Train digital baseband signal recovery method, device and equipment and storage medium

Abstract

The invention relates to the technical field of signal processing, and discloses a train digital baseband signal recovery method, device and equipment and a storage medium, and the method comprises the steps: obtaining a plurality of sample train digital baseband signals, and dividing the sample train digital baseband signals into training samples, test samples and verification samples; adding various noises to any sample train digital baseband signal in the training sample to obtain a disturbed train digital baseband signal; processing the disturbed train digital baseband signal to obtain an amplitude time-frequency diagram; carrying out Poisson sampling on the amplitude time-frequency diagram to obtain a pulse sequence; and de-noising the pulse sequence by adopting a pulse neural network to obtain a digital baseband signal of the target train. According to the method, various noises are added to the training sample, a complex noise interference scene of an actual train communication environment is simulated, and a clear target train digital baseband signal is obtained by adopting the pulse neural network for denoising, so that high-quality recovery of the signal is realized, and the robustness and generalization ability are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a method, device, equipment and storage medium for recovering a train digital baseband signal. Background Art

[0002] In high-speed mobile environments, such as railway communication systems, it is crucial to ensure reliable signal transmission. However, the digital baseband signal of train communication faces a variety of interferences in actual transmission, including pulse interference, pantograph offline noise interference, multipath interference, co-frequency interference and various types of noise interference, which will lead to signal quality deterioration, increase bit error rate and reduce communication reliability. Therefore, how to suppress interference and restore the digital baseband signal of the train is a problem that needs to be solved.

[0003] Existing technologies mainly rely on convolutional neural networks and other traditional filtering technologies to suppress interference and obtain clear train digital baseband signals. However, convolutional neural networks are triggered by levels, and signals and interference are transmitted in the form of levels, which makes convolutional neural networks have certain limitations in dealing with dynamically changing electromagnetic interference and real-time signal processing. Especially in complex and changeable noise environments, its anti-interference performance and real-time response capabilities are insufficient, resulting in the inability to restore clear train digital baseband signals and poor communication reliability. Summary of the invention

[0004] In view of this, the present invention provides a train digital baseband signal recovery method, device, equipment and storage medium to solve the problem that under complex and changeable noise environments, the convolutional neural network has insufficient anti-interference performance and real-time response capability, resulting in the inability to recover clear train digital baseband signals and poor communication reliability.

[0005] In a first aspect, the present invention provides a method for recovering a train digital baseband signal, the method comprising:

[0006] Acquire multiple sample train digital baseband signals, and divide the multiple sample train digital baseband signals into training samples, test samples, and verification samples;

[0007] For any sample train digital baseband signal in the training samples, multiple noises are added to the sample train digital baseband signal to obtain a disturbed train digital baseband signal;

[0008] Process the digital baseband signal of the disturbed train to obtain the amplitude time-frequency diagram;

[0009] Perform Poisson sampling on the amplitude time-frequency diagram to obtain a pulse train;

[0010] The pulse neural network is used to denoise the pulse sequence and obtain the digital baseband signal of the target train.

[0011] The train digital baseband signal recovery method provided by the embodiment of the present invention obtains multiple sample train digital baseband signals and reasonably divides them into training samples, test samples and verification samples, thereby laying a foundation for subsequent model training and evaluation. For the training samples, various noises are added to simulate the complex noise interference scenario in the actual train communication environment. The train digital baseband signal is processed and a pulse sequence is obtained by Poisson sampling. The pulse neural network is used for denoising to obtain a clear target train digital baseband signal, thereby achieving high-quality signal recovery, significantly enhancing the transmission quality and stability of the train digital baseband signal in a complex noise environment, and enhancing the robustness and generalization ability.

[0012] In an optional implementation, the disturbed train digital baseband signal is processed to obtain an amplitude time-frequency diagram, including:

[0013] Preprocessing the disturbed train digital baseband signal to obtain a preprocessed train digital baseband signal;

[0014] Perform short-time Fourier transform on the preprocessed train digital baseband signal to obtain the amplitude time-frequency diagram.

[0015] The train digital baseband signal recovery method provided in an embodiment of the present invention obtains an amplitude time-frequency diagram by preprocessing the disturbed train digital baseband signal and then performing a short-time Fourier transform, thereby capturing the frequency composition of the signal in different time periods and providing rich frequency domain information for subsequent processing.

[0016] In an optional implementation, Poisson sampling is performed on the amplitude time-frequency diagram to obtain a pulse sequence, including:

[0017] For each pixel in the amplitude time-frequency graph, determining the pulse emission probability of each pixel;

[0018] Divide the total time of the amplitude time-frequency diagram into multiple time steps;

[0019] In the first time step, a random number is generated for each pixel;

[0020] Traverse each pixel of the amplitude time-frequency graph, and for any pixel, emit a pulse in the first time step if the random number of the pixel is not greater than the probability of pulse emission;

[0021] The above process of determining whether to transmit a pulse in a time step is repeated until a plurality of time steps are traversed and completed, and a pulse sequence is formed based on the pulse transmission conditions in the plurality of time steps.

[0022] The train digital baseband signal recovery method provided by the embodiment of the present invention reduces data complexity by extracting the high-frequency information of the amplitude time-frequency diagram in a probabilistic form to form a discrete pulse sequence, while retaining key information to adapt to the input format of the pulse neural network.

