Multi-node cooperative interference identification method based on interference-to-noise ratio adaptive node selection

Abstract

The invention discloses a multi-node cooperative interference identification method based on interference-to-noise ratio adaptive node selection, and belongs to the technical field of wireless communication anti-interference. The method comprises the following steps: constructing a multi-node cooperative interference identification network; training an interference identification model based on a convolutional neural network (CNN) and sharing network parameters; each collaborative sensing node independently performs interference identification and calculates an interference to noise ratio; the recognition result and the interference-noise ratio are transmitted back to the center node; the center node selects a high-quality node based on an adaptive interference-noise ratio threshold; and performing weighted fusion on the identification results of the selected nodes to obtain a global identification result. According to the invention, by introducing an interference-to-noise ratio sensing node selection mechanism and an adaptive threshold adjustment strategy, the negative influence of low-quality nodes on the overall recognition performance is effectively avoided, the recognition accuracy of an interference recognition algorithm under a low interference-to-noise ratio is improved, and the interference recognition performance under a complex electromagnetic environment is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication signal processing, and particularly relates to a multi-node cooperative interference identification method based on adaptive node selection of a jamming-to-noise ratio. BACKGROUND

[0002] Due to the openness and broadcast nature of a wireless channel, the wireless communication signals transmitted by the wireless channel are vulnerable to attacks from malicious jamming, which is a main factor threatening the survivability of communication. In view of this problem, people have proposed various anti-jamming technologies, such as frequency hopping, direct sequence spread spectrum, and anti-jamming methods based on game theory, optimal power control methods, and the like. It is worth noting that the premise for effectively implementing these anti-jamming technologies is accurate interference identification.

[0003] In a wireless communication network, a single-node interference identification method uses one node to complete the reception, processing, and identification of a jamming signal, and the performance of the method is easily affected by a channel environment. Therefore, technical personnel have borrowed from a cognitive radio cooperative spectrum sensing method, and through a multi-sensing node cooperative interference identification manner, the interference identification performance in a low jamming-to-noise ratio (JNR) situation is improved.

[0004] Existing multi-node cooperative interference identification methods mainly adopt a center decision or a hard decision manner. In the center decision method, a center node receives original signal information from M cooperative cognitive nodes, and transmits the information back to the center node for processing and neural network training. In the hard decision method, each cooperative cognitive node uses a trained neural network to make a classification decision, independently obtains an identification result, and transmits the identification result back to the center node. According to the sensing information of each cooperative cognitive node, the center node performs data fusion, and according to a majority decision criterion, obtains a global identification result.

[0005] However, the existing technology has the following deficiencies:

[0006] (1) In an actual scene, the JNRs of different cooperative nodes are different, and the identification result of a high JNR node will be affected by a low JNR node, resulting in a decrease in the overall identification accuracy;

[0007] (2) The existing method performs equal-weight fusion on the identification results of all nodes, without considering the difference in signal quality of the nodes;

[0008] (3) There is a lack of a dynamic adjustment mechanism for node selection, and the mechanism cannot adapt to a complex and changeable electromagnetic environment. SUMMARY

[0009] The present application aims to overcome the deficiencies of the prior art, and provides a multi-node cooperative interference identification method based on adaptive node selection of a jamming-to-noise ratio, which improves the interference signal identification accuracy.

[0010] The purpose of the present application is achieved by the following scheme:

[0011] A multi-node cooperative interference identification method based on dry noise ratio adaptive node selection, characterized by comprising the following steps:

[0012] S1, a multi-node cooperative interference identification system model is constructed, including 1 center node and M cooperative cognitive nodes;

[0013] S2, training an interference identification model based on a convolutional neural network to obtain trained network parameters;

[0014] S3, sharing the trained network parameters to each cooperative cognitive node;

[0015] S4, each cooperative cognitive node independently performs interference identification and simultaneously calculates the JNR value of the node;

[0016] S5, the cooperative node returns the identification result and the JNR value of the node to the center node;

[0017] S6, the center node performs adaptive node selection based on the JNR threshold to select high-quality nodes;

[0018] S7, the selected node identification results are weighted and fused to obtain the global identification result.

