Signal receiving method, communication node and storage medium

Abstract

The invention provides a signal receiving method, a communication node and a storage medium. According to the method, a received signal is input to a receiver to obtain an output signal, and the received signal comprises a data pilot symbol and a data symbol; the receiver comprises the following cascaded neural networks: a first neural network for pre-demodulating a data pilot symbol of the received signal, and a second neural network for performing channel estimation on the data pilot symbol and a data symbol of the received signal, and a third neural network for equalizing data symbols of the received signal; and demodulating the output signal to obtain received data.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, for example, to a signal receiving method, a communication node and a storage medium. Background Art

[0002] With the continuous improvement of economic and social development needs, infrastructure has developed rapidly, and the demand for mobile communications in high-speed mobile scenarios such as high-speed railways and subways has increased. Therefore, the efficiency and reliability of wireless communications in high-speed scenarios have become a research hotspot in the domestic and foreign communications industry.

[0003] Different from the more extensive low-speed mobile scenarios, the wireless channels in high-speed scenarios have characteristics such as high Doppler frequency deviation and fast time variation that cause signal distortion, which requires more accurate channel estimation and brings greater challenges to the design of communication systems. Most of the existing channel estimation and equalization technologies suitable for high-speed mobile scenarios are based on the traditional Multiple Input Multiple Output (MIMO) + Orthogonal Frequency Division Multiplexing (OFDM) wireless communication receiving architecture, which is achieved by obtaining channel state information and frequency domain estimation and interpolation of pilots. The diversity gain of MIMO space division multiplexing and the anti-interference ability of OFDM orthogonal subcarriers bring good wireless signal transmission performance, but the large number of pilot insertions and code division multiplexing occupy data transmission resources, which affects the throughput of the entire signal system to a certain extent, and the efficiency of receiving signal processing is low. Summary of the invention

[0004] The present application provides a signal receiving method, a communication node and a storage medium.

[0005] The present application provides a signal receiving method, including:

[0006] Inputting a received signal into a receiver to obtain an output signal, wherein the received signal includes a data pilot symbol and a data symbol; the receiver includes the following cascaded neural networks: a first neural network for pre-demodulating the data pilot symbol of the received signal, a second neural network for performing channel estimation on the data pilot symbol and the data symbol of the received signal, and a third neural network for equalizing the data symbol of the received signal;

[0007] The output signal is demodulated to obtain received data.

[0008] An embodiment of the present application also provides a communication node, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned signal receiving method when executing the program.

[0009] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned signal receiving method is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flowchart of a signal receiving method provided by an embodiment;

[0011] Figure 2 A schematic diagram of a two-stream signal data frame structure provided by an embodiment;

[0012] Figure 3 A schematic diagram of a signal transmission system based on a cascade neural network provided by an embodiment;

[0013] Figure 4 A block diagram of a high-speed receiver based on a cascaded neural network with dual drive of a data model provided by an embodiment;

[0014] Figure 5 A schematic diagram of the structure of a signal receiving device provided by an embodiment;

[0015] Figure 6 A schematic diagram of the hardware structure of a communication node provided by an embodiment. DETAILED DESCRIPTION

[0016] The present application is described below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should be noted that, in the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other arbitrarily. It should also be noted that, for ease of description, only the parts related to the present application rather than all structures are shown in the accompanying drawings.

[0017] A signal receiving method provided by the present application is used to improve signal throughput and signal receiving performance. In recent years, artificial intelligence (AI) technologies such as machine learning and deep learning have broken through the limitations of traditional communication modules in system modeling and model approximation, and have made progress in multiple directions such as channel estimation, channel equalization, high-level user scheduling and network deployment optimization. AI technology builds a more flexible and reliable network model based on the processing and training of a large amount of data to achieve system performance optimization. Specific to the AI ​​design of the physical layer, there are mainly two mainstream routes: end-to-end communication link design based on AI technology and communication module algorithm design based on AI technology. The communication module algorithm design based on AI technology refers to replacing some modules in traditional transmitters and receivers with deep neural network models. It is also divided into two design ideas: data-driven and data model dual-driven. The data-driven solution is to replace multiple modules of the wireless communication system with a deep learning network as a whole, while the data model dual-driven is to replace a certain module or train related parameters with a deep learning network without changing the original wireless communication system model results to achieve performance improvement. In high-speed scenarios, this application can use deep learning neural networks to replace the original wireless channel receiver architecture to improve its performance gain. At the same time, it combines the transmitter's precoding and pre-equalization processing to achieve pilot-free transmission and improve system throughput.

[0018] Figure 1 A flowchart of a signal receiving method provided by an embodiment is shown in FIG. Figure 1 As shown, the method provided in this embodiment includes steps 110 to 120.

[0019] In step 110, a received signal is input into a receiver to obtain an output signal, wherein the received signal includes data pilot symbols and data symbols; the receiver includes the following cascaded neural networks: a first neural network for pre-demodulating the data pilot symbols of the received signal, a second neural network for performing channel estimation on the data pilot symbols and data symbols of the received signal, and a third neural network for equalizing the data symbols of the received signal.

[0020] In step 120, the output signal is demodulated to obtain received data.

[0021] In this embodiment, it is assumed that in a pilot-free communication system, the received signal can be represented by Y. Figure 2 A schematic diagram of a two-stream signal data frame structure provided by an embodiment, such as Figure 2 As shown, the data pilot symbol included in the received signal can be expressed as Y PW , the data symbol can be expressed as Y DataThe received data has been normalized by default, for OFDM signals to be detected through multipath fading channels. The receiver performs detection at the symbol granularity. Figure 3 A schematic diagram of a signal transmission system based on a cascade neural network is provided in an embodiment. The transmitting signal of the transmitter is divided into two parts: a data pilot symbol and a data symbol. Figure 3 As shown, both the data pilot symbol and the data symbol can carry the transmitted data information, and can be configured into different modulation modes according to actual performance. Figure 4 A block diagram of a high-speed receiver based on a cascaded neural network with dual drive of a data model is provided in an embodiment, such as Figure 4 As shown, the receiver can use three cascaded deep learning neural networks to implement the symbol-level processing flow, wherein the first neural network can be used to implement pre-demodulation of data pilot symbols, the second neural network can be used to implement channel estimation of data pilot symbols and data symbols, and the third neural network can be used to implement MIMO equalization of data symbols. The receiver in this embodiment can be dual-driven by data models. It can realize the signal detection function of traditional communication receivers in high-speed scenarios. The design of MIMO channel estimation and MIMO channel equalization AI modules can simultaneously support single-stream / multi-stream signal configuration, and can obtain a certain degree of performance improvement compared to traditional receivers in high-speed scenarios; the input data introduces the concept of data pilot symbols, which avoids the problem of pilot symbols occupying data symbol resources in traditional communication systems and improves the throughput of the system. At the same time, data pilot symbols and data symbols flexibly support the configuration of their respective symbol numbers and modulation modes, and can adapt to the performance requirements in high-speed communication scenarios. In addition, the structure of the neural network in this embodiment includes but is not limited to a deep neural network, a convolutional neural network, a residual neural network, or a residual neural network with an attention mechanism.

[0022] In one embodiment, the second neural network includes a first sub-network for performing channel estimation on data pilot symbols of the received signal and a second sub-network for performing channel estimation on data symbols of the received signal;

[0023] The input of the second sub-network includes data symbols of the received signal and the output of the third neural network.

