A distributed compressive sensing system and method for wideband communication anti-jamming

Abstract

The application discloses a kind of distributed compressed sensing systems and methods for broadband communication anti-interference, comprising, sensor, be configured as, the sub-band signal of broadband signal to be perceived is perceived, and the short-time spectrum of sub-band is obtained;Encoder, be configured as, according to the short-time spectrum of sub-band, compressed encoding operation is carried out, and the encoding vector is obtained;Binary module, be configured as, according to the encoding vector, binary operation is carried out, and binary spectrum information compression frame is obtained;Decoder, be configured as, according to binary spectrum information compression frame, spectrum decoding operation is carried out, and the reconstruction short-time spectrum of sub-band is obtained;Fusion center, be configured as, according to the reconstruction short-time spectrum of sub-band, broadband spectrum is obtained, to realize the distributed sensing and compressed reconstruction of broadband signal to be perceived.The application can effectively reduce the high-speed A / D requirement of broadband sensing, and there is no obvious performance loss when being used for intelligent anti-interference decision.

Description

Technical Field

[0001] This invention relates to a distributed compressed sensing system and method for anti-interference in broadband communication, belonging to the field of wireless communication technology. Background Technology

[0002] With the rapid development of information technology, various electromagnetic interference technologies are widely used in military wireless countermeasures. How to effectively counter electromagnetic interference is a key issue that military wireless communication needs to address. In communication countermeasures, both sides need to make rapid and scientific decisions to maximize the benefits of the countermeasure. Traditional anti-jamming communication technologies, represented by frequency hopping communication, mainly improve the system's anti-jamming capability by increasing the frequency hopping rate and bandwidth. They have strong anti-jamming capabilities and are difficult to intercept, thus establishing frequency hopping communication networks that are widely used in the communication field. Currently, most military radio stations use pure frequency hopping technology to improve the network's resistance to human interference in communication-constrained environments. However, traditional frequency hopping communication anti-jamming technologies have weak anti-tracking interference capabilities and poor adaptability, making them difficult to apply, especially in dynamic environments with limited communication resources and in wideband frequency hopping scenarios with large operational space. Furthermore, with the increasing intelligence of wireless jamming technologies, fixed-strategy frequency hopping anti-jamming technologies can no longer achieve dynamic anti-jamming optimization performance.

[0003] Currently, the main approach to intelligent interference signals is based on deep reinforcement learning for anti-jamming intelligent decision-making. For details, please refer to the paper: Xin Liu, et al., “Anti-jamming Communications Using Spectrum Waterfall: A Deep Reinforcement Learning Approach”, IEEE Communication Letters, vol.22, no.5, May.2018. This method first discretizes the mixed signal of the user and the interference, then performs overall sensing to obtain the spectrum, which serves as the environmental state. The neural network, after multiple learning and training iterations based on this environmental state, can output the optimal anti-jamming strategy. However, this method is only applicable to theoretical narrowband communication scenarios, characterized by low requirements for high-speed A / D conversion in spectrum sensing and a small action space, which is beneficial for the learning of the decision-making algorithm. In practical broadband communication scenarios, the action space is large, and the requirements for high-speed A / D conversion in spectrum sensing are high. Traditional anti-jamming decision-making algorithms struggle to meet these requirements, resulting in a significant decrease in anti-jamming performance, an inability to make rapid decisions, and limited applicability to practical anti-jamming scenarios.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a distributed compressed sensing system and method for broadband communication anti-interference, which can effectively reduce the high-speed A / D requirements of broadband sensing, and has no significant performance loss when used for intelligent anti-interference decision-making, making traditional anti-interference decision-making algorithms applicable to broadband communication anti-interference scenarios.

[0006] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0007] On one hand, this invention discloses a distributed compressed sensing system for anti-interference in broadband communication, comprising a sensor, an encoder, a binarization module, a decoder, and a fusion center. The number of the sensor, encoder, binarization module, and decoder is the same, and there are multiple sensor, encoder, binarization module, and decoder. One sensor is connected to a corresponding encoder, one encoder is connected to a corresponding binarization module, and one binarization module is connected to a corresponding decoder. Multiple decoders are connected to a fusion center.