[0019] Further, in step S1, the multi-node cooperative interference identification system model is as follows:

[0020] Assuming that the multi-node cooperative interference identification network is composed of 1 center node and M cooperative cognitive nodes, the signal received by the i-th cognitive node can be represented as:

[0021] ;

[0022] wherein, is a communication signal, is an interference signal, is a noise signal, and the noise signal obeys a complex Gaussian distribution with a mean of 0 and a variance of N is the number of signal sampling points.

[0023] Further, in step S1, the sample vector received by the M cognitive nodes at the nth moment can be represented as: The received information is summarized to the center node, and is used to represent the received information matrix composed of M cognitive nodes within the cognitive time period, i.e.

[0024] ;

[0025] Further, in step S2, before training the neural network model, the received signal needs to be pre-processed, and after receiving the information from each collaborative cognitive node, the center node will aggregate the information to the neural network input. The input layer will perform FFT transformation on the received signal matrix, i.e. , the real part and the imaginary part of the FFT of the received signal matrix can be represented as and , respectively. and are taken as the input of the neural network model in the form of pages, i.e. the neural network input is a two-page matrix.

[0026] Further, in step S2, after pre-processing the received signal, the input neural network, the network model is trained until the network reaches a convergent state.

[0027] Further, in step S3, the trained network parameters are shared with each collaborative cognitive node.

[0028] Further, in step S4, since the trained network parameters are shared with each collaborative cognitive node in step S3, each collaborative cognitive node independently identifies the received interference signal, in addition, each collaborative cognitive node calculates the JNR value, and performs spectrum analysis on the received discrete sequence , calculates the power spectral density, and then calculates the interference power and the noise power according to the power spectral density, respectively.

[0029] Further, in step S5, each node packs the identification result and value and returns it to the center node through the control channel, and the data packet format is: [node ID | identification result | JNR value | time stamp];

[0030] Further, in step S6, the high-quality nodes are screened based on the JNR threshold value, and the specific method is as follows:

[0031] (1) Initialization of JNR threshold value:

[0032] Set the initial JNR threshold value

[0033] (2) Adaptive adjustment of JNR threshold value:

[0034] The center node dynamically adjusts the JNR threshold value based on the historical identification performance, and the specific adjustment strategy is as follows:

[0035] a) Performance evaluation period setting:

[0036] Set the evaluation period The system recognition accuracy in each evaluation period ;

[0037] b) Threshold adjustment decision:

[0038] If , it indicates that the current threshold setting is too conservative, and the threshold can be appropriately lowered to include more nodes. The threshold is: ;

[0039] If , it indicates that the current threshold setting is too low, and the threshold needs to be raised to exclude low-quality nodes. The threshold is: ;

[0040] If , it indicates that the current threshold setting is reasonable, and the threshold remains unchanged.

[0041] Wherein, is the target accuracy, is the tolerance range, and are the threshold adjustment steps, is the evaluation period index.

[0042] c) Threshold boundary constraint:

[0043] To prevent excessive adjustment of the threshold, set the threshold range constraint:

[0044] ;

[0045] Wherein, is the minimum threshold, is the maximum threshold.