[0024] In this embodiment, the second neural network includes a first subnetwork and a second subnetwork, the first subnetwork can be used to implement channel estimation of data pilot symbols, and the second subnetwork can be used to implement MIMO channel estimation of data symbols. The second subnetwork and the third neural network can form an iterative feedback loop structure, the input of the second subnetwork can include the data symbol of the received signal and the output of the third neural network, the demodulated data of the current data symbol is used as a known signal to perform channel estimation of the current data symbol position, and then the updated channel estimation result on the current symbol and the next symbol received data are used as the input of the equalizer network to perform the next symbol received data processing, which can better adapt to the characteristics of fast time-varying signals in high-speed channels and improve the performance of receivers in high-speed scenarios.

[0025] In one embodiment, the signal receiving method further includes:

[0026] Determine a first real number array according to the real part and the imaginary part of the data pilot symbol of the received signal;

[0027] The first real number array is rearranged based on the processing sub-band as the granularity to obtain first sub-band received data as the input of the first neural network.

[0028] In this embodiment, the data pilot symbol Y of the first symbol (m=0) is extracted from the received frequency domain signal Y. PW , the pre-demodulation neural network is a real neural network, so it is necessary to change Y PW The real and imaginary parts of are separated to form a set of real number arrays, which are determined as the first real number array, such as: [Re(Y PW ),Im(Y PW )], where Re(·) and Im(·) are operations to take the real part and the imaginary part respectively. Define the processing granularity of the neural network model as X resource units (RE) as a processing subband, and rearrange the first real number array with the processing subband as the granularity to N first subband receiving data Net 1,in =[N×X×2] ka , as the input of the first neural network.

[0029] In one embodiment, the signal receiving method further includes:

[0030] Generate bit stream data to be modulated, wherein the bit stream data passes through a wireless fading channel to obtain a sample receiving signal;

[0031] Determine a second real number array according to the real part and the imaginary part of the data pilot symbol of the sample received signal;

[0032] Rearranging the second real number array based on the processing sub-band as the granularity to obtain sample sub-band received data;

[0033] Modulating the bit stream data to obtain a real received signal;

[0034] Determine a third real number array according to the real part and the imaginary part of the data pilot symbol of the real received signal;

[0035] Rearranging the third real number array based on the processing sub-band as the granularity to obtain label sub-band received data;

[0036] The first neural network is trained according to the sample sub-band received data and the label sub-band received data.

[0037] In this embodiment, Figure 3 As shown, the bit stream data X to be modulated by the transmitting end data pilot symbol is generated bit The bit stream data passes through the wireless fading channel to obtain the sample received signal received by the receiving end. According to the data pilot symbol data Y of the sample received signal PW , separate the real part and the imaginary part of the data pilot symbol to obtain N*X second real number arrays, and the composition of a single second real number array is: [Re(Y PW ), Im(Y PW )], rearrange the second real number array based on the processing sub-band as the granularity to obtain sample sub-band received data as training samples.

[0038] The bit stream data X of the pilot symbol of the transmitting end bit Modulation generates the real received signal X PW , according to the data pilot symbol of the real received signal, the real part and the imaginary part of the data pilot symbol are separated to form N*X third real number arrays, wherein the jth third real number array is as follows: X′ PW,j =[Re(X PW ) Im(X PW )], the third real number array is rearranged based on the processing subband as the granularity, and N*X third real number arrays X′ are PW As the label of the training sample, the label subband received data is obtained. In this embodiment, under different fading channels, multiple sets of training samples and labels are generated and input into the first neural network for training to obtain the estimated value of the transmitted signal (The order is the same as the label data), and then with the label X′ PW Calculate the average loss function together.

[0039] After training is completed, the first neural network is deployed at the receiving end such as Figure 4 During the online signal detection, the first real number array Net obtained by processing and arranged in sub-bands is used. 1,in, as the input of the pre-demodulated first neural network. After the pre-demodulated first neural network is tested online, the output of the first neural network is obtained.

[0040] In one embodiment, the loss function used to train the first neural network includes at least one of the following: binary cross entropy loss function, Euclidean distance, and cosine distance.

[0041] In this embodiment, the loss function includes but is not limited to: a binary cross entropy loss function, using Euclidean distance or cosine distance as a loss function, or combining Euclidean distance and cosine distance as a loss function, etc. Based on the loss function, the Adam optimizer is used to train the neural network. When the loss function converges to a certain degree, the network training is completed.

[0042] In one embodiment, it further includes:

[0043] Performing a hard decision on the output of the first neural network based on a data pilot symbol modulation method to obtain a first decision result;

[0044] Determine a fourth real number array according to the real part and the imaginary part of the data pilot symbol of the received signal and the real part and the imaginary part of the first decision result;

[0045] The fourth real number array is rearranged based on the processing sub-band as the granularity to obtain second sub-band received data as the input of the first sub-network.

[0046] In this embodiment, based on the data pilot symbol modulation mode Q m,PW Input to the first neural network A hard decision is made to obtain a first decision result; then, the received data pilot signal Y corresponding to each receiving antenna is used PW The real part and the imaginary part of and the real part and the imaginary part of the first judgment result constitute a set of real number arrays, which are determined as the fourth real number array: The processing granularity size of the neural network model X RE is defined as the size of the processing subband, and the data pilot symbols received on all antennas are The output of the first neural network is rearranged with the processing sub-band as the granularity, and the data is rearranged into N second sub-bands receiving data. 2-1,in =[N×X×4] ka , as the input of the first sub-network, for channel estimation of data pilot symbols (m=0).

[0047] In one embodiment, it further includes:

[0048] Generate data pilot symbol modulation data, wherein the data pilot symbol modulation data passes through a wireless fading channel to obtain a sample data pilot symbol;

[0049] Determine a fifth real number array according to the data pilot symbol modulation data and the real part and the imaginary part of the sample data pilot symbol;

[0050] Rearranging the fifth real number array based on the processing sub-band as the granularity to obtain sample sub-band received data;

[0051] Determine a sixth real number array according to the real part and the imaginary part of the ideal channel through which the sample data pilot symbol passes;

[0052] Rearranging the sixth real number array based on the processing sub-band as the granularity to obtain label sub-band received data;

[0053] The first sub-network is trained according to the sample sub-band received data and the label sub-band received data.

[0054] In this embodiment, Figure 3 As shown, the transmitter data pilot symbol modulation data X is generated PW , the data pilot symbol modulated data passes through the high-speed wireless fading channel, and the received sample data pilot symbol data Y is obtained at the receiving end PW , determine N*X fifth real number arrays as training samples according to the data pilot symbol modulation data and the real part and imaginary part of the sample data pilot symbol, wherein a single fifth real number array is composed of: [Re(Y PW ),Im(Y PW ),Re(X PW ),Im(X PW )]. The fifth real number array is rearranged based on the processing sub-band as the granularity to obtain sample sub-band received data.