[0008] The sensor is configured to sense the subband signal of the broadband signal to be sensed and obtain the short-time spectrum of the subband.

[0009] The encoder is configured to perform compression coding operations based on the short-time spectrum of the sub-band to obtain a coding vector;

[0010] The binarization module is configured to perform a binarization operation based on the encoding vector to obtain a binary spectrum information compressed frame.

[0011] The decoder is configured to perform spectrum decoding operations based on the binary spectrum information compressed frame to obtain the reconstructed short-time spectrum of the subband;

[0012] The fusion center is configured to obtain a broadband spectrum based on the reconstructed short-time spectrum of the sub-band, so as to realize distributed sensing and compressed reconstruction of the broadband signal to be sensed.

[0013] Furthermore, the bandwidth of the broadband signal to be sensed is [F L F U The bandwidth of the broadband signal to be sensed is B = F. U -F L ;

[0014] The number of perceptrons, encoders, binarization modules and decoders is the same, and the number is set to M, where M>1;

[0015] M sensors are independently distributed. The m-th sensor among the M independent sensors corresponds to a sub-band signal of the broadband signal to be sensed, i.e., a sub-band (F). L +(m-1)(F U -F L ) / M, F L +m(F U -F L The electromagnetic spectrum signal within the range of ) / M) is used to obtain the short-time spectrum of a corresponding sub-band, m=1,2,…,M.

[0016] Furthermore, the expression for the short-time spectrum of the sub-band is as follows:

[0017]

[0018] in, Let M represent the short-time spectrum of the subband obtained by the m-th perceptron within the time interval [t, t+Δt], where 1≤m≤M; Let l represent the l-th discrete spectrum sample value obtained by the m-th perceptron, where l = 1, 2, ..., L, l represents the index of the discrete spectrum sample value, and L represents the length of the short-time spectrum.

[0019] Furthermore, the expression for the discrete spectrum sample values ​​is as follows:

[0020]

[0021] Where log(·) represents the logarithmic function; l represents the index of the discrete spectrum sample value, l = 1, 2, ..., L, where L represents the length of the short-time spectrum; Δf represents the spectral resolution, Δf = b / L, where b represents the subband bandwidth; f m Indicates the center frequency of the sub-band; It is the power spectral density at the sensor receiver, determined by the independent variable. power spectral density The function is calculated; df represents the derivative with respect to frequency f.

[0022] Furthermore, the encoder is constructed from a deep neural network, including a first convolutional block, a second convolutional block, a third convolutional block, a first fully connected layer, a second fully connected layer, and a third fully connected layer connected in sequence.

[0023] The expression for the encoded vector is:

[0024]

[0025] Among them, F E Indicates compression encoding operation; L represents the short-time spectrum of the subband obtained by the m-th perceptron within the time interval [t, t+Δt]; L′ represents the length of the compressed encoding vector. This represents the encoded vector obtained by the m-th encoder;

[0026] The length L′ of the compressed encoded vector is 48.

[0027] Furthermore, the expression for the binary spectrum information compressed frame is:

[0028]

[0029] Among them, F B This represents the binarization operation; This represents the compressed frame of binary spectral information obtained by the m-th binarization module.

[0030] Furthermore, the decoder is constructed from a deep neural network, including a fourth fully connected layer, a fifth fully connected layer, a sixth fully connected layer, a fourth convolutional block, a fifth convolutional block, a sixth convolutional block, and a tenth convolutional layer connected in sequence.

[0031] The expression for the reconstructed short-time spectrum of the sub-band is:

[0032]

[0033] in, F represents the reconstructed short-time spectrum of the subband obtained by the m-th decoder; D This indicates a spectrum decoding operation.

[0034] Furthermore, the broadband spectrum s obtained by the fusion center t The expression is:

[0035]

[0036] in, Let m represent the reconstructed short-time spectrum of the subband obtained by the m-th decoder, where m = 1, 2, ..., M.

[0037] In a second aspect, the present invention discloses a distributed compressed sensing method for broadband communication interference resistance, applicable to the distributed compressed sensing system for broadband communication interference resistance described in the first aspect, the method comprising:

[0038] Based on the sensor of the distributed compressed sensing system, the subband signal of the broadband signal to be sensed is obtained, and the short-time spectrum of the subband is obtained.

