Radar communication integrated signal identification method
Through the training of radar communication signal data and network model based on signal-to-noise ratio packets, the problem of single signal recognition type and low accuracy in complex electromagnetic environments is solved, and accurate identification and efficient calculation of multiple radar communication signals are achieved.
Patent Information
- Application Number
- CN202510058043.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-14
- Publication Date
- 2025-05-06
AI Technical Summary
When identifying integrated radar communication signals, existing neural networks have problems with single signal type and low accuracy, which are difficult to apply to complex electromagnetic environments.
By generating a variety of radar communication signal data, based on signal-to-noise ratio packets, a network model including an input layer, a CNN module, an LSTM module and a fully connected layer is established, and trained to identify the types of radar communication signals in the composite signal.
It realizes accurate identification of multiple radar communication signals in complex electromagnetic environments, improves identification accuracy and calculation efficiency, and is suitable for the identification of multiple radar communication signals.
Smart Images
Figure CN119936799A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of radar communication technology, and in particular to a radar communication integrated signal recognition method. Background Art
[0002] With the increasing application of radar and communication, the problem of frequency band overlap between radar and communication signals is becoming more and more serious, especially in complex electromagnetic environments, where it is difficult to identify non-cooperative signals. At present, the most promising solution is the processing and identification of integrated radar and communication signals, among which modulation signal recognition is a key technology that needs to be solved urgently.
[0003] At present, among the modulation signal recognition methods, the most common one is to identify the integrated modulation signal of radar communication based on characteristic parameters. The characteristic quantities used for characteristic parameter recognition mainly include cyclic autocorrelation function, cyclic spectrum, high-order cumulant, etc. It is necessary to select features and design threshold standards based on experience and a large amount of data analysis, and the computational complexity is relatively low. In recent years, neural networks have been increasingly used in radar communication signal recognition. Based on the autonomous learning ability of neural networks, they are good at discovering the potential structure and pattern of high-dimensional data, and are very suitable for radar communication signal recognition.
[0004] Neural network recognition of radar communication signals is divided into direct recognition and indirect recognition. Indirect recognition requires preprocessing of the signal to extract feature quantities and then input them into the neural network to identify the modulation type; direct recognition directly inputs the IQ data of the radar communication signal into the neural network for recognition. The advantage is that it is based on data, does not require manual control or feature selection, and has better versatility and robustness.
[0005] However, existing neural networks for direct recognition of radar communication integrated signals generally have the problem of single recognition signal type and low accuracy, making them difficult to apply in complex electromagnetic environments. Summary of the invention
[0006] In view of the above analysis, an embodiment of the present invention aims to provide a radar communication integrated signal recognition method to solve the common problems of existing neural networks in recognizing a single signal type and low accuracy.
[0007] On the one hand, an embodiment of the present invention provides a radar communication integrated signal recognition method, the method specifically comprising:
[0008] Generate multiple radar communication signal data, and mark the category of each radar communication signal data, each radar communication signal data and its marked category as a sample; group the radar communication signal samples based on the signal-to-noise ratio, establish a data set for each group of radar communication signal samples, and establish a training set and a validation set based on each data set;
[0009] Establishing a network model, the network model includes an input layer, a first CNN module, a second CNN module, a first LSTM module, a second LSTM module, a fully connected layer, and an output layer connected in sequence; dividing the radar communication signal samples in each training set into multiple batches and sequentially inputting them into the network model for training, and after each training cycle, inputting the corresponding verification set into the network model and outputting the recognition accuracy;
[0010] When the recognition accuracy of each verification set reaches the preset threshold, the training is completed;
[0011] After collecting a composite signal composed of a mixture of multiple radar communication signals and inputting it into the trained network model, the types of radar communication signals and the corresponding radar communication signals included in the composite signal are identified.
[0012] Based on the further improvement of the above method, the radar communication signal is the original IQ data.
[0013] Based on further improvements of the above method, the number of radar communication signal samples in each batch is 64.
[0014] Based on the further improvement of the above method, the first CNN module includes a sequence folding layer, a first CNN module convolution layer, a first CNN module normalization layer, a first CNN module Relu layer, and a first CNN module pooling layer connected in sequence, wherein:
[0015] The sequence folding layer is used to convert each batch of input data sequences into a data matrix;
[0016] After the first CNN module convolutional layer is used to extract the local features of each element in the data matrix,
[0017] Then, it is processed in sequence by the normalization layer of the first CNN module, the Relu layer of the first CNN module, and the pooling layer of the first CNN module to generate a first feature matrix and output it to the second CNN module.
