Signal detection and model training method, device, equipment and storage medium

Abstract

The present invention provides a signal detection and model training method, apparatus, device, and storage medium. The method comprises: demodulating a received signal to obtain corresponding demodulated data; extracting the real and imaginary parts of the demodulated data to form real data corresponding to the demodulated data; inputting the real data into a pre-trained target deep learning network model to obtain an output result; and obtaining a detection result of the original transmitted signal based on the output result. By employing a deep learning network model for channel estimation and equalization, the present invention adaptively learns communication channel state information, avoids the influence of imaginary interference, and effectively improves the accuracy of signal detection.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a signal detection and model training method, device, equipment and storage medium. Background Art

[0002] With the rapid development of communication technology, we have entered the 5G era. In this era, compared to the OFDM (Orthogonal Frequency Division Multiplexing) technology of LTE (Long Term Evolution), the FBMC (Filter Bank Multi-Carrier) system does not require a cyclic prefix, has low out-of-band leakage, and high spectral efficiency, thus offering superior stability, time-frequency focusing, and higher spectrum utilization. However, because the FBMC system only meets strict orthogonality in the real domain and has inherent imaginary interference, the channel estimation methods used in the OFDM system cannot be directly used to implement signal detection in the FBMC system.

[0003] To address this issue, the existing technology generally uses least squares channel estimation, MMSE equalization algorithm and noise estimation method required for MMSE equalization based on a given pilot structure to achieve channel estimation and equalization. However, the existing channel estimation and equalization methods have low accuracy in detecting the transmission signal. Summary of the Invention

[0004] Embodiments of the present invention provide a signal detection and model training method, apparatus, device, and storage medium to solve the problem of low accuracy of transmission signal detection results in the prior art.

[0005] In a first aspect, an embodiment of the present invention provides a signal detection method, comprising:

[0006] Demodulating the received signal to obtain corresponding demodulated data;

[0007] Extracting a real part and an imaginary part of the demodulated data to form real number data corresponding to the demodulated data;

[0008] Inputting the real number data into a pre-trained target deep learning network model to obtain an output result;

[0009] According to the output result, a detection result of the original transmitted signal is obtained.

[0010] In a second aspect, an embodiment of the present invention provides a signal detection model training method, comprising:

[0011] Obtaining a training data set, wherein the training data set includes training input data and training supervision data;

[0012] The pre-established deep learning network is trained using the training data set. When the loss of the training output result relative to the training supervision data reaches a second preset condition, the training is terminated to obtain a trained target deep learning network model.

[0013] In a third aspect, an embodiment of the present invention provides a signal detection device, including:

[0014] A demodulation module, used to demodulate the received signal to obtain corresponding demodulated data;

[0015] An extraction module, configured to extract a real part and an imaginary part from the demodulated data to form real data corresponding to the demodulated data;

[0016] A detection module is used to input the real number data into a pre-trained target deep learning network model to obtain an output result;

[0017] The processing module is used to obtain the detection result of the original transmission signal according to the output result.

[0018] In a fourth aspect, an embodiment of the present invention provides an electronic device, comprising: a memory, a transceiver, and at least one processor;

[0019] The processor, the memory and the transceiver are interconnected via a circuit;

[0020] The memory stores computer-executable instructions; the transceiver is used to receive a signal sent by a transmitting end;

[0021] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method described in the first aspect and various possible designs of the first aspect, and executes the method described in the second aspect and various possible designs of the second aspect.

[0022] In a fifth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer execution instructions. When a processor executes the computer execution instructions, the method described in the first aspect and various possible designs of the first aspect are implemented, as well as the method described in the second aspect and various possible designs of the second aspect are implemented.

[0023] The signal detection and model training method, apparatus, device and storage medium provided by the embodiments of the present invention adopt a deep learning network model for channel estimation and equalization, adaptively learn communication channel state information, avoid the influence of imaginary interference, and effectively improve the accuracy of signal detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0025] Figure 1 A schematic flow chart of a signal detection method provided in one embodiment of the present invention;

[0026] Figure 2 A block diagram showing the overall structure of a communication system according to an embodiment of the present invention;

[0027] Figure 3 A schematic structural diagram of a communication system with a channel classification function provided by one embodiment of the present invention;

[0028] Figure 4 A schematic diagram of the communication system structure of a training process provided by one embodiment of the present invention;

[0029] Figure 5 A schematic diagram of the communication system structure during testing or actual application provided by one embodiment of the present invention;

[0030] Figure 6 A schematic diagram showing the comparison results of the LSTM network model for short packets provided by one embodiment of the present invention with DNN and other traditional methods under the same channel conditions;

[0031] Figure 7 A schematic diagram showing the comparison results of the LSTM network model with long packets and encoding provided by one embodiment of the present invention, and the DNN and traditional methods under the same channel conditions;

[0032] Figure 8 A flowchart of a model training method for signal detection provided by one embodiment of the present invention;

[0033] Figure 9 A schematic structural diagram of a signal detection device provided by an embodiment of the present invention;

[0034] Figure 10 A schematic diagram of an exemplary structure of a signal detection device provided by an embodiment of the present invention;

[0035] Figure 11 A schematic diagram of the structure of a signal detection model training device provided by one embodiment of the present invention;

[0036] Figure 12 A schematic structural diagram of an electronic device provided by an embodiment of the present invention.

[0037] The above drawings illustrate specific embodiments of the present invention, which will be described in more detail below. These drawings and the accompanying description are not intended to limit the scope of the present invention in any way, but rather to illustrate the concept of the present invention to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0039] First, the terms involved in the present invention are explained:

[0040] LSTM: Long Short-Term Memory, also known as long short-term memory network, is a time recurrent neural network and a special RNN (Recurrent Neural Network). Compared with ordinary RNNs, LSTM can perform better in longer sequences.

[0041] In addition, the terms "first," "second," etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. In the description of the following embodiments, "plurality" means two or more, unless otherwise specifically defined.

[0042] The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. The embodiments of the present invention will be described below with reference to the accompanying drawings.

[0043] An embodiment of the present invention provides a signal detection method for detecting an original transmitted signal in an FBMC system. The embodiment is performed by a device signal detection apparatus, which can be provided in an electronic device, such as a receiver or other implementable computer device.

