CSI into two or more CSI partitions; encode the CSI partitions separately, wherein for each CSI partition the encoder of the same or of a different trained algorithm or computing system is used; and transmit CSI feedback comprising the encoded CSI partitions and CSI partition information to the second communication device, the CSI partition information indicating how to reassemble the encoded CSI partitions.

9. First communication device according to embodiment 8, wherein circuitry is configured to control the partitioning operation such that for each CSI partition an encoder with matching configuration of a trained algorithm or computing system is available.

10. First communication device according to any preceding embodiment, wherein the circuitry is configured to compute a reconstruction error as a difference between the determined CSI and reconstructed CSI data obtained by subsequent application of an encoder and the corresponding decoder to the determined CSI; and use the reconstruction error to determine if the determined CSI shall be encoded using the encoder of the trained algorithm or computing system or using a different encoder and/or if the trained algorithm or computing system shall be updated.

11. First communication device according to embodiment 10, wherein the circuitry is configured to compare the reconstruction error to a first error threshold to determine if the determined CSI shall be encoded using the encoder of the trained algorithm or computing system or using a different encoder, in particular to determine that the determined CSI shall be encoded using the encoder of the trained algorithm or computing system if the reconstruction error is below the first error threshold; and/or compare the reconstruction error to a second error threshold to determine if the trained algorithm or computing system shall be updated, in particular to determine that the trained algorithm or computing system shall be updated if the reconstruction error is above the second error threshold.

12. First communication device according to any preceding embodiment, wherein the circuitry is configured to use a learning system, a neural network, a convolutional neural network or a U-net-like network as trained algorithm or computing system.

13. Second communication device configured to communicate with a first communication device, the second communication device comprising circuitry configured to: transmit a data unit to the first communication device; receive channel state information (CSI) feedback comprising encoded CSI from the first communication device; and decode the encoded CSI using a decoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data, the decoder corresponding to an encoder of the trained algorithm or computing system that has been used for encoding the CSI.

14. Second communication device according to embodiment 13, wherein the circuitry is configured to receive the decoder from the first communication device; or receive a decoder indication indicating the decoder from the first communication device and select the decoder from a plurality of decoder based on the received decoder information.

15. Second communication device according to any of embodiments 13 to 14, wherein the circuitry is configured to use a decoder of the trained algorithm or computing system received from the first communication device; or a decoder selected based on a system indication received from the first communication device, the system indication indicating the encoder used for encoding by the first communication device or indicating a corresponding decoder to be used by the second communication device for decoding or indicating the trained algorithm or computing system from which the encoder has been used.

16. Second communication device according to any of embodiments 13 to 15, wherein the circuitry is configured to receive, from the first communication device, settings information comprising channel sounding settings and/or announcement settings, and to transmit the data unit according to the channel sounding settings, the channel sounding settings comprising one or more of an allowed spatial mapping matrix, cyclic shift delay values, a number of active transmit antennas, an allowed bandwidth, and an allowed format of the data unit; and/or a channel sounding announcement according to the announcement settings, the announcement settings comprising one or more of a supported feedback bandwidth, one or more supported compression ratios, and a recommended quantization of the encoded CSI.

17. Second communication device according to any of embodiments 13 to 16, wherein the circuitry is configured to receive configuration information indicating the configuration of two or more trained algorithms or computing systems; and transmit a selection request to the first communication device indicating a desired trained algorithm or computing system. 18. Second communication device according to any of embodiments 13 to 17, wherein the circuitry is configured to transmit capability information to the first communication device, the capability information indicating a communications standard supported by the second communication device and/or capabilities of the second communication device.

19. Second communication device according to any of embodiments 13 to 18, wherein the circuitry is configured to receive CSI feedback comprising encoded CSI partitions of the encoded CSI and CSI partition information from the first communication device, the CSI partition information indicating how to reassemble the encoded CSI partitions; decode the CSI partitions separately, wherein for each CSI partition the decoder of the same or of a different trained algorithm or computing system is used; and reassemble the decoded CSI partitions into the encoded CSI based on the CSI partition information.