[0039] The encoder of the distributed compressed sensing system performs compression coding operations based on the short-time spectrum of the sub-band to obtain the coding vector.

[0040] Based on the binarization module of the distributed compressed sensing system, a binarization operation is performed according to the encoding vector to obtain a binary spectrum information compressed frame.

[0041] The decoder based on the distributed compressed sensing system performs spectrum decoding operation on the compressed frame according to the binary spectrum information to obtain the reconstructed short-time spectrum of the sub-band.

[0042] Based on the fusion center of the distributed compressed sensing system, the broadband spectrum is obtained according to the reconstructed short-time spectrum of the sub-band, so as to realize the distributed sensing and compressed reconstruction of the broadband signal to be sensed.

[0043] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0044] The distributed compressed sensing system for broadband communication anti-interference proposed in this invention, through the cooperation of a sensor, encoder, binarization module, decoder and fusion center, can effectively reduce the high-speed sampling requirements of broadband sensing, and has no significant performance loss when used for intelligent anti-interference decision-making, making traditional anti-interference decision-making algorithms applicable to broadband communication anti-interference scenarios. Attached Figure Description

[0045] Figure 1 An overall structural diagram of a distributed compressed sensing system for broadband communication anti-interference provided in an embodiment;

[0046] Figure 2 The structural diagram of the encoder, binarization module, and decoder provided in the embodiment;

[0047] Figure 3 A schematic diagram of spectral compression distortion at different output bit counts provided for the embodiments;

[0048] Figure 4 A normalized throughput diagram illustrating the spectrum compression anti-interference performance provided for an embodiment;

[0049] Figure 5 This is a schematic diagram of five interference modes across the entire communication band in the broadband spectrum provided for the embodiment. Detailed Implementation

[0050] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0051] Example 1

[0052] This embodiment 1 provides a distributed compressed sensing system for anti-interference in broadband communication, including a sensor, an encoder, a binarization module, a decoder, and a fusion center. The number of sensors, encoders, binarization modules, and decoders is the same, and there are multiple sensors, encoders, binarization modules, and decoders. One sensor is connected to a corresponding encoder, one encoder is connected to a corresponding binarization module, and one binarization module is connected to a corresponding decoder. Multiple decoders are connected to a fusion center.

[0053] The sensor is configured to sense the subband signal of the broadband signal to be sensed and obtain the short-time spectrum of the subband.

[0054] The encoder is configured to perform compression coding based on the short-time spectrum of the subband to obtain a coded vector;

[0055] The binarization module is configured to perform binarization operations based on the encoding vector to obtain a binary spectrum information compressed frame.

[0056] The decoder is configured to compress the frame based on the binary spectral information, perform spectral decoding operations, and obtain the reconstructed short-time spectrum of the subband.

[0057] The fusion center is configured to obtain the broadband spectrum based on the reconstructed short-time spectrum of the sub-band, so as to realize distributed sensing and compressed reconstruction of the broadband signal to be sensed.

[0058] The technical concept of this invention is as follows: Addressing the limitation of traditional anti-interference decision-making algorithms in broadband communication anti-interference scenarios, this invention utilizes a distributed compressed sensing system for broadband communication anti-interference to perform distributed sensing of broadband signals, obtaining the final broadband spectrum. This effectively reduces the high-speed A / D requirements for broadband sensing and enables its application in broadband communication anti-interference decision-making, thus making traditional anti-interference decision-making algorithms suitable for broadband communication anti-interference scenarios.

[0059] like Figure 1 As shown, this embodiment first constructs a broadband communication anti-interference environment based on the jammer model and the channel model. According to the distributed architecture, multiple sensors are used to perform distributed sensing of the broadband signal, with each sensor only needing to sense one sub-band, thus obtaining the sub-band spectrum. Then, encoder-based spectrum compression technology is used to compress and encode the sub-band spectrum, resulting in a binary spectrum information compressed frame composed of binary bit streams. Finally, the spectrum is decoded and fused by the decoder in the information fusion center to obtain the final broadband spectrum, which can be used for broadband communication anti-interference decision-making.

[0060] The specific steps are as follows.

[0061] S1: The sensor senses the sub-band signal of the broadband signal to be sensed and obtains the short-time spectrum of the sub-band.

