WO2021109768A1

WO2021109768A1 - Decoding result determining method and device, storage medium, and electronic device

Info

Publication number: WO2021109768A1
Application number: PCT/CN2020/125569
Authority: WO
Inventors: 张恩溯; 李虎虎
Original assignee: 中兴通讯股份有限公司
Priority date: 2019-12-04
Filing date: 2020-10-30
Publication date: 2021-06-10
Also published as: CN112910600A

Abstract

The present application provides a decoding result determining method and device, a storage medium, and an electronic device. The method comprises: preprocessing data to be decoded received by a multiple input multiple output (MIMO) system to obtain target data; using a target data decoding model to process the target data to obtain probability information of bits corresponding to constellation points in the MIMO system, wherein the target data decoding model is obtained by training a predetermined neural network model by using a plurality of sets of data, and each set of data in the plurality of sets of data comprising: sample data and sample bit probability information; and on the basis of the probability information of bits corresponding to constellation points in the MIMO system, determining a decoding result.

Description

Method, device, storage medium and electronic device for determining decoding result

Cross-references to related applications

This application is filed based on a Chinese patent application with an application number of 201911229147.7 and an application date of December 04, 2019, and claims the priority of the Chinese patent application. The entire content of the Chinese patent application is hereby incorporated into this application by way of introduction.

Technical field

This application relates to the field of communications, and in particular, to a method, device, storage medium, and electronic device for determining a decoding result.

Background technique

Multiple Input Multiple Output (MIMO) technology is a hot spot technology of wireless communication in recent years. It greatly improves the channel capacity and transmission reliability of the system without increasing signal transmission power and bandwidth, and becomes the first The 4th Generation mobile communication technology (4G) and the 5th Generation mobile communication technology (5G) are key technologies for mobile communication. A key technology of MIMO technology is signal detection and decoding. The complexity of detection and decoding of received signals in MIMO systems far exceeds that of single-input-output systems. Each antenna of the MIMO system has to receive signals from this antenna as well as signals from other antennas. These signals overlap in time and frequency bands. . This requires the decoding system to have high data throughput, can process large-capacity, large-bandwidth data in a short time, and have low latency. This leads to high complexity of the MIMO received signal detection and decoding system. The core of the MIMO signal detection and decoding technology is the constellation point decision technology. It can be considered that the MIMO decoding process is the constellation point decision process.

The constellation point judgment of the MIMO system is complicated because the received signal is random, and the constellation point judgment is the classification of random signals. It is difficult to classify random signals with existing decoding methods, and the algorithm complexity is high. So far, there are some MIMO constellation point decision algorithms in the industry. Although their performance is better than equidistant decision, they are mostly in the theoretical stage and the algorithms are complex. The amount of calculation increases exponentially with the number of antennas and modulation types. The algorithms are all serial decoding, it is difficult to have high data throughput. Therefore, these algorithms are almost impossible to use.

In view of the large amount of calculation and high complexity of MIMO system constellation point decision in related technologies, no effective solution has been proposed at present.

Summary of the invention

The embodiments of the present application provide a method, a device, a storage medium, and an electronic device for determining a decoding result, so as to at least solve the problem of large calculation amount and high complexity of MIMO system constellation point judgment in related technologies.

The embodiment of the present application provides a method for determining a decoding result, including: preprocessing the to-be-decoded data received by a multiple-input multiple-output MIMO system to obtain target data; and using a target data decoding model to analyze the target data Processing to obtain the probability information of the bit corresponding to the constellation point in the MIMO system, where the target data decoding model is obtained by training a predetermined neural network model using multiple sets of data, and each of the multiple sets of data Each group of data includes: sample data and sample bit probability information; the decoding result is determined based on the probability information of the bit corresponding to the constellation point in the MIMO system.

The embodiment of the application provides a device for determining a decoding result, including: a first processing module, configured to preprocess the data to be decoded received by a multiple-input multiple-output MIMO system to obtain target data; and a second processing module, It is used to process the target data using the target data decoding model to obtain the probability information of the bit corresponding to the constellation point in the MIMO system, where the target data decoding model is to use multiple sets of data to perform a predetermined neural network model After training, each group of data in the multiple groups of data includes: sample data and sample bit probability information; a determining module is used to determine a decoding result based on the probability information of the bit corresponding to the constellation point in the MIMO system.

