Intelligent channel decoding method and system based on complex neural network

Abstract

The application discloses an intelligent channel decoding method and system based on a complex neural network, which can directly process complex signals in practical application, improve the performance of an intelligent channel decoder by using the advantages of complex signals, and optimize an intelligent communication scheme. After being trained by sufficient data, the channel decoder based on the complex neural network can obtain an intelligent channel decoding system based on the complex neural network, and decoding after convolution classification can reduce signaling overhead while ensuring communication quality. The application can solve the problem of intelligent and reliable information transmission in a complex signal space, and can be further applied to intelligent terminals and devices to improve the compatibility and upgradability of the intelligent terminals and devices and enhance the reliability of time-varying wireless channel information transmission. Meanwhile, the application can effectively reduce the energy consumption of a system, thereby bringing ecological benefits.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and more specifically, to an intelligent channel decoding method and system based on complex neural networks. Background Technology

[0002] Wireless signals based on the classical Maxwell's electromagnetic field equations have both real and imaginary parts in their transmission space. However, existing neural network architectures do not consider the design of a two-dimensional signal space. Whether it's the transmission mechanism proposed by Professor Jin Shi based on digital image processing or the equalizer designed by Professor Li Ye based on deep learning, both employ deep learning neural network architectures based on real numbers. Most existing intelligent communication solutions can only process real number sequences, requiring specific signal forms; or they separate complex number signal sequences into real and imaginary parts for processing. However, compared to the original problem, this doubles the dimensionality of the generated algorithm and may corrupt special data structures.

[0003] In fields such as wireless communication and audio processing, complex numbers are generated naturally or through design, so there is a certain correlation between the real and imaginary parts of a signal. Using neural networks to directly process complex signals can better extract signal features. Moreover, in the signal processing process, constructing analytic signals can reduce the bandwidth by half, and the calculations are simpler when representing real signals. Therefore, we extend the processing object of neural networks to the complex domain to improve the performance of intelligent channel decoders.

[0004] Existing technologies provide a deep learning-based intelligent communication method. This method enables intelligent communication modules based on a cognitive loop model to autonomously perceive and recognize communication signals through a "learning-perception-decision-response" cognitive loop model. A robust convolutional neural network model is trained and integrated into a software-defined radio (SDR) host computer to achieve real-time recognition. Various modulation signal samples containing random noise are pre-generated, and a deep learning framework is built for training. The trained model and its parameters are saved to form the required automatic classifier. The trained model is integrated into the receiver's SDR computer, and the identified modulation scheme is then demodulated using appropriate methods. Summary of the Invention

[0005] To overcome the shortcomings of existing channel decoders that cannot directly process complex signals, this invention provides an intelligent channel decoding method and system based on complex neural networks, which can directly process complex signals and improve the performance of the channel decoder by utilizing the processed complex signals.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0007] This invention provides an intelligent channel decoding method based on complex neural networks, the method comprising:

[0008] S1: Obtain the complex original information bit sequence and perform preprocessing to obtain the received codeword sequence;

[0009] S2: The channel decoder based on the complex neural network is trained and optimized using the received codeword sequence to obtain the optimized channel decoder based on the complex neural network.

[0010] S3: Obtain the complex information bit sequence to be decoded, input it into the optimized channel decoder based on complex neural network for decoding, and obtain the decoding result of the complex information bit sequence to be decoded.

[0011] Preferably, the specific method of S2 is as follows:

[0012] S21: Construct a channel decoder based on a complex neural network, including a classifier network unit and a decoder network unit connected in sequence;

[0013] S22: The classifier network unit is used to perform feature extraction, normalization, pooling and classification operations on the received codeword sequence in sequence to obtain the predicted classification vector and the corresponding predicted category label of the received codeword sequence.

[0014] S23: Set the classification loss function, calculate the classification loss function value based on the predicted class label and the true class label, and when the classification loss function value reaches the minimum value, input the predicted classification vector of the optimized received codeword sequence and the corresponding predicted class label into the decoder network unit;

[0015] S24: The decoder network unit decodes the predicted classification vector and the corresponding predicted category label of the optimized received codeword sequence and outputs the predicted information bit sequence.

[0016] S25: Set the decoding loss function, calculate the decoding loss function value based on the predicted information bit sequence and the real information bit sequence, and adjust the parameters of the constructed complex neural network model;

[0017] S26: When the decoding loss function reaches its minimum value, the trained complex neural network model is obtained.