[0018] Based on the further improvement of the above method, the second CNN module includes a second CNN module convolution layer, a second CNN module normalization layer, a second CNN module Relu layer, a second CNN module pooling layer, a sequence unfolding layer, and an unfolding layer connected in sequence, wherein:
[0019] After the first feature matrix is further processed by the convolution layer of the second CNN module to extract local features, it is processed by the normalization layer of the second CNN module, the Relu layer of the second CNN module, and the pooling layer of the second CNN module to generate the second feature matrix;
[0020] The sequence unfolding layer is used to restore each element in the second feature matrix to the same position as in the input data sequence to obtain a third feature matrix;
[0021] The expansion layer is used to expand the third feature matrix to obtain a first data sequence and output it.
[0022] Based on the further improvement of the above method, the number of convolution kernels of the first CNN module convolution layer and the second CNN module convolution layer is 256, and the size of each convolution kernel is 3.
[0023] Based on the further improvement of the above method, the first data sequence is successively passed through the first LSTM module and the second LSTM module to extract long-term dependencies and timing pattern features to obtain the second data sequence, wherein the first LSTM module and the second LSTM module are both provided with a Dropout layer, wherein the dropout rate of each Dropout layer is set to 20%.
[0024] Based on the further improvement of the above method, the initial value of the learning rate of the network model is set to 0.001; after every 5 trainings, the learning rate is reduced by 20%.
[0025] Based on the further improvement of the above method, the number of neurons in the fully connected layer of the network model is set to the number of radar communication signal types used for training.
[0026] Based on further improvement of the above method, the preset threshold is 99%.
[0027] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:
[0028] 1. The generated various radar communication signal data are grouped into multiple data sets based on the signal-to-noise ratio, and each data set is used to train the established network model, so that the network model can fully learn the characteristics of various radar communication signals of each signal-to-noise ratio and have better generalization.
[0029] 2. The network model has a simple structure and does not need to extract the characteristic quantities of radar communication signals. It can directly identify and output each signal modulation type included in the composite radar communication signal IQ data under a complex electromagnetic environment. It is suitable for the identification of various radar communication signals under complex electromagnetic environments, with high computational efficiency and high recognition accuracy.
[0030] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. In the entire drawings, the same reference symbols represent the same components;
[0032] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION
[0033] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.
[0034] A specific embodiment of the present invention discloses a radar communication integrated signal recognition method, such as Figure 1 shown.
[0035] Generate multiple radar communication signal data, and mark the category of each radar communication signal data, each radar communication signal data and its marked category as a sample; group the radar communication signal samples based on the signal-to-noise ratio, establish a data set for each group of radar communication signal samples, and establish a training set and a validation set based on each data set;
[0036] Establishing a network model, the network model includes an input layer, a first CNN module, a second CNN module, a first LSTM module, a second LSTM module, a fully connected layer, and an output layer connected in sequence; dividing the radar communication signal samples in each training set into multiple batches and sequentially inputting them into the network model for training, and after each training cycle, inputting the corresponding verification set into the network model and outputting the recognition accuracy;
[0037] When the recognition accuracy of each validation set reaches the preset threshold, the training is completed;
[0038] After collecting a composite signal composed of a mixture of multiple radar communication signals and inputting it into the trained network model, the types of radar communication signals and the corresponding radar communication signals included in the composite signal are identified.
[0039] The first step is to generate a variety of radar communication signal samples and mark the categories of each radar communication signal data. Each radar communication signal data and its marked category is regarded as a sample; the radar communication signal samples are grouped based on the signal-to-noise ratio, and a data set is established for each group of radar communication signal samples. A training set and a validation set are established based on each data set.
[0040] Specifically, the category of the radar communication signal can be one-hot encoded. For example, for 10 types of radar communication signals, each bit in the code represents a type of radar communication signal. The code (1, 0, 0, 0, 0, 0, 0, 0, 0) indicates that the type of the signal is 2FSK, and (0, 1, 0, 0, 0, 0, 0, 0, 0) indicates that the type of the signal is MSK.
[0041] In the field of deep learning, the premise of training the required network model is to build a high-quality data set. Based on the application of the network model in this embodiment to the integrated signal recognition of composite radar communication in a complex electromagnetic environment, the actual scene is first simulated to randomly generate multiple types of radar communication signal data.
[0042] Preferably, this embodiment randomly generates 10 typical radar communication modulation signals, and the signal types include: 2FSK, MSK, QPSK, π / 4-DQPSK, 16QAM, LFM, NLFM, Costas frequency coding, Barker coding, and Frank coding.