[0044] like Figure 1 FIG. 1 is a flow chart of a signal detection method provided in this embodiment, which includes:

[0045] Step 101: demodulate the received signal to obtain corresponding demodulated data.

[0046] Specifically, the received signal is the signal to be sent from the transmitting end of the FBMC system and reaches the receiving end after being transmitted through the communication channel. During the transmission process of the communication channel, the signal to be sent is processed by the multipath channel, Doppler effect and Gaussian white noise, and finally the received signal at the receiving end is obtained; the receiving end demodulates the received signal to obtain demodulated data in the frequency domain. The demodulated data includes frequency domain data at each time-frequency grid point. In practical applications, the demodulated data can be expressed as a complex matrix. If N represents the number of FBMC symbols and M represents the number of subcarriers, the demodulated data is an M×N complex matrix.

[0047] The demodulation operation of the received signal includes serial-to-parallel conversion, inverse filtering, fast Fourier transform and deletion of the initial phase.

[0048] Step 102: extract the real part and the imaginary part of the demodulated data to form real data corresponding to the demodulated data.

[0049] Specifically, after obtaining the demodulated data corresponding to the received signal, it is necessary to extract the real and imaginary parts of each complex element in the demodulated data, and form corresponding real data with the extracted real and imaginary parts. The real data meets the input requirements of the deep learning network model. The deep learning network model is used to further restore the original transmitted signal. The original transmitted signal refers to the original information bit stream to be transmitted by the transmitter. The original information bit stream needs to be modulated, take the real and imaginary parts, insert the pilot, add the initial phase, inverse fast Fourier transform, filtering, parallel-to-serial conversion, etc. at the transmitter to obtain the corresponding signal to be sent. The signal to be sent is transmitted through the communication channel to obtain the received signal.

[0050] For example, the real part and the imaginary part can be arranged in a row to form a vector to be input into a deep learning network model.

[0051] Step 103: input the real number data into the pre-trained target deep learning network model to obtain the output result.

[0052] Specifically, the target deep learning network model is a model obtained by training based on a large amount of training data, which is used for channel estimation and equalization to accurately restore the original transmitted signal. The input of the target deep learning network model is the real data formed by the real and imaginary parts of the demodulated data, and the output result is the parallel data of the restored original transmitted signal.

[0053] The target deep learning network model is obtained through pre-training. Specifically, a training data set can be obtained and used to train the pre-established deep learning network. When the loss of the training output result relative to the training supervision data reaches a second preset condition, the training is terminated to obtain the trained target deep learning network model, wherein the training data set includes training input data and training supervision data.

[0054] In practical applications, the randomly generated training information bit stream can be processed by the transmitter and then sent to the receiver via a communication channel. The receiver demodulates and extracts the real part and the imaginary part to obtain the training input data, and uses the training information bit stream as the training supervision data. If the transmitter encodes the training information bit stream, the encoding result can be used as the training supervision data at the receiver. The second preset condition can be set according to actual needs. Specifically, for example, the training information bit stream can be randomly generated, the pilot is a determined information bit, the training information bit stream is modulated to obtain a second modulation result, and similarly, the pilot also needs to be modulated to obtain a first modulation result; the modulation of the training information bit stream and the pilot is in no particular order, and the specific modulation method can be set according to actual needs, such as using QAM modulation; after modulation, the real data to be sent can be determined based on the first modulation result of the pilot and the second modulation result of the training information bit stream. The first modulation result and the second modulation result are both complex symbols, and it is necessary to convert them into corresponding real data symbols by extracting the real part and the imaginary part, and the real data symbols corresponding to the pilot and the real data symbols of the training information bit stream form the real data to be sent; that is, the real data to be sent includes the real part of the pilot, the imaginary part of the pilot and the real data data of the training information bit stream, the real part and the imaginary part of the pilot are the real part and the imaginary part of the first modulation result corresponding to the pilot, and the real data of the training information bit stream is The real and imaginary parts of the second modulation result; after obtaining the real data to be sent, an initial phase is added to the real data to be sent, so that the real and imaginary parts of the same complex number are staggered by one phase and remain orthogonal, and an inverse fast Fourier transform is performed after adding the initial phase to obtain a second transformation result, and the second transformation result is further filtered. Specifically, a prototype filter can be used for filtering to obtain a filtering result, and the filtering result is converted from parallel to serial to obtain a training signal to be sent. The training signal to be sent is input into the communication channel for transmission and then reaches the receiving end, and the receiving end obtains a training received signal. The receiving end demodulates the training received signal to obtain corresponding training demodulated data, extracts the real and imaginary parts of the training demodulated data, and forms training real data as training input data and inputs it into a deep learning network (such as an LSTM network). The encoding result of the training information bit stream is used as training supervision data to supervise the end of training. After the training is completed, the trained target deep learning network model can be obtained.

[0055] Step 104: Obtain the detection result of the original transmitted signal according to the output result.

[0056] Specifically, after obtaining the output result of the target deep learning network model, the detection result of the original transmitted signal can be obtained based on the output result, such as performing parallel-to-serial conversion on the output result to obtain the original transmitted signal.

[0057] Optionally, before the original information bit stream is modulated at the transmitting end, the original information bit stream can also be encoded, and then subsequent modulation and other processing can be performed. Correspondingly, at the receiving end, the output result of the target deep learning network model is the undecoded original transmission signal, and the output result needs to be decoded to obtain the decoding result. Since the output result is parallel data, in order to obtain a result consistent with the original transmission signal, parallel-to-serial conversion is also required to obtain the detection result of the original transmission signal, that is, the restored original transmission signal. The decoding method is adapted to the encoding method in the transmitting end processing process, and can be set according to actual needs. For example, the convolutional code method is used for encoding, and the Viterbi decoding algorithm is used for decoding.

[0058] It can be understood that for a communication system that includes an encoding processing process, during the training of the target deep learning network model, the sending end also needs to encode the training information bit stream before performing subsequent related processing, and the receiving end needs to perform corresponding decoding operations. The details will not be repeated here.

[0059] The signal detection method provided in this embodiment adopts a deep learning network model to perform channel estimation and equalization, adaptively learns communication channel state information, avoids the influence of imaginary part interference, and effectively improves the accuracy of signal detection.

[0060] In order to make the technical solution of the present invention clearer, another embodiment of the present invention further supplements the method provided in the above embodiment.