[0055] Get the full bandwidth ideal channel H through which the sample data pilot symbol passes PW , the sixth real number array is rearranged based on the processing subband as the granularity to form N*X out The sixth real number group H′ PW , where the jth real number array structure is as follows: H′ PW,j =[Re(H PW ),Im(H PW )],N*X out *2 real number arrays H′ pw The label subband receives data as training samples. The output granularity of the neural network here is: X in both single and double stream configurations. out =BW / N. Under different fading channels, multiple groups of corresponding training samples and labels are generated, and the first sub-network is trained according to the sample sub-band received data and the label sub-band received data to obtain the channel estimation value (The order is the same as the label data), and then with the label H′ pwThe loss function is calculated together. The loss function includes but is not limited to: Mean Square Error (MSE) loss function, using Euclidean distance or cosine distance as the loss function, or combining Euclidean distance and cosine distance as the loss function. Based on the loss function, the Adam optimizer is used to train the neural network. When the loss function converges to a certain degree, the network training is completed.

[0056] After training is completed, the first sub-network is deployed at the receiving end as Figure 4 During the online signal detection, the second sub-band received data Net arranged according to the sub-band is obtained. 2-1,in , as the input of the first sub-network. After the first sub-network is tested online, the output of the first sub-network is obtained. Channel estimation results.

[0057] In the embodiment of the present application, the ideal channel can be understood as an accurate channel estimation result obtained under known channel conditions. The full bandwidth can be understood as the total bandwidth occupied by the current scheduling signal (sample data pilot symbol or sample data symbol) in the frequency domain, that is, all processing subbands.

[0058] In one embodiment, the processing subband includes X resource elements RE;

[0059] In the case of a single-stream configuration, X = BW / N; in the case of a dual-stream configuration, X = BW / N / 2;

[0060] Wherein, BW is the total number of REs of the signal transmission bandwidth, and N is the total number of data symbols of the received signal.

[0061] In one embodiment, the signal receiving method further includes:

[0062] Determine a seventh real number array according to the real part and the imaginary part of the frequency domain data of the data symbol and the real part and the imaginary part of the channel estimation result output by the second neural network;

[0063] The seventh real number array is rearranged based on the processing sub-band as the granularity to obtain third sub-band received data as the input of the third neural network.

[0064] In this embodiment, according to the data symbol Y Data The real and imaginary parts of the frequency domain data and the channel estimation results output by the second neural network The real part and the imaginary part of the real number array are used to determine the seventh real number array; the seventh real number array is rearranged with the processing sub-band as the granularity to obtain the third sub-band received data Net 3,in , as the input of the third neural network, used for channel equalization of the first data symbol (m=1).

[0065] In one embodiment, for a first data symbol of the received signal, the input of the third neural network includes frequency domain data of the first data symbol and a channel estimation result of a data pilot symbol of the received signal;

[0066] For data symbols of the received signal other than the first data symbol, the input of the third neural network includes frequency domain data of the corresponding data symbol and a channel estimation result of the previous data symbol.

[0067] In this embodiment, the third neural network and the second sub-network are a cyclic iteration architecture with symbol m as the granularity. When m=1, the first data symbol is equalized, and the frequency domain data of the first data symbol is used as the And the channel estimation result of the received signal data pilot symbol As the input of the third neural network; when m>1, the subsequent data symbols are equalized in sequence according to the symbol, and the frequency domain data of data symbol m is used as ... The channel estimation output obtained by the second sub-network corresponding to the previous data symbol m-1 is As the input of the third neural network. The neural network input is constructed in the same way in the above two cases. Taking m>1 as an example, the input data real number array Net is constructed in the following way 3,in :Single stream configuration composition Understanding MIMO dual-stream configuration components

[0068] The processing granularity of the neural network model is defined as X REs, which is the size of the processing subband. and Data is rearranged based on the processing sub-band granularity into N sub-band data. Single stream configuration input: Net 3,in =[N×X×4], MIMO dual-stream configuration input: Net 3,in =[N×X×12].

[0069] In one embodiment, the channel equalization result of the third neural network for the mth data symbol of the received data is used to update the channel estimation result of the second sub-network for the mth symbol;

[0070] The channel estimation result after the m-th symbol is updated and the m+1-th data symbol of the received data are used as inputs of the third neural network.

[0071] In this embodiment, the channel equalization result of the mth data symbol is used to update the channel estimation of the current symbol, and the updated channel estimation result of the current second sub-network for the mth symbol is obtained by calculation and processing, and the channel estimation result after the update of the mth symbol is combined with the received data of the m+1th data symbol, and the combination is input into the channel equalization network to perform the equalization calculation of the m+1th symbol until the channel equalization of all data symbols is completed. Subsequently, the demodulation and decoding of the traditional receiver is performed.

[0072] In one embodiment, the method further includes:

[0073] Generate data symbol modulation data, wherein the data symbol modulation data passes through a wireless fading channel to obtain sample data symbols;

[0074] Determine an eighth real number array according to the real part and the imaginary part of the sample data symbol and the real part and the imaginary part of the ideal channel through which the sample data symbol passes;

[0075] Rearranging the eighth real number array based on the processing sub-band as the granularity to obtain sample sub-band received data;

[0076] Determine a ninth real number array according to the real part and the imaginary part of the data symbol modulation data;

[0077] Rearranging the ninth real number array based on the processing sub-band as the granularity to obtain label sub-band received data;

[0078] The third neural network is trained according to the sample sub-band received data and the label sub-band received data.

[0079] In this embodiment, Figure 3 As shown, the transmitting end performs signal processing to generate data symbols and modulate data The data symbol modulated data passes through the high-speed wireless fading channel to obtain the sample data symbol of data symbol m In addition, the ideal channel through which the sample data symbol m-1 passes is obtained According to the real part and the imaginary part of the sample data symbol and the real part and the imaginary part of the ideal channel through which the sample data symbol passes, N*X eighth real number arrays are obtained, and a single real number array single stream is composed of:

[0080] A single real array double stream consists of:

[0081] The eighth real number array is rearranged based on the processing sub-band as the granularity to obtain sample sub-band received data.

[0082] Then obtain the data symbol modulation data of the transmitting symbol m Separate its real and imaginary parts to form N*X ninth real number arrays, where the jth real number array has the following single stream composition: The dual streams are composed as follows: The ninth real number array is rearranged with the processing subband as the granularity to obtain the label subband received data as the training label of the third neural network. Data,m As the label of the training sample. Under different fading channels, multiple sets of corresponding training samples and labels are generated and input into the third neural network for training to obtain the estimated value of the transmitted signal. (The order is the same as the label data), and then with the label X′ Data,m The loss function is calculated together. The loss function includes but is not limited to: mean square error (MSE) loss function, using Euclidean distance or cosine distance as the loss function, or combining Euclidean distance and cosine distance as the loss function. Based on the loss function, the Adam optimizer is used to train the neural network. When the loss function converges to a certain degree, the network training is completed.

[0083] After the third neural network is trained, it can be deployed at the receiving end such as Figure 4 During the online signal detection, a seventh real number array is determined according to the real part and the imaginary part of the frequency domain data of the data symbol and the real part and the imaginary part of the channel estimation result output by the second neural network; the seventh real number array is rearranged according to the granularity of the processing subband to obtain the third subband received data Net after rearrangement according to the subband. 3,in , as the input of the third neural network. After the third neural network performs online detection, the third neural network output of symbol m is obtained It is a set of real number arrays that need to be combined into complex values ​​for hard decision processing of the modulation constellation position.