[0043] Preferably, the signal-to-noise ratio of the generated radar communication signal samples ranges from 15 to 20 dB, 10 signal-to-noise ratios are selected, and 1,000 signal samples are randomly selected under each signal-to-noise ratio as a group to establish a data set, each data set includes 10 types of radar communication signal data, and the radar communication signal samples in each data set are radar communication original IQ data sequences arranged by the 10 types of radar communication signal data.
[0044] Each modulation signal in each group is set with different parameters such as symbol sequence, bandwidth, pulse width, etc. Preferably, the radar communication modulation signal range is 5-100 Hz, and the pulse width range is 5-50 μs.
[0045] The order of radar communication signal samples in each data set is randomly disrupted, and then 60% of the data is taken out from each data set to establish a training set, and 20% of the data is used as a validation set; the remaining 20% of the data is used as a test set, among which the training set and validation set are used to train the network model, and the test set is used to evaluate the performance of the network model.
[0046] Step 2: Establish a network model, which includes an input layer, a first CNN module, a second CNN module, a first LSTM module, a second LSTM module, a fully connected layer, and an output layer connected in sequence; divide the radar communication signal data in each training set into multiple batches according to a preset data length, and input them into the network model for training in sequence. After each training is completed, input the corresponding verification set into the network model and output the recognition accuracy.
[0047] Specifically, the training of the network model is carried out in sequence according to the data sets grouped by signal-to-noise ratio. The training sets corresponding to each signal-to-noise ratio are all input into the network model once, and the network model recognizes and outputs the radar communication signal type recognition result. Then, the verification sets corresponding to the input training sets are all input into the network model to complete one recognition and classification, which is regarded as completing one training.
[0048] The training set is first input into the input layer of the network model, which reads each radar communication signal sample in turn and then inputs them into the network model in batches.
[0049] Furthermore, each radar communication signal sample is raw IQ data.
[0050] Preferably, the original IQ data includes three data fields, which respectively store the I component, Q component and signal type tag of the IQ data, so as to facilitate subsequent identification and processing of the network model.
[0051] Furthermore, the number of segmented training set data input into the network model is 64.
[0052] The purpose is to help speed up the training speed and efficiency of network models.
[0053] Furthermore, the first CNN module includes a sequence folding layer, a first CNN module convolution layer, a first CNN module normalization layer, a first CNN module Relu layer, and a first CNN module pooling layer connected in sequence, wherein:
[0054] The sequence folding layer is used to convert each batch of input data sequences into a data matrix;
[0055] After the first CNN module convolutional layer is used to extract the local features of each element in the data matrix,
[0056] Then, it is processed in sequence by the normalization layer of the first CNN module, the Relu layer of the first CNN module, and the pooling layer of the first CNN module to generate the first feature map and output it to the second CNN module.
[0057] Specifically, the sequence folding layer converts each batch of input data sequence into a data matrix, with the purpose of converting the sequence data into a data matrix that is easy for the CNN network to process, which helps to enhance the network model's extraction of local patterns and features of radar communication signal data and improve the network model's ability to understand radar communication signal data. Next, the convolution layer of the first CNN module extracts local features from each element in the data matrix output by the sequence folding layer to generate a feature map, which is then processed by the first CNN module Relu layer and the first CNN module pooling layer in turn to generate the first feature map.
[0058] Furthermore, the second CNN module includes a second CNN module convolution layer, a second CNN module normalization layer, a second CNN module Relu layer, a second CNN module pooling layer, a sequence unfolding layer, and an unfolding layer connected in sequence, wherein:
[0059] After the first feature map is further processed by the convolution layer of the second CNN module to extract local features, it is processed by the normalization layer of the second CNN module, the Relu layer of the second CNN module, and the pooling layer of the second CNN module to generate a second feature map;
[0060] The sequence unfolding layer is used to restore each element in the second feature map to the same position as in the input data sequence to obtain a third feature map;
[0061] The expansion layer is used to expand the third feature map to obtain a first data sequence and output it.
[0062] Furthermore, the number of convolution kernels of the first CNN module convolution layer and the second CNN module convolution layer is 256, and the size of each convolution kernel is 3.
[0063] Two convolution layers are the best choice for extracting local pattern features of radar communication signals. After the first feature map is processed by the second CNN module convolution layer to extract local features, it is processed by the second CNN module normalization layer, the second CNN module Relu layer, and the second CNN module pooling layer to generate the second feature map; each element in the second feature map already includes the local feature information of the input batch radar communication data, and the dimension of the second feature map has been restored to be consistent with the data matrix; then, the sequence unfolding layer restores each element in the second feature map to the same position as the input data sequence to obtain the third feature map, the purpose is to return each radar communication data that has been extracted through the convolution operation to the position when it was input, so that the subsequent expansion layer can expand the third feature map into the first data sequence with the same data order as the input data sequence, which is convenient for subsequent processing of the network model.