[0062] S1.1: In this embodiment, the bandwidth of the broadband signal to be sensed is [F L F U The bandwidth of the broadband signal to be sensed is B = F. U -F L F L and F U These represent the start frequency and the cutoff frequency, respectively.

[0063] The number of perceptrons, encoders, binarization modules and decoders is the same, and the number is set to M, where M>1;

[0064] M sensors are independently distributed. The m-th sensor corresponds to a sub-band signal with a bandwidth b = B / M corresponding to the broadband signal to be sensed, i.e., a sub-band (F). L +(m-1)(F U -F L ) / M, F L +m(F U -F L The electromagnetic spectrum signal within the range of ) / M) is used to obtain the short-time spectrum of a corresponding sub-band, m=1,2,…,M.

[0065] S1.2: In the time domain, according to S1.1, the m-th subband signal received by the perceptron m (m = 1, 2, ..., M) at time t can be expressed as:

[0066]

[0067] Where x(t) and x j (t) represent the baseband signals of the user and the jammer, respectively, ωt and Let represent the instantaneous angular frequencies of the user's and the jammer's baseband signals at time t, respectively. and Let n(t) represent the random initial phase of the baseband signals of the user and the jammer, respectively, and let g represent the additive random noise. u，m It is the channel power gain of the user signal from the transmitter to the m-th sensor, g j，m It is the channel power gain of the interference signal from the jammer to the m-th sensor.

[0068] S1.3, the bandwidth of the m-th sub-band signal that the sensor m can sense is [f m -b / 2,f m +b / 2], the sub-band center frequency generated by the local oscillator is f m =F L +(m-1 / 2)×b, the cutoff frequency of the low-pass filter is b / 2, and the ADC sampling rate is f. s =b. Then, according to y in step S1.2m (t) can be used to calculate the power spectral density (PSD) function at the receiver of the sensor m.

[0069]

[0070] Among them, U t (xf t () represents the power spectral density of the baseband signal; N represents the power spectral density of the interference signal; t (x) represents the power spectral density of the noise signal, which can be estimated using the P-Welch algorithm; f t This represents the center frequency of the baseband signal at time t; The center frequency of the interference signal at time t; g u，m This represents the channel power gain of the user signal from the transmitter to the m-th sensor; g j，m J represents the channel power gain of the interference signal from the jammer to the m-th sensor; J represents the total number of interference frequency points.

[0071] S1.4, Based on the power spectral density function in step S1.3 Make its independent variable Thus, the discrete spectrum sample values ​​are calculated. Discrete spectrum sample values The expression is as follows:

[0072]

[0073] Where log(·) represents the logarithmic function; / represents the index of the discrete spectrum sample value, l = 1, 2, ..., L, where L represents the length of the short-time spectrum; Δf represents the spectral resolution, Δf = b / L, where b represents the subband bandwidth; f m Indicates the center frequency of the sub-band; It is the power spectral density at the sensor receiver, determined by the independent variable in step S1.3. power spectral density The function is calculated; df represents the derivative with respect to frequency f.

[0074] The short-time spectrum of the m-th subband sensed within the time interval [t, t+Δt] can be expressed as:

[0075]

[0076] in, Let M represent the short-time spectrum of the subband obtained by the m-th perceptron within the time interval [t, t+Δt], where 1≤m≤M; Let l represent the l-th discrete spectrum sample value obtained by the m-th perceptron, where l = 1, 2, ..., L, l represents the index of the discrete spectrum sample value, and L represents the length of the short-time spectrum.

[0077] This completes the distributed sensing of broadband signals.

[0078] S2: Encoder, which performs compression coding operations based on the short-time spectrum of the subband to obtain the coded vector.

[0079] like Figure 2 As shown, the encoder is built from a deep neural network, including a first convolutional block, a second convolutional block, a third convolutional block, a first fully connected layer, a second fully connected layer, and a third fully connected layer connected in sequence.

[0080] The first convolutional block consists of a first convolutional layer and a first max pooling layer. The first convolutional layer has 32 channels, a 1×3 kernel, a stride of 1, and a padding of 1; the first max pooling layer has a 1×2 kernel and a stride of 1.