[0084] In one embodiment, it further includes:

[0085] Combining the outputs of the third neural network into complex values, performing hard decision of the modulation constellation position on the complex values, and obtaining a second decision result;

[0086] Determine a tenth real number array according to the real part and the imaginary part of the data symbol of the received signal and the real part and the imaginary part of the second decision result;

[0087] The tenth real number array is rearranged based on the processing sub-band as the granularity to obtain fourth sub-band received data as the input of the second sub-network.

[0088] In this embodiment, the output of the third neural network Combining into a complex value, performing a hard decision of the modulation constellation position on the complex value, and obtaining a second decision result. The real and imaginary parts of the data symbol and the output of the third neural network The real and imaginary parts of the second decision result determine the tenth real number array, a single stream consisting of: Dual flow composition: The processing granularity of the neural network model, X RE, is defined as the size of the processing sub-band. The tenth real number array is rearranged with the processing sub-band as the granularity to receive data in N fourth sub-bands. Single stream: Net 2,in =[N×X×4] ka , double stream: Net 2,in =[N×X×8] ka ; As the input of the second sub-network, used for channel estimation of symbol m.

[0089] In one embodiment, it further includes:

[0090] Generate data symbol modulation data, wherein the data symbol modulation data passes through a wireless fading channel to obtain sample data symbols;

[0091] Determine an eleventh real number array according to the real part and the imaginary part of the data symbol modulation data and the real part and the imaginary part of the sample data symbol;

[0092] Rearranging the eleventh real number array based on the processing sub-band as the granularity to obtain sample sub-band received data;

[0093] Determine a twelfth real number array according to the real part and the imaginary part of the ideal channel through which the sample data symbol passes;

[0094] Rearranging the twelfth real number array based on the processing sub-band as the granularity to obtain label sub-band received data;

[0095] The second sub-network is trained according to the sample sub-band received data and the label sub-band received data.

[0096] In this embodiment, Figure 3 As shown, the transmitter data symbol modulation data is generated Data symbol modulation data After a high-speed wireless fading channel, the sample data symbol data received by the receiving end is obtained According to the real and imaginary parts of the data symbol modulation data and the real and imaginary parts of the sample data symbol, N*X eleventh real number arrays are obtained as training samples, wherein a single real number array and a single stream are composed of: The dual stream consists of: The eleventh real number array is rearranged with the processing sub-band as the granularity to obtain sample sub-band received data as training samples.

[0097] According to the sample data symbol data Full bandwidth ideal channel The real part and the imaginary part of the twelfth real number array are determined, and the twelfth real number array is rearranged with the processing subband as the granularity to form N*Xout twelfth real number arrays H′ Data,m , where the j-th real number array is composed as follows:

[0098] Single stream: H′ Data,m,j =[Re(H Data ),Im(H Data )]

[0099] Double Stream:

[0100]

[0101] Among them, single stream [N*X out *2] / Dual Stream[N*X out *8] real number array H′ Data,m The label subband receives data as labels for training samples. The output granularity of the neural network is: X in both single and double stream configurations. out =BW / N. Under different fading channels, multiple sets of corresponding training samples and labels are generated according to the sample subband received data and the label subband received data, and are input into the second subnetwork for training to obtain channel estimation values (The order is the same as the label data), and then with the label H′ Data,m Calculate the loss function together. Based on the loss function, use the Adam optimizer to train the neural network. When the loss function converges to a certain degree, the network training is completed.

[0102] After training is completed, the second sub-network is deployed at the receiving end as Figure 4 During the online signal detection, the output of the third neural network is combined into a complex value, and a hard decision of the position of the modulation constellation diagram is performed on the complex value to obtain a second decision result; a tenth real number array is determined according to the real part and imaginary part of the data symbol of the received signal and the real part and imaginary part of the second decision result; the tenth real number array is rearranged with the processing subband as the granularity to obtain the fourth subband received data Net 2,in , as the input of the second sub-network. After the second sub-network is tested online, the output of the second sub-network is obtained. The channel estimation result of the mth symbol is used for channel equalization of the symbol m+1 data.

[0103] In one embodiment, the loss function used to train the second neural network or the third neural network includes at least one of the following: mean square error loss function, Euclidean distance, and cosine distance.

[0104] The signal receiving method is exemplarily described below through specific embodiments.

[0105] Embodiment 1

[0106] Assume a pilot-free communication system in a high-speed scenario. In the OFDM MIMO system, the user data is dual-stream, the transmitter has P = 2 antenna ports and Tx = 2 physical transmit antennas, and the receiver has Rx = 2 physical receive antennas. The number of RBs N for data pilot symbols and data symbol scheduling in the frequency domain is RB =24, these 24 RBs can be divided into N sub = 4 sub-bands, each sub-band RE data =6*12=72 subcarriers. The data pilot occupies N=1 OFDM symbol, and the corresponding modulation mode is Q m,PW =2, Quadrature Phase Shift Keying (QPSK) modulation. The data pilot symbols of different data streams are arranged in a staggered manner between streams in the frequency domain, that is, each layer occupies RE PW = 36 subcarriers; data occupies M = 13 OFDM symbols, corresponding to the modulation mode Q m,data =4,16 Quadrature Amplitude Modulation (QAM).

[0107] This embodiment introduces the construction of the input of the data pilot symbol pre-demodulation neural network Net1 (i.e., the first neural network), and obtains the frequency domain estimation value of the transmitted data pilot symbol through online detection of the data pilot symbol pre-demodulation neural network Net1. The data are normalized by default.

[0108] 1) Frequency domain data generation.

[0109] The frequency domain signal Y received by the two receiving antennas is divided into data pilot symbol frequency domain data Y PW and data symbol frequency domain data Y Data Since it is a two-stream MIMO system with staggered data pilot streams, the two receiving antennas RKa=1,2 will receive data sent from the two transmitting ports P=1,2 respectively (port 1 is on the even subcarriers, and port 2 is on the odd subcarriers). represents the data pilot frequency domain signal received by receiving antenna i from transmitting port j, then the data pilot symbol can be extracted and decomposed into the following matrix:

[0110]

[0111] Its dimension is P*Rx*(N*N sub )*RE PW , that is, 2*2*4*36, extract the receiving antennas corresponding to different ports for pre-demodulation processing, and the extraction matrix is The format is rearranged into 8*36 data groups.

[0112] 2) Construct neural network model Net1 input.

[0113] The input granularity of the data pilot symbol pre-demodulation neural network model Net1 is RE per subband per stream PW =36, divide the frequency domain data of the data pilot symbol according to the granularity, and separate the real and imaginary parts, according to [Re(Y PW ), Im(Y PW )] structure to form a real number array as the input Net of the pre-demodulation neural network Net1 1,in , whose dimension is 8*36*2.

[0114] 3) Generate training samples and labels, and perform model training on Net1.

[0115] Send bit stream X bit Proceed as Figure 3 The transmitting end signal is processed through the high-speed channel to obtain the frequency domain signal Y of the data pilot in step 1). PW , and construct a real number array as training samples in the format of step 2), with a dimension of 8*36*2. bit The generated data pilot X PW Similarly, follow step 1) to rearrange the real and imaginary parts to obtain X′ PW,j =[Re(X PW ) Im(X PW )], whose dimension is 8*36*2, and serves as the training label of neural network Net1.

