Depth receiver construction method and application system, receiver, equipment and medium

Abstract

The invention discloses a depth receiver construction method, an application system, a receiver, equipment and a medium, and the construction method comprises the steps: constructing an MU-MIMO OFDM uplink transmission system, so as to obtain an original training data set; a depth receiver is constructed, and the depth receiver is based on a multi-head self-attention mechanism; and performing multi-time iterative model training and convergence on the depth receiver by using the original training data set. A reliable training data set is obtained by designing a standard-meeting MU-MIMO OFDM uplink transmission system, and model training is carried out by constructing a depth receiver based on a multi-head self-attention mechanism, so that accurate decoding of data streams in a non-orthogonal superposition pilot frequency transmission scheme is effectively realized; therefore, there is no need to design a plurality of independent modules and solve the problem of cooperative work among the modules, and the bit error rate performance is superior to that of a traditional receiver scheme.

Description

Technical Field

[0001] The present application relates to the field of information and communication technology, and in particular to a method for constructing a deep receiver of multi-user multiple-input multiple-output (MU-MIMO) orthogonal frequency division multiplexing (OFDM) based on a non-orthogonal superimposed pilot transmission scheme and an application system, receiver, device, and medium. Background Art

[0002] In the fifth generation (5G) and future communication systems, machine learning technology has gradually become the core innovation driver for solving complex air interface problems.

[0003] With the development of wireless communication technology, the system has higher and higher requirements for accuracy, efficiency and dynamic adaptability. Traditional Orthogonal Frequency Division Multiplexing (OFDM) receivers rely on the cooperation between multiple independent modules, including channel estimation, channel equalization and signal demodulation.

[0004] The inventors found that although this modular design architecture can ensure the stability of each functional module, the interdependence between modules also brings significant technical limitations. Therefore, it is urgent to combine machine learning technology to solve the technical limitations brought by this modular design in order to provide a more optimized deep receiver. Summary of the invention

[0005] In order to solve the problem, this application proposes a deep receiver construction method and application system, receiver, device, medium, and the technical solution is as follows:

[0006] On the one hand, a method for constructing a deep receiver is provided, comprising the steps of:

[0007] Build a MU-MIMO OFDM uplink transmission system to obtain the original training data set;

[0008] Construct a deep receiver, where the deep receiver is based on a multi-head self-attention mechanism;

[0009] The deep receiver is trained with the original training dataset for multiple iterations and converges.

[0010] The construction of MU-MIMO OFDM uplink transmission system includes:

[0011] Base station (BS) deployment N t transmit antennas, user equipment (UE) deployed N r Root receiving antenna;

[0012] The transmission resource configuration of the system is to use S subcarriers in the frequency domain and T symbols in the time domain to achieve the transmission of K user bit stream data;

[0013] The transmitter sends the information bit stream B∈{0,1} K×T×S×M Execution 2 M Order modulation to obtain data symbols

[0014] The transmitter performs non-orthogonal superposition of pilot symbols and data symbols to obtain superposition symbols Right now:

[0015]

[0016] in, is the pilot symbol matrix of the kth user, 1≤k≤K is the index of the number of users, and are the weight matrices of the kth user data symbol and pilot symbol respectively;

[0017] The superimposed symbols are transmitted through the channel and sent to the receiving end to obtain the received signal, that is:

[0018]

[0019] in, For the r The received signal of the receiving antenna (1≤r≤N r ), the symbol ⊙ is the Hadamard product,

[0020] For the r The channel matrix of the root receiving antenna and the kth user, is complex valued additive white Gaussian noise (AWGN), and the variance of each element is σ 2 ;

[0021] Finally, the r The receiving signal Y of the receiving antenna r Cascade to obtain the final received signal

[0022] in, and The design needs to ensure that the superimposed symbols of each user satisfy the average energy of 1, that is:

[0023]

[0024] The depth receiver includes a depth receiver f(·, θ), and the depth receiver construction method includes the steps of:

[0025] To receive the signal and the pilot symbol matrix is the input of the deep receiver f(·,θ), performs implicit channel estimation and equalization, and calculates

[0026] Obtain pilot position information based on pilot symbol P, so that the deep receiver f(·, θ) can accurately separate data symbols from superimposed symbols;

[0027] Using the information bit stream B∈{0,1} K×T×S×M as labeled data to supervise the training of the deep receiver f(·,θ), where θ is the model parameter.