[0023] In an optional implementation, the pulse neural network includes an input layer, an encoder, a skip module, an intermediate layer, a decoder, and an output layer. The pulse neural network is used to denoise the pulse sequence to obtain a target train digital baseband signal, including:

[0024] The encoder of the pulse neural network is used to convolve the pulse sequence input into the input layer to obtain the spatial features of the pulse sequence;

[0025] The encoder of the pulse neural network is used to downsample the spatial features to obtain downsampled features;

[0026] The skip module of the spiking neural network is used to pass the spatial features to the decoder;

[0027] Repeat the above encoder and skip module processing until the encoding process is completed, and use the middle layer of the spiking neural network to pass the target downsampled features obtained by the encoder after multiple convolutions and downsampling to the decoder;

[0028] A decoder of a pulse neural network is used to upsample based on spatial features and target downsampled features to obtain upsampled features;

[0029] The decoder of the pulse neural network is used to convolve the upsampled features to obtain the convolved upsampled features;

[0030] Repeat the above decoder processing until the decoding process is completed and the target pulse sequence is output through the output layer;

[0031] The target pulse sequence is restored to obtain the target train digital baseband signal.

[0032] The train digital baseband signal recovery method provided by the embodiment of the present invention captures the spatial features of the pulse sequence through the encoder, reduces the data dimension by downsampling, extracts more abstract and noise-resistant feature representation, and reduces the computational burden at the same time. The jump connection transmits the fine spatial features extracted in the encoder stage to the decoder, which helps to improve the quality of the recovered signal. The up-sampled features, down-sampled features and spatial features are fused in the decoder so that the fused sequence has both low-level detail features and high-level abstract features, thereby improving the utilization rate of low-order features, being beneficial to back propagation in the training process of the pulse neural network model, and being able to repair the detail features lost due to the downsampling operation in the encoder stage.

[0033] In an optional implementation, recovering the target pulse sequence to obtain the target train digital baseband signal includes:

[0034] Decoding the target pulse sequence to obtain a time-frequency diagram of the target pulse sequence;

[0035] The time-frequency diagram of the target pulse sequence is converted into the time domain to obtain the digital baseband signal of the target train.

[0036] The train digital baseband signal recovery method provided in the embodiment of the present invention decodes the target pulse sequence and converts it into a time-frequency diagram in the time-frequency domain, thereby restoring the time-frequency characteristics of the signal. By converting the obtained time-frequency diagram back to the time domain, the target train digital baseband signal is successfully reconstructed. Through meticulous decoding and conversion steps, the distortion that may be introduced during the processing, transformation and denoising process is effectively suppressed, thereby improving the purity and accuracy of the final signal.

[0037] In an optional embodiment, the method further includes:

[0038] In the process of denoising based on the spiking neural network, adjusting the network parameters of the spiking neural network;

[0039] After processing the verification samples in the digital baseband signals of multiple sample trains, Poisson sampling is used to obtain the pulse sequence of the verification samples, and based on the pulse sequence of the verification samples, the performance of the adjusted spiking neural network is evaluated to obtain an evaluation result;

[0040] The above adjustment and evaluation process is repeated until the spiking neural network denoises all training samples, and the network parameters of the spiking neural network used to obtain the maximum evaluation result are used as the final network parameters of the spiking neural network.

[0041] The train digital baseband signal recovery method provided in the embodiment of the present invention ensures that the pulse neural network can achieve higher denoising accuracy under various noise conditions by continuously adjusting network parameters and iteratively optimizing based on verification sample feedback, thereby improving the quality of the final recovered signal.

[0042] In an optional embodiment, the method further includes:

[0043] After processing the test samples in the digital baseband signals of the multiple sample trains, Poisson sampling is adopted to obtain the pulse sequence corresponding to the test samples;

[0044] Based on the final network parameters of the spiking neural network, the pulse sequence corresponding to the test sample is denoised to obtain the target train digital baseband signal corresponding to the test sample;

[0045] The performance of the spiking neural network is determined based on the target train digital baseband signal corresponding to the test sample.

[0046] The train digital baseband signal recovery method provided in an embodiment of the present invention can quantitatively evaluate the denoising performance of the pulse neural network by comparing the target train digital baseband signal obtained based on the test sample with the original signal, providing an objective and quantifiable performance indicator, which is helpful to discover and correct potential hidden problems of the pulse neural network and can be used as a basis for subsequent iterative optimization.

[0047] In a second aspect, the present invention provides a train digital baseband signal recovery device, the device comprising:

[0048] An acquisition module is used to acquire multiple sample train digital baseband signals, where the multiple sample train digital baseband signals are divided into training samples, test samples and verification samples;

[0049] An adding module is used to add a variety of noises to any sample train digital baseband signal in the training samples to obtain a disturbed train digital baseband signal;

[0050] A processing module is used to process the digital baseband signal of the disturbed train to obtain an amplitude time-frequency diagram;

[0051] A sampling module is used to perform Poisson sampling on the amplitude time-frequency diagram to obtain a pulse sequence;

[0052] The denoising module is used to denoise the pulse sequence using a pulse neural network to obtain a digital baseband signal of a target train.

[0053] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, computer instructions are stored in the memory, and the processor executes the train digital baseband signal recovery method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0054] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the train digital baseband signal recovery method of the above-mentioned first aspect or any corresponding embodiment thereof.