[0018] Preferably, in S21, the classifier network unit includes a first complex convolutional layer, a first batch normalization layer, a first max pooling layer, a second complex convolutional layer, a second batch normalization layer, a second max pooling layer, a first fully connected layer, a second fully connected layer, and a third fully connected layer connected in sequence.

[0019] The output of the third fully connected layer is connected to the decoder network unit; the third fully connected layer outputs the predicted category label.

[0020] Preferably, in S21, the decoder network unit includes a fourth fully connected layer, a fifth fully connected layer, an addendum point, a sixth fully connected layer, and a seventh fully connected layer connected in sequence;

[0021] The output of the third fully connected layer is connected to the input of the fourth fully connected layer and the summing point, respectively; the seventh fully connected layer outputs a prediction information bit sequence.

[0022] Preferably, the specific method of S22 is as follows:

[0023] S221: After the received codeword sequence is processed by the first convolutional layer for feature extraction, the first convolutional feature matrix is ​​obtained;

[0024] S222: The first convolutional feature matrix is ​​normalized by the first batch of standard layers to obtain the first normalized feature matrix;

[0025] S223: The first normalized feature matrix is ​​obtained by pooling through the first max pooling layer to obtain the first pooled feature matrix;

[0026] S224: Repeat steps S221-S223 to obtain the second pooling feature matrix;

[0027] S225: The second pooling feature matrix is ​​classified through the first fully connected layer, the second fully connected layer and the third fully connected layer connected in sequence to obtain the predicted classification vector and the corresponding predicted category label of the received codeword sequence.

[0028] Preferably, in S23, the formula for the classification loss function is:

[0029] The formula for the classification loss is:

[0030]

[0031]

[0032] in, η is the predicted type label; l is the true category label; η is the first set of optimization parameters.

[0033] Preferably, in S24, the output predicted bit information sequence is:

[0034]

[0035] in, σ is the predicted classification vector of the received codeword sequence; σ4 is the activation function of the fourth fully connected layer; σ5 is the activation function of the fifth fully connected layer; σ6 is the activation function of the sixth fully connected layer; σ7 is the activation function of the seventh fully connected layer; θ is the second set of optimized parameters; Q1, Q2, Q3 and Q4 are the first set of parameters in the second set of optimized parameters; D1, D2, D3 and D4 are the second set of parameters in the second set of optimized parameters.

[0036] Preferably, in S25, the formula for the decoding loss function is:

[0037]

[0038]

[0039] in, denoted as b, which represents the predicted information bit sequence; b represents the actual information bit sequence; and θ represents the second set of optimized parameters.

[0040] Preferably, after step S26, the method further includes binary decoding of the predicted information bit sequence to obtain the BER value using the following formula:

[0041]

[0042] in, To predict the information bit sequence, It is a sequence of real information bits;

[0043]

[0044] Where |C| represents the number of codewords in codeword sequence C; K represents the number of information bits in a codeword; Predicted information bit sequence; b ik It is a complex sequence of original information bits.

[0045] The present invention also provides an intelligent channel decoding system based on a complex neural network for implementing the above method, the system comprising:

[0046] The data acquisition module acquires the complex raw information bit sequence and performs preprocessing to obtain the received codeword sequence;

[0047] The training module uses the received codeword sequence to train and optimize the constructed channel decoder based on the complex neural network, thereby obtaining the optimized channel decoder based on the complex neural network.

[0048] The decoding module acquires the complex information bit sequence to be decoded, inputs it into the optimized channel decoder based on a complex neural network for decoding, and obtains the decoding result of the complex information bit sequence to be decoded.

[0049] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0050] This invention can directly process complex signals in practical applications, leveraging the advantages of complex signals to improve the performance of intelligent channel decoders and optimize intelligent communication schemes. The channel decoder based on a complex neural network, after sufficient data training, can yield an intelligent channel decoding system based on a complex neural network. Decoding after convolutional classification can reduce signaling overhead while ensuring communication quality. It can solve the problem of intelligent and reliable information transmission in complex signal space, and can be further applied to intelligent terminals and devices, improving their compatibility and scalability, and enhancing the reliability of time-varying wireless channel information transmission. Simultaneously, this invention can effectively reduce system energy consumption, thereby bringing ecological benefits. Attached Figure Description

[0051] Figure 1 This is a flowchart of the intelligent channel decoding method based on complex neural networks described in Example 1;

[0052] Figure 2 This is a schematic diagram of the classifier network unit described in Example 2;

[0053] Figure 3 This is a schematic diagram of the decoder network unit described in Example 2;

[0054] Figure 4 This is a schematic diagram of the intelligent channel decoding system based on complex neural networks described in Example 3. Detailed Implementation

[0055] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.