[0061] As an implementable method, in order to further improve the accuracy of signal detection, based on the above embodiment, optionally, real number data is input into the pre-trained target deep learning network model to obtain the output result, including: inputting real number data into the target LSTM network model to obtain the output result.

[0062] Specifically, since all bits of the communication information bit stream are correlated and have memory, in order to better utilize the correlation between the bits, this embodiment uses a trained target LSTM network model as a deep learning network model for channel estimation and equalization for signal detection, thereby improving the accuracy of the FBMC system and further improving the accuracy of signal detection.

[0063] The target LSTM network model is obtained based on a large amount of training data. Since the LSTM network can learn long-term dependent information, it can well learn the correlation between all bits in the communication information bit stream, thereby effectively improving the accuracy of signal detection.

[0064] Furthermore, in one embodiment, in order to better perform channel estimation and equalization, demodulating the received signal to obtain corresponding demodulated data may specifically include:

[0065] The received signal is converted from serial to parallel to obtain a corresponding parallel signal; the parallel signal is inversely filtered using an inverse prototype filter to obtain an inverse filtering result; the inverse filtering result is fast Fourier transformed to obtain a first transformation result; the initial phase of the first transformation result is deleted to obtain demodulated data.

[0066] Specifically, the received signal is transmitted serially, and needs to be converted from serial to parallel first to obtain a parallel signal. The parallel signal is inversely filtered using an inverse prototype filter to obtain an inverse filtering result. The inverse filtering result is further subjected to a fast Fourier transform (FFT) to obtain a first transformation result, and the initial phase is deleted from the first transformation result to obtain demodulated data. The initial phase is deleted because the initial phase is added before the inverse fast Fourier transform is performed at the transmitting end. Specifically, at the transmitting end, the original information bit stream to be transmitted is modulated (such as QAM), the real and imaginary parts are taken, the pilot is inserted, the initial phase is added, the inverse fast Fourier transform (IFFT) is performed, the filtering is performed, the parallel to serial conversion is performed, etc. to obtain the corresponding signal to be transmitted. In actual applications, the original information bit stream can also be encoded before modulation. The signal to be transmitted is transmitted through the communication channel to the receiving end, and the receiving end receives the received signal. Specifically, OQAM (Offset Quadrature Amplitude Modulation) can be used. Modulation, offset quadrature amplitude modulation) modulation preprocessing separates the real and imaginary parts of a complex number into two pure real numbers (which can be called real symbols and imaginary symbols, respectively), thereby processing the real and imaginary parts of the complex signal separately, and setting the time interval to the symbol period T / 2. The FBMC system uses OQAM modulation technology to ensure orthogonality between carriers. The specific modulation principle is existing technology and will not be repeated here.

[0067] For example, Figure 2 As shown in the figure, it is a block diagram of the overall structural principle of the communication system provided by this embodiment, wherein Real() and Imag() represent taking the real part and taking the imaginary part, j^(n+2m) and j^(n+2m+1) represent adding the initial phase, j^-(n+2m) and j^-(n+2m+1) represent deleting the initial phase, and Rounding represents the Round function processing process, that is, the result of rounding off according to the specified number of decimal places, converting each decimal value of the output result of the target LSTM network model into 0 or 1; it should be noted that, in this example, the target LSTM network model outputs a group of decimals between (0,1), indicating the probability of each transmitted bit being 1 or 0. After Rounding, the recovered undecoded original transmission signal is obtained, and then the original transmission signal can be obtained by decoding (Decoding) and parallel-to-serial conversion.

[0068] It can be understood that in practical applications, Rounding can also be used as part of the target LSTM network model. The output result of the target LSTM network model is the undecoded parallel original transmission signal, which can be obtained by decoding and parallel-to-serial conversion.

[0069] In another embodiment, in order to further improve the accuracy of the signal detection result, the detection result of the original transmitted signal is obtained according to the output result, including: using a Viterbi decoding algorithm to decode the output result and perform parallel-to-serial conversion to obtain the detection result of the original transmitted signal.

[0070] Specifically, before the transmitter modulates the original information bit stream, it encodes the original information bit stream to obtain the encoding result, and then performs subsequent modulation, takes the imaginary and real parts, inserts pilots, adds initial phases, performs inverse fast Fourier transform, filters, and performs parallel-to-serial conversion. Correspondingly, at the receiving end, the output result of the target deep learning network model is the undecoded original transmission signal, so it is necessary to decode the output result to obtain the decoding result. Since the output result is parallel data, in order to obtain a result consistent with the original transmission signal, it is also necessary to perform parallel-to-serial conversion to obtain the detection result of the original transmission signal, that is, the restored original transmission signal. The decoding method is adapted to the encoding method in the processing process of the transmitter. The embodiment of the present invention adopts a convolutional code method for encoding and a Viterbi decoding algorithm for decoding.

[0071] During the convolutional code encoding process, the input information bits are grouped and encoded. The encoded output bits of each code group are related not only to the information bits of that group, but also to the information bits of other groups at the previous moment. Similarly, during the convolutional code decoding process, decoding information is obtained not only from the group received at the current moment, but also from related information from the previous and subsequent groups. Because the correlation between each group is fully utilized during the convolutional code encoding process, the convolutional code has a very good performance gain. However, in a communication system, when a signal is transmitted in a channel, it will be affected by various noise interferences due to the non-ideal conditions of the channel itself, thus generating bit errors. Therefore, the embodiment of the present invention performs convolutional encoding on the transmitted information bit stream to generate a convolutional code, and transmits it after modulation and other processing, which can effectively reduce the bit error rate and thus improve the accuracy of the signal detection results.

[0072] In actual applications, if a vehicle-mounted communication system enters a tunnel from an open environment, the communication channel scenarios are completely different. In order to enable the communication system to switch scenarios in the shortest possible time, the embodiment of the present invention also pre-trains models of multiple channel environments. When the communication system switches from one channel scenario to another, seamless signal detection can be achieved.

[0073] As another practicable manner, in order to improve the universality of the communication system, the method may further include:

[0074] Obtain characteristic data of the target channel of the current transmission signal; input the characteristic data of the target channel into a trained channel classification model to obtain the scene type of the target channel; obtain corresponding target LSTM parameters according to the scene type of the target channel; and use the network model based on the target LSTM parameters as the target LSTM network model.