[0055] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, wherein the computer instructions are used to enable a computer to execute the train digital baseband signal recovery method of the above-mentioned first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 is a flow chart of a train digital baseband signal recovery method according to an embodiment of the present invention;

[0058] Figure 2 is a flowchart of another train digital baseband signal recovery method according to an embodiment of the present invention;

[0059] Figure 3 is a schematic diagram of a pulse residual block according to an embodiment of the present invention;

[0060] Figure 4 is a schematic diagram of a downsampling module according to an embodiment of the present invention;

[0061] Figure 5 is the workflow of a pulse neural network according to an embodiment of the present invention;

[0062] Figure 6 is a flowchart of another train digital baseband signal recovery method according to an embodiment of the present invention;

[0063] Figure 7 is a flowchart of another train digital baseband signal recovery method according to an embodiment of the present invention;

[0064] Figure 8 is a structural block diagram of a train digital baseband signal recovery device according to an embodiment of the present invention;

[0065] Fig. 9 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0067] The digital baseband signal of train communication faces a variety of noise interferences in actual transmission, which will lead to the deterioration of signal quality, increase the bit error rate, and reduce the communication reliability. Therefore, how to suppress interference and restore the digital baseband signal of the train is a problem that needs to be solved. The existing technology mainly relies on convolutional neural networks and other traditional filtering technologies to suppress interference. Its anti-interference performance and real-time response capabilities are insufficient, resulting in the inability to restore clear digital baseband signals of the train and poor communication reliability. The method for recovering the digital baseband signal of the train provided in an embodiment of the present invention adds a variety of noises to the training samples, simulates the complex noise interference scenarios in the actual train communication environment, and obtains a clear target digital baseband signal of the train by denoising through a pulse neural network, thereby achieving high-quality recovery of the signal, significantly enhancing the transmission quality and stability of the digital baseband signal of the train in a complex noise environment, and enhancing the robustness and generalization ability.

[0068] According to an embodiment of the present invention, an embodiment of a method for recovering a digital baseband signal of a train is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0069] In this embodiment, a train digital baseband signal recovery method is provided, which can be used in electronic equipment. Figure 1 : is a flow chart of a method for recovering a train digital baseband signal according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0070] Step S101, obtain multiple sample train digital baseband signals, and the multiple sample train digital baseband signals are divided into training samples, test samples and verification samples. Specifically, it is assumed that the number of sample train digital baseband signals is 3000, and each sample train digital baseband signal has 1 channel and 1000 time steps. The 3000 sample train digital baseband signals are randomly divided into training samples, test samples and verification samples in a ratio of 8:1:1, that is, the training samples include 2400 sample train digital baseband signals, and the test samples and verification samples respectively include 300 sample train digital baseband signals. Optionally, the number, structure and division ratio of the sample train digital baseband signals are only examples, and the embodiments of the present invention are not limited to this.

[0071] Step S102, for any sample train digital baseband signal in the training sample, add multiple noises to the sample train digital baseband signal to obtain the disturbed train digital baseband signal. Specifically, the training sample is used for signal recovery. Before signal recovery, in order to simulate the interference in the actual train communication environment and better conform to the actual working conditions, multiple noises are added to the sample train digital baseband signal in the training sample, such as random noise, pulse interference, pantograph network offline interference, etc.

[0072] Step S103, the disturbed train digital baseband signal is processed to obtain an amplitude time-frequency diagram. Specifically, the disturbed train digital baseband signal after adding noise interference is processed to obtain an amplitude time-frequency diagram suitable for subsequent processing.

[0073] Step S104, Poisson sampling is performed on the amplitude time-frequency diagram to obtain a pulse sequence. Specifically, the pulse sequence can intuitively display the time domain and frequency domain characteristics of the signal, and the pulse sequence is generated based on the amplitude time-frequency diagram by using Poisson sampling, which can effectively characterize the amplitude time-frequency diagram with fewer sampling points, significantly reduce the amount of calculation, and can naturally filter out some noise, so that the pulse sequence can be directly processed by the pulse neural network to improve the processing efficiency.

[0074] Step S105, using a pulse neural network to denoise the pulse sequence to obtain a digital baseband signal of the target train. Specifically, the pulse neural network simulates the activity of biological neurons and processes discrete pulse events to transmit and process information. The pulse neural network is triggered by pulses, and the signal is transmitted in the form of discrete pulses. Since noise interference is usually transmitted in the form of a level, the pulse triggering mechanism of the pulse neural network enables it to naturally resist such interference, thereby overcoming the limitations of existing convolutional neural networks in interference suppression, significantly enhancing the denoising ability, and obtaining a clearer and more accurate digital baseband signal of the target train.

[0075] The train digital baseband signal recovery method provided by the embodiment of the present invention obtains multiple sample train digital baseband signals and reasonably divides them into training samples, test samples and verification samples, thereby laying a foundation for subsequent model training and evaluation. For the training samples, various noises are added to simulate the complex noise interference scenario in the actual train communication environment. The train digital baseband signal is processed and a pulse sequence is obtained by Poisson sampling. The pulse neural network is used for denoising to obtain a clear target train digital baseband signal, thereby achieving high-quality signal recovery, significantly enhancing the transmission quality and stability of the train digital baseband signal in a complex noise environment, and enhancing the robustness and generalization ability.

[0076] In this embodiment, a method for recovering a train digital baseband signal is provided, which can be used in the above-mentioned electronic device. Figure 2is a flow chart of another train digital baseband signal recovery method according to an embodiment of the present invention, such as Figure 2 As shown, the process includes the following steps:

[0077] Step S201, obtain multiple sample train digital baseband signals, and divide the multiple sample train digital baseband signals into training samples, test samples and verification samples. For details, please refer to Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0078] Step S202: for any sample train digital baseband signal in the training sample, add multiple noises to the sample train digital baseband signal to obtain a disturbed train digital baseband signal. Figure 1 Step S102 of the illustrated embodiment will not be described in detail here.

