WO2025134284A1 - Receiver and wireless communication program - Google Patents

Abstract

This disclosure relates to a receiver and a wireless communication program. This receiver receives data from a transmitter, and comprises a processor and a memory in which a program to be executed by the processor is stored. The processor is configured so as to execute: processing for performing a plurality of types of learning by means of an autoencoder (AE) on the basis of a plurality of types of data for training; processing for storing a plurality of types of training results; processing for determining the compensation factor required for a received data signal by using a plurality of types of trained models generated on the basis of the training results; and processing for compensating the received data signal in accordance with the determined compensation factor. The data for training is constituted by teacher data corresponding to reception data and training data corresponding to the reception data, and the corresponding compensation factor is associated with the data for training.

Description

Receiver and wireless communication program

This disclosure relates to a receiver and a wireless communication program.

In order to improve communication quality, there is a technology that outputs the original signal based on the signal that has changed during the communication process. For example, Non-Patent Document 1 discloses a method of outputting the original signal based on a pilot signal that has been affected by amplifier distortion and a multiple regression model obtained by pre-learning. In this method, the input/output characteristics of the amplifier are derived, and a reference signal to be used for demodulation is generated based on the derived characteristics, thereby improving communication quality.

T. Tanaka, K. Kuriyama, H. Hasegawa, and M. Miyagi, “An estimation method of input/output characteristics of a transmitter amplifier using multiple regression analysis,” IEICE General Conference B-5-81, 2023.

However, the above method uses pilot signals, which increases the amount of shared information and reduces communication speed. Furthermore, there is an issue that compensation performance can be significantly degraded if the training data used in pre-learning and the received signal differ significantly due to changes in the type or performance of the wireless equipment used in communication.

In order to solve the above problems, the primary objective of this disclosure is to provide a receiver that can provide appropriate compensation and improve communication speed even when the type or performance of the wireless device used in communication changes.

A second objective of this disclosure is to provide a wireless communication program that can provide appropriate compensation and improve communication speeds even when the type or performance of the wireless device used in communication changes.

The first aspect of the present disclosure is a receiver that receives data from a transmitter, comprising a processor and a memory that stores a program to be executed by the processor, and configured to execute a process of performing multiple types of learning using an autoencoder (AE) based on multiple types of learning data, a process of storing the multiple types of learning results, a process of determining a compensation factor required by a received data signal using multiple types of learning models generated based on the learning results, and a process of compensating the received data signal according to the determined compensation factor, and the learning data is preferably teacher data corresponding to the received data and learning data corresponding to the received data, and the receiver is linked to the corresponding compensation factor.

A second aspect of the present disclosure is a wireless communication program to be executed by a receiver having a processor and a memory, the program being stored in the memory and computer-readable, and including a program for causing the processor to execute a process of performing multiple types of learning using an autoencoder (AE) based on multiple types of learning data, a process of storing the multiple types of learning results, a process of determining the compensation factor required by the received data signal using multiple types of learning models generated based on the learning results, and a process of compensating the received data signal according to the determined compensation factor, and the learning data is preferably teacher data corresponding to the received data and learning data corresponding to the received data, and the wireless communication program is linked to the corresponding compensation factor.

The first and second aspects of the present disclosure make it possible to provide appropriate compensation and improve communication speed even when the type or performance of the wireless device used in communication changes.

FIG. 1 is a diagram illustrating a configuration example of a wireless communication system according to a first embodiment of the present disclosure. FIG. 2 is a diagram illustrating a hardware configuration of a receiver according to a first embodiment of the present disclosure. FIG. 11 is a diagram illustrating a pre-processing according to the first embodiment of the present disclosure. FIG. 2 is a diagram showing a process during communication according to the first embodiment of the present disclosure.

First embodiment
1 is a diagram illustrating a configuration example of a wireless communication system according to a first embodiment of the present disclosure. The wireless communication system 100 includes a transmitter 20. The transmitter 20 transmits a signal to a receiver 40.

The transmitter 20 has an information generating unit 21. The information generating unit 21 generates information bits for the information to be transmitted to the receiver 40. The information generating unit 21 may also have a function of adding an error correction code or an interleaving function.

The generated information bits are transmitted to the data signal modulation unit 22. The data signal modulation unit 22 modulates the information bits into a data signal. The modulation method used here is, for example, quadrature amplitude modulation (QAM).

The data signal obtained by the modulation is transmitted to the D/A conversion unit 23. The D/A conversion unit 23 converts the digitally modulated data signal into an analog signal. At this time, an I signal and a Q signal are generated.

The analog data signal is transmitted to the IQ signal modulation unit 24. The IQ signal modulation unit 24 performs quadrature modulation on the I and Q signals.

The quadrature modulated data signal is transmitted to the amplifier 25. The amplifier 25 amplifies the data signal and transmits it to the receiver 40.

The receiver 40 includes an amplifier unit 41. The amplifier unit 41 amplifies the data signal received from the transmitter 20.

The amplified data signal is transmitted to the IQ signal demodulation unit 42. The IQ signal demodulation unit 42 demodulates the received data signal into an I signal and a Q signal.