20. First communication method of a first communication device configured to communicate with a second communication device, the first communication method comprising: receiving a data unit transmitted by the second communication device or a third communication device; determining channel state information (CSI) based on the received data unit; encoding the determined CSI using an encoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data; and transmitting CSI feedback comprising the encoded CSI to the second communication device.

21. Second communication method of a second communication device configured to communicate with a first communication device, the second communication method comprising: transmitting a data unit to the first communication device; receiving channel state information (CSI) feedback comprising encoded CSI from the first communication device; and decoding the encoded CSI using a decoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data, the decoder corresponding to an encoder of the trained algorithm or computing system that has been used for encoding the CSI.

22. A non-transitory computer-readable recording medium that stores therein a computer program product, which, when executed by a processor, causes the method according to embodiment 20 or 21 to be performed.

23. A computer program comprising program code means for causing a computer to perform the steps of said method according to embodiment 20 or 21 when said computer program is carried out on a computer.

Claims

2. First communication device according to claim 1, wherein the trained algorithm or computing system comprises an encoder and decoder and is trained to minimize a reconstruction error between a plurality of CSI training data sets and reconstructed CSI data obtained by subsequent application of an encoder and the corresponding decoder to the plurality of CSI training data sets.

3. First communication device according to claim 1, wherein the CSI comprises a plurality of the CSI values for one or more channels between transmit antennas of the second or third communication device and receive antennas of the first communication device; and wherein the circuitry is configured to decompose the CSI into scalar values and/or to decompose CSI values into their real and imaginary parts or into their magnitude and phase before encoding them by the encoder.

4. First communication device according to claim 1, wherein the circuitry is configured to transmit, to the second communication device, a decoder of the trained algorithm or computing system; or a system indication indicating the encoder used for encoding by the first communication device or indicating a corresponding decoder to be used by the second communication device for decoding or indicating the trained algorithm or computing system from which the encoder has been used.

5. First communication device according to claim 1, wherein the circuitry is configured to transmit, to the second communication device, settings information comprising: channel sounding settings to be observed by the second communication device for transmitting the data unit, the channel sounding settings comprising one or more of an allowed spatial mapping matrix, cyclic shift delay values, a number or a range of numbers of active transmit antennas, an allowed bandwidth, and an allowed format of the data unit; and/or announcement settings to be observed by the second communication device for transmitting a channel sounding announcement, the announcement settings comprising one or more of a supported feedback bandwidth, one or more supported compression ratios, and a recommended quantization of the encoded CSI.

6. First communication device according to claim 1, wherein the circuitry is configured to select one of a set of two or more trained algorithms or computing systems, each being trained for a different configuration, a configuration comprising one or more of a number of subcarriers, a number of receive antennas of the first communication device, a number or a range of numbers of transmit antennas of the second communication device, bandwidth, guard bands, a number of pilot signals, carrier spacing, and compression ratio; use the encoder of the selected trained algorithm or computing system for encoding the determined CSI, and/or transmit the decoder of the selected trained algorithm or computing system or a system indication indicating the encoder or the decoder of the selected trained algorithm or computing system or indicating the selected trained algorithm or computing system.

7. First communication device according to claim 6, wherein the circuitry is configured to select the one trained algorithm or computing system by transmitting configuration information indicating the configuration of the two or more trained algorithms or computing systems, receiving a selection request from the second communication device indicating a desired trained algorithm or computing system, and selecting the trained algorithm or computing system based on the selection request; or determining a communications standard supported by the second communication device and/or capabilities of the second communication device, in particular by receiving corresponding capability information from the second communication device, and selecting the trained algorithm or computing system based on the determined communications standard and/or capabilities.

8. First communication device according to claim 1, wherein the circuitry is configured to partition the determined CSI into two or more CSI partitions; encode the CSI partitions separately, wherein for each CSI partition the encoder of the same or of a different trained algorithm or computing system is used; and transmit CSI feedback comprising the encoded CSI partitions and CSI partition information to the second communication device, the CSI partition information indicating how to reassemble the encoded CSI partitions.