[0079] Step S203, processing the disturbed train digital baseband signal to obtain an amplitude time-frequency diagram.

[0080] Specifically, the above step S203 includes:

[0081] Step S2031, preprocessing the disturbed train digital baseband signal to obtain a preprocessed train digital baseband signal. Specifically, filtering, normalizing, smoothing and other operations are performed on the disturbed train digital baseband signal to facilitate subsequent Fourier transform. Optionally, other preprocessing operations may also be used, which are not limited in the embodiment of the present invention.

[0082] Step S2032, short-time Fourier transform is performed on the pre-processed train digital baseband signal to obtain an amplitude time-frequency diagram. Specifically, for the pre-processed train digital baseband signal x(n), a time window of length W is used, and the time window moves in steps of R each time. The pre-processed train digital baseband signal x(n) within the range of [nW, W] is intercepted, and then short-time Fourier transform is performed to obtain the expression of the amplitude time-frequency diagram as shown in the following formula (1). Among them, the amplitude spectrum of the amplitude time-frequency diagram at time n can be expressed by the following formula (2), and the phase spectrum can be expressed by the following formula (3). The amplitude time-frequency diagram after short-time Fourier transform is conjugate symmetric about the positive and negative frequency axes, the amplitude spectrum is even symmetric, and the phase spectrum is odd symmetric. By performing short-time Fourier transform on the pre-processed train digital baseband signal, an amplitude time-frequency diagram is obtained, which captures the frequency composition of the signal in different time periods and provides rich frequency domain information for subsequent processing.

[0083]

[0084] Among them, X m(ω) represents the short-time Fourier transform result centered at mR, that is, the expression of the amplitude time-frequency diagram centered at mR; n represents time; W represents the length of the time window; x(n) represents the expression of the disturbed train digital baseband signal at time n; w() represents the window function; R represents the moving step length of the window function each time when moving and intercepting; e -jwn is a complex exponential function, representing the frequency component.

[0085] M n (ω)=|X n (ω)|(2)

[0086] Among them, M n (ω) represents the time-frequency domain amplitude spectrum of the amplitude time-frequency diagram at time n; X n (ω) represents the short-time Fourier transform result at time n, that is, the expression of the amplitude time-frequency diagram at time n.

[0087]

[0088] Among them, P n (ω) represents the phase spectrum of the amplitude time-frequency diagram at time n; Im() represents the imaginary part; Re() represents the real part; X n (ω) represents the short-time Fourier transform result at time n, that is, the expression of the amplitude time-frequency diagram at time n.

[0089] Step S204: Perform Poisson sampling on the amplitude time-frequency diagram to obtain a pulse sequence.

[0090] Specifically, the above step S204 includes:

[0091] Step S2041: for each pixel in the amplitude time-frequency diagram, determine the pulse emission probability of each pixel. Specifically, the pulse emission probability of each pixel can be determined by the following formula (4).

[0092]

[0093] Where i represents the index of the time point of any pixel in the amplitude time-frequency graph; j represents the index of the frequency of any pixel in the amplitude time-frequency graph; p ij represents the pulse emission probability of the pixel with i and j as index; S(t i ,f j ) represents the amplitude of the pixel with i and j as index in the amplitude time-frequency diagram, t i represents the time point corresponding to i, f j represents the frequency corresponding to j; S max Indicates the maximum value of the amplitudes corresponding to all pixels in the amplitude time-frequency diagram.

[0094] Step S2042, dividing the total time of the amplitude time-frequency diagram into a plurality of time steps. Specifically, the horizontal axis of the amplitude time-frequency diagram represents time, and based on the maximum value of the horizontal axis, that is, the total time, a plurality of time steps of equal length are obtained.

[0095] Step S2043, in the first time step, a random number is generated for each pixel. Specifically, in the first time step, a random number is generated for each pixel in the interval [0, 1] using random sampling of a Poisson process.

[0096] Step S2044, traverse each pixel of the amplitude time-frequency graph, and for any pixel, if the random number of the pixel is not greater than the pulse emission probability, emit a pulse in the first time step. Specifically, traverse each pixel in the amplitude time-frequency graph, compare the pulse emission probability and the random number for each pixel, and emit the pulse corresponding to the pixel if the random number is not greater than the pulse emission probability. The pulse emission position can be represented by (Δt, f), Δt represents the current time step, and f represents the frequency of the pixel.

[0097] Step S2045, repeat the above process of determining whether to emit a pulse in a time step until multiple time steps are traversed and completed, and a pulse sequence is formed based on the pulse emission conditions in multiple time steps. Specifically, the above steps S2043 and S2044 are performed in each time step, and in step S2044, for each pixel, a judgment is made as to whether a pulse is emitted based on the pulse emission probability and the random number. The pulse emission conditions in each time step, that is, based on the pulse emission position of each pulse, form a pulse sequence. By extracting the high-frequency information of the amplitude time-frequency diagram in a probabilistic form to form a discrete pulse sequence, the data complexity is reduced, while retaining key information to adapt to the input format of the pulse neural network.

[0098] Step S205, using a pulse neural network to denoise the pulse sequence to obtain a digital baseband signal of a target train, the pulse neural network includes an input layer, an encoder, a skip module, an intermediate layer, a decoder and an output layer.