[0056] To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions;

[0057] It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.

[0058] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0059] Example 1

[0060] This embodiment provides an intelligent channel decoding method based on complex neural networks, such as... Figure 1 As shown, the method includes:

[0061] S1: Obtain the complex original information bit sequence and perform preprocessing to obtain the received codeword sequence;

[0062] S2: The channel decoder based on the complex neural network is trained and optimized using the received codeword sequence to obtain the optimized channel decoder based on the complex neural network.

[0063] S3: Obtain the complex information bit sequence to be decoded, input it into the optimized channel decoder based on complex neural network for decoding, and obtain the decoding result of the complex information bit sequence to be decoded.

[0064] Example 2

[0065] This embodiment provides an intelligent channel decoding method based on complex neural networks, the method comprising:

[0066] S1: Obtain the complex original information bit sequence and perform preprocessing to obtain the received codeword sequence;

[0067] S2: The channel decoder based on the complex neural network is trained and optimized using the received codeword sequence to obtain the optimized channel decoder based on the complex neural network.

[0068] S21: Construct a channel decoder based on a complex neural network, including a classifier network unit and a decoder network unit connected in sequence;

[0069] like Figure 2 As shown, the classifier network unit includes a first complex convolutional layer, a first batch normalization layer, a first max pooling layer, a second complex convolutional layer, a second batch normalization layer, a second max pooling layer, a first fully connected layer, a second fully connected layer, and a third fully connected layer connected in sequence.

[0070] The output of the third fully connected layer is connected to the decoder network unit; the third fully connected layer outputs the predicted category label.

[0071] like Figure 3 As shown, the decoder network unit includes a fourth fully connected layer, a fifth fully connected layer, an addendum point, a sixth fully connected layer, and a seventh fully connected layer connected in sequence.

[0072] The output of the third fully connected layer is connected to the input of the fourth fully connected layer and the summing point, respectively; the seventh fully connected layer outputs a prediction information bit sequence.

[0073] S22: The classifier network unit is used to perform feature extraction, normalization, pooling and classification operations on the received codeword sequence in sequence to obtain the predicted classification vector and the corresponding predicted category label of the received codeword sequence.

[0074] The input dimension of the complex convolutional layer is initialized to (8,1,1), the kernel dimension is (5,1), the output space dimension is (8,64,1), the number of groups is 2, and the activation function is the non-complex analytic function f(z)=tanh(|z|)exp(iarg(z)), where z is the output data after grouped convolution.

[0075] S221: After the received codeword sequence is processed by the first convolutional layer for feature extraction, the first convolutional feature matrix is ​​obtained;

[0076] Specifically, this involves first using a complex convolution function to calculate the input sample data. The corresponding eigenvalues: The output matrix x is the feature matrix of the data, and the size of x is [N, C, H, W]. N is the batch size, C is the channel dimension, and H and W are the spatial height and width axes.

[0077] S222: The first convolutional feature matrix is ​​normalized by the first batch of standard layers to obtain the first normalized feature matrix;

[0078] The feature matrix x of the data is obtained from the complex convolutional layer, and the feature x is normalized. The data feature x obtained from the previous layer is then used as input, and its shape is reshaped by dividing the channel dimension C into G groups, resulting in the input feature matrix shape being [N,G,C / / G,H,W]:

[0079] x'=f(x)

[0080] The function of f(x) is to reshape the shape, updating x to x'.

[0081] Calculate the mean and variance of the features within each group:

[0082]

[0083]

[0084] Where S i It is the set for calculating the mean and standard deviation of the i-th group in sequence, and m is the size of this set.