[0075] Specifically, for channels in different scenarios, different LSTM parameters can be trained and stored, so that the original transmission signal can be accurately detected for transmissions on different channels. In actual applications, the characteristic data of the target channel of the current transmission signal can be obtained. The characteristic data of the target channel is the channel parameters of the target channel. Based on the trained channel classification model, the scenario type of the target channel can be identified, and the LSTM parameters of the corresponding scenario type (called target LSTM parameters) can be obtained and passed into the corresponding network to form an LSTM network model as the target LSTM network model. The original transmission signal is further obtained based on the output of the target LSTM network model.

[0076] The channel classification model can be set according to actual needs.

[0077] Furthermore, in order to improve classification accuracy, an embodiment of the present invention adopts a deep neural network model as a channel classification model.

[0078] The deep neural network model includes an input layer, a hidden layer, and an output layer. The number of hidden layers can be set according to actual needs. The layers are fully connected. The number of neurons in each layer is set according to actual needs. The number of neurons in the output layer is the number of scene types. For example, if there are 14 scene types, the output layer has 14 neurons.

[0079] Exemplary communication coverage propagation scenarios include: indoor wireless communication (such as indoor offices and residential communities), indoor to outdoor wireless communication, urban microcells, poor urban microcells, poor urban macrocells, suburban village macrocells, outdoor to indoor wireless communication, fixed feeders, urban macrocells, rural mobile networks, etc. The maximum delay and / or number of multipaths for different scenario types are different, as shown in Table 1 below, where LOS (Line Of Sight) and NLOS (Not Line Of Sight) refer to line-of-sight transmission and non-line-of-sight transmission of wireless signals, respectively; therefore, these scenarios can be classified and divided into a certain number of scenario types.

[0080] Table 1

[0081]

[0082] Exemplarily, the above scenarios are divided into 14 scenario types. The channel classification model includes an input layer, a first hidden layer, a second hidden layer and an output layer. The number of neurons in each layer is, for example, 32, 128, 64 and 14 respectively.

[0083] For example, Figure 3 As shown in the figure, it is a structural diagram of a communication system with a channel classification function provided by this embodiment. The communication system includes a channel classification module (i.e., the gray part in the figure), wherein the channel decision module is a deep neural network model for identifying the scene type of the channel. LSTM1-LSTM14 represent LSTM parameters corresponding to 14 scene types. These LSTM parameters are all pre-trained. After identifying the scene type of the channel, the corresponding LSTM parameters are obtained from the 14 LSTM parameters and transmitted to the LSTM network to form a target LSTM network model for estimating and equalizing the currently transmitted signal to restore the original transmitted signal.

[0084] In one embodiment, in order to ensure the accuracy of the classification result of the channel classification model, the method further includes:

[0085] Acquire channel training feature data and corresponding label data; train a pre-established deep neural network based on the channel training feature data and the corresponding label data. If the loss reaches a first preset condition, terminate the training and obtain a deep neural network model.

[0086] Specifically, in order for the channel classification model to accurately classify the channel, it is necessary to pre-train the channel classification model through a large amount of training data. Specifically, the channel training feature data and the corresponding label data of each scene type can be obtained. The channel training feature data includes channel parameter information of various scene types, and the label data is the actual scene type of the channel. The channel training feature data will be input into a pre-established deep neural network according to preset rules. The loss will be calculated using a preset loss function based on the prediction results output by the network and the corresponding label data. When the loss reaches the first preset condition, the training can be terminated to obtain a trained channel classification model. The preset loss function can adopt any feasible loss function according to actual needs, such as the cross entropy loss function, and can be set according to actual needs. This embodiment does not limit it.

[0087] Exemplarily, the total sample data includes 400 txt files, 300 of which can be used as training data and 100 as test data; the channel training feature data includes 300 txt files, each txt file includes a matrix of size 10000*32, that is, 10000 rows, and 32 numbers in each row. During training, the 32 numbers in each row of each file can be used as one input, and each row can be input into the deep neural network in turn. For each input of 32 numbers, the deep neural network starts from the first layer and calculates the activation value layer by layer, and finally outputs the prediction result. Then, the prediction result can be calculated by Boolean function, and different vectors are obtained by using different combinations of 0 and 1. Each vector represents a classification. The prediction result is compared with the label data, and the loss is calculated. If the loss does not meet the first preset condition, the training continues. If the loss meets the first preset condition, the training is ended to obtain a trained channel classification model; the first preset condition can be set according to actual needs.

[0088] As another feasible approach, in order to ensure the channel estimation and equalization effect of the deep learning network model, it is necessary to perform training optimization in advance, that is, the method also includes:

[0089] A training data set is obtained, where the training data set includes training input data and training supervision data; a pre-established deep learning network is trained using the training data set, and when the loss of the training output result relative to the training supervision data reaches a second preset condition, the training is terminated to obtain a trained target deep learning network model.

[0090] Specifically, the randomly generated training information bit stream can be processed by the sending end and then sent to the receiving end via the communication channel. The receiving end demodulates and extracts the real part and the imaginary part to obtain the training input data, and uses the training information bit stream as the training supervision data; if the sending end encodes the training information bit stream, the encoding result can be used as the training supervision data at the receiving end; the second preset condition can be set according to actual needs.

[0091] Optionally, since the transmission content is unknown in a real communication system, the training process can also be judged by the accuracy of the recovered pilot to determine whether the training process is completed, that is, the training supervision data is the pilot data.

[0092] For example, Figure 4 As shown in FIG, a schematic diagram of the communication system structure of the training process provided in this embodiment is shown in FIG. Figure 5As shown, a schematic diagram of the communication system structure of the test or actual application process provided by this embodiment is shown; wherein, the deep learning network includes 4 layers of LSTM blocks (LSTM Block) and a fully connected layer (FC). The number of LSTM cells (LSTMCell) included in the LSTM Block can be set according to actual needs. In this example, the number of LSTM cells included in each LSTM Block is 384 / 700 / 256 / 64 respectively, and the number of neurons included in the fully connected layer is 16. The loss calculation module (Calculate Cost) extracts the recovered pilot data according to the output of the deep learning network. Here, the fully connected layer outputs a set of decimals between (0,1). The loss calculation module can convert each decimal into 0 or 1 according to a certain rule, such as through a sigmoid function, and then compare the conversion result with the corresponding pilot label value to determine whether it is equal to the pilot label value. If they are equal, it means that the pilot bit is recovered correctly. Otherwise, the recovery is incorrect. The average error rate of the pilot recovery is calculated accordingly and sent to the training network module (Training Net). The training network module Net receives the cost calculated by the loss calculation module and determines whether the cost value meets the second preset condition. If so, the current network parameters obtained from the deep learning network are used as the trained LSTM parameters, thereby forming a trained LSTM network model and storing it; if the cost does not meet the second preset condition, it means that further training is needed, then Training Net updates the current network parameters according to the preset rules and transmits the new network parameters to the deep learning network, continues optimization training, and so on, until the cost value meets the second preset condition.