[0116] In different high-speed channel scenarios, multiple sets of training samples are generated in the above manner and input into the data pilot pre-demodulation neural network model Net1 for training to obtain the frequency domain estimation value of the transmitted data pilot signal. Will and training labels X′ PW Calculate the loss function. The loss function here uses the binary cross entropy loss function. Based on the loss function, the Adam optimizer is used to train the neural network until the model converges.

[0117] 4) Perform online detection processing on the neural network model Net1.

[0118] After training is completed, Net1 is deployed at the receiving end as follows Figure 4 Perform online signal detection at the position shown. 1,in As the input of the pre-demodulation neural network, the neural network is a convolutional neural network (CNN), and the pre-demodulation output result is obtained Its dimension is 8*36. The output is based on the data pilot symbol modulation mode Q m,PW After hard decision is made, it is used as input of the channel estimation neural network Net2-1 for subsequent processing.

[0119] Embodiment 2

[0120] Assume a pilot-free communication system in a high-speed scenario. In the OFDM MIMO system, the user data is dual-stream, the transmitter has P = 2 antenna ports and Tx = 2 physical transmit antennas, and the receiver has Rx = 2 physical receive antennas. The number of RBs N for data pilot symbols and data symbol scheduling in the frequency domain is RB =24, these 24 RBs can be divided into N sub = 4 sub-bands, each sub-band RE data =6*12=72 subcarriers. The data pilot occupies N=1 OFDM symbol, and the corresponding modulation mode is Q m,PW =2 (QPSK modulation), the data pilot symbols of different data streams are arranged in a staggered manner between streams in the frequency domain, that is, each layer occupies RE PW = 36 subcarriers; data occupies M = 13 OFDM symbols, corresponding to the modulation mode Q m,data =4 (16QAM modulation)

[0121] This embodiment introduces the construction of the input of the data pilot symbol channel estimation neural network Net2-1 (i.e., the first sub-network), and obtains the channel estimation result of the data pilot symbol through online detection of the data pilot symbol channel estimation neural network Net2-1. The data are normalized by default.

[0122] 1) Build the neural network model Net2-1 input.

[0123] The received data pilot frequency domain data Y PW And the pre-demodulation model Net1 output The hard decision result is used as the input of the data pilot channel estimation neural network Net2-1. The input granularity of the neural network is RE per subband per stream. PW =36, divide the frequency domain data of the data pilot symbol according to the granularity, and separate its real and imaginary parts, according to The structure of the real number array Net is 16*36*4 2-1,in , Net 2-1,in As the input of the data pilot channel estimation neural network Net2-1.

[0124] 2) Generate training samples and labels, and perform model training on Net2-1.

[0125] according to Figure 3 After the transmitting end signal is processed and passed through the high-speed channel, the frequency domain signal Y of the data pilot in step 1) is obtained PW , and press [Re(Y PW ), Im(Y PW ), Re(Y PW ), Im(Y PW )] format to construct a real number array as a training sample, whose dimension is 16*36*4. In addition, the ideal channel H with full scheduling bandwidth 24RB through which the data pilot symbol passes is obtained. ideal,PW , and rearrange according to the real and imaginary parts to obtain the real number array H′ ideal,PW =[Re(H ideal,PW ) Im(H ideal,PW )], whose dimension is 16*72*2, and serves as the training label of the neural network Net2-1.

[0126] In different scenarios, multiple sets of training samples are generated in the above manner and input into the data pilot channel estimation neural network model Net2-1 for training to obtain the channel estimation value of the sent data pilot signal. Will and training labels H′ ideal,PW The mean square error (MSE) loss function is input to calculate the loss. The Adam optimizer is then used to train the neural network until the model converges.

[0127] 3) Perform online detection and processing on the neural network model Net2-1.

[0128] After training is completed, Net2-1 is deployed at the receiving end as follows Figure 4 Perform online signal detection at the position shown. 2-1,in As the input of the data pilot channel estimation neural network, the neural network is a CNN convolutional neural network, and the output result after channel estimation and interpolation is obtained Its dimension is 16*72*2. This result is used for channel equalization of the first data symbol.

[0129] Embodiment 3

[0130] Assume a pilot-free communication system in a high-speed scenario. In the OFDM MIMO system, the user data is dual-stream, the transmitter has P = 2 antenna ports and Tx = 2 physical transmit antennas, and the receiver has Rx = 2 physical receive antennas. The number of RBs N for data pilot symbols and data symbol scheduling in the frequency domain is RB =24, these 24 RBs can be divided into N sub = 4 sub-bands, each sub-band RE data =6*12=72 subcarriers. The data pilot occupies N=1 OFDM symbol, and the corresponding modulation mode is Q m,PW =2 (QPSK modulation), the data pilot symbols of different data streams are arranged in a staggered manner between streams in the frequency domain, that is, each layer occupies RE PW = 36 subcarriers; data occupies M = 13 OFDM symbols, corresponding to the modulation mode Q m,data =4 (16QAM modulation).

[0131] This embodiment introduces the construction of the input of the data symbol channel estimation neural network Net2-2 (i.e., the second sub-network), and obtains the channel estimation result of the data symbol through the online detection of the data symbol channel estimation neural network Net2-2. The data are normalized by default.

[0132] 1) Build the neural network model Net2-2 input.

[0133] The frequency domain data of the currently received data symbol m (dimension is 2*4*72, complex number) and the output of the channel equalization neural network model Net3 The hard decision result (dimension is 2*4*72, complex number) is used as the input of the data symbol channel estimation neural network Net2-2. The input granularity of the neural network is RE per subband per stream data =72, divide the data symbol frequency domain data according to the granularity, and separate its real and imaginary parts, according to The structure is rearranged to form a 4*72*8 real number array Net 2-2,in , Net 2-2,in As the input of the data symbol channel estimation neural network Net2-2.

[0134] 2) Generate training samples and labels, and perform model training on Net2-2.

[0135] according to Figure 3 Structure, generate the transmitter data symbol modulation data After a high-speed wireless fading channel, the received data symbol data is obtained at the receiving end. and press The format of constructing a real number array as a training sample has a dimension of 4*72*8. In addition, the ideal channel through which the data symbol passes is obtained. And rearrange according to the real and imaginary parts to get a real number array Its dimension is 4*72*8 and it serves as the training label for the neural network Net2-2.

[0136] In different scenarios, multiple sets of training samples are generated in the above manner to input the data symbol channel estimation neural network model Net2-2 for training, and the channel estimation value of the transmitted data symbol m is obtained. Will and training labels H′ ideal,Data The mean square error (MSE) loss function is input to calculate the loss. The Adam optimizer is then used to train the neural network until the model converges.

[0137] 3) Perform online detection and processing on the neural network model Net2-2.

[0138] After training is completed, Net2-2 is deployed at the receiving end as follows Figure 4 Perform online signal detection at the position shown. 2-2,in As the input of the data symbol channel estimation neural network, the neural network is a CNN convolutional neural network, and the output result after channel estimation and interpolation is obtained Its dimension is 4*72*8. The result is used for channel equalization of the m+1th data symbol.