[0064] Furthermore, the first data sequence is sequentially passed through the first LSTM module and the second LSTM module to extract long-term dependencies and timing pattern features to obtain a second data sequence, wherein the first LSTM module and the second LSTM module are both provided with a Dropout layer, wherein the dropout rate of each Dropout layer is set to 20%.
[0065] Next, the first LSTM module and the second LSTM module extract long-term dependencies and temporal pattern features from the first data sequence in turn. This process is performed serially. Based on the LSTM module's ability to process long-term dependencies and temporal pattern features, the correlation of each radar communication signal data in the input data sequence is further identified, thereby further improving the network model's ability to recognize and learn different types of radar communication signal data. Among them, the first LSTM module and the second LSTM module are both set with a Dropout layer, and the Dropout layer discard rate is set to 20% in order to prevent overfitting and improve the generalization ability of the model.
[0066] Each element in the second data sequence obtained after being processed sequentially by the first LSTM module and the second LSTM module includes both the local features of the IQ data and the long-term dependencies and timing pattern features of the IQ data. Further, the second data sequence is input into the fully connected layer.
[0067] Furthermore, the number of neurons in the fully connected layer of the network model is set to the number of radar communication signal types used for training.
[0068] In this embodiment, the number of neurons is 10, and the local features, long-term dependencies, and timing pattern features of each element of the second data sequence are mapped to each radar communication signal data in the second data sequence based on the fully connected layer, so that the network model ultimately outputs the probability of each radar communication signal.
[0069] When each batch of data in the training set is input into the network model to complete one training, the type of each radar communication signal sample obtained is identified, and then compared with the signal type label of each radar communication signal sample to calculate the cross entropy loss function. The model parameters are updated according to the loss function. When all batches of data have completed training, it is considered that a training cycle is completed, and the corresponding verification set is input into the network model for recognition to obtain the recognition accuracy.
[0070] Furthermore, the initial value of the learning rate of the network model is set to 0.001; after each 5 trainings, the learning rate is reduced by 20%.
[0071] Each training set is input into the network model multiple times for training. After every 5 training times, the learning rate is reduced to 80% of the learning rate of the last completed learning round. The initial learning rate is 100%. After 5 training times, the learning rate is reduced to 80% of the initial learning rate, that is, 0.0008. After another 5 training times, the learning rate is reduced to 0.00064, and so on.
[0072] Step 3: When the recognition accuracy of each verification set reaches the preset threshold, the training is completed.
[0073] Furthermore, the preset threshold is 99%.
[0074] The training of the network model for each training set and validation set is carried out sequentially, including:
[0075] S1: Take a set of training sets and validation sets as the current grouping;
[0076] S2: Use the current group to train the network model. After each training cycle, input the current group verification set into the network model to identify and output the recognition accuracy.
[0077] S3: Determine whether the recognition accuracy reaches the preset threshold. If yes, complete the training of the network model for the current group, take the next group as the current group, set the network model learning rate to the initial learning rate, and return to execute S2-S3;
[0078] Otherwise, after adjusting the parameters of the network model, return to execute S2-S3.
[0079] When the recognition accuracy of each verification set reaches the preset threshold, the training of the network model is completed.
[0080] Step 4: Collect a composite signal composed of a mixture of multiple radar communication signals and input it into the trained network model to identify the types of radar communication signals and corresponding radar communication signals included in the composite signal.
[0081] After the network model is trained, it can be put into practical application to collect composite signals mixed with multiple radar communication signals in complex electromagnetic environments. Among them, it only needs to collect the composite signal with original IQ data and input it into the network model to identify the type of radar communication signal and the corresponding radar communication signal included in the composite signal.
[0082] In this embodiment, the performance of the network model is evaluated by applying each test set to simulate a composite signal in a complex electromagnetic environment.
[0083] Specifically, the confusion matrix is used to observe the classification accuracy of each radar communication signal.
[0084] Table 1 shows the recognition accuracy of 10 radar communication signals under 4 signal-to-noise ratios of this embodiment using the network model. It can be seen that the recognition rate is as high as 99.6% under high signal-to-noise ratio conditions. As the signal-to-noise ratio decreases, the recognition performance decreases, but even when the signal-to-noise ratio is negative, the recognition accuracy still reaches a high level of 87.67%.
