Morse signal fast decoding method of improved CTC

Abstract

The invention discloses an improved CTC Morse signal fast decoding method, which comprises the following steps: carrying out short-time Fourier transform on extracted Morse signal data to obtain a time-frequency matrix, extracting one-dimensional features of Morse signals in the time-frequency matrix, and forming a training set by tags corresponding to the one-dimensional features; constructing a deep learning framework, wherein the deep learning framework comprises a deep residual shrinkage module, a full connection layer and a softmax structure; a middle layer of the deep residual shrinkage module is output through a full connection layer and a softmax structure to serve as a sub-model, CTC loss of the sub-model is calculated and weighted with CTC loss of a full model to serve as a loss function, and a network model is trained; and inputting the extracted one-dimensional features of the Morse signal time-frequency matrix into the trained network model, and performing CTC-based decoding post-processing on the output to obtain a decoding result. According to the method, the CTC loss of the full model and the loss of the sub-model are combined for training, so that the calculation overhead is very low, and the decoding accuracy can be effectively improved.

Description

Technical Field

[0001] The invention relates to a fast decoding method for Morse signals of an improved CTC, belonging to the technical field of automatic decoding of Morse telegraphs. Background Art

[0002] Morse code is a communication method that uses a combination of dots (short signals) and dashes (long signals) to represent letters, numbers, and symbols. It requires very little bandwidth and is suitable for low-bandwidth environments. In addition, the rules of Morse code are simple and can be mastered after a short period of training. In the event of failure of modern communication systems, Morse code is an important means of emergency communication. Morse code can still maintain a high recognition rate in a noisy environment because its signal characteristics are obvious and it has strong anti-interference ability. Morse code can be transmitted over long distances through radio waves and is particularly suitable for long-distance communications, such as maritime and aviation fields. With the development of radio communications, Morse code has also become an important coding method in radio communications and is widely used in military and civilian communications. However, traditional Morse code decoding mainly relies on the operator's auditory recognition ability. The operator needs to undergo special training to master the sending and receiving of Morse code. In implementation, the operator may cause decoding errors due to fatigue or lack of concentration, and the operator's decoding speed is limited and cannot meet the needs of high-speed communication. Therefore, the research on end-to-end automatic Morse signal decoding methods is very valuable and necessary.

[0003] Traditional machine learning methods such as support vector machines (SVM) and clustering algorithms can achieve automatic decoding of Morse signals under high signal-to-noise ratios, but these methods are sensitive to noise, and the decoding accuracy is greatly reduced at low signal-to-noise ratios. At present, the mainstream automatic decoding method is based on deep learning. After preprocessing the extracted Morse signal data, short-time Fourier transform or wavelet transform is used to obtain the time-frequency feature map of the Morse signal as network input, and deep learning models such as convolutional neural networks (CNN), recurrent neural networks (RNN) or long short-term memory networks (LSTM) are used to extract features. The CTC (Connectionist Temporal Classification) loss function is used during the training process. Its greedy search algorithm can solve the problem that the length of the input feature sequence is greater than the length of the output message during decoding.

[0004] It is worth mentioning that in the current decoding methods based on deep learning, the network input is a time-frequency diagram obtained from a short message, which has a small amount of information and is easy to train and reason. However, it is impossible to separate and splice from long messages to short messages. In addition, the time-frequency diagram input to the network is of fixed size, which has great limitations on the length of the message and cannot realize the decoding of Morse signals of variable length. In addition, due to the extremely small bandwidth occupied by Morse signals, the frequency bands without Morse signal information in the time-frequency diagram occupy a large area. These redundant information undoubtedly make the calculation complexity of the reasoning process higher, which may lead to limited real-time processing capabilities and occupy unnecessary computing resources. In complex practical application scenarios, decoding may face problems such as low signal-to-noise ratio and large-scale fading of Morse signals, which affect the extraction of Morse signal features and make the decoding results incorrect. Therefore, it is necessary to design an automatic decoding method with certain noise robustness and anti-fading capabilities. The deep learning structure needs to supervise the features extracted by networks of different depths to ensure the reliability and accuracy of the decoding results. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a fast decoding method for Morse signals based on an improved CTC, which can effectively extract the deep features of the model signal and improve the decoding accuracy in scenarios such as low Morse signal-to-noise ratio and large-scale fading.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] A method for fast decoding of Morse signals by improving CTC comprises the following steps:

[0008] Step 1, obtain the Morse signal and perform short-time Fourier transform to obtain a time-frequency matrix, extract the one-dimensional features of the Morse signal in the time-frequency matrix, and form a training set with the one-dimensional features of the Morse signal and the label corresponding to the Morse signal;