[0099] Specifically, the above step S205 includes:

[0100] Step S2051, using the encoder of the pulse neural network, convolve the pulse sequence input by the input layer to obtain the spatial features of the pulse sequence. Specifically, the encoder contains 3 pulse residual blocks, Figure 3 is a schematic diagram of a pulse residual block according to an embodiment of the present invention, such as Figure 3As shown in the figure, each pulse residual block includes a 3*3 convolution layer, a batch normalization layer, and an activation function layer. The batch normalization layer normalizes the features extracted by the convolution layer. The activation function layer generally uses the Atan function, the pulse neuron threshold is set to 1, and the inverse tangent function is used as the gradient replacement function of the pulse emission function during back propagation. The convolution operation performed by the pulse residual block in the encoder can capture the pattern and edge information of the local area in the pulse sequence, which is the basis for noise reduction and feature extraction.

[0101] Step S2052: downsample the spatial features using the pulse neural network encoder to obtain downsampled features. Specifically, downsample the spatial features obtained by the pulse residual block of the encoder, usually using a downsampling module. Figure 4 is a schematic diagram of a downsampling module according to an embodiment of the present invention, such as Figure 4 As shown in the figure, each downsampling module uses a 3*3 convolutional layer with a convolution kernel step size of 2, which halves the spatial dimension of the input spatial features and increases their depth. It also performs nonlinear processing and enhancement of the features through batch normalization layers and activation function layers, reduces the spatial resolution while retaining important information, helps reduce the amount of computation, and extracts higher-level feature representations.

[0102] Step S2053, using the skip module of the pulse neural network to transfer the spatial features to the decoder. Specifically, the spatial features obtained from each pulse residual block are transferred to the corresponding position of the decoder through the skip module, so that the spatial features participate in the decoding process of the decoder.

[0103] Step S2054, repeat the processing of the above encoder and jump module until the encoding process is completed, and use the intermediate layer of the pulse neural network to pass the target downsampling features obtained by the encoder after multiple convolutions and downsampling to the decoder. Specifically, after each pulse residual block is convolved, the skip module is used to pass the spatial features obtained by convolution to the decoder. At the same time, the spatial features obtained by convolution are also used for downsampling. The encoder repeats the above process until the encoding is completed, and the target downsampling features after 3 downsamplings are obtained, and the target downsampling features are passed to the decoder through the intermediate layer. Among them, the intermediate layer includes a batch normalization layer, a two-dimensional convolution layer, and an Atan activation function layer.

[0104] Step S2055, using a decoder of a pulse neural network, upsampling is performed based on the spatial features and the target downsampling features to obtain upsampling features. Specifically, the decoder upsamples the target downsampling features of the encoder, and usually uses an upsampling module for upsampling. The structure of the upsampling module is referenced Figure 4As shown in Figure 1, the structure is consistent with the down-sampling module, but the convolution kernel stride of the convolution layer of the up-sampling module is 1 to restore the abstract feature information and enhance the representation ability of the features through batch normalization layers and pulse neurons.

[0105] Step S2056, using the decoder of the pulse neural network, convolve the upsampled features to obtain the upsampled features after convolution. Specifically, the decoder also contains 3 pulse residual blocks, and the structure is consistent with the pulse residual block in the encoder. Since the jump module transmits the spatial features in the encoding process to the decoder, when the decoder uses the pulse residual block for convolution, it can not only consider the upsampled features obtained based on the target downsampled features, but also consider the spatial features, and fuse the upsampled features, downsampled features and spatial features in the decoder, so that the fused sequence has both low-level detail features and high-level abstract features, which improves the utilization rate of low-order features, is beneficial to the back propagation in the training process of the pulse neural network model, and can repair the detail features lost due to the downsampling operation in the encoder stage.

[0106] Step S2057, repeat the processing of the above decoder until the decoding process is completed, and output the target pulse sequence through the output layer. Specifically, since the decoder includes 3 pulse residual blocks and 3 upsampling processes, when the last pulse residual block completes the convolution, the final feature is output to the output layer, and the output layer processes the output feature and finally outputs the target pulse sequence. Optionally, the process of converting the output feature into a pulse sequence by the output layer is a prior art and will not be repeated here.

[0107] In some optional embodiments, Figure 5 is a flowchart of a pulse neural network according to an embodiment of the present invention, such as Figure 5 As shown in the figure, the input layer inputs the pulse sequence to the encoder, and the encoder extracts features through the pulse residual block and the downsampling module. After each convolution of the pulse residual block, the spatial features obtained by the convolution are passed to the decoder through the jump connection layer. The encoder passes the target downsampling features obtained by 3 convolutions and 3 downsamplings to the decoder through the intermediate layer. The decoder fuses the target downsampling features and the spatial features obtained by each convolution through the pulse residual block and the upsampling module, and inputs the features after 3 convolutions and 3 upsamplings to the output layer. Optionally, the input layer and the output layer are Figure 5 Not shown in FIG.

[0108] Step S2058, recover the target pulse sequence to obtain the target train digital baseband signal.

[0109] In some optional implementations, the above step S2058 includes:

[0110] Step a1, decode the target pulse sequence to obtain a time-frequency diagram of the target pulse sequence. Specifically, based on the above step S204, it can be known that Poisson sampling is used to convert the amplitude time-frequency diagram into a pulse sequence, so a decoding method corresponding to Poisson sampling is used for decoding to convert the pulse sequence into a time-frequency diagram. Optionally, the process of converting the pulse sequence into a time-frequency diagram is a prior art and will not be repeated here.

[0111] Step a2, convert the time-frequency diagram of the target pulse sequence to the time domain to obtain the target train digital baseband signal. Specifically, by using the amplitude-frequency characteristics and phase-frequency characteristics obtained by short-time Fourier transform, the inverse transform of short-time Fourier transform can be used to restore the amplitude and phase characteristics of the signal in the entire frequency domain, and then restore the time domain characteristics to restore the clean target train digital baseband signal.