[0028] The input data is concatenated and reshaped to adjust the input shape, and then the input is mapped to a high-dimensional feature space through linear transformation. Finally, position encoding based on sine and cosine functions is added to obtain an output containing the relative position information of the input data in the time-frequency domain. The input data includes and the pilot symbol matrix

[0029] The input data is subjected to a layer normalization (LN) and the query Q, key K and value V matrices are obtained through three linear transformations. The attention score is then calculated using the scaled dot product attention mechanism. The output of the attention is then processed through a linear layer and weighted with the network input data to obtain a weighted result.

[0030] The weighted result is normalized once and then input into a multi-layer perceptron for feature extraction. The result of the input layer normalization is then weighted with the output of the multi-layer perceptron to obtain the output of a single-layer Pre-LN Transformer encoder block. N layers of encoder blocks are stacked repeatedly to obtain the output of an N-layer Pre-LN Transformer encoder.

[0031] The output of the N-layer Pre-LN Transformer encoder is mapped to the log-likelihood ratio (LLR) of each resource element (RE) through linear transformation and dimension reshaping.

[0032] After obtaining the original training data set, data enhancement is performed based on the original training data set to obtain training data sets for different scenarios; data enhancement includes any one or more of the following: angle rotation, time domain transformation, and data flipping.

[0033] On the other hand, a deep receiver constructed using any of the above methods is provided, and the deep receiver is used to achieve accurate decoding of data streams in a non-orthogonal superimposed pilot transmission scheme.

[0034] On the other hand, an application system of a depth receiver is provided, the system comprising:

[0035] A transmitter, used to implement coded transmission of data streams in a non-orthogonal superposition pilot transmission scheme;

[0036] The above-mentioned deep receiver is used to receive the data sent by the transmitter and accurately decode the sent data.

[0037] On the other hand, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above methods when executing the computer program.

[0038] On the other hand, a readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, any of the above-mentioned methods for constructing a deep receiver is implemented.

[0039] The beneficial effects of this application are: proposing a deep receiver construction method, constructing a MU-MIMO OFDM uplink transmission system to obtain the original training data set; constructing a deep receiver, wherein the deep receiver is based on a multi-head self-attention mechanism; and using the original training data set to perform multiple iterations of model training on the deep receiver and converge. By designing a standard MU-MIMO OFDM uplink transmission system to obtain a reliable training data set, and by constructing a deep receiver based on a multi-head self-attention mechanism for model training, accurate decoding of data streams in non-orthogonal superimposed pilot transmission schemes is effectively achieved, thereby eliminating the need to design multiple independent modules and solve the problem of collaborative work between modules, and the bit error rate performance is better than that of traditional receiver schemes. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of an embodiment of a method for constructing a deep receiver of the present application;

[0041] Figure 2 A flowchart of another embodiment of the method for constructing a deep receiver of the present application;

[0042] Figure 3 This is a schematic diagram of the architecture of an embodiment of a MU-MIMO OFDM uplink transmission system of the present application;

[0043] Figure 4 A schematic diagram of a training architecture of a deep receiver according to an embodiment of the present application;

[0044] Figure 5 A schematic diagram showing a BER performance comparison between a deep receiver of the present application and different receiver solutions;

[0045] Figure 6 A schematic diagram showing a comparison of throughput performance between a deep receiver of the present application and different receiver solutions in an embodiment;

[0046] Figure 7 A schematic diagram showing the BER performance comparison of an embodiment of the deep receiver of the present application under a flexible and diverse data enhancement strategy;

[0047] Figure 8 This is a schematic diagram of BER performance comparison of an embodiment of the deep receiver of the present application under different parameter settings;

[0048] Fig. 9 It is a schematic diagram of a framework of an embodiment of an electronic device provided by the present application;

[0049] Fig.10 It is a schematic diagram of a framework of an embodiment of a computer-readable storage medium provided by the present application. DETAILED DESCRIPTION

[0050] In order to facilitate the understanding of the present application, the present application is described in more detail below in conjunction with the accompanying drawings and specific embodiments. The preferred embodiments of the present application are provided in the accompanying drawings. However, the present application can be implemented in many different forms and is not limited to the embodiments described in this specification. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive.

[0051] It should be noted that, unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. For example, the term "multiple" includes two or more.