[0046] (3) Threshold adjustment based on environmental perception:

[0047] In addition to performance-based threshold adjustment, the system can also fine-tune the threshold based on the characteristics of the electromagnetic environment. First, evaluate the environmental complexity by calculating the variance of each node JNR:

[0048] ;

[0049] Then, based on the environmental complexity, the threshold is modified as follows:

[0050] If , it indicates that the environmental complexity is high, and the node quality difference is large, so the threshold can be appropriately raised: ;

[0051] If , it indicates that the environment is relatively uniform, and the threshold can be appropriately lowered: ;

[0052] wherein, and is an environmental complexity threshold, and is an adjustment coefficient;

[0053] (4) Node selection

[0054] Based on the self-adaptive adjustment of the JNR threshold, a node set satisfying the condition is selected:

[0055] ;

[0056] Further, in step S7, the selected node recognition result is weighted and fused, including weight calculation and weighted voting, and the specific method is as follows:

[0057] (1) Weight calculation

[0058] For the selected node , the weight is calculated as:

[0059] ;

[0060] wherein, is the JNR value of the node, and is the JNR threshold.

[0061] (2) Weighted voting

[0062] The recognition result of each node is weighted according to the weight , and the final recognition result is:

[0063] ;

[0064] wherein, is the interference category, and is an indicator function. BRIEF DESCRIPTION OF DRAWINGS

[0065] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained according to these drawings without creative labor.

[0066] Figure 1 is a step flow chart of the method used in the embodiments of the present application;

[0067] Figure 2 Structure diagram of convolutional neural network in an embodiment of the present application;

[0068] Figure 3 Performance comparison diagram of the present application and hard decision based cooperative interference identification in an embodiment of the present application. DETAILED DESCRIPTION

[0069] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0070] In one embodiment, as shown in Figure 1 , a multi-node cooperative interference identification method based on adaptive node selection of JNR is provided, comprising the following steps:

[0071] S1, a multi-node cooperative interference identification system model is constructed, including 1 center node and M cooperative cognitive nodes;

[0072] S2, an interference identification model based on convolutional neural network is trained to obtain trained network parameters;

[0073] S3, the trained network parameters are shared to each cooperative cognitive node;

[0074] S4, each cooperative cognitive node independently performs interference identification and simultaneously calculates the JNR value of the node;

[0075] S5, the cooperative node transmits the identification result and the JNR value of the node back to the center node;

[0076] S6, the center node performs adaptive node selection based on the JNR threshold to filter high-quality nodes;

[0077] S7, the selected node identification result is weighted and fused to obtain the global identification result.

[0078] In further other embodiments, the present application first constructs a system model, assuming that a multi-node cooperative interference identification network is composed of 1 center node and M cooperative cognitive nodes, and the signal received by the i-th cognitive node can be expressed as:

[0079] ;

[0080] wherein, is a communication signal, is an interference signal, is a noise signal, and the noise signal is subject to a mean of 0 and a variance of , where N is the number of signal samples.

[0081] The sample vector received by the M cognitive nodes at the n th moment can be expressed as: The received information is summarized to the center node, and the input of the neural network model is expressed as: , where R is the received information matrix of the M cognitive nodes in the cognitive time period, and

[0082] ;

[0083] In further other embodiments, before training the neural network model, the input received signal needs to be preprocessed, which includes communication signals, interference signals and noise signals. The communication signal adopts QPSK signal, and the interference signal is six common malicious interference patterns, which are single tone interference, multi-tone interference, narrowband interference, wideband interference, comb interference and sweep frequency interference. The channel is an additive white Gaussian noise channel. After receiving the information of each cooperative cognitive node, the center node summarizes the information to the neural network input. The input layer performs FFT transformation on the received signal matrix, that is: , then the real part and the imaginary part of the FFT of the received signal matrix can be expressed as:

[0084] ;

[0085] ;

[0086] and are taken as the input of the neural network model in the form of pages, that is, the neural network input is a two-page matrix.

[0087] The structure of the neural network is shown in Figure 2 . After preprocessing the neural network input, it passes through 6 convolution modules, each of which contains a convolution layer, a batch normalization layer and a max pooling layer. Finally, an average pooling layer and a fully connected layer are used for classification output. There are 6 convolution layers, and the number of convolution kernels is 16, 24, 32, 48, 64 and 96 respectively. ReLU is used as the activation function of each hidden layer:

[0088] ;

[0089] The network uses Aadm optimization algorithm, and cross-entropy is used as the loss function:

[0090] ;

[0091] , where represents the true interference classification label, represents the predicted interference classification label. ​

[0092] In further other embodiments, the trained network parameters are shared to each cooperative sensing node.