[0112] In some optional embodiments, Figure 6 is a flow chart of another train digital baseband signal recovery method according to an embodiment of the present invention, such as Figure 6 As shown, noise is added to the sample train digital baseband signal in the training sample to obtain the disturbed train digital baseband signal. Then, the disturbed train digital baseband signal is processed to obtain the amplitude time-frequency diagram. Then, the pulse sequence is obtained by Poisson sampling the amplitude time-frequency diagram. Then, the pulse sequence is input into the pulse neural network for denoising, and the denoised target pulse sequence is output. Then, the target pulse sequence is decoded to obtain the time-frequency diagram of the target pulse sequence. Finally, the time-frequency diagram of the target pulse sequence is converted to the time domain to obtain a clear target train digital baseband signal.

[0113] The train digital baseband signal recovery method provided by the embodiment of the present invention obtains multi-sample train digital baseband signals and reasonably divides them into training samples, test samples and verification samples, thereby laying a foundation for subsequent model training and evaluation. For the training samples, various noises are added to simulate the complex noise interference scenario in the actual train communication environment. The train digital baseband signal is processed and sampled by Poisson sampling to obtain a pulse sequence, and denoising is performed by a pulse neural network to obtain a clear target train digital baseband signal, thereby achieving high-quality signal recovery, significantly enhancing the transmission quality and stability of the train digital baseband signal in a complex noise environment, and enhancing the robustness and generalization ability.

[0114] In this embodiment, a method for recovering a train digital baseband signal is provided, which can be used in the above-mentioned electronic device. Figure 7 is a flow chart of another train digital baseband signal recovery method according to an embodiment of the present invention, such as Figure 7 As shown, the process includes the following steps:

[0115] Step S701, obtain multiple sample train digital baseband signals, and divide the multiple sample train digital baseband signals into training samples, test samples and verification samples. For details, please refer to Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0116] Step S702: for any sample train digital baseband signal in the training sample, add multiple noises to the sample train digital baseband signal to obtain a disturbed train digital baseband signal. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0117] Step S703, process the disturbed train digital baseband signal to obtain an amplitude time-frequency diagram. Figure 2 Step S203 of the illustrated embodiment will not be described in detail here.

[0118] Step S704, perform Poisson sampling on the amplitude time-frequency diagram to obtain a pulse sequence. Figure 2 Step S204 of the illustrated embodiment will not be described in detail here.

[0119] Step S705: Use a pulse neural network to denoise the pulse sequence to obtain a target train digital baseband signal. Figure 2 Step S205 of the illustrated embodiment will not be described in detail here.

[0120] Step S706, in the process of denoising based on the pulse neural network, adjust the network parameters of the pulse neural network. Specifically, when the pulse neural network denoises the training sample, the network parameters in the pulse neural network can be adjusted once per time step. Optionally, the embodiment of the present invention does not limit the time length of each time step. By adjusting the network parameters of the pulse neural network, the network parameters can be optimized so that the pulse neural network has good performance.

[0121] Step S707, after processing the verification samples in the digital baseband signals of the multiple sample trains, Poisson sampling is used to obtain the pulse sequence of the verification samples, and based on the pulse sequence of the verification samples, the performance of the adjusted pulse neural network is evaluated to obtain an evaluation result. Specifically, referring to the above steps S703 and S704, the verification samples are processed and the pulse sequence is generated. After each parameter adjustment of the pulse neural network, the pulse sequence of one or more verification samples is input into the pulse neural network, and based on the comparison of the verification samples before and after denoising the pulse neural network, the denoising performance of the pulse neural network is evaluated to obtain an evaluation result. Among them, the evaluation result can be a signal-to-noise ratio, that is, the larger the evaluation result, the better the denoising performance of the pulse neural network. Optionally, other evaluation indicators can also be used to represent the evaluation result, and the embodiment of the present invention is not limited to this.

[0122] Step S708, repeat the above adjustment and evaluation process until the pulse neural network denoises all training samples, and the network parameters of the pulse neural network used when the maximum evaluation result is obtained are used as the final network parameters of the pulse neural network. Specifically, the above steps S706 and S707 are performed in the process of the pulse neural network processing the training samples. In the process of the pulse neural network denoising the training samples, the network parameters of the pulse neural network are continuously adjusted, and the verification samples are used for verification after each adjustment to obtain the corresponding evaluation results after each adjustment. When the pulse neural network has completed denoising for all training samples, the pulse neural network has made multiple network parameter adjustments at this time, and the network parameters corresponding to the maximum evaluation result in the evaluation results obtained after each adjustment are used as the final network parameters of the pulse neural network, that is, the performance of the pulse neural network is the best under the final network parameters, ensuring that the pulse neural network can achieve higher denoising accuracy under various noise conditions and improve the quality of the restored signal.

[0123] Step S709, after processing the test samples in the multiple sample train digital baseband signals, Poisson sampling is adopted to obtain a pulse sequence corresponding to the test samples. Specifically, referring to the above steps S703 and S704, the test samples are processed and a pulse sequence is generated.

[0124] Step S710, based on the final network parameters of the spiking neural network, denoise the pulse sequence corresponding to the test sample to obtain the target train digital baseband signal corresponding to the test sample. Specifically, the pulse sequence corresponding to the test sample is input into the spiking neural network with determined network parameters to obtain the target train digital baseband signal corresponding to the test sample, providing support for verifying the generalization ability of the spiking neural network.