[0052] The following are some explanations of terms used in this application:

[0053] 3GPP: Third Generation Partnership Project

[0054] UMa: Urban Macro Cell

[0055] MU: Multi-User

[0056] MU-MIMO: Multi-User Multiple Input Multiple Output

[0057] OFDM: Orthogonal Frequency Division Multiplexing

[0058] MIMO: Multiple Input Multiple Output

[0059] Goodput: Actual throughput

[0060] Sionna: Open source simulation library for deep learning of physical layer communications

[0061] BER: Bit Error Rate

[0062] QAM: Quadrature Amplitude Modulation

[0063] BPSK: Binary Phase Shift Keying

[0064] As mentioned before, traditional receivers are modular and have their own limitations.

[0065] The solution of this application is based on a deep learning model, which directly learns the entire OFDM receiver processing from input to output, and no longer relies on manually designed modular processes. This not only effectively simplifies the system design, but also greatly improves the overall performance of the system by optimizing the joint processing of the entire transmission link. At the same time, the end-to-end learning architecture has stronger adaptability and can more flexibly respond to complex communication scenarios and equipment requirements.

[0066] Specifically, in this field, relevant research has also demonstrated the feasibility of this path:

[0067] Reference 1H.Ye,GYLi and B.-H.Juang,“Power of deep learning for channelestimation and signal detection in OFDM systems,”IEEE Wireless Commun.Lett.,vol.7,no.1,pp.114-117,Feb.2018. The application of deep learning in OFDM systems was studied. The channel estimation and signal detection performance were improved by end-to-end training of deep neural networks, especially in complex nonlinear channels, showing broad communication potential.

[0068] Reference 2 M. Honkala, D. Korpi and JMJ Huttunen, "DeepRx: Fully convolutional deep learning receiver," IEEE Trans. Wirel. Commun., vol. 20, no. 6, pp. 3925-3940, June 2021. A fully convolutional deep learning receiver is proposed, which does not require explicit channel estimation and directly performs end-to-end optimization from received signals to decoded data. This method adapts to complex channel conditions and significantly improves communication performance and robustness.

[0069] Reference 3 F.Ait Aoudia and J.Hoydis, “End-to-end learning for OFDM:From neural receivers to pilotless communication,” IEEE Trans.Wirel.Commun., vol.21, no.2, pp.1049-1063, Feb.2022. An end-to-end deep learning method is proposed, which directly replaces the traditional receiver and channel estimation module with a neural network, reduces the pilot overhead, and improves the communication efficiency and system robustness.

[0070] Based on the above background, this application innovatively proposes a new method for constructing a deep receiver, which involves a method for constructing a multi-user multiple-input multiple-output (MU-MIMO) orthogonal frequency division multiplexing (OFDM) deep receiver based on a non-orthogonal superimposed pilot transmission scheme, which improves the throughput of the MU-MIMO OFDM communication system running in a time-selective fading channel. It mainly includes constructing a MU-MIMO OFDM uplink transmission system based on the UMa channel model in 3GPP, generating an experimental data set according to the non-orthogonal superimposed pilot transmission scheme, and designing a novel deep receiver based on the multi-head self-attention mechanism according to the characteristics of the superimposed data to achieve the separation and reception of multi-layer bit stream information. The robustness of the deep receiver is also improved by designing a flexible data enhancement strategy. Finally, the deep receiver is trained and designed based on the above generated data set and data enhancement method.

[0071] The following are specific embodiments:

[0072] On the one hand, reference Figure 1 As shown, the present application provides a method for constructing a deep receiver, comprising the steps of:

[0073] Build a MU-MIMO OFDM uplink transmission system to obtain the original training data set;

[0074] Construct a deep receiver, where the deep receiver is based on a multi-head self-attention mechanism;

[0075] The deep receiver is trained with the original training dataset for multiple iterations and converges.