[0009] Step 2, constructing a deep learning network model, including a deep residual shrinkage module, a first to a second fully connected layer, and a first to a second softmax function, wherein the deep residual shrinkage module, the first fully connected layer, and the first softmax function are connected in sequence, and the deep residual shrinkage module includes a one-dimensional convolution input layer and a plurality of basic modules connected in sequence, and a second fully connected layer and a second softmax function are connected in sequence after the mth basic module, where m=2,…,n-1, and n is the number of basic modules;

[0010] Step 3, the output of the mth basic module after passing through the second fully connected layer and the second softmax function is used as the sub-model, and the CTC loss of the sub-model is calculated. The output of the last basic module after passing through the first fully connected layer and the first softmax function is used as the full model, and the CTC loss of the sub-model and the CTC loss of the full model are weighted as the loss function. The deep learning network model is trained using the training set to obtain the Morse signal decoding model.

[0011] Step 4, extracting the one-dimensional features of the time-frequency matrix of the Morse signal to be decoded, inputting it into the Morse signal decoding model, performing CTC decoding post-processing on the output of the model, and obtaining the decoding result of the Morse signal to be decoded.

[0012] As a preferred solution of the present invention, in step 1, the one-dimensional features of the Morse signal in the time-frequency matrix are extracted, specifically as follows:

[0013] Take the square value of the short-time Fourier transform of the Morse signal x(n) to obtain the time-frequency matrix X(k,l) of x(n), where k and l represent the column and row of the matrix, i.e., the frequency axis and time axis index, respectively. The calculation formula is:

[0014]

[0015] Where w(n) is the window function, N is the length of the window function, L is the window step, and K is the number of discrete Fourier transform points;

[0016] The one-dimensional feature of the Morse signal is the row vector corresponding to the carrier frequency coordinate in the time-frequency matrix, that is, X(k c ,l), where k c Satisfy k c =N·f c / f s , f s is the Morse signal sampling frequency, f c is the carrier frequency of the Morse signal.

[0017] As a preferred solution of the present invention, in step 2, the basic module includes a first convolutional layer, a second convolutional layer, an SE channel attention mechanism, and a soft threshold processing unit connected in sequence, and the SE channel attention mechanism includes an average pooling layer, a third fully connected layer, a fourth fully connected layer, and a sigmoid activation function;

[0018] The input x of the basic module is obtained after the first convolution layer and the second convolution layer After the average pooling layer, it is reduced to a one-dimensional vector X. The one-dimensional vector X passes through the third fully connected layer and the fourth fully connected layer in turn, and then is activated by the sigmoid activation function to obtain a one-dimensional vector Y. Multiply X and Y to get the threshold T of each channel. T is a 1*1*C vector, and C is The number of channels; then use the threshold T Perform soft thresholding, the calculation formula is as follows:

[0019]

[0020] Among them, y thresholded for The result after soft thresholding using threshold T; finally, the cross-layer identity connection method is used to take y thresholded After adding it to the input x and passing through the ReLU activation function, it becomes the output of the basic module.

[0021] As a preferred solution of the present invention, in step 3, the loss function is calculated as follows:

[0022]

[0023] in, represents the loss function used in training, w is the weight, is the loss function of the whole model, is the loss function of the sub-model.

[0024] As a preferred solution of the present invention, in step 4, the CTC decoding post-processing is specifically as follows:

[0025] The number of input channels of the first fully connected layer is the number of channels of the Morse signal deep feature sequence output by the deep residual contraction module, and the number of output channels is the number of categories of the label characters plus 1. The probability distribution matrix of each character at each time step is obtained by the first softmax function.

[0026] Using the greedy decoding algorithm, in the probability distribution matrix output by the first softmax function, the character with the largest probability value in each time step is output, and then the decoding result is obtained after removing empty characters and repeated characters.

[0027] A computer device comprises a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, the steps of the improved CTC Morse signal fast decoding method are implemented.

[0028] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the improved CTC Morse signal fast decoding method are implemented.

[0029] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0030] 1. The present invention reduces the use of redundant information by extracting a one-dimensional feature sequence of Morse signals as the input of the network model, can effectively save computing resources, improve the reasoning speed, and can realize the decoding of Morse signals of any length.

[0031] 2. The present invention integrates the SE (Squeeze-and-Excitation) channel attention mechanism with soft thresholding processing, which can effectively reduce noise interference and make the model noise-robust when facing extreme noise.

[0032] 3. The present invention can extract the deep features of the sequence by stacking many basic modules of the deep residual shrinkage network, effectively reducing the impact of fading on the decoding results.