[0093] In further other embodiments, due to the trained network parameters are shared to each cooperative sensing node, each cooperative sensing node independently identifies the received interference signal, in addition, each cooperative sensing node calculates the JNR value, and performs spectrum analysis on the received discrete sequence , calculates the power spectral density:

[0094] ;

[0095] wherein, is the frequency domain representation of the signal, is the number of FFT points. Set the frequency threshold , divide the spectrum into interference frequency band and noise frequency band . Calculate the interference power and noise power respectively:

[0096] ;

[0097] ;

[0098] wherein, is the frequency resolution.

[0099] Calculate JNR:

[0100] ;

[0101] In further other embodiments, each node packs the identification result and value, and returns it to the center node through the control channel, and the data packet format is: [node ID | identification result | JNR value | time stamp];

[0102] In further other embodiments, the high-quality nodes are screened based on the JNR threshold value, and the specific method is as follows:

[0103] (1) Initialization of JNR threshold value:

[0104] Set the initial JNR threshold value , wherein is set to -10dB according to system requirements and historical experience in this embodiment;

[0105] (2) Adaptive adjustment of JNR threshold value:

[0106] The center node dynamically adjusts the JNR threshold value based on the historical identification performance, and the specific adjustment strategy is as follows:

[0107] a) Performance evaluation period setting:

[0108] Set evaluation period In this embodiment, 100 identification periods are set, and the system identification accuracy is counted in each evaluation period ;

[0109] b) Threshold adjustment decision:

[0110] If , it indicates that the current threshold setting is too conservative, and the threshold can be appropriately lowered to include more nodes, then the threshold is: ;

[0111] If , it indicates that the current threshold setting is too low, and the threshold needs to be raised to exclude low-quality nodes, then the threshold is: ;

[0112] If , it indicates that the current threshold setting is reasonable, and the threshold remains unchanged;

[0113] Wherein, is the target accuracy, which is set to 90% in this embodiment, is the tolerance range, which is set to 2% in this embodiment, and are the threshold adjustment steps, which are set to 1dB in this embodiment, is the evaluation period index;

[0114] c) Threshold boundary constraint:

[0115] To prevent excessive adjustment of the threshold, the threshold range constraint is set as:

[0116] ;

[0117] Wherein, is the minimum threshold, which is set to -15dB in this embodiment, is the maximum threshold, which is set to 0dB in this embodiment;

[0118] (3) Threshold adjustment based on environmental perception:

[0119] In addition to performance-based threshold adjustment, the system can also fine-tune the threshold based on the electromagnetic environment characteristics. First, the environmental complexity is evaluated by calculating the variance of each node JNR to evaluate the environmental complexity:

[0120] ;

[0121] Then, based on the environmental complexity, the threshold is corrected as follows:

[0122] If , it indicates that the environment complexity is high, and the node quality difference is large, so the threshold is appropriately increased: ;

[0123] If , it indicates that the environment is relatively uniform, so the threshold can be appropriately reduced: ;

[0124] wherein, and are the environment complexity thresholds, which are set to 5dB and 2dB respectively in the embodiment, and are adjustment coefficients.

[0125] (4) Node selection

[0126] Based on the adaptively adjusted JNR threshold, the node set meeting the condition is selected:

[0127] ;

[0128] In further other embodiments, the selected node recognition result is weighted and fused, including weight calculation and weighted voting, and the specific method is as follows:

[0129] (1) Weight calculation

[0130] For the selected node , the weight is calculated as:

[0131] ;

[0132] wherein, is the JNR value of the node, and is the JNR threshold.