[0076] The construction of MU-MIMO OFDM uplink transmission system includes:

[0077] Base station (BS) deployment N t transmit antennas, user equipment (UE) deployed N r Root receiving antenna;

[0078] The transmission resource configuration of the system is to use S subcarriers in the frequency domain and T symbols in the time domain to achieve the transmission of K user bit stream data;

[0079] The transmitter sends the information bit stream B∈{0,1} K×T×S×M Execution 2 M Order modulation to obtain data symbols

[0080] The transmitter performs non-orthogonal superposition of pilot symbols and data symbols to obtain superposition symbols Right now:

[0081]

[0082] in, is the pilot symbol matrix of the kth user, 1≤k≤K is the index of the number of users, and are the weight matrices of the kth user data symbol and pilot symbol respectively;

[0083] The superimposed symbols are transmitted through the channel and sent to the receiving end to obtain the received signal, that is:

[0084]

[0085] in, For the r The received signal of the receiving antenna (1≤r≤N r ), the symbol ⊙ is the Hadamard product, is the channel matrix of the rth receiving antenna and the kth user, is complex valued additive white Gaussian noise (AWGN), and the variance of each element is σ 2 ;

[0086] Finally, the r The receiving signal Y of the receiving antenna r Cascade to obtain the final received signal

[0087] Specifically, please refer to Figure 3 The figure shows the architecture of the MU-MIMO OFDM uplink transmission system. Channel is the communication channel, the left side is the transmitter, and the right side is the receiver. The figure shows a typical K users sending information stream B through a wireless base station, and then decoding the pilot symbol P and the received signal Y at the receiving end to achieve deep reception, and obtain the estimated decoding information after hard decision. process.

[0088] On this basis, the MU-MIMO OFDM deep receiver construction method based on the non-orthogonal superposition pilot transmission scheme proposed in this application jointly optimizes channel estimation, channel equalization and demodulation tasks by utilizing the powerful data representation ability of deep learning to achieve information bit stream High-precision reception.

[0089] in,

[0090] and The design needs to ensure that the superimposed symbols of each user satisfy the average energy of 1, that is:

[0091] E{|S k | 2}=1.

[0092] The pilot symbol matrix P is generated by randomly sampling a binary phase shift keying (BPSK) constellation based on a zero-mean pseudo-random sequence to achieve zero average energy of the transmission signal.

[0093] In order to reduce mutual interference caused by superimposed symbols of each layer, the superimposed positions of pilot symbols of each layer are staggered in the time domain dimension, and a block pilot design method is adopted.

[0094] For further information, please refer to Figure 4 , the depth receiver includes a depth receiver f(·, θ), and the depth receiver construction method includes the steps of:

[0095] To receive the signal and the pilot symbol matrix is the input of the deep receiver f(·,θ), performs implicit channel estimation and equalization, and calculates

[0096] Obtain pilot position information based on pilot symbol P, so that the deep receiver f(·, θ) can accurately separate data symbols from superimposed symbols;

[0097] Using the information bit stream B∈{0,1} K×T×S×M as labeled data to supervise the training of the deep receiver f(·,θ), where θ is the model parameter.

[0098] Among them, the information bit stream decoded from the received signal for:

[0099]

[0100] Wherein, p(·) represents a hard decision process, which is defined as p(x)=0 when x<0 and p(x)=1 when x≥0.

[0101] The input data is concatenated and reshaped to adjust the input shape, and then the input is mapped to a high-dimensional feature space through linear transformation. Finally, position encoding based on sine and cosine functions is added to obtain an output containing the relative position information of the input data in the time-frequency domain. The input data includes and the pilot symbol matrix

[0102] The input data is subjected to a layer normalization (LN) and the query Q, key K and value V matrices are obtained through three linear transformations. The attention score is then calculated using the scaled dot product attention mechanism. The output of the attention is then processed through a linear layer and weighted with the network input data to obtain a weighted result.

[0103] The weighted result is normalized once and then input into a multi-layer perceptron for feature extraction. The result of the input layer normalization is then weighted with the output of the multi-layer perceptron to obtain the output of a single-layer Pre-LN Transformer encoder block. N layers of encoder blocks are stacked repeatedly to obtain the output of an N-layer Pre-LN Transformer encoder.

[0104] The output of the N-layer Pre-LN Transformer encoder is mapped to the log-likelihood ratio (LLR) of each resource element (RE) through linear transformation and dimension reshaping.

[0105] The deep receiver uses binary cross entropy loss to calculate and B∈{0,1} L×T×S×M The error between them, the loss function is:

[0106]

[0107] Where N b is the batch size,

[0108] The loss function is optimized using the Adaptive Moment (Adam) optimizer.

[0109] Specifically, in a multi-user scenario using a non-orthogonal superimposed pilot transmission scheme, the parallel transmission of multiple bit streams within the same channel can lead to signal coupling between antennas and users, resulting in significant interference. To address this challenge, a deep receiver using a multi-head self-attention (MHSA) mechanism is developed. Figure 4 As shown in Figure 2, the MHSA mechanism processes the input signal features from different angles and subspaces through multiple attention heads, enhancing the capture of different user signal features and thus reducing interference between users.