9. First communication device according to claim 1, wherein the circuitry is configured to compute a reconstruction error as a difference between the determined CSI and reconstructed CSI data obtained by subsequent application of an encoder and the corresponding decoder to the determined CSI; and use the reconstruction error to determine if the determined CSI shall be encoded using the encoder of the trained algorithm or computing system or using a different encoder and/or if the trained algorithm or computing system shall be updated.

10. First communication device according to claim 9, wherein the circuitry is configured to compare the reconstruction error to a first error threshold to determine if the determined CSI shall be encoded using the encoder of the trained algorithm or computing system or using a different encoder, in particular to determine that the determined CSI shall be encoded using the encoder of the trained algorithm or computing system if the reconstruction error is below the first error threshold; and/or compare the reconstruction error to a second error threshold to determine if the trained algorithm or computing system shall be updated, in particular to determine that the trained algorithm or computing system shall be updated if the reconstruction error is above the second error threshold.

11. Second communication device configured to communicate with a first communication device, the second communication device comprising circuitry configured to: transmit a data unit to the first communication device; receive channel state information (CSI) feedback comprising encoded CSI from the first communication device; and decode the encoded CSI using a decoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data, the decoder corresponding to an encoder of the trained algorithm or computing system that has been used for encoding the CSI.

12. Second communication device according to claim 11 , wherein the circuitry is configured to receive the decoder from the first communication device; or receive a decoder indication indicating the decoder from the first communication device and select the decoder from a plurality of decoder based on the received decoder information.

13. Second communication device according to claim 11 , wherein the circuitry is configured to use a decoder of the trained algorithm or computing system received from the first communication device; or a decoder selected based on a system indication received from the first communication device, the system indication indicating the encoder used for encoding by the first communication device or indicating a corresponding decoder to be used by the second communication device for decoding or indicating the trained algorithm or computing system from which the encoder has been used.

14. Second communication device according to claim 11 , wherein the circuitry is configured to receive, from the first communication device, settings information comprising channel sounding settings and/or announcement settings, and to transmit the data unit according to the channel sounding settings, the channel sounding settings comprising one or more of an allowed spatial mapping matrix, cyclic shift delay values, a number of active transmit antennas, an allowed bandwidth, and an allowed format of the data unit; and/or a channel sounding announcement according to the announcement settings, the announcement settings comprising one or more of a supported feedback bandwidth, one or more supported compression ratios, and a recommended quantization of the encoded CSI.

15. Second communication device according to claim 11 , wherein the circuitry is configured to receive configuration information indicating the configuration of two or more trained algorithms or computing systems; and transmit a selection request to the first communication device indicating a desired trained algorithm or computing system.

16. Second communication device according to claim 11 , wherein the circuitry is configured to transmit capability information to the first communication device, the capability information indicating a communications standard supported by the second communication device and/or capabilities of the second communication device.

17. Second communication device according to claim 11 , wherein the circuitry is configured to receive CSI feedback comprising encoded CSI partitions of the encoded CSI and CSI partition information from the first communication device, the CSI partition information indicating how to reassemble the encoded CSI partitions; decode the CSI partitions separately, wherein for each CSI partition the decoder of the same or of a different trained algorithm or computing system is used; and reassemble the decoded CSI partitions into the encoded CSI based on the CSI partition information.

18. First communication method of a first communication device configured to communicate with a second communication device, the first communication method comprising: receiving a data unit transmitted by the second communication device or a third communication device; determining channel state information (CSI) based on the received data unit; encoding the determined CSI using an encoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data; and transmitting CSI feedback comprising the encoded CSI to the second communication device.

19. Second communication method of a second communication device configured to communicate with a first communication device, the second communication method comprising: transmitting a data unit to the first communication device; receiving channel state information (CSI) feedback comprising encoded CSI from the first communication device; and decoding the encoded CSI using a decoder of a trained algorithm or computing system that has been trained on a plurality of CSI training data, the decoder corresponding to an encoder of the trained algorithm or computing system that has been used for encoding the CSI.

20. A non-transitory computer-readable recording medium that stores therein a computer program product, which, when executed by a processor, causes the method according to claim 18 or 19 to be performed.