The demodulated data signal is transmitted to the A/D conversion unit 43. The A/D conversion unit 43 digitizes the analog data signal for digital demodulation.

The digitized data signal is transmitted to the channel equalization unit 44. The channel equalization unit 44 obtains an estimate of the originally transmitted signal by inversely calculating the amplitude and phase information of the channel response based on the data signal.

The data signal from which the estimated value has been obtained is transmitted to the determination unit 45. The determination unit 45 generates multiple learning models based on the learning results of multiple AEs (Autoencoders) obtained in pre-processing. The determination unit 45 then uses the multiple learning models to determine the compensation factor required by the received data signal.

Compensation factors are factors that cause imperfections in the RF circuit, for example, and specifically, nonlinear distortion of the amplifier, IQ imbalance, phase noise, and combinations of these. Furthermore, compensation factors may include "no compensation required" which is used when compensation is not required. Details of the pre-processing and determination method will be described later.

The data signal is transmitted to compensation unit 46a or 46b depending on the determined compensation factor. Compensation units 46a and 46b compensate the received data signal. Compensation units 46a and 46b differ in compensation target and calculation capability. Also, while an embodiment in which there are two types of compensation units, compensation units 46a and 46b, is shown here, there may be three or more types.

The compensated data signal is transmitted to the information detection unit 47. The information detection unit 47 detects information bits from the data signal. Depending on the function of the information generation unit 21, the information detection unit 47 may also have a function of decoding error correction codes or a deinterleaving function.

Next, the pre-processing will be described. The wireless communication system 100 includes a pre-processing unit 60. The pre-processing unit 60 may be included in each receiver 40. Alternatively, the pre-processing unit 60 may be connected to multiple receivers 40 to perform pre-processing collectively.

The pre-processing unit 60 has a learning data generation unit 61. The learning data generation unit 61 generates learning data to be used in AE. Here, multiple types of learning data are generated depending on the compensation factor.

The generated multiple types of learning data are transmitted to the AE learning unit 62. The AE learning unit 62 performs multiple learning operations using the AE based on the multiple types of learning data.

The multiple learning results are transmitted to the learning result storage unit 48 of the receiver 40. The learning result storage unit 48 stores the multiple types of learning results obtained by the pre-processing unit 60. The aforementioned determination unit 45 generates multiple types of learning models based on the multiple learning results.

FIG. 2 is a diagram showing the hardware configuration of a receiver according to the first embodiment of the present disclosure. Each function of the receiver 40 may be configured in whole or in part by hardware such as a PLD (Programmable Logic Device) or an FPGA (Field Programmable Gate Array), or may be configured as a program executed by a processor such as a CPU.

For example, the receiver 40 can be realized using a computer and a program, and the program can be recorded on a storage medium or provided over a network.

As shown in FIG. 2, the receiver 40 has an input unit 400, an output unit 401, a communication unit 402, a CPU 403, a memory 404, and a HDD 405 connected via a bus 406, and functions as a computer. The receiver 40 is also capable of inputting and outputting data to and from a computer-readable storage medium 407.

The input unit 400 is, for example, a keyboard and a mouse. The output unit 401 is, for example, a display device such as a display.

The communication unit 402 is, for example, a communication interface that communicates with the wireless device to be controlled.

The CPU 403 controls each component of the receiver 40 and performs predetermined processing. The memory 404 and HDD 405 store data, etc.

The storage medium 407 is capable of storing programs and the like that cause the receiver 40 to execute the functions of the receiver 40. Note that the architecture that constitutes the receiver 40 is not limited to the example shown in FIG. 2.

FIG. 3 is a diagram showing pre-processing according to the first embodiment of the present disclosure. First, the learning data generation unit 61 generates learning data. The learning data is, for example, a pair of learning data 2 and teacher data 4. The learning data 2 corresponds to received data, and is, for example, a complex signal that has changed due to a specific compensation factor. The teacher data 4 corresponds to received data, and is, for example, a complex signal that has changed due to the same compensation factor as the learning data 2. In other words, the learning data 2 and teacher data 4 in this disclosure are the same signal.

Furthermore, the learning data generation unit 61 links the corresponding compensation factors to the learning data. As described above, the learning data generation unit 61 generates multiple types of learning data that are each linked to multiple types of compensation factors.

Next, the AE learning unit 62 performs multiple types of learning by the AE based on multiple types of learning data. First, the AE learning unit 62 performs a first learning by a learning network using the AE 10 based on the learning data, and generates a learning model. This learning network is, for example, a network with a machine learning layer configuration. This learning model makes it possible to extract feature quantities 16 that arise between the encoder process 12 and the decoder process 14 when the AE 10 outputs teacher data 4 from the learning data 2.

Then, the AE learning unit 62 performs a second learning process using the extracted feature quantities 16. Specifically, learning is performed using the extracted feature quantities 16 as learning data, and compensation factors and compensation programs corresponding to the corresponding compensation factors as teacher data. This learning is performed for multiple types of compensation factors, so multiple types of learning results are obtained.