[0110] The specific design is as follows: first, the time-frequency domain in the resource grid (RG) is used as the sequence information in the Transformer encoder, and the receiving antenna dimension N of Y is r , the user dimension K of P and the real and imaginary dimensions are concatenated as feature dimensions. Use linear transformation to map feature dimensions to high-dimensional feature space Then, an N-layer Pre-LNTransformer encoder block is used to implement implicit mapping of channel estimation and channel equalization. Finally, a linear transformation is used to map the features to Due to the parallel computing capability of the MHSA mechanism, the position information in the time-frequency domain is lost. Position encoding (PE) based on sine and cosine functions is used to record the relative position, which is given by

[0111]

[0112] Where l and i represent the position index and feature dimension index respectively.

[0113] Assume that the input of the MHSA mechanism is Then the query, key, and value matrices of the hth head can be obtained through unbiased linear transformation, which are as well as

[0114] The MHSA mechanism can be expressed as

[0115] MHSA(Q,K,V)=Concat(head 1 ,…,head H )W O ,

[0116] Where H represents the number of heads, is the output linear transformation matrix. Since the pilot positions superimposed on each layer of data symbols are different, the inter-layer interference problem is solved by setting H = L, and a multi-layer perceptron (MLP) is used to extract the spatial features of each layer.

[0117] Each head uses scaled dot product attention to calculate the attention score of the feature and update the weight of each head feature. The h-th head can be expressed as:

[0118]

[0119] in,

[0120] where x i and L a are the i-th component and dimension of the input vector x, respectively.

[0121] Among them, reference Figure 2As shown in the figure, after obtaining the original training data set, data enhancement is performed based on the original training data set to obtain training data sets for different scenarios; data enhancement includes any one or more of the following: angle rotation, time domain transformation, and data flipping. This strategy can be dynamically adjusted to meet the specific needs of various scenarios. Depending on the data characteristics and model requirements, each method can be applied alone or in combination. The specific contents are as follows:

[0122] (1) Angle rotation: Applying different small angle rotations to all symbols in an OFDM frame can better handle the phase error and shift of the model. The specific steps are as follows:

[0123] Y aug =exp(j·α aug )·Y,

[0124] in, Represents the rotation angle matrix.

[0125] (2) Time domain shuffling: The received signal Pilot Matrix and information bit stream B∈{0,1} K×T×S×M Transforming according to the same rules in the time domain can help deep receivers better adapt to different channel environments, avoid dependence on fixed timing patterns, and enhance robustness to time-varying channels and interference. Specifically, for the received signal Y, the shuffle operation can be expressed as:

[0126] Y aug [:,k,:,:,:]=Y[:,π(k),:,:,:],

[0127] Where π(k) represents the random permutation of the user dimension K, ensuring that each sequence is different.

[0128] (3) Data flipping: for the pilot matrix and the information bitstream matrix B∈{0,1} K×T×S×M Symmetric flipping the user number dimension does not change their statistical properties, such as amplitude and phase. This operation can prevent the model from being overly dependent on a specific user order and enhance the model to focus on overall features rather than specific orders. The symmetric flipping operation can be mathematically expressed as:

[0129] P aug [:,k,:,:,:]=P[:,Kk-1,:,:,:],

[0130] B aug [:,k,:,:,:]=B[:,Kk-1,:,:,:].

[0131] During the data augmentation process, these methods are triggered randomly with a certain probability, allowing the dataset to contain a wider range of variations, thereby enhancing the robustness of the model.

[0132] Specifically, this application implements a MU-MIMO OFDM system using the UMa channel model from the Sionna open source library to evaluate the throughput gain of deep receivers and non-orthogonal superimposed pilot transmission schemes in the communication system.

[0133] Among them, the detailed parameter settings of the communication system are shown in Table 1.

[0134] For the proposed deep receiver, the parameters are configured as follows: N = 6, d model =256, H = K = 4, and the hidden dimension of the linear layer in MLP is set to 512 (d model ×2).

[0135] During the training process, the learning rate lr = 1×10 -3 , batch size N b =20, and a total of 100 epochs are used.