[0093] In practical applications, the specific number of layers of the LSTM network model can be set according to actual needs and is not limited to the above-mentioned 4-layer LSTM block and one fully connected layer. For example, it can also be 1 layer, 2 layers or 3 layers of LSTM blocks.

[0094] It can be understood that for different scenario types, the training process of the corresponding deep learning network model is consistent. The embodiment of the present invention can train the corresponding deep learning network model according to the above training method for each scenario type and store it corresponding to each scenario type. Subsequently, the corresponding deep learning network model can be selected as the target deep learning network model for channel estimation and equalization based on the scenario type identified based on the channel classification model, thereby effectively improving the universality of the communication system.

[0095] Furthermore, in order to obtain a training data set for a deep learning network model, the method may further include:

[0096] Obtain a training information bit stream and a corresponding pilot; encode the training information bit stream to obtain an encoding result; modulate the pilot to obtain a first modulation result, and modulate the encoding result to obtain a second modulation result; determine the real data to be sent based on the first modulation result and the second modulation result, the real data to be sent including the real part of the pilot, the imaginary part of the pilot and the real data of the training information bit stream; add an initial phase to the real data to be sent and perform an inverse fast Fourier transform to obtain a second transformation result; filter the second transformation result to obtain a filtering result; perform parallel-to-serial conversion on the filtering result to obtain a training signal to be sent; transmit the training signal to be sent through a communication channel to obtain a training received signal; demodulate the training received signal to obtain corresponding training demodulated data; extract the real part and the imaginary part of the training demodulated data to form training real data corresponding to the training demodulated data; use the training real data as training input data and use the encoding result as training supervision data.

[0097] Specifically, the training information bit stream can be randomly generated, and the pilot is a determined information bit. In order to reduce the bit error rate, the training information bit stream is encoded to obtain a coding result. For example, the training information bit stream is encoded using a convolutional coding method to obtain a coding result, and the coding result is a convolutional code; after encoding, the coding result can be modulated to obtain a second modulation result. Similarly, the pilot also needs to be modulated to obtain a first modulation result; the modulation order of the coding result and the pilot is not particular, and the specific modulation method can be set according to actual needs, such as using QAM modulation, which can be 4QAM modulation, 16QAM modulation, 64QAM modulation, 2 56QAM modulation, and so on; after modulation, the real data to be sent can be determined according to the first modulation result of the pilot and the second modulation result of the coding result. The first modulation result and the second modulation result are both complex symbols, and they need to be converted into corresponding real data symbols by extracting the real part and the imaginary part. The real data symbols corresponding to the pilot and the real data symbols of the training information bit stream form the real data to be sent; that is, the real data to be sent includes the real part of the pilot, the imaginary part of the pilot and the real data of the training information bit stream. The real part of the pilot and the imaginary part of the pilot are the real part and imaginary part of the first modulation result corresponding to the pilot, and the real data of the training information bit stream are the real part and imaginary part of the second modulation result.

[0098] Exemplarily, the real data to be sent is an M×N data block on the frequency domain time coordinate, the first two columns are the pilot real part and the pilot imaginary part respectively, and the following columns are the training information bit stream real data.

[0099] After obtaining the real data to be sent, an initial phase is added to the real data to be sent, with the aim of staggering the real and imaginary parts of the same complex number by one phase and maintaining orthogonality. After adding the initial phase, an inverse fast Fourier transform is performed to obtain a second transform result, and the second transform result is further filtered. Specifically, a prototype filter can be used for filtering to obtain a filtering result, and the filtering result is converted from parallel to serial to obtain a training signal to be sent. The training signal to be sent is input into the communication channel for transmission and reaches the receiving end, whereupon the receiving end obtains a training received signal. The receiving end demodulates the training received signal to obtain corresponding training demodulated data, extracts the real and imaginary parts of the training demodulated data, and forms training real data as training input data, which is input into a deep learning network (such as an LSTM network), and the encoding result of the training information bit stream is used as training supervision data to supervise the end of training. The specific processing of the training received signal at the receiving end is consistent with the above-mentioned processing process of the received signal, and will not be described in detail.

[0100] In practical applications, the pilot may also be encoded and then subsequently processed. The specific settings may be made according to actual needs and are not limited in this embodiment.

[0101] For example, the learning rate of the LSTM network can be set to 0.001, the optimization function can adopt the Adam optimization algorithm, and can also adopt the RMSProp (Root Mean Square Prop) optimization algorithm, and the output layer can adopt the sigmoid function as the activation function.

[0102] In practical applications, the FBMC baseband equivalent transmission signal (i.e., the signal to be transmitted at the transmitter) can be expressed as s(t):

[0103]

[0104] Among them, N represents the number of FBMC symbols, M represents the number of subcarriers, and a m,n is the real-valued data symbol (i.e., complex symbol) sent on the m-th subcarrier in the n-th FBMC symbol, g m,n (t) represents the odd function at the time-frequency grid coordinate (m,n), which can be obtained by the following time-frequency transformation:

[0105]

[0106] Where v0 represents the interval between subcarriers, τ0 represents the time offset between the real and imaginary parts of adjacent symbols, τ0v0 = 1 / 2; the initial phase

[0107] Furthermore, the received signal at the receiving end can be expressed as r(t):

[0108]

[0109] Among them, h m,n is the channel value at the time-frequency grid coordinate (m,n), Denotes a convolution operation, w(t) denotes Gaussian white noise, and the signal transmission and reception principles of the FBMC communication system are prior art and will not be described in detail here.