[0081] The second convolutional block consists of a second convolutional layer and a second max pooling layer. The second convolutional layer has 64 channels, a 1×3 kernel, a stride of 1, and a padding of 1; the second max pooling layer has a 1×2 kernel and a stride of 1.

[0082] The third convolutional block consists of a third convolutional layer and a third max pooling layer. The third convolutional layer has 128 channels, a 1×3 kernel, a stride of 1, and padding of 1; the third max pooling layer has a 1×2 kernel and a stride of 1.

[0083] After a series of convolutional blocks, three fully connected layers (Fc) are used to obtain the information embedding vector. In the encoder, the activation function of the third fully connected layer is tanh, with a dimension of L′ to generate L′ outputs in the range [-1, 1]. The activation functions of the first and second fully connected layers are ReLU, with dimensions of 1024 and 512, respectively.

[0084] The short-time spectrum of the subband obtained by the sensor As input to the encoder, it outputs an encoded vector of length L′, and the expression for the encoded vector is:

[0085]

[0086] Among them, F E Indicates compression encoding operation; L represents the short-time spectrum of the subband obtained by the m-th perceptron within the time interval [t, t+Δt]; L′ represents the length of the compressed encoding vector. This represents the encoded vector obtained by the m-th encoder;

[0087] The length L′ of the compressed encoding vector is set to 48.

[0088] S3: Binarization module, which performs binarization operation based on the encoding vector to obtain a binary spectrum information compressed frame.

[0089] The encoded vector output by the encoder is used as the input to the binarization module to obtain the binary spectral information compressed frame, as shown in the following expression:

[0090]

[0091] Where FB represents the binarization operation; This represents the compressed frame of binary spectral information obtained by the m-th binarization module.

[0092] Similar to the AlexNet neural network, this invention uses randomized binarization to map the encoder output to L′ bits. Binarization allows for control of the compression ratio by simply constraining the number of bits in the output, and the transmission of the bit vector over a wireless channel is both serializable and deserializable. The randomized binarization function F... B (x), x∈[-1,1], is defined as:

[0093] F B (x)=x+σ(x)∈{-1,1}

[0094] Where σ(x) is the quantization noise, and its value is 1-x or 1+x.

[0095] This completes the compression of the subband spectrum after distributed sensing.

[0096] S4: Decoder, compresses frames based on binary spectral information, performs spectral decoding operations, and obtains the reconstructed short-time spectrum of the subband.

[0097] like Figure 2 As shown, the decoder is built from a deep neural network, consisting of a fourth fully connected layer, a fifth fully connected layer, a sixth fully connected layer, a fourth convolutional block, a fifth convolutional block, a sixth convolutional block, and a tenth convolutional layer connected in sequence. The fourth, fifth, and sixth convolutional blocks can convert the high-order representations generated by the three fully connected layers into a 1×L vector.

[0098] The dimensions of the fourth, fifth, and sixth fully connected layers are 512, 1024, and 3200, respectively.

[0099] The fourth convolutional block consists of a fourth convolutional layer and a fifth convolutional layer. The fourth convolutional layer has 64 channels, a 1×3 kernel, a stride of 1, and padding of 1. The fifth convolutional layer has 64 channels, a 1×3 kernel, a stride of 2, and padding of 1.

[0100] The fifth convolutional block includes a sixth convolutional layer and a seventh convolutional layer. The sixth convolutional layer has 32 channels, a 1×3 kernel, a stride of 1, and padding of 1. The seventh convolutional layer has 32 channels, a 1×3 kernel, a stride of 2, and padding of 1.

[0101] The sixth convolutional block consists of an eighth convolutional layer and a ninth convolutional layer. The eighth convolutional layer has 16 channels, a 1×3 kernel, a stride of 1, and padding of 1. The ninth convolutional layer has 16 channels, a 1×3 kernel, a stride of 2, and padding of 1.

[0102] The tenth convolutional layer has 1 channel, 1×3 kernel, 1 stride, and 1 padding.

[0103] The activation functions of the fourth, fifth, sixth, seventh, eighth, and ninth convolutional layers are all ReLU, while the tenth convolutional layer does not use an activation function.