[0136] Table 1 Communication system parameter settings

[0137]

[0138]

[0139] exist Figure 5 In this paper, the error performance of the proposed deep receiver is compared with that of traditional receivers, including least squares (LS) + linear minimum mean square error (LMMSE), perfect channel state information (CSI) + LMMSE, and deep receivers based on residual neural network (ResNet) [2], [3] and multi-layer perceptron (MLP) [1]. Traditional receiving algorithms use orthogonal pilot schemes because they cannot decode signals under the NOSIP scheme.

[0140] To ensure a fair comparison, the BER performance of the proposed deep receiver is evaluated under orthogonal pilot and NOSIP schemes. Other deep receiver schemes are designed based on the NOSIP protocol.

[0141] like Figure 5 As shown in Figure 1, the existing deep receiver cannot decode the signal in the MU-MIMO OFDM system based on the NOSIP scheme. In contrast, the deep receiver proposed in this application can achieve accurate decoding under both the orthogonal pilot and NOSIP schemes. Although the bit error rate performance of the orthogonal pilot scheme is better than that of the NOSIP scheme, the latter significantly improves the utilization rate of the OFDM multiplexing rate.

[0142] In order to highlight the performance gain of the non-orthogonal superposition pilot scheme, Figure 6 The system throughput performance of different pilot transmission schemes is given in . Among them, the throughput of the MU-MIMO OFDM system is:

[0143]

[0144] Where c is the code rate and ρ is the ratio of data symbols in RG. When non-orthogonal SIP or no pilot scheme is used, ρ = 1. When an orthogonal pilot scheme is used, ρ = 3 / 4.

[0145] The good input upper bound of the non-orthogonal SIP scheme is about 35.91% higher than that of the orthogonal pilot scheme. Compared with the “LS+LMMSE” baseline scheme, the proposed Trans-Rx scheme based on orthogonal pilot and non-orthogonal SIP has good throughput improvements of about 7.01% and 37.15%, respectively.

[0146] Figure 7 The BER performance of the deep receiver under different data augmentation methods is shown. The three data augmentation methods of angle rotation (rotation), time dimension transformation (shuffling) and data flipping (flipping) all effectively improve the BER performance of the deep receiver. Among them, "deep receiver-NOSIP-augmentation" represents the BER performance of the deep receiver when the three data augmentation methods are used at the same time, and the BER performance of the deep receiver reaches the best effect at this time.

[0147] Figure 8 The error performance of the deep receiver under different parameter settings is shown, including d model , N, and with or without PE. Due to the powerful parallel processing capability of the MHSA mechanism, without PE, the model cannot effectively capture the relative position relationship of the elements in the input sequence, resulting in the loss of temporal structure information. This causes the proposed deep receiver to be unable to correctly recover the input information during the decoding process, resulting in a sharp drop in BER performance. model ∈{64,128,256}, the BER performance of the proposed deep receiver also increases with d model When N=6, the bit error rate performance of the deep receiver under this system setting is the best.

[0148] On the other hand, a deep receiver constructed using any of the above methods is provided, and the deep receiver is used to achieve accurate decoding of data streams in a non-orthogonal superimposed pilot transmission scheme.

[0149] On the other hand, an application system of a depth receiver is provided, the system comprising:

[0150] A transmitter, used to implement coded transmission of data streams in a non-orthogonal superposition pilot transmission scheme;

[0151] The above-mentioned deep receiver is used to receive the data sent by the transmitter and accurately decode the sent data.

[0152] On the other hand, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above methods when executing the computer program.

[0153] Specifically, see Fig. 9 The electronic device described in the embodiment of the present application may specifically include a processor 210 and a memory 220. The memory 220 is coupled to the processor 210.

[0154] The processor 210 is used to control the operation of the electronic device. The processor 210 may also be referred to as a CPU.

[0155] (Central Processing Unit). The processor 210 may be an integrated circuit chip with signal processing capabilities. The processor 210 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor 210 may also be any conventional processor, etc.

[0156] The memory 220 is used to store computer programs, and may be a RAM, a ROM, or other types of storage devices. Specifically, the memory may include one or more computer-readable storage media, which may be non-transitory. The memory may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory is used to store at least one program code.

[0157] The processor 210 is used to execute the computer program stored in the memory 220 to implement the methods described in the various method embodiments of the present application.