[0110] In an exemplary embodiment, the beneficial effects of the present invention are described in conjunction with simulation experiment results.

[0111] The simulation conditions in this example use the new wireless system channel model "WINNER II channel models" standard, with a system bandwidth of 2.5 GHz, 64 subcarriers, a Hermite prototype filter, an overlap factor of 4, and OQAM modulation. The simulation uses a (2,1,3) convolutional code with a code rate of 1 / 2, generator polynomial coefficients C1 = (1,1,1), C2 = (1,0,1), and Viterbi decoding; Figure 6 As shown in FIG, the comparison results of the LSTM network model of the short packet provided in this embodiment and the DNN and other traditional methods under the same channel conditions are shown in FIG. Figure 7 As shown in FIG, a schematic diagram of the comparison results of the long packet and encoded LSTM network model provided in this embodiment with the DNN and traditional methods under the same channel conditions; in the figure, the horizontal axis is the signal-to-noise ratio (SNR), in decibels dB, and the vertical axis is the bit error rate (BER); Pilots LS (OFDM) is the LS (least squares) channel estimation method under the existing OFDM system, and its calculation method is that the pilot signal at the transmitting end is H T , the pilot signal received by the receiving end is H R , then the estimated impact value of the channel is: Pilots Deep Learning (OFDM) is a channel estimation method using a five-layer DNN network structure under the existing OFDM system. The number of neurons in each layer is: 256, 500, 250, 120, 16, for a total of five layers; Pilots MMSE (OFDM) is an MMSE (least mean square) channel estimation method under the existing OFDM system; Pilots demapping (FBMC) is an LS (least square) channel estimation method under the existing FBMC system, with an overlap coefficient of O=4 and a Hermite filter as the prototype filter; Pilots DL-CE (FBMC) is a channel estimation method using a five-layer DNN network structure under the existing FBMC system: the number of neurons in each layer is: 384, 700, 300, 100, 16, for a total of five layers, an overlap coefficient of O=4, and a Hermite filter as the prototype filter; Pilots LSTM (FBMC) is the simulation result of the channel estimation method of the short packet LSTM network under the FBMC system of the present invention. Each packet length is 100*384, and the five-layer LSTM network structure has the following characteristics: the number of neurons in each layer is: 384, 700, 256, 100, 16, and there are five layers in total. The overlapping coefficient is O=4, and the prototype filter is a Hermite filter. Pilots LSTM long (FBMC) is the simulation result of the long packet under the FBMC system of the present invention. Each packet length is 800*384, and the five-layer LSTM network structure has the following characteristics: the packet length is 800*384, and the number of neurons in each layer is: 384, 700, 256, 100, 16, and there are five layers in total. The overlapping coefficient is O=4, and the prototype filter is a Hermite filter. Pilots LSTM long decode(FBMC) is the simulation result of long packets of convolutional coding and Viterbi decoding under the FBMC system of the present invention. Each packet is 800*384 long and has a five-layer LSTM network structure with 384, 700, 256, 100, and 16 neurons per layer. The overlap coefficient is O=4 and the prototype filter is a Hermite filter.

[0112] Obviously, by Figure 6 and Figure 7 It can be seen that under the same bit error rate BER, the signal-to-noise ratio of LSTM is 3.74dB to 5.27dB lower than that of DNN, the short packet LSTM is 0.67dB to 2.14dB lower than the long packet LSTM, and the long packet encoded LSTM is 0.71dB to 2.94dB lower than the long packet LSTM. In summary, the signal detection method based on LSTM of the present invention has better effect than the existing method. In a complex wireless channel environment, the signal detection method based on the LSTM network model proposed in the present invention effectively improves the accuracy. Under the condition of adding convolutional code, the method of the present invention further improves the accuracy.

[0113] It should be noted that each implementable method in this embodiment can be implemented separately, or can be implemented in any combination without conflict, and the present invention does not limit this.

[0114] The signal detection method provided in this embodiment performs channel estimation and equalization through an LSTM network model, thereby effectively improving detection accuracy. It also effectively reduces the bit error rate by encoding the original transmitted signal, thereby further improving detection accuracy. It can also identify multiple channels through a channel classification model, and provide a corresponding LSTM network model for each channel to perform channel estimation and equalization, thereby improving the universality of the communication system.

[0115] Yet another embodiment of the present invention provides a signal detection model training method for training and obtaining the deep learning network model required by the above embodiment.

[0116] like Figure 8 FIG. 1 is a flow chart of a model training method for signal detection provided in this embodiment, which specifically includes:

[0117] Step 301: Obtain a training data set, where the training data set includes training input data and training supervision data.

[0118] Step 302: Use the training data set to train the pre-established deep learning network. When the loss of the training output result relative to the training supervision data reaches a second preset condition, the training is terminated to obtain a trained target deep learning network model.

[0119] The specific operations of the above steps have been described in detail in the above embodiments and will not be repeated here.

[0120] Furthermore, obtaining the training data set may specifically include:

[0121] Obtain a training information bit stream and a corresponding pilot; encode the training information bit stream to obtain an encoding result; modulate the pilot to obtain a first modulation result, and modulate the encoding result to obtain a second modulation result; determine real data to be transmitted based on the first modulation result and the second modulation result, the real data to be transmitted including the real part of the pilot, the imaginary part of the pilot, and the real data of the training information bit stream; add an initial phase to the real data to be transmitted and perform an inverse fast Fourier transform to obtain a second transformation result; filter the second transformation result to obtain a filtering result; perform parallel-to-serial conversion on the filtering result to obtain a training signal to be transmitted; transmit the training signal to be transmitted through a communication channel to obtain a training received signal; demodulate the training received signal to obtain corresponding training demodulated data; extract the real part and imaginary part of the training demodulated data to form training real data corresponding to the training demodulated data; use the training real data as training input data, and use the encoding result as training supervision data. For specific operations, please refer to the above embodiment and will not be repeated here.

[0122] In one embodiment, the method may further include:

[0123] Acquire channel training feature data and corresponding label data; train a pre-established deep neural network based on the channel training feature data and the corresponding label data. If the loss reaches a first preset condition, terminate the training and obtain a trained deep neural network model for channel scene type detection.

[0124] Yet another embodiment of the present invention provides a signal detection device for executing the method of the above embodiment.