[0125] Step S711, based on the target train digital baseband signal corresponding to the test sample, determine the performance of the pulse neural network. Specifically, when determining the performance of the pulse neural network, it can be quantified using indicators such as accuracy, recall, and F1 score. This process is a prior art and will not be repeated here. By denoising the pulse sequence of the test sample, the performance of the pulse neural network on unknown data is tested, and the generalization ability of the pulse neural network is verified.

[0126] The train digital baseband signal recovery method provided by the embodiment of the present invention obtains multiple sample train digital baseband signals and reasonably divides them into training samples, test samples and verification samples, thereby laying a foundation for subsequent model training and evaluation. For the training samples, various noises are added to simulate the complex noise interference scenario in the actual train communication environment. The train digital baseband signal is processed and a pulse sequence is obtained by Poisson sampling. The pulse neural network is used for denoising to obtain a clear target train digital baseband signal, thereby achieving high-quality signal recovery, significantly enhancing the transmission quality and stability of the train digital baseband signal in a complex noise environment, and enhancing the robustness and generalization ability.

[0127] In this embodiment, a train digital baseband signal recovery device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0128] This embodiment provides a train digital baseband signal recovery device, such as Figure 8 As shown, including:

[0129] The acquisition module 801 is used to acquire multiple sample train digital baseband signals, where the multiple sample train digital baseband signals are divided into training samples, test samples and verification samples.

[0130] The adding module 802 is used to add a variety of noises to any sample train digital baseband signal in the training samples to obtain a disturbed train digital baseband signal.

[0131] The processing module 803 is used to process the disturbed train digital baseband signal to obtain an amplitude time-frequency diagram.

[0132] The sampling module 804 is used to perform Poisson sampling on the amplitude time-frequency diagram to obtain a pulse sequence.

[0133] The denoising module 805 is used to denoise the pulse sequence using a pulse neural network to obtain a target train digital baseband signal.

[0134] In some optional implementations, the processing module 803 includes:

[0135] The preprocessing unit is used to preprocess the disturbed train digital baseband signal to obtain a preprocessed train digital baseband signal.

[0136] The transform unit is used to perform short-time Fourier transform on the pre-processed train digital baseband signal to obtain an amplitude time-frequency diagram.

[0137] In some optional implementations, the sampling module 804 includes:

[0138] The first determination unit is used to determine the pulse emission probability of each pixel in the amplitude time-frequency diagram.

[0139] The division unit is used to divide the total time of the amplitude time-frequency diagram into multiple time steps.

[0140] The generator is used to generate random numbers for each pixel in the first time step.

[0141] The traversal unit is used for traversing each pixel of the amplitude time-frequency diagram, and for any pixel, when the random number of the pixel is not greater than the pulse emission probability, a pulse is emitted in the first time step.

[0142] The second determination unit is used to repeat the above process of determining whether to transmit a pulse in a time step until a plurality of time steps are traversed and completed, and form a pulse sequence based on the pulse transmission conditions in the plurality of time steps.

[0143] In some optional embodiments, the spiking neural network includes an input layer, an encoder, a skip module, an intermediate layer, a decoder, and an output layer, and the denoising module 805 includes:

[0144] The first convolution unit is used to use the encoder of the pulse neural network to convolve the pulse sequence input by the input layer to obtain the spatial characteristics of the pulse sequence.

[0145] The down-sampling unit is used for down-sampling the spatial features by using an encoder of a pulse neural network to obtain down-sampled features.

[0146] The first transmission unit is used to transmit the spatial features to the decoder by using a skip module of a pulse neural network.

[0147] The second transfer unit is used to repeat the processing of the above encoder and jump module until the encoding process is completed, and adopts the middle layer of the pulse neural network to transfer the target down-sampling features obtained by the encoder after multiple convolutions and down-sampling to the decoder.

[0148] The upsampling unit is used for a decoder using a pulse neural network to upsample based on the spatial feature and the target downsampled feature to obtain an upsampled feature.

[0149] The second convolution unit is used to convolve the upsampled features using a decoder of a pulse neural network to obtain the convolved upsampled features.

[0150] The output unit is used to repeat the processing of the above decoder until the decoding process is completed and the target pulse sequence is output through the output layer.

[0151] The recovery unit is used to recover the target pulse sequence to obtain the target train digital baseband signal.

[0152] In some optional implementations, the recovery unit includes:

[0153] The decoding subunit is used to decode the target pulse sequence to obtain a time-frequency diagram of the target pulse sequence.

[0154] The conversion subunit is used to convert the time-frequency diagram of the target pulse sequence into the time domain to obtain the target train digital baseband signal.

[0155] In some optional embodiments, the device further comprises:

[0156] The adjustment module is used to adjust the network parameters of the spiking neural network during the denoising process based on the spiking neural network.

[0157] The evaluation module is used to process the verification samples in the digital baseband signals of multiple sample trains and obtain the pulse sequence of the verification samples by using Poisson sampling. Based on the pulse sequence of the verification samples, the performance of the adjusted pulse neural network is evaluated to obtain the evaluation result.

[0158] The first determination module is used to repeat the above adjustment and evaluation process until the spiking neural network denoises all training samples, and the network parameters of the spiking neural network used when the maximum evaluation result is obtained are used as the final network parameters of the spiking neural network.

[0159] In some optional embodiments, the device further comprises:

[0160] The second determination module is used to process the test samples in the multiple sample train digital baseband signals and obtain the pulse sequence corresponding to the test samples by using Poisson sampling.