[0139] Embodiment 4

[0140] Assume a pilot-free communication system in a high-speed scenario. In the OFDM MIMO system, the user data is dual-stream, the transmitter has P = 2 antenna ports and Tx = 2 physical transmit antennas, and the receiver has Rx = 2 physical receive antennas. The number of RBs N for data pilot symbols and data symbol scheduling in the frequency domain is RB =24, these 24 RBs can be divided into N sub = 4 sub-bands, each sub-band RE data =6*12=72 subcarriers. The data pilot occupies N=1 OFDM symbol, and the corresponding modulation mode is Q m,PW =2 (QPSK modulation), the data pilot symbols of different data streams are arranged in a staggered manner between streams in the frequency domain, that is, each layer occupies RE PW = 36 subcarriers; data occupies M = 13 OFDM symbols, corresponding to the modulation mode Q m,data =4 (16QAM modulation).

[0141] This embodiment introduces constructing the input of the channel equalization neural network Net3 (i.e., the third neural network), and obtaining the channel equalization result through online detection of the channel equalization neural network Net3. The data are normalized by default.

[0142] 1) Build the neural network model Net3 input.

[0143] The frequency domain data of the currently received data symbol m (dimension is 2*4*72, complex number) and the output obtained by the neural network model Net2-2 of the previous data symbol m-1 (The dimension is 2*2*4*72, complex number, the first data symbol uses the channel estimation result of the data pilot symbol ) is used as the input of the channel equalization neural network Net3. The input granularity of the neural network is RE per subband per stream. data =72, divide the data symbol frequency domain data according to the granularity, and separate its real and imaginary parts, according to The structure is rearranged to form a 4*72*12 real number array Net 3,in , Net 3,in As the input of the channel equalization neural network Net3.

[0144] 2) Generate training samples and labels and perform model training on Net3.

[0145] according to Figure 3 After the transmitting end signal is processed and passed through the high-speed channel, the frequency domain signal of the data symbol m in step 1) is obtained. In addition, the ideal channel through which data symbol m-1 passes is obtained and press The real and imaginary parts of the format are rearranged to construct a real number array as a training sample, whose dimension is 4*72*12. Then obtain the frequency domain modulation data of the transmitting symbol m Its dimension is 4*72*4, which serves as the training label of the neural network Net3.

[0146] In different high-speed channel scenarios, multiple sets of training samples are generated in the above manner and input into the channel equalization neural network model Net3 for training to obtain the channel equalization result estimation value of the transmitted data symbol m Will and training labels The mean square error (MSE) loss function is input to calculate the loss. The Adam optimizer is then used to train the neural network until the model converges.

[0147] 3) Perform online detection and processing on the neural network model Net3.

[0148] After training is completed, Net3 is deployed on the receiving end as follows Figure 4 Perform online signal detection at the position shown. 3,in As the input of the channel equalization neural network, the neural network is a CNN convolutional neural network, and the output of the equalization network is obtained Its dimensions are 4*72*4.

[0149] The output is a 4*72 group structure. Each group can be combined into a complex value on 2 ports. The combined 4*72*2 complex value data is processed by hard decision of the modulation constellation position, and then the real and imaginary parts of the hard decision result are re-expanded to obtain a real number array with a dimension of 4*72*4. This array is used for channel estimation of the m+1th data symbol and subsequent receiver processing.

[0150] The embodiment of the present application also provides a signal receiving device. Figure 5 FIG. 1 is a schematic diagram of a signal receiving device provided by an embodiment. Figure 5 As shown, the signal receiving device includes:

[0151] The input module 210 is configured to input a received signal into a receiver to obtain an output signal, wherein the received signal includes a data pilot symbol and a data symbol; the receiver includes the following cascaded neural networks: a first neural network for pre-demodulating the data pilot symbol of the received signal, a second neural network for performing channel estimation on the data pilot symbol and the data symbol of the received signal, and a third neural network for equalizing the data symbol of the received signal.

[0152] The demodulation module 220 is configured to demodulate the output signal to obtain received data.

[0153] The signal receiving device of this embodiment is a high-speed receiver system based on a cascaded neural network driven by a data model. It can realize the signal detection function of a traditional communication receiver in a high-speed scenario. The design of the MIMO channel estimation and MIMO channel equalization AI modules can simultaneously support single-stream / multi-stream signal configurations, and can achieve a certain degree of performance improvement compared to traditional receivers in high-speed scenarios. The input data introduces the concept of data pilot symbols, which avoids the problem of pilot symbols occupying data symbol resources in traditional communication systems and improves the system throughput. At the same time, data pilot symbols and data symbols flexibly support the configuration of their respective symbol numbers and modulation methods, and can adapt to the performance requirements in high-speed communication scenarios.

[0154] In one embodiment, the second neural network includes a first sub-network for performing channel estimation on data pilot symbols of the received signal and a second sub-network for performing channel estimation on data symbols of the received signal;

[0155] The input of the second sub-network includes data symbols of the received signal and the output of the third neural network.

[0156] In one embodiment, the signal receiving device further includes:

[0157] A first determination module is configured to determine a first real number array according to the real part and the imaginary part of the data pilot symbol of the received signal;

[0158] The first rearrangement module is configured to rearrange the first real number array based on a processing sub-band as a granularity to obtain first sub-band received data as an input of the first neural network.

[0159] In one embodiment, the signal receiving device further includes:

[0160] A first generating module is configured to generate bit stream data to be modulated, wherein the bit stream data is passed through a wireless fading channel to obtain a sample receiving signal;

[0161] A second determination module is configured to determine a second real number array according to the real part and the imaginary part of the data pilot symbol of the sample received signal;

[0162] A second rearrangement module is configured to rearrange the second real number array based on a processing subband as a granularity to obtain sample subband received data;

[0163] A modulation module, configured to modulate the bit stream data to obtain a real received signal;

[0164] A third determination module is configured to determine a third real number array according to the real part and the imaginary part of the data pilot symbol of the real received signal;

[0165] A third rearrangement module is configured to rearrange the third real number array based on a processing subband as a granularity to obtain label subband received data;

[0166] The first training module is configured to train the first neural network according to the sample sub-band received data and the label sub-band received data.

[0167] In one embodiment, the loss function used to train the first neural network includes at least one of the following: binary cross entropy loss function, Euclidean distance, and cosine distance.

[0168] In one embodiment, the device further comprises:

[0169] A first decision module is configured to perform a hard decision on the output of the first neural network based on a data pilot symbol modulation mode to obtain a first decision result;

[0170] a fourth determination module, configured to determine a fourth real number array according to the real part and the imaginary part of the data pilot symbol of the received signal and the real part and the imaginary part of the first decision result;

[0171] The fourth rearrangement module is configured to rearrange the fourth real number array based on the granularity of the processing sub-band to obtain second sub-band received data as the input of the first sub-network.

[0172] In one embodiment, the device further comprises:

[0173] A second generating module is configured to generate data pilot symbol modulation data, wherein the data pilot symbol modulation data is passed through a wireless fading channel to obtain a sample data pilot symbol;

[0174] A fifth determination module, configured to determine a fifth real number array according to the data pilot symbol modulation data and the real part and the imaginary part of the sample data pilot symbol;

[0175] a fifth rearrangement module, configured to rearrange the fifth real number array based on a processing subband as a granularity to obtain sample subband received data;

[0176] A sixth determination module, configured to determine a sixth real number array according to the real part and the imaginary part of the ideal channel through which the sample data pilot symbol passes;

[0177] a sixth rearrangement module, configured to rearrange the sixth real number array based on a processing sub-band as a granularity to obtain label sub-band received data;

[0178] The second training module is configured to train the first sub-network according to the sample sub-band received data and the label sub-band received data.