[0158] In some embodiments, the electronic device may further include: a peripheral device interface 230 and at least one peripheral device. The processor 210, the memory 220 and the peripheral device interface 230 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface 230 via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 240, a display screen 250, an audio circuit 260 and a power supply 270.

[0159] The peripheral device interface 230 may be used to connect at least one peripheral device related to I / O (Input / output) to the processor 210 and the memory 220. In some embodiments, the processor 210, the memory 220, and the peripheral device interface 230 are integrated on the same chip or circuit board; in some other implementations, any one or two of the processor 210, the memory 220, and the peripheral device interface 230 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

[0160] The radio frequency circuit 240 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 240 communicates with communication networks and other communication devices through electromagnetic signals, and the radio frequency circuit 240 is the communication circuit of the electronic device. The radio frequency circuit 240 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. Optionally, the radio frequency circuit 240 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc. The radio frequency circuit 240 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes but is not limited to: the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and / or

[0161] WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 240 may also include NFC (Near Field Communication) related circuits, which are not limited in the present application.

[0162] The display screen 250 is used to display the UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 250 is a touch display screen, the display screen 250 also has the ability to collect touch signals on the surface or above the surface of the display screen 250. The touch signal can be input to the processor 210 as a control signal for processing. At this time, the display screen 250 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, the display screen 250 can be one, set on the front panel of the electronic device; in other embodiments, the display screen 250 can be at least two, respectively set on different surfaces of the electronic device or in a folding design; in other embodiments, the display screen 250 can be a flexible display screen, set on a curved surface or a folding surface of the electronic device. Even, the display screen 250 can also be set to a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 250 can be made of materials such as LCD (Liquid Crystal Display), OLED (Organic Light Emitting Diode, organic light-emitting diode).

[0163] The audio circuit 260 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals and input them into the processor 210 for processing, or input them into the radio frequency circuit 240 to achieve voice communication. For the purpose of stereo acquisition or noise reduction, there may be multiple microphones, which are respectively arranged in different parts of the electronic device. The microphone may also be an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert the electrical signal from the processor 210 or the radio frequency circuit 240 into sound waves. The speaker may be a traditional film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for purposes such as ranging. In some embodiments, the audio circuit 260 may also include a headphone jack.

[0164] The power supply 270 is used to power various components in the electronic device. The power supply 270 can be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 270 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged through a wired line, and a wireless rechargeable battery is a battery that is charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0165] For a detailed description of the functions and execution processes of each functional module or component in the intelligent control platform device embodiment of the present application, reference can be made to the description in the above-mentioned method embodiments of the present application, which will not be repeated here.

[0166] In the several embodiments provided in the present application, it should be understood that the disclosed intelligent control platform device and method can be implemented in other ways. For example, the various embodiments of the intelligent control platform device described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0167] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0168] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0169] On the other hand, a readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, any of the above-mentioned methods for constructing a deep receiver is implemented.

[0170] See also Fig.10 , if the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium 300. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions / computer programs to enable an intelligent control platform device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks or optical disks, and electronic devices such as computers, mobile phones, laptops, tablet computers, cameras, etc. having the above-mentioned storage media.

[0171] The description of the execution process of the program data in the computer-readable storage medium can refer to the description in the above-mentioned method embodiments of the present application, which will not be repeated here.

[0172] The above descriptions are merely embodiments of the present application and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

[0173] Those skilled in the art can understand that in the above method of the specific implementation, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic. In summary, the present application has the following beneficial effects:

[0174] The deep receiver construction method of this application is based on a deep learning model, which directly learns the entire OFDM receiver processing process from input to output, and no longer relies on manually designed modular processes. This not only effectively simplifies the system design, but also greatly improves the overall performance of the system by optimizing the joint processing of the entire transmission link. At the same time, the end-to-end learning architecture has stronger adaptability and can more flexibly respond to complex communication scenarios and equipment requirements.

[0175] Furthermore, a reliable training dataset is obtained by designing a standard-compliant MU-MIMO OFDM uplink transmission system. The deep receiver with multi-head self-attention (MHSA) mechanism processes the input signal features from different angles and subspaces to enhance the capture of different user signal features, thereby reducing interference between users.

[0176] Furthermore, flexible and diverse data augmentation strategies are provided, including angle rotation, time domain shuffling, and data flipping. The strategy can be dynamically adjusted to meet the specific needs of various scenarios. Depending on the data characteristics and model requirements, each method can be applied alone or in combination.