[0085] Table 1 Recognition accuracy under different signal-to-noise ratios
[0086] SNR / dB -5 0 5 10 Accuracy / % 87.67 88.33 98.33 99.60
[0087] The present embodiment discloses a radar communication integrated signal recognition method, which first generates a plurality of radar communication signal data, groups the radar communication signal data based on the signal-to-noise ratio, establishes a data set for each group, and establishes a corresponding training set and a verification set for training a network model based on each data set; then establishes a network model consisting of an input layer, a first CNN module, a second CNN module, a first LSTM module, a second LSTM module, a fully connected layer, and an output layer connected in sequence, uses each group of training sets to train the network model, and uses a corresponding verification set to input the network model output recognition accuracy after each training to evaluate the training effect; when the recognition accuracy of each group of verification sets exceeds a preset threshold, the network model training is completed; a composite signal consisting of a plurality of radar communication signals is directly input into the trained network model to identify the type of radar communication signals included in the composite signal and the corresponding radar communication signals, which solves the problem of single type and low accuracy in identifying radar communication signals in a complex electromagnetic environment compared to the prior art.
[0088] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.
[0089] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.
Claims
1. A radar communication integrated signal recognition method, characterized in that: The method specifically comprises: Generate multiple radar communication signal data, and mark the category of each radar communication signal data, each radar communication signal data and its marked category as a sample; group the radar communication signal samples based on the signal-to-noise ratio, establish a data set for each group of radar communication signal samples, and establish a training set and a validation set based on each data set; Establishing a network model, the network model includes an input layer, a first CNN module, a second CNN module, a first LSTM module, a second LSTM module, a fully connected layer, and an output layer connected in sequence; dividing the radar communication signal samples in each training set into multiple batches and sequentially inputting them into the network model for training, and after each training cycle, inputting the corresponding verification set into the network model and outputting the recognition accuracy; When the recognition accuracy of each verification set reaches the preset threshold, the training is completed; After collecting a composite signal composed of a mixture of multiple radar communication signals and inputting it into the trained network model, the types of radar communication signals and the corresponding radar communication signals included in the composite signal are identified.
2. The radar communication integrated signal recognition method according to claim 1, characterized in that: The radar communication signal is raw IQ data.
3. The radar communication integrated signal recognition method according to claim 2, characterized in that: The number of radar communication signal samples in each batch is 64.
4. The radar communication integrated signal recognition method according to claim 3, characterized in that: The first CNN module includes a sequence folding layer, a first CNN module convolution layer, a first CNN module normalization layer, a first CNN module Relu layer, and a first CNN module pooling layer connected in sequence, wherein: The sequence folding layer is used to convert each batch of input data sequences into a data matrix; After the first CNN module convolutional layer is used to extract the local features of each element in the data matrix, Then, it is processed in sequence by the normalization layer of the first CNN module, the Relu layer of the first CNN module, and the pooling layer of the first CNN module to generate the first feature map and output it to the second CNN module.
5. The radar communication integrated signal recognition method according to claim 4, characterized in that: The second CNN module includes a second CNN module convolution layer, a second CNN module normalization layer, a second CNN module Relu layer, a second CNN module pooling layer, a sequence unfolding layer, and an unfolding layer connected in sequence, wherein: After the first feature map is processed by the second CNN module convolution layer to extract local features, it is processed by the second CNN module normalization layer, the second CNN module Relu layer, and the second CNN module pooling layer in sequence to generate a second feature map; The sequence unfolding layer is used to restore each element in the second feature map to the same position as in the input data sequence to obtain a third feature map; The expansion layer is used to expand the third feature map to obtain a first data sequence and output it.
6. The radar communication integrated signal recognition method according to claim 5, characterized in that: The number of convolution kernels of the first CNN module convolution layer and the second CNN module convolution layer is 256, and the size of each convolution kernel is 3.
7. A radar communication integrated signal recognition method according to claim 6, characterized in that: The first data sequence is sequentially passed through the first LSTM module and the second LSTM module to extract long-term dependencies and timing pattern features to obtain a second data sequence, wherein the first LSTM module and the second LSTM module are both provided with a Dropout layer, wherein the dropout rate of each Dropout layer is set to 20%.
8. The radar communication integrated signal recognition method according to claim 7, characterized in that: The initial value of the learning rate of the network model is set to 0.001; after every 5 trainings, the learning rate is reduced by 20%.
9. The radar communication integrated signal recognition method according to claim 8, characterized in that: The number of neurons in the fully connected layer of the network model is set to the number of radar communication signal types used for training.
10. The radar communication integrated signal recognition method according to claim 1, characterized in that: The preset threshold is 99%.