[0033] 4. The present invention supervises the features extracted by networks of different depths through weighted intermediate CTC loss and final layer CTC loss, which not only improves the training efficiency but also enables the model to focus on useful features earlier, thereby achieving a higher decoding accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is the overall flow chart of the method for fast decoding of Morse signals of CTC improved by the present invention;

[0035] Figure 2 This is an example of the time-frequency diagram of the Morse signal;

[0036] Figure 3 This is an example of a visualization of a 1D Morse signal sequence;

[0037] Figure 4 It is the basic module structure diagram;

[0038] Figure 5 It is the CTC loss calculation structure diagram;

[0039] Figure 6 It is the overall process of inference decoding. DETAILED DESCRIPTION

[0040] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0041] like Figure 1 As shown, the present invention proposes a fast decoding method of Morse signal of improved CTC, and the specific steps are as follows:

[0042] Step 1: Perform short-time Fourier transform on the extracted Morse signal data to obtain a time-frequency matrix, extract the one-dimensional features of the Morse signal in the time-frequency matrix as the input of the neural network, and use the corresponding labels to form a training set.

[0043] In step 1, the method for extracting the one-dimensional features of the Morse signal in the time-frequency matrix is ​​as follows:

[0044] Take the square value of the short-time Fourier transform of the Morse signal x(n) to obtain its time-frequency matrix X(k,l), and visualize the matrix as shown below Figure 2 As shown, k and l represent the column (frequency axis) and row (time axis) index of the matrix respectively, and the calculation formula is:

[0045]

[0046] Where w(n) is the added window function, N is the length of the window function, K is the number of discrete Fourier transform points, and L is the window step size. The one-dimensional feature of the Morse signal is the row vector corresponding to the carrier frequency coordinate in the time-frequency matrix, that is, X(k c ,l). Among them, k c Satisfy k c =N·f c / f s , f s is the Morse signal sampling frequency, f c is the carrier frequency of the Morse signal. The visualization results are as follows Figure 3 As shown, it can be seen that the Morse signal sequence after extracting the one-dimensional feature is a row of the time-frequency matrix.

[0047] Step 2: construct a deep learning framework, which includes a deep residual shrinkage module, the first to second fully connected layers, and the first to second softmax functions.

[0048] The deep learning framework in step 2 is as follows Figure 5 As shown, the description is as follows:

[0049] The deep residual shrinkage module consists of a one-dimensional convolution input layer and many basic modules. The basic module is composed of three parts, including the residual network module, the SE (Squeeze-and-Excitation) channel attention mechanism, and the soft threshold processing unit. The one-dimensional sequence of the Morse signal is input into the deep residual shrinkage network to obtain its high-channel deep features, and the output is recorded as the Morse signal deep feature sequence.

[0050] In the first fully connected layer and the first softmax function, the number of input channels of the first fully connected layer is the number of channels of the Morse signal deep feature sequence output by the deep residual contraction module, and the number of output channels is the number of categories of the label characters + 1. The probability distribution matrix of each character at each time step is obtained by the first softmax function.

[0051] The basic module structure is as follows Figure 4 As shown, the description is as follows:

[0052] Note that the input x of the basic module is x after two convolutional layers: According to the SE channel attention mechanism, It is globally average pooled and reduced to a 1*1*C vector. It is then input into two fully connected layers and activated by the sigmoid function. The 1*1*C vector at this time is multiplied by the 1*1*C vector after global average pooling to obtain the threshold of each channel, denoted as T, where T is a 1*1*C vector and C is Then use the threshold T to Perform soft thresholding, the calculation formula is as follows:

[0053]

[0054] Among them, y thresholded for The result after soft thresholding using threshold T. Finally, the cross-layer identity connection method is used to take y thresholded After adding it to the input x and passing through the ReLU activation function, it becomes the output of the basic module.

[0055] Step 3: The basic module in the middle of the deep residual contraction module is output through the second fully connected layer and the second softmax function as a sub-model. The CTC loss of the sub-model is calculated and weighted with the CTC loss of the full model as the loss function to train the network model.

[0056] In the middle layer of several layers of the basic module stacked in the deep residual contraction module, the output is passed through the second fully connected layer and the second softmax function. The model output by the middle layer is recorded as the sub-model, and the loss function is The loss function of the entire model is The total loss function of the training is the weighted sum of the two, and the calculation formula is as follows:

[0057]

[0058] in, is the loss function used in the final training, and w is the weight.

[0059] Step 4: Input the one-dimensional features of the extracted Morse signal time-frequency matrix into the trained network model, and perform post-processing based on CTC decoding on the network output to obtain the decoding result of the Morse signal.

[0060] CTC decoding post-processing uses a greedy decoding algorithm to take the character output with the largest probability value in each time step from the probability distribution matrix output by the first softmax function, and then removes empty characters and repeated characters to obtain the decoding result. The complete reasoning process from Morse signal audio to decoding results is as follows: Figure 6 shown.