The embodiment of the present application also provides a computer-readable storage medium in which a computer program is stored, wherein the computer program is configured to execute the steps in the foregoing method embodiment when running.

An embodiment of the present application also provides an electronic device, including a memory and a processor, the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in the foregoing method embodiment.

Description of the drawings

The drawings described here are used to provide a further understanding of the application and constitute a part of the application. The exemplary embodiments and descriptions of the application are used to explain the application, and do not constitute an improper limitation of the application. In the attached picture:

FIG. 1 is a block diagram of the hardware structure of a mobile terminal of a method for determining a decoding result according to an embodiment of the present application;

Fig. 2 is a flowchart of a method for determining a decoding result according to an embodiment of the present application;

Fig. 3 is a flowchart of training a predetermined neural network according to an embodiment of the present application;

Fig. 4 is a specific flow chart of training a predetermined neural network according to an embodiment of the present application;

Fig. 5 is a structural block diagram of a device for determining a decoding result according to an embodiment of the present application;

Fig. 6 is a flowchart of obtaining a decoding result according to an embodiment of the present application;

Fig. 7 is a structural block diagram of a device for determining constellation points based on deep learning according to an embodiment of the present application;

Fig. 8 is a structural block diagram of a judgment unit according to an embodiment of the present application.

Detailed ways

Hereinafter, the present application will be described in detail with reference to the drawings and in conjunction with the embodiments. It should be noted that the embodiments in the application and the features in the embodiments can be combined with each other if there is no conflict.

It should be noted that the terms "first" and "second" in the specification and claims of the application and the above-mentioned drawings are used to distinguish similar objects, and not necessarily used to describe a specific sequence or sequence.

Deep learning has been a hot research topic in recent years and has a wide range of applications in artificial intelligence. The reasoning process of deep learning can be regarded as the process of solving the maximum likelihood solution. Through training, the decision of random signals has a high accuracy rate. Its decision performance is equivalent to Maximum Likelihood Estimation (MLE). In other current MIMO constellation point decision algorithms; at the same time, the implementation complexity of the deep learning model is much smaller than that of MLE. MLE can hardly use hardware under 16 quadrature amplitude modulation (16 Quadrature Amplitude Modulation, referred to as 16QAM) and above modulation methods. Realization, using deep learning for the constellation point decision of the MIMO system can greatly reduce the computational complexity while obtaining decoding performance consistent with MLE.

The application will be described below in conjunction with embodiments.

The method embodiments provided in the embodiments of the present application may be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, FIG. 1 is a hardware structural block diagram of a mobile terminal of a method for determining a decoding result according to an embodiment of the present application. As shown in FIG. 1, the mobile terminal 10 may include one or more (only one is shown in FIG. 1) processor 102 (the processor 102 may include but is not limited to a microcontroller (Micro Control Unit, referred to as MCU)) or may A processing device such as a Programmable Logic Device (PLD for short) and a memory 104 for storing data. In one embodiment, the above-mentioned mobile terminal may also include a transmission device 106 for communication functions and an input and output device 108. Those of ordinary skill in the art can understand that the structure shown in FIG. 1 is only for illustration, and does not limit the structure of the above-mentioned mobile terminal. For example, the mobile terminal 10 may also include more or fewer components than those shown in FIG. 1 or have a different configuration from that shown in FIG. 1.

The memory 104 may be used to store computer programs, for example, software programs and modules of application software, such as the computer programs corresponding to the method for determining the decoding result in the embodiment of the present application. The processor 102 runs the computer programs stored in the memory 104, Thereby, various functional applications and data processing are executed, that is, the above-mentioned method is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely provided with respect to the processor 102, and these remote memories may be connected to the mobile terminal 10 through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

The transmission device 106 is used to receive or send data via a network. The above-mentioned specific example of the network may include a wireless network provided by the communication provider of the mobile terminal 10. In one example, the transmission device 106 includes a network interface controller (Network Interface Controller, NIC for short), which can be connected to other network devices through a base station to communicate with the Internet. In an example, the transmission device 106 may be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet in a wireless manner.