[0104] After receiving these binary spectral information compressed frames, M decoders perform spectral decoding to obtain the reconstructed short-time spectrum of the subband, expressed as:

[0105]

[0106] in, F represents the reconstructed short-time spectrum of the subband obtained by the m-th decoder; D Indicates a spectrum decoding operation; Let l represent the value of the l-th reconstructed discrete spectrum sample obtained by the m-th decoder, where l = 1, 2, ..., L.

[0107] S5: Fusion Center, which obtains the broadband spectrum based on the reconstructed short-time spectrum of the sub-band, in order to realize distributed sensing and compressed reconstruction of the broadband signal to be sensed.

[0108] The broadband spectrum s obtained by the fusion center t The expression is:

[0109]

[0110] in, Let m represent the reconstructed short-time spectrum of the subband obtained by the m-th decoder, where m = 1, 2, ..., M.

[0111] like Figure 1 As shown, the broadband spectrum obtained by the fusion center can be used as an environmental state input to the DRL network for training. The DRL network outputs appropriate frequency points to the transmitter as a decision for broadband communication anti-interference.

[0112] It should be noted that the encoder and decoder in this embodiment are pre-built and trained encoders and decoders with optimal network parameters.

[0113] The specific training method involves using the mean squared error loss (MSELoss) as the loss function L for both the encoder and decoder. MSE :

[0114]

[0115] Where N is the number of training samples. Then, gradient descent can be used to update the parameters of the encoder and decoder neural networks based on the loss function to find the optimal network parameters.

[0116] A well-trained encoder and decoder with optimal network parameters can minimize the distortion introduced by the encoder, thereby improving the accuracy of the entire system.

[0117] for Figure 1 The distributed sensing shown in the diagram involves each sensor sending a compressed narrowband signal spectrum to the information fusion center every Δt. The information fusion center then reconstructs the narrowband signal spectrum and fuses these spectra into a wideband spectrum. The information fusion center employs an end-to-end training procedure, with its input being the full-band short-time spectrum collected by all sensors. Typically, the communication overhead required for the compressed narrowband spectrum is much less than that required for the directly sensed narrowband spectrum; however, compression can lead to severe distortion in the reconstructed short-time spectrum and affect anti-interference performance. Assuming the bit length representing each compressed narrowband short-time spectrum is L′, the communication overhead is proportional to L′, and the distortion from spectrum compression can be expressed as:

[0118]

[0119] in It is reconstructed from L′ bits. The l-th element in.

[0120] Clearly, as L′ decreases, distortion increases, which degrades the anti-interference performance. To explore the relationship between compression ratio and distortion, this invention trained several encoders with different output lengths L′ on the same dataset and calculated the average distortion for each spectral compression model. Figure 3 The effect of L′ on spectral compression distortion is shown. It can be seen that distortion decreases as the number of encoder output bits increases. When L′≥48, the distortion no longer decreases significantly. Therefore, the optimal choice for the required number of output bits is L′=48, at which point the distortion is approximately 0.06.

[0121] Please see Figure 4To investigate the impact of using the reconstructed short-time spectrum on the anti-interference performance of the proposed scheme, L′ = 48 bits were used to reconstruct the short-time spectrum. The normalized throughput is defined as:

[0122]

[0123] Where S0 is the initial state, π * This is the optimal strategy.

[0124] Figure 4 A comparison of the interference-resistant normalized throughput based on Deep Reinforcement Learning (DRL) is presented, using signals trained with uncompressed sensing, signals quantized with 4 bits, and signals trained with a distributed compressed sensing system of L′=48. Simulation results show that as the number of training iterations increases, the normalized throughput in all three scenarios gradually converges to a common upper limit of 0.98. This means that neither the direct quantization method nor our proposed distributed compressed sensing system of L′=48 causes any significant performance degradation. Furthermore, the communication overhead of the distributed compressed sensing system proposed in this invention is approximately (4×L) / L′≈17 times lower than that of the direct quantization method.

[0125] for Figure 4 The results in this implementation case demonstrate the use of a robust DRL decision-making algorithm based on Deep Q-learning Network (DQN). DQN typically uses deep convolutional neural networks to fit action values.

[0126]

[0127] in θ is the instantaneous reward value, and θ is the network parameter. The network is implemented by two Conv layers with 16 and 32 filters, and two Fc layers with 2048 and 50 nodes. Except for the output layer, each layer uses the ReLU activation function.