[0085] Normalize the features within each group by subtracting the mean and dividing by the standard deviation to obtain the normalized features:

[0086]

[0087] The normalized features are restored to their original shape, which becomes [N, C, H, W]:

[0088] x = shape_func(x_gn)

[0089] The normalized features are scaled and translated, and multiplication and addition operations are performed using gamma and beta:

[0090] h=γx+β

[0091] S223: The first normalized feature matrix is ​​obtained by pooling through the first max pooling layer to obtain the first pooled feature matrix;

[0092] After obtaining the normalized feature matrix y, perform max pooling:

[0093]

[0094]

[0095] S224: Repeat steps S221-S223 to obtain the second pooling feature matrix;

[0096] S225: The second pooling feature matrix is ​​classified through the first fully connected layer, the second fully connected layer and the third fully connected layer connected in sequence to obtain the predicted classification vector and the corresponding predicted category label of the received codeword sequence.

[0097] S23: Set the classification loss function, calculate the classification loss function value based on the predicted class label and the true class label, and when the classification loss function value reaches the minimum value, input the predicted classification vector of the optimized received codeword sequence and the corresponding predicted class label into the decoder network unit;

[0098] The formula for the classification loss is:

[0099]

[0100]

[0101] in, η is the predicted type label; l is the true category label; η is the first set of optimization parameters.

[0102] S24: The decoder network unit decodes the predicted classification vector and the corresponding predicted category label of the optimized received codeword sequence and outputs the predicted information bit sequence.

[0103] In S24, the output sequence of predicted bit information is:

[0104]

[0105] in, σ is the predicted classification vector of the received codeword sequence; σ4 is the activation function of the fourth fully connected layer; σ5 is the activation function of the fifth fully connected layer; σ6 is the activation function of the sixth fully connected layer; σ7 is the activation function of the seventh fully connected layer; θ is the second set of optimized parameters; Q1, Q2, Q3 and Q4 are the first set of parameters in the second set of optimized parameters; D1, D2, D3 and D4 are the second set of parameters in the second set of optimized parameters.

[0106] S25: Set the decoding loss function, calculate the decoding loss function value based on the predicted information bit sequence and the real information bit sequence, and adjust the parameters of the constructed complex neural network model;

[0107] The formula for the decoding loss function is:

[0108]

[0109]

[0110] in, denoted as b, which represents the predicted information bit sequence; b represents the actual information bit sequence; and θ represents the second set of optimized parameters.

[0111] S26: When the decoding loss function reaches its minimum value, the trained complex neural network model is obtained.

[0112] It also includes binary decoding of the predicted information bit sequence, and the formula for obtaining the BER value is:

[0113]

[0114] in, To predict the information bit sequence, It is a sequence of real information bits;

[0115]

[0116] Where |C| represents the number of codewords in codeword sequence C; K represents the number of information bits in a codeword; Predicted information bit sequence; b ik It is a complex sequence of original information bits.

[0117] S3: Obtain the complex information bit sequence to be decoded, input it into the optimized channel decoder based on complex neural network for decoding, and obtain the decoding result of the complex information bit sequence to be decoded.

[0118] Example 3

[0119] This embodiment provides an intelligent channel decoding system based on a complex neural network, used to implement the methods of Embodiments 1 and 2, such as... Figure 4As shown, the system includes:

[0120] The data acquisition module acquires the complex raw information bit sequence and performs preprocessing to obtain the received codeword sequence;

[0121] The training module uses the received codeword sequence to train and optimize the constructed channel decoder based on the complex neural network, thereby obtaining the optimized channel decoder based on the complex neural network.

[0122] The decoding module acquires the complex information bit sequence to be decoded, inputs it into the optimized channel decoder based on a complex neural network for decoding, and obtains the decoding result of the complex information bit sequence to be decoded.

[0123] The same or similar labels correspond to the same or similar parts;