[0133] (2) Weighted voting

[0134] The recognition result of each node is weighted according to the weight , and the final recognition result is:

[0135] ;

[0136] wherein, is the interference category, is the indication function.

[0137] Experimental verification and performance analysis

[0138] In order to verify the effectiveness of the method of the present application, simulation experiments are carried out.Figure 3 A performance comparison diagram of the application and the hard decision based cooperative jamming recognition is given. Figure 3 From the experimental results, it can be seen that under the condition of very low JSR (JSR=-15dB), the multi-node cooperative jamming recognition method based on the JNR adaptive node selection proposed in the application has an identification accuracy of about 9% higher than the traditional hard decision method. This proves that by introducing the node selection mechanism and the adaptive threshold adjustment strategy of JNR perception, the negative influence of low-quality nodes on the overall recognition performance is effectively avoided, and the jamming recognition performance in complex electromagnetic environment is significantly improved.

[0139] The application is not limited to the above-mentioned embodiments, and those skilled in the art can make several improvements and refinements without departing from the principles of the application, and these improvements and refinements are also considered within the protection scope of the application.

Claims

1. A multi-node cooperative interference identification method based on interference-to-noise ratio adaptive node selection, characterized in that: The following steps are involved: S1, build a multi-node cooperative interference recognition network, including 1 central node and M cooperative recognition nodes; S2, train the CNN-based interference recognition model to obtain the trained network parameters; S3, shares the trained network parameters with each collaborative sensing node; S4, each collaborative sensing node independently identifies interference and calculates the node's JNR value at the same time; S5, the collaborative node transmits the recognition result and the node’s JNR value back to the central node; S6, the central node performs adaptive node selection and screens high-quality nodes based on the JNR threshold; S7, performing weighted fusion on the selected node recognition results to obtain a global recognition result.

2. The method according to claim 1, characterized in that In step 1, the signal received by the i-th cognitive node Expressed as: ； in, For communication signals, is the interference signal, is a noise signal with a mean of 0 and a variance of Complex Gaussian distribution, N is the number of signal sampling points; The sample vector received by M cognitive nodes at time n can be expressed as: 。 3. The method according to claim 1, characterized in that In step 2, the input received signal is preprocessed, the received signal matrix is ​​subjected to FFT transformation, and the real and imaginary parts of the FFT are used as inputs to the neural network model in the form of pages; the neural network model includes 6 convolution modules, and the number of convolution kernels is 16, 24, 32, 48, 64, and 96 respectively.

4. The method according to claim 1, wherein The method for calculating the JNR value in step 4 includes: performing spectrum analysis on the received discrete sequence to calculate the power spectrum density; calculating the interference power and noise power ; Calculate the JNR value: 。 5. The method according to claim 1, wherein The method for screening high-quality nodes based on the JNR threshold in step 6 includes: Setting the initial JNR threshold Dynamically adjust the JNR threshold based on historical recognition performance: like , then the threshold is: ; like , then the threshold is: ; like , the threshold remains unchanged; in, is the target accuracy, For the tolerance range, and is the threshold adjustment step size, is the evaluation period index; Set the threshold range constraint: ; Based on the adaptively adjusted JNR threshold, select the node set: 。 6. The method according to claim 5, characterized in that Also includes threshold adjustment based on context awareness: Calculate the variance of JNR of each node To assess the complexity of the environment: ； like , then increase the threshold: ; like , then lower the threshold: ; in, and is the environmental complexity threshold, and is the adjustment factor.

7. The method according to claim 1, characterized in that The method for weighted fusion of the selected node recognition results in step 7 includes: Weight calculation: ； in, for The JNR value of the node, is the JNR threshold; Weighted Voting: ； in, is the interference category, is the indicator function.

8. The method according to any one of claims 1 to 7, characterized in that The interference signals include six types: single-tone interference, multi-tone interference, narrowband interference, broadband interference, comb interference and swept-frequency interference.

Images