[0179] In one embodiment, the processing subband includes X resource elements RE;

[0180] In the case of a single-stream configuration, X = BW / N; in the case of a dual-stream configuration, X = BW / N / 2;

[0181] Wherein, BW is the total number of REs of the signal transmission bandwidth, and N is the total number of data symbols of the received signal.

[0182] In one embodiment, the signal receiving device further includes:

[0183] a seventh determination module, configured to determine a seventh real number array according to the real part and the imaginary part of the frequency domain data of the data symbol and the real part and the imaginary part of the channel estimation result output by the second neural network;

[0184] The seventh rearrangement module is configured to rearrange the seventh real number array based on the processing sub-band as the granularity to obtain third sub-band received data as the input of the third neural network.

[0185] In one embodiment, for a first data symbol of the received signal, the input of the third neural network includes frequency domain data of the first data symbol and a channel estimation result of a data pilot symbol of the received signal;

[0186] For data symbols of the received signal other than the first data symbol, the input of the third neural network includes frequency domain data of the corresponding data symbol and a channel estimation result of the previous data symbol.

[0187] In one embodiment, the channel equalization result of the third neural network for the mth data symbol of the received data is used to update the channel estimation result of the second sub-network for the mth symbol;

[0188] The channel estimation result after the m-th symbol is updated and the m+1-th data symbol of the received data are used as inputs of the third neural network.

[0189] In one embodiment, the signal receiving device further includes:

[0190] A third generating module is configured to generate data symbol modulation data, wherein the data symbol modulation data passes through a wireless fading channel to obtain a sample data symbol;

[0191] An eighth determination module, configured to determine an eighth real number array according to the real part and the imaginary part of the sample data symbol and the real part and the imaginary part of the ideal channel through which the sample data symbol passes;

[0192] an eighth rearrangement module, configured to rearrange the eighth real number array based on a processing sub-band as a granularity to obtain sample sub-band received data;

[0193] A ninth determination module, configured to determine a ninth real number array according to the real part and the imaginary part of the data symbol modulation data;

[0194] a ninth rearrangement module, configured to rearrange the ninth real number array based on a processing subband as a granularity to obtain label subband received data;

[0195] The third training module is configured to train the third neural network according to the sample sub-band received data and the label sub-band received data.

[0196] In one embodiment, the device further comprises:

[0197] A second decision module is configured to combine the outputs of the third neural network into complex values, perform hard decision of the modulation constellation position on the complex values, and obtain a second decision result;

[0198] a tenth determination module, configured to determine a tenth real number array according to the real part and the imaginary part of the data symbol of the received signal and the real part and the imaginary part of the second decision result;

[0199] The tenth rearrangement module is configured to rearrange the tenth real number array based on the processing sub-band as the granularity to obtain fourth sub-band received data as the input of the second sub-network.

[0200] In one embodiment, it further includes:

[0201] A fourth generating module is configured to generate data symbol modulation data, wherein the data symbol modulation data passes through a wireless fading channel to obtain a sample data symbol;

[0202] an eleventh determination module, configured to determine an eleventh real number array according to the real part and the imaginary part of the data symbol modulation data and the real part and the imaginary part of the sample data symbol;

[0203] an eleventh rearrangement module, configured to rearrange the eleventh real number array based on a processing sub-band as a granularity to obtain sample sub-band received data;

[0204] A twelfth determination module, configured to determine a twelfth real number array according to the real part and the imaginary part of the ideal channel through which the sample data symbol passes;

[0205] A twelfth rearrangement module is configured to rearrange the twelfth real number array based on a processing sub-band as a granularity to obtain label sub-band received data;

[0206] The fourth training module is configured to train the second sub-network according to the sample sub-band received data and the label sub-band received data.

[0207] In one embodiment, the loss function used to train the second neural network or the third neural network includes at least one of the following: mean square error loss function, Euclidean distance, and cosine distance.

[0208] The signal receiving device proposed in this embodiment and the signal receiving method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to any of the above embodiments, and this embodiment has the same beneficial effects as the signal receiving method.

[0209] The embodiment of the present application also provides a communication node. Figure 6 A hardware structure diagram of a communication node provided by an embodiment is shown in FIG. Figure 6As shown, the communication node provided by the present application includes a processor 310 and a memory 320; the processor 310 in the communication node may be one or more, Figure 6 A processor 310 is taken as an example; the memory 320 is configured to store one or more programs; the one or more programs are executed by the one or more processors 310, so that the one or more processors 310 implement the signal receiving method as described in the embodiment of the present application.

[0210] The communication source node further includes: a communication device 330 , an input device 340 and an output device 350 .

[0211] The processor 310, memory 320, communication device 330, input device 340 and output device 350 in the communication node may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0212] The input device 340 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the electronic device. The output device 350 may include a display device such as a display screen.

[0213] The communication device 330 may include a receiver and a transmitter. The communication device 330 is configured to perform information transmission and reception communication according to the control of the processor 310.

[0214] The memory 320, as a computer-readable storage medium, may be configured to store software programs, computer executable programs, and modules, such as program instructions / modules corresponding to the signal receiving method described in the embodiment of the present application (e.g., input module 210, demodulation module 220 in the signal receiving device). The memory 320 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the communication node, etc. In addition, the memory 320 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 320 may further include a memory remotely arranged relative to the processor 310, and these remote memories may be connected to the communication node via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0215] The present application also provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the signal receiving method described in any one of the embodiments of the present application is implemented. The signal receiving method is applied to a communication node, and includes:

[0216] Inputting a received signal into a receiver to obtain an output signal, wherein the received signal includes a data pilot symbol and a data symbol; the receiver includes the following cascaded neural networks: a first neural network for pre-demodulating the data pilot symbol of the received signal, a second neural network for performing channel estimation on the data pilot symbol and the data symbol of the received signal, and a third neural network for equalizing the data symbol of the received signal;

[0217] The output signal is demodulated to obtain received data.

[0218] The computer storage medium of the embodiment of the present application can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (Random Access Memory, RAM), a read-only memory (Read Only Memory, ROM), an erasable programmable read-only memory (ErasableProgrammable Read Only Memory, EPROM), a flash memory, an optical fiber, a portable CD-ROM, an optical storage device, a magnetic storage device, or any suitable combination of the above. The computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, a device or a device or used in combination with it.

[0219] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to: electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0220] The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, radio frequency (RF), etc., or any suitable combination of the above.

[0221] The computer program code for performing the operation of the present application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and also conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect through the Internet).

[0222] The above description is merely an exemplary embodiment of the present application and is not intended to limit the protection scope of the present application.

[0223] It will be appreciated by those skilled in the art that the term user terminal covers any suitable type of wireless user equipment, such as a mobile phone, a portable data processing device, a portable web browser or a vehicle-mounted mobile station.

[0224] In general, various embodiments of the present application can be implemented in hardware or dedicated circuits, software, logic or any combination thereof. For example, some aspects can be implemented in hardware, while other aspects can be implemented in firmware or software that can be executed by a controller, microprocessor or other computing device, although the present application is not limited thereto.

[0225] Embodiments of the present application may be implemented by executing computer program instructions by a data processor of a mobile device, for example in a processor entity, or by hardware, or by a combination of software and hardware. The computer program instructions may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages.