In this embodiment, a method for determining a decoding result is provided. FIG. 2 is a flowchart of the method for determining a decoding result according to an embodiment of the present application. As shown in FIG. 2, the flow includes the following steps.

Step S202: Preprocessing the to-be-decoded data received by the multiple-input multiple-output MIMO system to obtain target data.

Step S204: Use the target data decoding model to process the target data to obtain the probability information of the bits corresponding to the constellation points in the MIMO system, where the target data decoding model uses multiple sets of data to perform a predetermined neural network The model is obtained by training, and each of the multiple sets of data includes: sample data and sample bit probability information.

Step S206: Determine the decoding result based on the probability information of the bit corresponding to the constellation point in the MIMO system.

In the above-mentioned embodiment, deep learning is used to make MIMO system constellation point judgment. Compared with traditional judgment methods, using the solution in this application under the premise of close performance can greatly simplify the calculation amount and complexity, and effectively solve related technologies. The MIMO system constellation point decision has a large amount of calculation and high complexity.

In one embodiment, the sample data is data obtained after the preprocessing of the sample data to be decoded, and the sample bit probability information is obtained after processing the sample data using the Maximum Likelihood Estimation MLE algorithm Information. In this embodiment, the MLE algorithm is actually used to generate the expected value. It should be noted that the traditional deep learning model training needs to calibrate the label of the input value, that is, the expected value. In the MIMO constellation point judgment, the channel state is different, and the soft information of the judgment is different. If the expected value is de-calibrated, the change of the channel state cannot be reflected immediately. The output value of the deep learning model (that is, the target data decoding model mentioned above) and The difference in expected value is not accurate, and the trained model is not optimal; and deep learning model training requires a lot of input data, and label calibration is also a heavy task. In the embodiment of this application, the MLE algorithm is used to generate the expected value. One is because the expected value is generated by the MLE algorithm, which contains channel state information. This channel state information is consistent with the channel state in the deep learning model, so deep learning The comparison between the output of the model and the expected value is more accurate, and the training is closer to reality. Second, the heavy work of calibrating labels can be avoided, saving a lot of time and cost. In addition, the trained deep learning model uses Application Specific Integrated Circuit (ASIC for short), Field Programmable Gate Array (FPGA for short), and Graphics Processing Unit (FPGA for short) as the reasoning part. (Referred to as GPU) and other technologies are implemented into a device, which can be used for receivers in the communication field to determine the constellation points of the received signal. Compared with MLE, the implementation device proposed in the embodiment of the present application has fewer hardware resources, greater data throughput, and more flexible implementation.

In one embodiment, preprocessing the to-be-decoded data received by the multiple-input multiple-output MIMO system to obtain the target data includes: determining a channel matrix H, where H is an R×T channel matrix, and R is the MIMO system The number of receiving antennas, T is the number of transmitting antennas of the MIMO system; Singular Value Decomposition (SVD) is performed on the channel matrix H to obtain USV, where U is the unitary matrix and S is the diagonal matrix , V is a unitary matrix; matrix multiplication is performed on the data to be decoded and the conjugate transposed matrix U ^{H of U to obtain the target data.} The following is a detailed description of how to obtain the target data.

According to the principle of MIMO system:

Y=HX+N

Among them, X={x ₁ ,x ₂ ,...,x _T } is the symbol to be transmitted at the transmitting end, and T is the number of transmitting antennas. Y={y ₁ ,y ₂ ,...,y _R } is the symbol received by the receiving end, and R is the number of receiving antennas. H is the channel matrix of R×T. N={n ₁ ,n ₂ ,...,n _R } is noise. SVD decomposition of the channel matrix H can be obtained:

Y=USVX+N

Among them, U is a unitary matrix, S is a diagonal matrix, and V is a unitary matrix. X will be pre-processed before launch, multiplying the conjugate device matrix V ^{H of} V; then multiplying both sides of the equation by the conjugate transposed matrix U ^{H of} U to obtain:

U ^H Y＝U ^H USVV ^H X+U ^H N

U ^H Y＝SX+U ^H N

Z=SX+N′

Among them, Z=U ^H Y, N′=U ^H N, because U is a unitary matrix, the statistical characteristics of N′ are consistent with N, and it is still noise. After the channel matrix H is decomposed into the diagonal matrix S, there is no interference between the antennas, and each receiving antenna can use the same deep learning model to make a decision.