The multiple types of learning results are transmitted to the learning result storage unit 48. The learning result storage unit 48 stores the multiple types of learning results. The judgment unit 45 generates multiple types of learning models based on the multiple types of learning results acquired from the learning result storage unit 48.

This learning model is a model that generates the same signal from a signal that has changed due to the associated compensation factor. This learning model is generated for each anticipated compensation factor.

In this embodiment, AE is used to generate the learning model. When imperfections in multiple RF circuits affect a signal, the effects of each are superimposed, resulting in complex changes in the signal. For this reason, even when machine learning is used for compensation, it becomes necessary to extract fine features. Therefore, in this embodiment, an autoencoder is used, which is less susceptible to overlearning and is suitable for extracting fine features.

In this embodiment, AE is used to determine the compensation factor required for a received data signal. For example, consider a method of determining whether an input signal is equal to any of the known signals before degradation. This method requires storing a large number of signals in order to prepare the signal points required for the determination. On the other hand, in this embodiment, information in the form of restoration results can be obtained using the AE learning model. Therefore, it is possible to determine whether compensation is required without having to store a large number of signals.

FIG. 4 is a diagram showing the process during communication according to the first embodiment of the present disclosure. During communication, first, information bits generated by the information generating unit 21 of the transmitter 20 are transmitted as a data signal 6. Before reaching the determining unit 45, the data signal 6 changes due to the influence of imperfections in the RF circuit. Therefore, what the determining unit 45 receives is the changed data signal 8.

Then, the determination unit 45 converts the data signal 8 using a learning model generated in advance. Specifically, the determination unit 45 inputs the received data signal 8 to an AE based on multiple types of learning models generated in advance.

First, the determination unit 45 inputs the received data signal 8 to an AE based on a learning model linked to the compensation factor to be determined. The learning model used here is a learning model that generates the same signal from a signal that has changed due to the linked compensation factor.

Therefore, if the input data signal is a signal that has changed due to the associated compensation factor, the characteristics required for restoration will be the same. As a result, a data signal identical to the input data signal will be output. On the other hand, if the input data signal is not a signal that has changed due to the associated compensation factor, the characteristics required for restoration will be different. As a result, a data signal different from the input data signal will be output.

Then, if the output signal is the same as the input signal, the determination unit 45 determines that the compensation factor linked to the learning model used for the conversion is the compensation factor required by the received data signal.

Similarly, if the output signal differs from the input signal, the determination unit 45 determines that the compensation factor linked to the learning model used for the conversion is not the compensation factor required by the received data signal. The determination unit 45 then inputs the received data signal 8 to an AE based on the learning model linked to the undetermined compensation factor. By repeating this process, the determination unit 45 can determine the compensation factor required by the received data signal 8 from among the possible compensation factors.

Then, the determination unit 45 transmits the data signal to the compensation unit according to the determined compensation factor. Here, three types of compensation units, compensation units 46a, 46b, and 46c, are shown. The compensation units 46a, 46b, and 46c compensate for the received data signal. The compensated data signal is transmitted to the information detection unit 47. The information detection unit 47 detects information bits from the data signal.

As described above, with this disclosure, the required compensation factor can be determined from the received data signal, eliminating the need to use a pilot signal and improving communication speed. In addition, because the compensation factor is determined in advance, only the necessary compensation can be performed, reducing processing. In other words, it is possible to reduce power consumption.

In this embodiment, the effect of imperfections in the RF circuit has been described as an example, but this is effective for all causes of signal changes, such as fading.

The learning model shown in this embodiment is just an example, and the network configuration or learning data is not limited. Furthermore, the configuration of the wireless communication system is not limited.

2 Learning data 4 Teacher data 6 Data signal 8 Data signal 20 Transmitter 40 Receiver

Claims (4)

A receiver for receiving data from a transmitter, comprising:
A processor and a memory storing a program to be executed by the processor,
The processor,
A process of performing multiple types of learning by an autoencoder (AE) based on multiple types of learning data;
A process of storing a plurality of types of learning results;
A process of determining a compensation factor required for a received data signal using a plurality of types of learning models generated based on the learning results;
and compensating the received data signal in response to the determined compensation factor;
The learning data is
The received data is training data corresponding to the received data,
The receiver to which the corresponding compensation factor is associated. The receiver according to claim 1 , wherein the compensation factor includes “no compensation required” that is used when no compensation is required. the compensation factor is a factor causing imperfections in the RF circuit;
3. The receiver according to claim 1, wherein the imperfections in the RF circuit are at least one of a nonlinear distortion, an IQ imbalance, and a phase noise of an amplifier. A wireless communication program to be executed by a receiver having a processor and a memory,
stored in the memory,
It is computer readable;
The processor,
A process of performing multiple types of learning by an autoencoder (AE) based on multiple types of learning data;
A process of storing a plurality of types of learning results;
A process of determining a compensation factor required for a received data signal using a plurality of types of learning models generated based on the learning results;
and a process of compensating the received data signal in response to the determined compensation factor,
The learning data is
The received data is training data corresponding to the received data,
The wireless communication program to which the applicable compensation factor is associated.

Images