[0061] Based on the same inventive concept, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the aforementioned improved CTC Morse signal fast decoding method are implemented.

[0062] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the aforementioned improved CTC Morse signal fast decoding method are implemented.

[0063] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0064] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0065] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0066] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0067] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A fast decoding method for Morse signals based on improved CTC, characterized in that: The steps include: Step 1, obtain the Morse signal and perform short-time Fourier transform to obtain a time-frequency matrix, extract the one-dimensional features of the Morse signal in the time-frequency matrix, and form a training set with the one-dimensional features of the Morse signal and the label corresponding to the Morse signal; Step 2, constructing a deep learning network model, including a deep residual shrinkage module, a first to a second fully connected layer, and a first to a second softmax function, wherein the deep residual shrinkage module, the first fully connected layer, and the first softmax function are connected in sequence, and the deep residual shrinkage module includes a one-dimensional convolution input layer and a plurality of basic modules connected in sequence, and a second fully connected layer and a second softmax function are connected in sequence after the mth basic module, where m=2,…,n-1, and n is the number of basic modules; Step 3, the output of the mth basic module after passing through the second fully connected layer and the second softmax function is used as the sub-model, and the CTC loss of the sub-model is calculated. The output of the last basic module after passing through the first fully connected layer and the first softmax function is used as the full model, and the CTC loss of the sub-model and the CTC loss of the full model are weighted as the loss function. The deep learning network model is trained using the training set to obtain the Morse signal decoding model. Step 4, extracting the one-dimensional features of the time-frequency matrix of the Morse signal to be decoded, inputting it into the Morse signal decoding model, performing CTC decoding post-processing on the output of the model, and obtaining the decoding result of the Morse signal to be decoded.

2. The improved CTC Morse signal fast decoding method according to claim 1, characterized in that: In step 1, the one-dimensional features of the Morse signal in the time-frequency matrix are extracted, as follows: Take the square value of the short-time Fourier transform of the Morse signal x(n) to obtain the time-frequency matrix X(k,l) of x(n), where k and l represent the column and row of the matrix, i.e., the frequency axis and time axis index, respectively. The calculation formula is: Where w(n) is the window function, N is the length of the window function, L is the window step, and K is the number of discrete Fourier transform points; The one-dimensional feature of the Morse signal is the row vector corresponding to the carrier frequency coordinate in the time-frequency matrix, that is, X(k c ,l), where k c Satisfy k c =N·f c / f s , f s is the Morse signal sampling frequency, f c is the carrier frequency of the Morse signal.

3. The improved CTC Morse signal fast decoding method according to claim 1, characterized in that: In the step 2, the basic module includes a first convolutional layer, a second convolutional layer, an SE channel attention mechanism, and a soft thresholding processing unit connected in sequence, and the SE channel attention mechanism includes an average pooling layer, a third fully connected layer, a fourth fully connected layer, and a sigmoid activation function; The input x of the basic module is obtained after the first convolution layer and the second convolution layer After the average pooling layer, it is reduced to a one-dimensional vector X. The one-dimensional vector X passes through the third fully connected layer and the fourth fully connected layer in turn, and then is activated by the sigmoid activation function to obtain a one-dimensional vector Y. Multiply X and Y to get the threshold T of each channel. T is a 1*1*C vector, and C is The number of channels; then use the threshold T Perform soft thresholding, the calculation formula is as follows: Among them, y thresholded for The result after soft thresholding using threshold T; finally, the cross-layer identity connection method is used to take y thresholded After adding it to the input x and passing through the ReLU activation function, it becomes the output of the basic module.

4. The improved CTC Morse signal fast decoding method according to claim 1, characterized in that: In step 3, the loss function is calculated as follows: in, represents the loss function used in training, w is the weight, is the loss function of the whole model, is the loss function of the sub-model.

5. The improved CTC Morse signal fast decoding method according to claim 1, characterized in that: In step 4, CTC decoding post-processing is specifically as follows: The number of input channels of the first fully connected layer is the number of channels of the Morse signal deep feature sequence output by the deep residual contraction module, and the number of output channels is the number of categories of the label characters plus 1. The probability distribution matrix of each character at each time step is obtained by the first softmax function. Using the greedy decoding algorithm, in the probability distribution matrix output by the first softmax function, the character with the largest probability value in each time step is output, and then the decoding result is obtained after removing empty characters and repeated characters.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the computer program, the steps of the method for fast decoding of Morse signals of improved CTC according to any one of claims 1 to 5 are implemented.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for improving the fast decoding of Morse signals of CTC according to any one of claims 1 to 5 are implemented.

Images