[0161] The third determination module is used to denoise the pulse sequence corresponding to the test sample based on the final network parameters of the pulse neural network to obtain the target train digital baseband signal corresponding to the test sample.

[0162] The fourth determination module is used to determine the performance of the pulse neural network based on the target train digital baseband signal corresponding to the test sample.

[0163] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0164] The train digital baseband signal recovery device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0165] The embodiment of the present invention also provides a computer device having the above Figure 8 The train digital baseband signal recovery device shown.

[0166] See also Fig. 9 , Fig. 9 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Fig. 9 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Fig. 9 A processor 10 is taken as an example.

[0167] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0168] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0169] The memory 20 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 computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device 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.

[0170] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0171] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0172] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0173] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0174] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for recovering a train digital baseband signal, characterized in that: The method comprises: Acquire a plurality of sample train digital baseband signals, wherein the plurality of sample train digital baseband signals are divided into training samples, test samples and verification samples; For any sample train digital baseband signal in the training samples, adding multiple noises to the sample train digital baseband signal to obtain a disturbed train digital baseband signal; Processing the disturbed train digital baseband signal to obtain an amplitude time-frequency diagram; Performing Poisson sampling on the amplitude time-frequency diagram to obtain a pulse sequence; The pulse sequence is denoised using a pulse neural network to obtain a target train digital baseband signal.

2. The method according to claim 1, characterized in that The step of processing the disturbed train digital baseband signal to obtain an amplitude time-frequency diagram includes: Preprocessing the disturbed train digital baseband signal to obtain a preprocessed train digital baseband signal; Performing short-time Fourier transform on the preprocessed train digital baseband signal to obtain the amplitude time-frequency diagram.

3. The method according to claim 1, characterized in that: The step of performing Poisson sampling on the amplitude time-frequency diagram to obtain a pulse sequence comprises: For each pixel in the amplitude time-frequency diagram, determining a pulse emission probability of each pixel; Dividing the total time of the amplitude time-frequency diagram into a plurality of time steps; In the first time step, a random number is generated for each pixel; Traversing each pixel of the amplitude time-frequency graph, for any pixel, when the random number of the pixel is not greater than the pulse emission probability, emitting a pulse in the first time step; The above process of determining whether to transmit a pulse in a time step is repeated until the multiple time steps are traversed to the end, and the pulse sequence is formed based on the pulse transmission conditions in the multiple time steps.

4. The method according to claim 1, characterized in that: The pulse neural network includes an input layer, an encoder, a skip module, an intermediate layer, a decoder and an output layer. The pulse neural network is used to denoise the pulse sequence to obtain a target train digital baseband signal, including: Using the encoder of the pulse neural network, convolving the pulse sequence input by the input layer to obtain the spatial features of the pulse sequence; Downsampling the spatial features using the encoder of the pulse neural network to obtain downsampled features; Using a skip module of the spiking neural network, the spatial features are transmitted to the decoder; Repeat the processing of the encoder and the skip module until the encoding process is completed, and use the middle layer of the spiking neural network to pass the target down-sampled features obtained by the encoder through multiple convolutions and down-sampling to the decoder; Using the decoder of the pulse neural network, upsampling is performed based on the spatial feature and the target downsampled feature to obtain an upsampled feature; Using the decoder of the pulse neural network, convolving the up-sampled features to obtain convolved up-sampled features; Repeat the processing of the decoder until the decoding process is completed, and output the target pulse sequence through the output layer; The target pulse sequence is restored to obtain the target train digital baseband signal.

5. The method according to claim 4, characterized in that The step of recovering the target pulse sequence to obtain the target train digital baseband signal includes: Decoding the target pulse sequence to obtain a time-frequency diagram of the target pulse sequence; The time-frequency diagram of the target pulse sequence is converted into the time domain to obtain the target train digital baseband signal.

6. The method according to claim 1, characterized in that The method further comprises: In the process of denoising based on the spiking neural network, adjusting the network parameters of the spiking neural network; After processing the verification samples in the multiple sample train digital baseband signals, Poisson sampling is used to obtain a pulse sequence of the verification samples, and based on the pulse sequence of the verification samples, the performance of the adjusted pulse neural network is evaluated to obtain an evaluation result; The above adjustment and evaluation process is repeated until the spiking neural network denoises all training samples, and the network parameters of the spiking neural network used when the maximum evaluation result is obtained are used as the final network parameters of the spiking neural network.

7. The method according to claim 6, characterized in that The method further comprises: After processing the test samples in the multiple sample train digital baseband signals, Poisson sampling is adopted to obtain a pulse sequence corresponding to the test samples; Based on the final network parameters of the spiking neural network, denoising the pulse sequence corresponding to the test sample to obtain a target train digital baseband signal corresponding to the test sample; The performance of the pulse neural network is determined based on the target train digital baseband signal corresponding to the test sample.

8. A train digital baseband signal recovery device, characterized in that: The device comprises: An acquisition module is used to acquire a plurality of sample train digital baseband signals, wherein the plurality of sample train digital baseband signals are divided into training samples, test samples and verification samples; An adding module, configured to add a plurality of noises to any sample train digital baseband signal in the training samples to obtain a disturbed train digital baseband signal; A processing module, used for processing the disturbed train digital baseband signal to obtain an amplitude time-frequency diagram; A sampling module, used for performing Poisson sampling on the amplitude time-frequency diagram to obtain a pulse sequence; The denoising module is used to denoise the pulse sequence using a pulse neural network to obtain a digital baseband signal of a target train.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the train digital baseband signal recovery method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the train digital baseband signal recovery method according to any one of claims 1 to 7.

Images