After the above-mentioned processing, the obtained Z is the above-mentioned target data.

In one embodiment, before the target data is processed using the target data decoding model to obtain the probability information of the bit corresponding to the constellation point in the MIMO system, the method further includes: acquiring each of the multiple sets of data The sample data and the sample bit probability information in the set of data; use the sample data and the sample bit probability information in each set of data in the multiple sets of data to train the predetermined neural network model, To obtain the target data decoding model.

In an embodiment, training the predetermined neural network model using the sample data and the sample bit probability information in each of the multiple sets of data to obtain the target data decoding model includes : Input the sample data into the predetermined neural network model to obtain sample output data; when it is determined that the difference between the sample output data and the sample bit probability information is greater than a predetermined threshold, use the predetermined neural network model The back propagation model of the network model repeatedly executes the process of adjusting the weight values of each layer in the predetermined neural network model until the difference between the adjusted predetermined neural network output data and the sample bit probability information is less than Or equal to the predetermined threshold; the predetermined neural network model obtained after the final adjustment is determined as the target data decoding model. The process of training the predetermined neural network in this embodiment can be seen in FIG. 3, and includes the following steps.

S302: Establish training samples, that is, establish the above-mentioned multiple sets of data.

S304: Generate a deep neural network model, where the model is an initial model, that is, a model that has not been trained, and corresponds to the aforementioned predetermined neural network model.

S306: Train the deep neural network model by using the above-mentioned training samples.

S308: Determine whether the difference between the model output result and the expected value is less than ε (the ε corresponds to the above-mentioned predetermined threshold, the ε is a small value, generally ε<0.01), if it is determined to be less than ε, go to S310, otherwise , Go to S306.

S310: The deep neural network training is completed, that is, the aforementioned target data decoding model is obtained.

The following describes how to train the neural network in conjunction with a specific embodiment.

As shown in Fig. 4, when the neural network model training is performed, the following steps are specifically included.

S402. Establish a training sample to generate a set of values Z=X+N, where X is the constellation point under different modulation modes (in this specific embodiment, X corresponds to SX in the foregoing description), and N is random noise (in this In the specific embodiment, N corresponds to N′ in the foregoing description). Input Z into the MLE algorithm for constellation point judgment to obtain the expected value X'. The MLE algorithm is the traditional optimal algorithm for MIMO constellation point judgment, but because the MLE algorithm has to traverse all possible transmission vectors, its complexity varies with the number of antennas and modulation The order increases exponentially and cannot be used for receivers in the communications field. But because of its optimal performance in MIMO constellation point judgment, it can be used to generate the expected value X'in deep learning model training.

S404: Establish a decoding model based on deep learning, and establish a forward propagation model and a back propagation model through the neural network in the deep learning.

S406: Perform training of the decoding model, and input the Z generated in step S402 into the forward propagation model established in step S404 to generate an output result.

S408: The output result of the forward propagation model is compared with the expected value X', and then the compared difference is sent to the back propagation model, and the weight parameter of the forward propagation model is adjusted through the back propagation model.

S410: When the difference between the output result of the forward propagation model and the expected value is less than ε, the training ends, and the deep neural network model is obtained.

In one embodiment, determining the decoding result based on the probability information of the bit corresponding to the constellation point in the MIMO system includes one of the following: using a decoding device to decode the probability information of the bit corresponding to the constellation point in the MIMO system to obtain The decoding result; performing a hard decision on the probability information of the bit corresponding to the constellation point in the MIMO system to obtain the decoding result. In this embodiment, the deep learning decoding model obtained in FIG. 4 (the deep neural network model can be used as the deep learning decoding model) can be used for decoding. The received signal of the MIMO receiver is used as the input signal of the decoding model, and the output is the decoding soft information, that is, the probability value of each constellation point. Input the decoded soft information into the decoder for decoding, or directly make a hard decision on the decoded soft information to obtain the decoded result.