Claims

1. A method for constructing a deep receiver, characterized in that: Includes steps: Build a MU-MIMO OFDM uplink transmission system to obtain the original training data set; Constructing a deep receiver, wherein the deep receiver is based on a multi-head self-attention mechanism; The deep receiver is trained with the original training data set for multiple iterations and converged.

2. The method for constructing a deep receiver according to claim 1, characterized in that: The construction of the MU-MIMO OFDM uplink transmission system includes: Base station (BS) deployment N t transmit antennas, user equipment (UE) deployed N r Root receiving antenna; The transmission resource configuration of the system is to use S subcarriers in the frequency domain and T symbols in the time domain to achieve the transmission of K user bit stream data; The transmitter sends the information bit stream B∈{0,1} K×T×S×M Execution 2 M Order modulation to obtain data symbols The transmitter performs non-orthogonal superposition of pilot symbols and data symbols to obtain superposition symbols Right now: in, is the pilot symbol matrix of the kth user, 1≤k≤K is the index of the number of users, and are the weight matrices of the kth user data symbol and pilot symbol respectively; The superimposed symbols are transmitted through the channel and sent to the receiving end to obtain the received signal, that is: in, For the r The received signal of the receiving antenna (1≤r≤N r ), the symbol ⊙ is the Hadamard product, is the channel matrix of the rth receiving antenna and the kth user, is complex valued additive white Gaussian noise (AWGN), and the variance of each element is σ 2 ; Finally, the r The receiving signal Y of the receiving antenna r Cascade to obtain the final received signal 3. The method for constructing a deep receiver according to claim 2, characterized in that: and The design needs to ensure that the superimposed symbols of each user satisfy the average energy of 1, that is: E{|S k | 2 }=1.

4. The method for constructing a deep receiver according to claim 2, characterized in that: The depth receiver includes a depth receiver f(·, θ), and the depth receiver construction method includes the steps of: To receive the signal and the pilot symbol matrix is the input of the deep receiver f(·, θ), performs implicit channel estimation and equalization, and calculates Obtain pilot position information based on pilot symbol P, so that the deep receiver f(·, θ) can accurately separate data symbols from superimposed symbols; Using the information bit stream B∈{0,1} K×T×S×M as labeled data to supervise the training of the deep receiver f(·,θ), where θ is the model parameter.

5. The method for constructing a deep receiver according to claim 4, characterized in that: Includes steps: The input data is concatenated and reshaped to adjust the input shape, and then the input is mapped to a high-dimensional feature space through a linear transformation, and finally a position encoding based on sine and cosine functions is added to obtain an output containing the relative position information of the input data in the time-frequency domain; wherein the input data includes and the pilot symbol matrix Perform a layer normalization (LN) on the input data, obtain the query Q, key K and value V matrices through three linear transformations, and then use the scaled dot product attention mechanism to calculate the attention score. The output of the attention is then processed through a linear layer and weighted with the network input data to obtain a weighted result; The weighted result is subjected to a layer normalization, and then input into a multi-layer perceptron for feature extraction, and then the direct result of the input layer normalization is weighted with the output of the multi-layer perceptron to obtain the output of a single-layer Pre-LN Transformer encoder block, and N layers of encoder blocks are repeatedly stacked in this way to obtain the output of an N-layer Pre-LN Transformer encoder; The output of the N-layer Pre-LN Transformer encoder is mapped to the log-likelihood ratio (LLR) of each resource element (RE) through linear transformation and dimension reshaping.

6. The method for constructing a deep receiver according to claim 1, characterized in that: Also includes the steps: After obtaining the original training data set, data enhancement is performed according to the original training data set to obtain training data sets for different scenarios; the data enhancement includes any one or more of the following: angle rotation, time domain transformation, and data flipping.

7. A deep receiver constructed using the method according to any one of claims 1 to 6, characterized in that: The deep receiver is used to achieve accurate decoding of data streams in a non-orthogonal superimposed pilot transmission scheme.

8. An application system of a depth receiver, characterized in that: The system comprises: A transmitter, used to implement coded transmission of data streams in a non-orthogonal superposition pilot transmission scheme; The deep receiver as described in claim 7 is used to receive the transmitted data of the transmitter and accurately decode the transmitted data.

9. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 6 when executing the computer program.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for constructing a deep receiver according to any one of claims 1 to 6 is implemented.

Images