[0128] To update the DQN network, the agent will collect and transfer data. And store it in the experience replay pool, then randomly select a mini-batch of transfers to calculate the loss:

[0129]

[0130] In the simulation, both the user and the jammer operate in the frequency band [100MHz, 200MHz], i.e., bandwidth B = 100MHz. The distributed sensing system has M = 5 sensors. Each sensor performs partial frequency band sensing (bandwidth b = b / M = 20MHz) at Δf = 100kHz (i.e., L = b / Δf = 200), and sends a compressed sub-band short-time spectrum of length L′ to the information fusion center every Δt = 1ms. The agent retains the reconstructed spectrum data for T = 200ms (i.e., NT = T / ΔT = 200). Therefore, the spectrum waterfall S... n The size is 1000×200. The bandwidth of the legitimate signal is set to 2MHz, and its center frequency jumps once every τ = 10ms (i.e., Nτ = τ / Δt = 10) in 2MHz steps. The number of actions is |A| = 50. The agent can select an action from A = {101MHz, 103MHz, ..., 199MHz} in each time slot. The frequency switching cost is set to λ = 0.05. Both the legitimate and interference signals are shaped using a root-raised cosine pulse shaping filter with a roll-off factor of α = 0.4. The transmit power of the legitimate signal is set to 0dBm.

[0131] To evaluate the robustness of the deep reinforcement learning-based frequency selection scheme in the proposed distributed sensing architecture, the hyperparameters during training were set as follows: discount factor γ = 0.95, learning rate β = 10. -4 The transfer sampling batch size is mini-batch = 256, the experience replay pool size is 1000, and the target network update frequency is 1000.

[0132] Figure 4 As can be seen, with the increase of the number of training iterations, the normalized throughput of the proposed scheme can gradually converge to the upper limit of 0.98. The anti-interference performance of the proposed method is very close to that of the anti-interference scheme using ideal spectrum and SINR-based reward.

[0133] Figure 5 Five interference modes of the broadband spectrum across the entire communication band are presented in this implementation case 1, as follows:

[0134] (1) Full-band interference: The bandwidth of the interference signal is the entire selected frequency band, and the interference power is 40dBm;

[0135] (2) Frequency sweeping interference: the frequency sweeping speed is 0.5 GHz / s and the interference power is 50 dBm;

[0136] (3) Single-tone interference: The interference bandwidth is 5MHz, the interference power is 40dBm, and the jammer changes the center frequency every 20ms;

[0137] (4) Switching comb interference: An interference signal is emitted every 2MHz. The center frequency of the interference signal changes every 100ms. The interference power is 40dBm.

[0138] (5) Follower Interference: The interference center frequency is the same as the user's nearest communication frequency. If the user's signal is not within the range of the follower interference, the jammer selects a random frequency for interference. The interference bandwidth is 5MHz and the interference power is 50dBm.

[0139] The communication frequency band is divided into five 20MHz sub-bands (i.e., 100MHz–120MHz, 120MHz–140MHz, 140MHz–160MHz, 160MHz–180MHz, and 180MHz–200MHz), with each interference mode affecting one sub-band. Therefore, this application employs a distributed compressed sensing system for broadband communication anti-interference, which effectively reduces the high-speed sampling requirements of broadband sensing for different interference modes, without significant performance loss when used for intelligent anti-interference decision-making.

[0140] Example 2

[0141] This embodiment provides a distributed compressed sensing method for broadband communication interference immunity, applicable to the distributed compressed sensing system for broadband communication interference immunity in Embodiment 1. The method includes:

[0142] Based on the sensor of the distributed compressed sensing system, the sub-band signal of the broadband signal to be sensed is obtained and the short-time spectrum of the sub-band is obtained.

[0143] The encoder based on the distributed compressed sensing system performs compression coding operation based on the short-time spectrum of the sub-band to obtain the coding vector;

[0144] The binarization module based on the distributed compressed sensing system performs binarization operation on the encoded vector to obtain a binary spectrum information compressed frame.

[0145] The decoder based on the distributed compressed sensing system compresses the frame according to the binary spectrum information, performs spectrum decoding operation, and obtains the reconstructed short-time spectrum of the sub-band.