It should be noted that the implementation of the above-mentioned MIMO decoding method may be implemented by hardware (for example, FPGA, ASIC, etc.). Of course, it can also be implemented in software (for example, CPU, MCU, etc.). Either way, the computational complexity can be significantly reduced. Among them, when the above method is implemented by hardware, it will have greater advantages in processing speed, implementation complexity, and data throughput.

Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of hardware, or by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to enable a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the method described in each embodiment of the present application.

In this embodiment, a device for determining a decoding result is also provided, which is used to implement the above-mentioned embodiment, and what has been described will not be repeated. As used below, the term "module" can implement a combination of software and/or hardware with predetermined functions. Although the devices described in the following embodiments are preferably implemented by software, implementation by hardware or a combination of software and hardware is also possible and conceived.

Fig. 5 is a structural block diagram of a device for determining a decoding result according to an embodiment of the present application. As shown in Fig. 5, the device includes:

The first processing module 52 is used to preprocess the to-be-decoded data received by the MIMO system to obtain target data; the second processing module 54 is used to process the target data using the target data decoding model , Obtain the probability information of the bit corresponding to the constellation point in the MIMO system, where the target data decoding model is obtained by using multiple sets of data to train a predetermined neural network model +, each of the multiple sets of data Both include: sample data and sample bit probability information; the determining module 56 is used to determine the decoding result based on the probability information of the bit corresponding to the constellation point in the MIMO system.

In one embodiment, the sample data is data obtained after the preprocessing of the sample data to be decoded, and the sample bit probability information is obtained after processing the sample data using the Maximum Likelihood Estimation MLE algorithm Information.

In an embodiment, the first processing module 52 is configured to preprocess the to-be-decoded data received by the MIMO system in the following manner to obtain target data: determine the channel matrix H, where H is R ×T channel matrix, R is the number of receiving antennas of the MIMO system, T is the number of transmitting antennas of the MIMO system; SVD decomposition is performed on the channel matrix H to obtain USV, where U is the unitary matrix and S is the pair Angle matrix, V is a unitary matrix; matrix multiplication is performed on the data to be decoded and the conjugate transposed matrix U ^{H of U to obtain the target data.}

In an embodiment, the device is also used to process the target data using the target data decoding model to obtain the probability information of the bit corresponding to the constellation point in the MIMO system, to obtain the data in the multiple sets of data. The sample data and the sample bit probability information in each set of data; use the sample data and the sample bit probability information in each set of data in the multiple sets of data to train the predetermined neural network model , To obtain the target data decoding model.

In an embodiment, the device may use the sample data and the sample bit probability information in each of the multiple sets of data to train the predetermined neural network model in the following manner to obtain all The target data decoding model: input the sample data into the predetermined neural network model to obtain sample output data; when it is determined that the difference between the sample output data and the sample bit probability information is greater than a predetermined threshold , Using the back propagation model of the predetermined neural network model to repeatedly execute the process of adjusting the weight value of each layer in the predetermined neural network model until the adjusted predetermined neural network output data and the sample bit probability information The difference between is less than or equal to the predetermined threshold; the predetermined neural network model obtained after the final adjustment is determined as the target data decoding model.

In an embodiment, the determining module 56 may determine the decoding result based on the probability information of the bit corresponding to the constellation point in the MIMO system in one of the following ways: use a decoding device to map the constellation point in the MIMO system to the bit Decoding the probability information of to obtain the decoding result; and performing a hard decision on the probability information of the bit corresponding to the constellation point in the MIMO system to obtain the decoding result.

It should be noted that each of the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following manner, but not limited to this: the above modules are all located in the same processor; or, the above modules can be combined in any combination. The forms are located in different processors.

The application will be described below in conjunction with specific embodiments. In this embodiment, the following steps are included.

S11. Establishing a training sample set is to establish a set Z _M = X _J + N _M , where M is the number of samples and J is the number of constellation points. In order to train a model that meets the requirements, the value of M is very large. _{The constellation point X J} to be transmitted by the transmitter is selected from the constellation points of a certain modulation mode, and it is necessary to traverse all the constellation points under the modulation mode. The noise _NM is a random quantity, and a noise model that meets certain statistical characteristics, such as Gaussian white noise, can be selected.