[0124] The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0125] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A smart channel decoding method based on complex neural networks, characterized in that, The method includes: S1: Obtain the complex original information bit sequence and perform preprocessing to obtain the received codeword sequence; S2: The channel decoder based on the complex neural network is trained and optimized using the received codeword sequence to obtain the optimized channel decoder based on the complex neural network. S3: Obtain the complex information bit sequence to be decoded, input it into the optimized channel decoder based on complex neural network for decoding, and obtain the decoding result of the complex information bit sequence to be decoded; The specific method of S2 is as follows: S21: Construct a channel decoder based on a complex neural network, including a classifier network unit and a decoder network unit connected in sequence; S22: The classifier network unit is used to perform feature extraction, normalization, pooling and classification operations on the received codeword sequence in sequence to obtain the predicted classification vector and the corresponding predicted category label of the received codeword sequence. S23: Set the classification loss function, calculate the classification loss function value based on the predicted class label and the true class label, and when the classification loss function value reaches the minimum value, input the predicted classification vector of the optimized received codeword sequence and the corresponding predicted class label into the decoder network unit; S24: The decoder network unit decodes the predicted classification vector and the corresponding predicted category label of the optimized received codeword sequence and outputs the predicted information bit sequence. S25: Set the decoding loss function, calculate the decoding loss function value based on the predicted information bit sequence and the real information bit sequence, and adjust the parameters of the constructed complex neural network model; S26: When the decoding loss function reaches its minimum value, the trained complex neural network model is obtained.

2. The intelligent channel decoding method based on complex neural networks according to claim 1, characterized in that, In S21, the classifier network unit includes a first complex convolutional layer, a first batch normalization layer, a first max pooling layer, a second complex convolutional layer, a second batch normalization layer, a second max pooling layer, a first fully connected layer, a second fully connected layer, and a third fully connected layer connected in sequence. The output of the third fully connected layer is connected to the decoder network unit; the third fully connected layer outputs the predicted category label.

3. The intelligent channel decoding method based on complex neural networks according to claim 2, characterized in that, In S21, the decoder network unit includes a fourth fully connected layer, a fifth fully connected layer, an addendum point, a sixth fully connected layer, and a seventh fully connected layer connected in sequence. The output of the third fully connected layer is connected to the input of the fourth fully connected layer and the summing point, respectively; the seventh fully connected layer outputs a prediction information bit sequence.

4. The intelligent channel decoding method based on complex neural networks according to claim 2, characterized in that, The specific method of S22 is as follows: S221: After the received codeword sequence is processed by the first convolutional layer for feature extraction, the first convolutional feature matrix is ​​obtained; S222: The first convolutional feature matrix is ​​normalized by the first batch of standard layers to obtain the first normalized feature matrix; S223: The first normalized feature matrix is ​​obtained by pooling through the first max pooling layer to obtain the first pooled feature matrix; S224: Repeat steps S221-S223 to obtain the second pooling feature matrix; S225: The second pooling feature matrix is ​​classified through the first fully connected layer, the second fully connected layer and the third fully connected layer connected in sequence to obtain the predicted classification vector and the corresponding predicted category label of the received codeword sequence.

5. The intelligent channel decoding method based on complex neural networks according to claim 1, characterized in that, In S23, the formula for the classification loss function is: The formula for the classification loss is: in, η is the predicted type label; l is the true category label; η is the first set of optimization parameters.

6. The intelligent channel decoding method based on complex neural networks according to claim 1, characterized in that, In S24, the output sequence of predicted bit information is: in, The predicted classification vector; This is the activation function for the fourth fully connected layer; This is the activation function for the fifth fully connected layer; This is the activation function for the sixth fully connected layer; This is the activation function for the seventh fully connected layer; ; , , and The first parameter set in the second set of optimization parameters; , , and This is the second parameter set in the second optimization parameter set.

7. The intelligent channel decoding method based on complex neural networks according to claim 1, characterized in that, In S25, the formula for the decoding loss function is: in, b is the actual information bit sequence; θ is the second set of optimized parameters.

8. The intelligent channel decoding method based on complex neural networks according to claim 1, characterized in that, Following step S26, the method further includes binary decoding of the predicted information bit sequence to obtain the BER value using the following formula: in, To predict the information bit sequence, It is a sequence of real information bits; Where |C| represents the number of codewords in the codeword sequence C; K represents the number of information bits in the codeword; Predict the information bit sequence; It is a complex sequence of original information bits.

9. A smart channel decoding system based on a complex neural network, used to implement the method described in any one of claims 1-8, characterized in that, The system includes: The data acquisition module acquires the complex raw information bit sequence and performs preprocessing to obtain the received codeword sequence; The training module uses the received codeword sequence to train and optimize the constructed channel decoder based on the complex neural network, thereby obtaining the optimized channel decoder based on the complex neural network. The decoding module acquires the complex information bit sequence to be decoded, inputs it into the optimized channel decoder based on a complex neural network for decoding, and obtains the decoding result of the complex information bit sequence to be decoded.

Images