[0125] like Figure 9 FIG. 3 is a schematic diagram of the structure of the signal detection device provided in this embodiment. The device 30 includes: a demodulation module 31 , an extraction module 32 , a detection module 33 and a processing module 34 .

[0126] Among them, the demodulation module is used to demodulate the received signal to obtain the corresponding demodulated data; the extraction module is used to extract the real part and the imaginary part of the demodulated data to form the real data corresponding to the demodulated data; the detection module is used to input the real data into the pre-trained target deep learning network model to obtain the output result; the processing module is used to obtain the detection result of the original transmitted signal based on the output result.

[0127] Specifically, the demodulation module receives the receiving signal sent by other modules (such as the receiving module), demodulates the receiving signal to obtain the corresponding demodulated data, and sends it to the extraction module. The extraction module extracts the real part and the imaginary part of the demodulated data, and forms the real data corresponding to the demodulated data, and sends it to the detection module. The detection module inputs the real data into the pre-trained target deep learning network model, obtains the output result and sends it to the processing module. The processing module obtains the detection result of the original transmitted signal based on the output result.

[0128] Regarding the device in this embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and can achieve the same technical effect, so it will not be elaborated here.

[0129] In order to make the device of the present invention more clear, another embodiment of the present invention further supplements the device provided in the above embodiment.

[0130] As an implementable method, in order to further improve the accuracy of the detection results, based on the above embodiment, optionally, the detection module is specifically used to input real number data into the target LSTM network model to obtain output results.

[0131] Furthermore, the demodulation module is specifically configured to:

[0132] The received signal is converted from serial to parallel to obtain a corresponding parallel signal; the parallel signal is inversely filtered using an inverse prototype filter to obtain an inverse filtering result; the inverse filtering result is fast Fourier transformed to obtain a first transformation result; the initial phase of the first transformation result is deleted to obtain demodulated data.

[0133] In practical applications, the demodulation module may further include a serial-to-parallel conversion submodule, an inverse filtering submodule, a fast Fourier transform submodule, and a deletion submodule.

[0134] Among them, the serial-to-parallel conversion submodule is used to perform serial-to-parallel conversion on the received signal to obtain the corresponding parallel signal; the inverse filtering submodule is used to perform inverse filtering on the parallel signal using an inverse prototype filter to obtain an inverse filtering result; the fast Fourier transform submodule is used to perform fast Fourier transform on the inverse filtering result to obtain a first transformation result; the deletion submodule is used to delete the initial phase of the first transformation result to obtain demodulated data.

[0135] As another feasible manner, in order to further improve the accuracy of the detection result, optionally, the processing module is specifically configured to:

[0136] The Viterbi decoding algorithm is used to decode the output result and perform parallel-to-serial conversion to obtain the detection result of the original transmitted signal.

[0137] like Figure 10 , which is a schematic diagram of an exemplary structure of the signal detection device provided in this embodiment.

[0138] As another feasible manner, in order to improve the universality of the communication system, the apparatus further includes a first acquisition module 35 , a classification module 36 and a determination module 37 .

[0139] Among them, the first acquisition module is used to obtain the characteristic data of the target channel of the current transmission signal; the classification module is used to input the characteristic data of the target channel into the trained channel classification model to obtain the scene type of the target channel; the determination module is used to obtain the corresponding target LSTM parameters according to the scene type of the target channel, and use the network model based on the target LSTM parameters as the target LSTM network model.

[0140] Specifically, the first acquisition module can obtain the characteristic data of the target channel from the configuration file of the target channel and send it to the classification module. The classification module inputs the characteristic data of the target channel into the trained channel classification model, obtains the scene type of the target channel, and sends it to the determination module. The determination module obtains the corresponding target LSTM parameters according to the scene type of the target channel, and transfers the target LSTM parameters to the LSTM network to form an LSTM network model as the target LSTM network model.

[0141] Furthermore, in order to improve the accuracy of the classification results, the channel classification model is a deep neural network model.

[0142] Furthermore, in order to ensure the accuracy of the classification results, the acquisition module is also used to obtain channel training feature data and corresponding label data; the classification module is also used to train the pre-established deep neural network based on the channel training feature data and the corresponding label data. If the loss reaches the first preset condition, the training is terminated to obtain a deep neural network model.

[0143] As another practicable manner, the device further includes a second acquisition module 38 and a second training module 39 .

[0144] Among them, the second acquisition module is used to obtain a training data set, which includes training input data and training supervision data; the second training module is used to use the training data set to train a pre-established deep learning network. When the loss of the training output result relative to the training supervision data reaches a second preset condition, the training is terminated to obtain a trained target deep learning network model.

[0145] In one embodiment, for the transmission of the channel corresponding to the target deep learning network model, the target deep learning network model trained by the second training module can be sent to the detection module and directly used for the detection of the original transmitted signal.

[0146] In another embodiment, for a communication system with multiple scenario types, the target deep learning network model or model parameters trained by the second training module can be sent to the determination module, and the deep learning network models corresponding to various scenario types can be trained, and the model parameters can be sent to the determination module, so that the determination module can obtain the deep learning network model of the corresponding scenario type as the target deep learning network model for signal detection under different channel scenario types.

[0147] It should be noted that the various implementable methods in this embodiment may be implemented separately, or may be combined in any combination without conflict to implement the present invention without limitation.

[0148] Regarding the device in this embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and can achieve the same technical effect, so it will not be elaborated here.

[0149] In another embodiment of the present invention, a signal detection model training device can also be provided for training deep learning network models and / or channel classification models.

[0150] like Figure 11 , which is a structural diagram of the signal detection model training device provided in this embodiment, the device includes a third acquisition module 41 and a third training module 42.

[0151] Among them, the third acquisition module is used to obtain a training data set, which includes training input data and training supervision data; the third training module is used to use the training data set to train a pre-established deep learning network. When the loss of the training output result relative to the training supervision data reaches a second preset condition, the training is terminated to obtain a trained target deep learning network model.

[0152] Specifically, the third acquisition module can obtain a training data set from a preset storage area and send it to a third training module. The third training module uses the training data set to train a pre-established deep learning network. When the loss of the training output result relative to the training supervision data reaches a second preset condition, the training is terminated to obtain a trained target deep learning network model.