S12: Establish a deep neural network model, the input port is 1, the input is the symbol z to be decoded, and the output port is the probability information of each bit, which is sent to the decoding module as soft information. According to different modulation methods, the number of output ports is different. For example, in the debugging mode of 16QAM, there are 4 outputs, which represent the probability information of each bit.

S13, the model training, a training sample set _{_{Z M = {z 0, z}} 1, ..., z M} to the MLE algorithm unit, an output expectation value _{_{X 'M = {x' 0}} , x 1 ', ..., x' _M }; At the same time, _{input the set Z M to} the deep neural network model established in step 1, and the model will output a series of corresponding symbols {o ₁ ,o ₂ ,…,o _M } to solve {o ₁ ,o ₂ ,…, o _M } and the expected value {x' ₀ ,x ₁ ',…,x' _M }, establish a back propagation model based on the difference, and modify the weight value layer by layer through the back propagation model. When {o ₁ , o ₂ ,…,o _M } and the expected value {x' ₀ ,x ₁ ',…,x' _M } when the difference is less than ε (ε is a very small value, generally ε<0.01), the training ends, and the most Optimal weight parameters, these weight parameters will be stored in the storage unit of the device proposed in this patent.

After the above steps S11-S13, the training of the deep neural network model is completed. The above operations are the training part in the embodiment of this application. Model training can be done by machines with powerful computing capabilities such as large servers. The trained model can be implemented with FPGA, GPU, ASIC, etc., and becomes a device for constellation point judgment. The specific implementation method is shown in Figure 6, including the following steps:

S21: Perform SVD decomposition on the channel matrix H to obtain U ^H.

S22, using step S21 to generate U ^H and perform matrix multiplication with the input data Y to be decoded to obtain Z.

S23: Input Z into the deep neural network model trained in step S11-step S13, and the deep neural network model will output soft information of each bit. It is also possible to make a hard decision on the soft information of each bit output and directly output the decoding result.

The above steps S21-S23 are the reasoning part in the embodiment of this application. The inference part can be implemented with FPGA, GPU, ASIC, etc., used in the receiver part of the communication field to make constellation point judgment on the received signal.

Using the above method, a specific embodiment of this application also proposes a constellation point decision device based on deep learning.

The constellation point judgment device based on deep learning proposed in the specific embodiment of the present application includes a preprocessing unit, a storage unit, a judgment unit, an output unit, and a control unit, as shown in FIG. 7.

Among them, the preprocessing unit (corresponding to the aforementioned first processing module 52) first performs SVD decomposition on the channel matrix H to obtain U ^H , and then uses U ^H to perform matrix multiplication with the input data Y to be decoded to obtain Z:

Among them, the preprocessing unit mainly completes matrix operations, so it is composed of memory, multiplier and addition tree.

The storage unit mainly stores the weight parameters obtained by the training model. Because the structure of the deep neural network is complex, it will generate a huge amount of weight parameters. Therefore, the storage unit needs to be composed of a large-capacity memory with a high-speed interface, such as double-rate synchronous dynamic random Memory (Double Data Rate Synchronous Dynamic Random Access Memory, referred to as DDR SDRAM), solid state drive (Solid State Drive, referred to as SSD), etc. The weight parameter is written into the storage unit by software or other methods, and the judgment unit of this device will read the weight parameter from the storage unit.

The decision unit is used to implement a deep neural network model. The decision unit proposed in this application can implement neural network models with different layers and structures, which is very flexible. The decision unit is composed of a data buffer module, a multiply-accumulate matrix, and an activation unit, as shown in Figure 8.

Among them, the data cache module is used to store the input data and the calculated data of each layer. After the calculated data of each layer is placed in the cache, it can be used as the input data of the next layer to continue the calculation of the next layer. In this way, deep neural network models with different layers can be realized.

The multiplication and accumulation matrix is the core of the decision unit, because a neural network is a process of multiplying weight parameters and adding or accumulating different branches. The multiply-accumulate matrix is composed of a multiplier and an adder to form a matrix. No matter how many multiply and accumulate operations are needed in the model layer, it can be done through the multiply and accumulate matrix and data cache.