[0226] The block diagram of any logic flow in the drawings of the present application may represent program steps, or may represent interconnected logic circuits, modules and functions, or may represent a combination of program steps and logic circuits, modules and functions. A computer program may be stored in a memory. The memory may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as but not limited to read-only memory (ROM), random access memory (RAM), optical memory devices and systems (digital versatile discs (DVD) or compact disks (CD), etc. Computer-readable media may include non-transitory storage media. The data processor may be of any type suitable for the local technical environment, such as but not limited to general-purpose computers, special-purpose computers, microprocessors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FGPAs) and processors based on multi-core processor architectures.

[0227] By way of exemplary and non-limiting examples, a detailed description of exemplary embodiments of the present application has been provided above. However, various modifications and adjustments to the above embodiments will be apparent to those skilled in the art when considered in conjunction with the accompanying drawings and claims, but will not depart from the scope of the present application. Therefore, the proper scope of the present application will be determined according to the claims.

Claims

1. A signal receiving method, characterized in that: include: Inputting a received signal into a receiver to obtain an output signal, wherein the received signal includes a data pilot symbol and a data symbol; the receiver includes the following cascaded neural networks: a first neural network for pre-demodulating the data pilot symbol of the received signal, a second neural network for performing channel estimation on the data pilot symbol and the data symbol of the received signal, and a third neural network for equalizing the data symbol of the received signal; The output signal is demodulated to obtain received data.

2. The method according to claim 1, characterized in that The second neural network includes a first sub-network for performing channel estimation on the data pilot symbols of the received signal and a second sub-network for performing channel estimation on the data symbols of the received signal; The input of the second sub-network includes data symbols of the received signal and the output of the third neural network.

3. The method according to claim 1, characterized in that: Also includes: Determine a first real number array according to the real part and the imaginary part of the data pilot symbol of the received signal; The first real number array is rearranged based on the processing sub-band as the granularity to obtain first sub-band received data as the input of the first neural network.

4. The method according to claim 1, characterized in that Also includes: Generate bit stream data to be modulated, wherein the bit stream data passes through a wireless fading channel to obtain a sample receiving signal; Determine a second real number array according to the real part and the imaginary part of the data pilot symbol of the sample received signal; Rearranging the second real number array based on the processing sub-band as the granularity to obtain sample sub-band received data; Modulating the bit stream data to obtain a real received signal; Determine a third real number array according to the real part and the imaginary part of the data pilot symbol of the real received signal; Rearranging the third real number array based on the processing sub-band as the granularity to obtain label sub-band received data; The first neural network is trained according to the sample sub-band received data and the label sub-band received data.

5. The method according to claim 1, characterized in that The loss function used for training the first neural network includes at least one of the following: a binary cross entropy loss function, a Euclidean distance, and a cosine distance.

6. The method according to claim 2, characterized in that Also includes: Performing a hard decision on the output of the first neural network based on a data pilot symbol modulation method to obtain a first decision result; Determine a fourth real number array according to the real part and the imaginary part of the data pilot symbol of the received signal and the real part and the imaginary part of the first decision result; The fourth real number array is rearranged based on the processing sub-band as the granularity to obtain second sub-band received data as the input of the first sub-network.

7. The method according to claim 2, characterized in that Also includes: Generate data pilot symbol modulation data, wherein the data pilot symbol modulation data passes through a wireless fading channel to obtain a sample data pilot symbol; Determine a fifth real number array according to the data pilot symbol modulation data and the real part and the imaginary part of the sample data pilot symbol; Rearranging the fifth real number array based on the processing sub-band as the granularity to obtain sample sub-band received data; Determine a sixth real number array according to the real part and the imaginary part of the ideal channel through which the sample data pilot symbol passes; Rearranging the sixth real number array based on the processing sub-band as the granularity to obtain label sub-band received data; The first sub-network is trained according to the sample sub-band received data and the label sub-band received data.

8. The method according to claim 6, characterized in that The processing subband includes X resource elements RE; In the case of a single-stream configuration, X = BW / N; in the case of a dual-stream configuration, X = BW / N / 2; Wherein, BW is the total number of REs of the signal transmission bandwidth, and N is the total number of data symbols of the received signal.

9. The method according to claim 1, characterized in that: Also includes: Determine a seventh real number array according to the real part and the imaginary part of the frequency domain data of the data symbol and the real part and the imaginary part of the channel estimation result output by the second neural network; The seventh real number array is rearranged based on the processing sub-band as the granularity to obtain third sub-band received data as the input of the third neural network.

10. The method according to claim 9, characterized in that For a first data symbol of the received signal, the input of the third neural network includes frequency domain data of the first data symbol and a channel estimation result of a data pilot symbol of the received signal; For data symbols of the received signal other than the first data symbol, the input of the third neural network includes frequency domain data of the corresponding data symbol and a channel estimation result of the previous data symbol.

11. The method according to claim 2, characterized in that The channel equalization result of the third neural network for the mth data symbol of the received data is used to update the channel estimation result of the second sub-network for the mth symbol; The channel estimation result after the m-th symbol is updated and the m+1-th data symbol of the received data are used as inputs of the third neural network.

12. The method according to claim 1, characterized in that Also includes: Generate data symbol modulation data, wherein the data symbol modulation data passes through a wireless fading channel to obtain sample data symbols; Determine an eighth real number array according to the real part and the imaginary part of the sample data symbol and the real part and the imaginary part of the ideal channel through which the sample data symbol passes; Rearranging the eighth real number array based on the processing sub-band as the granularity to obtain sample sub-band received data; Determine a ninth real number array according to the real part and the imaginary part of the data symbol modulation data; Rearranging the ninth real number array based on the processing sub-band as the granularity to obtain label sub-band received data; The third neural network is trained according to the sample sub-band received data and the label sub-band received data.

13. The method according to claim 2, characterized in that Also includes: Combining the outputs of the third neural network into complex values, performing hard decision of the modulation constellation position on the complex values, and obtaining a second decision result; Determine a tenth real number array according to the real part and the imaginary part of the data symbol of the received signal and the real part and the imaginary part of the second decision result; The tenth real number array is rearranged based on the processing sub-band as the granularity to obtain fourth sub-band received data as the input of the second sub-network.

14. The method according to claim 2, characterized in that Also includes: Generate data symbol modulation data, wherein the data symbol modulation data passes through a wireless fading channel to obtain sample data symbols; Determine an eleventh real number array according to the real part and the imaginary part of the data symbol modulation data and the real part and the imaginary part of the sample data symbol; Rearranging the eleventh real number array based on the processing sub-band as the granularity to obtain sample sub-band received data; Determine a twelfth real number array according to the real part and the imaginary part of the ideal channel through which the sample data symbol passes; Rearranging the twelfth real number array based on the processing sub-band as the granularity to obtain label sub-band received data; The second sub-network is trained according to the sample sub-band received data and the label sub-band received data.

15. The method according to claim 1, characterized in that The loss function used to train the second neural network or the third neural network includes at least one of the following: mean square error loss function, Euclidean distance, and cosine distance.

16. A communication node, characterized in that: include: memory, and one or more processors; The memory is configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the signal receiving method according to any one of claims 1 to 15.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the signal receiving method as described in any one of claims 1 to 15 is implemented.

Images