[0153] In one embodiment, the third training module is specifically configured to:

[0154] Obtain a training information bit stream and a corresponding pilot; encode the training information bit stream to obtain an encoding result; modulate the pilot to obtain a first modulation result, and modulate the encoding result to obtain a second modulation result; determine the real data to be sent based on the first modulation result and the second modulation result, the real data to be sent including the real part of the pilot, the imaginary part of the pilot and the real data of the training information bit stream; add an initial phase to the real data to be sent and perform an inverse fast Fourier transform to obtain a second transformation result; filter the second transformation result to obtain a filtering result; perform parallel-to-serial conversion on the filtering result to obtain a training signal to be sent; transmit the training signal to be sent through a communication channel to obtain a training received signal; demodulate the training received signal to obtain corresponding training demodulated data; extract the real part and the imaginary part of the training demodulated data to form training real data corresponding to the training demodulated data; use the training real data as training input data and use the encoding result as training supervision data.

[0155] In practical applications, the third training module can also be divided into multiple sub-modules. The specific division method can be set according to actual needs and is not limited in this embodiment.

[0156] In one embodiment, the third acquisition module is also used to obtain channel training feature data and corresponding label data; the third training module is also used to train a pre-established deep neural network based on the channel training feature data and the corresponding label data. If the loss reaches a first preset condition, the training is terminated and the trained deep neural network model is obtained as a channel classification model for channel scene type detection.

[0157] Another embodiment of the present invention provides an electronic device for executing the method provided in the above embodiment. The electronic device may be a receiver or other implementable computer device.

[0158] like Figure 12 FIG. 5 is a schematic diagram of the structure of an electronic device provided in this embodiment. The electronic device 50 includes a memory 51 , a transceiver 52 and at least one processor 53 .

[0159] Among them, the processor, memory and transceiver are interconnected through circuits; the memory stores computer-executable instructions; the transceiver is used to receive signals sent by the receiving end; at least one processor executes the computer-executable instructions stored in the memory, so that at least one processor executes the method provided in any of the above embodiments.

[0160] Specifically, the transceiver receives the signal sent by the receiving end and sends it to the processor, and the processor reads and executes the computer execution instructions stored in the memory to implement the method provided in any of the above embodiments.

[0161] Optionally, the transceiver may also receive training data, channel configuration files and other data input by the user, which may be specifically set according to actual needs.

[0162] The electronic device provided by the present invention can be applied to any scenario based on the FBMC communication system, and is specifically applied to receiving-end channel estimation and equalization. In practical applications, the electronic device can refer to a receiver or a computer device independent of the receiver and connected to the receiver, and can be specifically set according to actual needs.

[0163] It should be noted that the electronic device of this embodiment can implement the method provided by any of the above embodiments and can achieve the same technical effects, which will not be described in detail here.

[0164] Yet another embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the method provided in any of the above embodiments is implemented.

[0165] It should be noted that the computer-readable storage medium of this embodiment can implement the method provided by any of the above embodiments and can achieve the same technical effects, which will not be described in detail here.

[0166] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0167] Other embodiments of the present invention will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the invention being indicated by the following claims.

[0168] It should be understood that the present invention is not limited to the exact construction described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof, which is limited only by the appended claims.

Claims

1. A signal detection method, characterized in that: include: Demodulating the received signal to obtain corresponding demodulated data; Extracting a real part and an imaginary part of the demodulated data to form real number data corresponding to the demodulated data; Inputting the real number data into a pre-trained target deep learning network model to obtain an output result; Obtaining a detection result of the original transmitted signal according to the output result; The signal detection method is applicable to FBMC communication systems with multiple scene switching. The signal detection method further comprises: for channels of different scenarios, pre-training a network model to obtain LSTM parameters corresponding to the scenario channels; Correspondingly, the inputting of the real number data into the pre-trained target deep learning network model specifically includes: obtaining the characteristic data of the target channel of the current transmission signal; inputting the characteristic data of the target channel into a trained channel classification model to obtain the scene type of the target channel, and the channel classification model is a deep neural network model; obtaining the corresponding target LSTM parameters according to the scene type of the target channel; and using the network model based on the target LSTM parameters as the target LSTM network model.

2. The method according to claim 1, characterized in that The demodulating the received signal to obtain corresponding demodulated data includes: Performing serial-to-parallel conversion on the received signal to obtain a corresponding parallel signal; Performing inverse filtering on the parallel signal using an inverse prototype filter to obtain an inverse filtering result; Performing a fast Fourier transform on the inverse filtering result to obtain a first transform result; An initial phase is deleted from the first transformation result to obtain the demodulated data.

3. The method according to claim 1, characterized in that Obtaining a detection result of the original transmitted signal according to the output result includes: The output result is decoded and parallel-to-serial converted using a Viterbi decoding algorithm to obtain a detection result of the original transmitted signal.

4. The method according to claim 1, wherein The method further comprises: Obtain channel training feature data and corresponding label data; Based on the channel training feature data and the corresponding label data, a pre-established deep neural network is trained. If the loss reaches a first preset condition, the training is terminated to obtain the deep neural network model.

5. A signal detection device, characterized in that: include: A demodulation module, used to demodulate the received signal to obtain corresponding demodulated data; An extraction module, configured to extract a real part and an imaginary part from the demodulated data to form real data corresponding to the demodulated data; A detection module is used to input the real number data into a pre-trained target deep learning network model to obtain an output result; A processing module, configured to obtain a detection result of the original transmitted signal according to the output result; The signal detection device is applicable to FBMC communication systems with multiple scene switching. The signal detection device further includes: a second training module for pre-training a network model of LSTM parameters corresponding to channels in different scenarios; A first acquisition module is used to acquire characteristic data of a target channel of a current transmission signal; A classification module, configured to input the characteristic data of the target channel into a trained channel classification model to obtain the scene type of the target channel, wherein the channel classification model is a deep neural network model; A determination module is used to obtain corresponding target LSTM parameters according to the scene type of the target channel, and use a network model based on the target LSTM parameters as the target LSTM network model.

6. An electronic device, characterized in that: include: memory, a transceiver, and at least one processor; The processor, the memory and the transceiver are interconnected via a circuit; The memory stores computer-executable instructions; The transceiver is used to receive the signal sent by the transmitting end; The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and when a processor executes the computer-executable instructions, the method according to any one of claims 1 to 4 is implemented.

Images