The activation unit implements the activation function in the deep neural network model.

The output unit is used to output the soft information of each bit. It is also possible to make a hard decision on the soft information of each bit output by the decision unit and directly output the decoding result.

The control unit is used to control the work of each unit, control the weighting coefficient and the multiplication and accumulation of the data stream, as well as the buffering and flow direction of the data stream.

In the following, the method and device of this patent are further described by taking the constellation point decision of the MIMO received signal with 4 receiving antennas and 64QAM modulation as an example.

Establish training samples. In the 64QAM modulation mode, the number of constellation points is 64, that is _{, the number of X J} sets is 64, X _J = {x ₀ ,x ₁ ,...,x ₆₃ }. noise

A Gaussian white noise with a mean value of E(x _i ), where i=0,1,2,...,63, and a variance of σ ^{2 =0.1.} The value of M is 64×10000, which means that each constellation point generates 10000 data, namely

For the first constellation point x ₀ , generate data

For the second constellation point x ₁ , generate data

For the third constellation point x ₂ , generate data

...

For the 64th constellation point x ₆₃ , generate data

Total data sample set

The forward propagation model and the back propagation model of the deep neural network are established. The forward propagation model has 1 input port, and the input is the data in the _{training sample set Z 64x10000.} For 64QAM, the model has 6 output ports, which are the probability information of each bit of the constellation point. At the same time, the MLE algorithm model is established, the input is also _{the data in the training sample set Z 64x10000} , and the output is also the probability information of each bit of the constellation point.

Perform model training, and send _{the data in the training sample set Z 64x10000} to the deep neural network model and the MLE model established in step 2 at the same time. MLE model output expected value x′ _i , i=0,1,2,…,64×10000-1 deep neural network model output data set o _i , i=0,1,2…,64×10000-1, compare 2 The results are sent to the back propagation model established in step 2, and the weight parameters in the forward propagation model are adjusted through the back propagation model until the difference between the expected value and the output value of the forward propagation model is less than ε = 0.005, The training is complete. Save the trained weight parameters.

Because it is a 4-antenna MIMO system, the channel matrix H is a 4×4 matrix. SVD decomposition is performed on H, and U ^H is also a 4×4 matrix.

Perform matrix multiplication on U ^H and the data Y to be decoded received by the receiver to obtain Z. Since Y is a 4×1 matrix, the result Z is a 4×1 matrix.

Z has 4 values, corresponding to 4 antennas. These 4 values are respectively input to the trained deep neural network, and then the probability information of each bit corresponding to each value is output. This probability information can be output to the decoding module for decoding, or the probability information can be hard-decided to directly output the decoding result.

The embodiment of the present application also provides a computer-readable storage medium in which a computer program is stored, wherein the computer program is configured to execute the steps in any one of the foregoing method embodiments when running.

Optionally, in this embodiment, the foregoing computer-readable storage medium may include, but is not limited to: U disk, Read-Only Memory (Read-Only Memory, ROM for short), Random Access Memory (Random Access Memory, for short) RAM), mobile hard disks, magnetic disks or optical disks and other media that can store computer programs.

The embodiment of the present application also provides an electronic device, including a memory and a processor, the memory is stored with a computer program, and the processor is configured to run the computer program to execute the steps in any of the foregoing method embodiments.

In an embodiment, the above-mentioned electronic device may further include a transmission device and an input-output device, wherein the transmission device is connected to the aforementioned processor, and the input-output device is connected to the aforementioned processor.

For specific examples in this embodiment, reference may be made to the examples described in the foregoing embodiment, and this embodiment will not be repeated here.

The method and device in the embodiments of the present application can greatly simplify the calculation amount and complexity of the constellation point decision of the traditional MIMO system. At the same time, the training part of the deep neural network model is optimized, so that the weight parameters trained are more accurate and closer to the real situation, and the time and labor cost of calibrating the label are eliminated.

Obviously, those skilled in the art should understand that the above-mentioned modules or steps of this application can be implemented by a general computing device, and they can be concentrated on a single computing device or distributed in a network composed of multiple computing devices. Above, in some exemplary embodiments, they can be implemented with program codes executable by a computing device, so that they can be stored in a storage device for execution by the computing device, and in some cases, they can be different from Here, the steps shown or described are executed in order, or they are respectively fabricated into individual integrated circuit modules, or multiple modules or steps of them are fabricated into a single integrated circuit module for implementation. In this way, this application is not limited to any specific combination of hardware and software.

The foregoing descriptions are only part of the embodiments of the present application, and are not used to limit the present application. For those skilled in the art, the present application may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the principles of this application shall be included in the protection scope of this application.

Claims

A method for determining the decoding result includes:

Preprocess the to-be-decoded data received by the multiple-input multiple-output MIMO system to obtain the target data;

Use the target data decoding model to process the target data to obtain the probability information of the bits corresponding to the constellation points in the MIMO system, where the target data decoding model is to use multiple sets of data to train a predetermined neural network model Obtained, each of the multiple sets of data includes: sample data and sample bit probability information;

The decoding result is determined based on the probability information of the bit corresponding to the constellation point in the MIMO system.
The method according to claim 1, wherein the sample data is data obtained by performing the preprocessing on the sample data to be decoded, and the sample bit probability information is the use of a maximum likelihood estimation MLE algorithm for the sample Information obtained after data processing.
The method according to claim 1, wherein preprocessing the to-be-decoded data received by the multiple-input multiple-output MIMO system to obtain the target data comprises:

Determine a channel matrix H, where H is an R×T channel matrix, R is the number of receiving antennas of the MIMO system, and T is the number of transmitting antennas of the MIMO system;

Perform singular value decomposition SVD on the channel matrix H to obtain USV, where U is a unitary matrix, S is a diagonal matrix, and V is a unitary matrix;

Perform matrix multiplication on the data to be decoded and the conjugate transposed matrix U H of U to obtain the target data.
The method according to claim 1, wherein before processing the target data using a target data decoding model to obtain probability information of bits corresponding to constellation points in the MIMO system, the method further comprises:

Acquiring the sample data and the sample bit probability information in each of the multiple sets of data;

The predetermined neural network model is trained using the sample data and the sample bit probability information in each of the multiple sets of data to obtain the target data decoding model.
The method according to claim 4, wherein the predetermined neural network model is trained using the sample data and the sample bit probability information in each of the multiple sets of data to obtain the target The data decoding model includes:

Input the sample data into the predetermined neural network model to obtain sample output data;

When it is determined that the difference between the sample output data and the sample bit probability information is greater than a predetermined threshold, the back propagation model of the predetermined neural network model is used to repeatedly perform the weighting of each layer in the predetermined neural network model Value adjustment processing until the difference between the adjusted predetermined neural network output data and the sample bit probability information is less than or equal to the predetermined threshold;

The predetermined neural network model obtained after the final adjustment is determined as the target data decoding model.
The method according to claim 1, wherein determining the decoding result based on the probability information of the bit corresponding to the constellation point in the MIMO system comprises one of the following:

Using a decoding device to decode the probability information of the bit corresponding to the constellation point in the MIMO system to obtain the decoding result;

A hard decision is made on the probability information of the bit corresponding to the constellation point in the MIMO system to obtain the decoding result.
A device for determining a decoding result includes:

The first processing module is used to preprocess the to-be-decoded data received by the multiple-input multiple-output MIMO system to obtain target data;

The second processing module is used to process the target data using the target data decoding model to obtain the probability information of the bit corresponding to the constellation point in the MIMO system, wherein the target data decoding model uses multiple sets of data pairs Obtained by training a predetermined neural network model, each of the multiple sets of data includes: sample data and sample bit probability information;

The determining module is used to determine the decoding result based on the probability information of the bit corresponding to the constellation point in the MIMO system.
8. The device according to claim 7, wherein the sample data is data obtained after the preprocessing of the sample data to be decoded, and the sample bit probability information is the use of the maximum likelihood estimation MLE algorithm for the sample Information obtained after data processing.
A computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the method described in any one of claims 1 to 6 when the computer program is run.
An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the method described in any one of claims 1 to 6.