[0146] Based on the fusion center of the distributed compressed sensing system, the broadband spectrum is obtained from the reconstructed short-time spectrum of the sub-band, so as to realize the distributed sensing and compressed reconstruction of the broadband signal to be sensed.

[0147] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0148] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0149] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0150] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0151] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A distributed compressed sensing system for anti-interference in broadband communication, characterized in that, It includes a perceptron, an encoder, a binarization module, a decoder, and a fusion center. The number of perceptrons, encoders, binarization modules, and decoders is the same, and there are multiple perceptrons, encoders, binarization modules, and decoders. One perceptron is connected to a corresponding encoder, one encoder is connected to a corresponding binarization module, and one binarization module is connected to a corresponding decoder. Multiple decoders are connected to one fusion center. The sensor is configured to sense the subband signal of the broadband signal to be sensed and obtain the short-time spectrum of the subband. The encoder is configured to perform compression coding operations based on the short-time spectrum of the sub-band to obtain a coding vector; The binarization module is configured to perform a binarization operation based on the encoding vector to obtain a binary spectrum information compressed frame. The decoder is configured to perform spectrum decoding operations based on the binary spectrum information compressed frame to obtain the reconstructed short-time spectrum of the subband; The fusion center is configured to obtain a broadband spectrum based on the reconstructed short-time spectrum of the sub-band, so as to realize distributed sensing and compressed reconstruction of the broadband signal to be sensed. The frequency band range of the broadband signal to be sensed is The bandwidth of the broadband signal to be sensed is ; The number of perceptrons, encoders, binarization modules and decoders is the same, and the number is set to M, where M>1; M sensors are independently distributed. The m-th sensor corresponds to a sub-band signal of the broadband signal to be sensed, i.e., a sub-frequency band. The electromagnetic spectrum signal within the range is used to obtain a corresponding sub-band's short-time spectrum. .

2. The distributed compressed sensing system for broadband communication anti-interference as described in claim 1, characterized in that, The expression for the short-time spectrum of the sub-band is as follows: ； in, This indicates that the m-th perceptron is in time interval The short-time spectrum of the sub-band obtained within the range, 1≤m≤M; This represents the value of the l-th discrete spectrum sample obtained by the m-th perceptron. , l represents the index of the discrete spectrum sample value, and L represents the length of the short-time spectrum.

3. The distributed compressed sensing system for broadband communication anti-interference as described in claim 2, characterized in that, The expression for the discrete spectrum sample values ​​is as follows: ； in, (·) denotes the logarithmic function; l denotes the index of the discrete spectrum sample value. L represents the length of the short-time spectrum; Indicates spectral resolution. , Indicates subband bandwidth; Indicates the center frequency of the sub-band; It is the power spectral density at the sensor receiver, determined by the independent variable. = power spectral density Obtained by function calculation; Indicates frequency The differential.

4. The distributed compressed sensing system for broadband communication anti-interference as described in claim 2, characterized in that, The encoder is constructed from a deep neural network, including a first convolutional block, a second convolutional block, a third convolutional block, a first fully connected layer, a second fully connected layer, and a third fully connected layer connected in sequence. The expression for the encoded vector is: ； in, Indicates compression encoding operation; This indicates that the m-th perceptron is in time interval The short-time spectrum of the sub-band obtained within; This indicates the length of the encoded vector after compression. This represents the encoded vector obtained by the m-th encoder; Wherein, the length of the compressed encoding vector is set. It is 48.

5. The distributed compressed sensing system for broadband communication anti-interference as described in claim 4, characterized in that, The expression for the binary spectrum information compressed frame is: ； in, This represents the binarization operation; This represents the compressed frame of binary spectral information obtained by the m-th binarization module.

6. The distributed compressed sensing system for broadband communication anti-interference as described in claim 5, characterized in that, The decoder is constructed from a deep neural network, including a fourth fully connected layer, a fifth fully connected layer, a sixth fully connected layer, a fourth convolutional block, a fifth convolutional block, a sixth convolutional block, and a tenth convolutional layer connected in sequence. The expression for the reconstructed short-time spectrum of the sub-band is: ； in, This represents the reconstructed short-time spectrum of the subband obtained by the m-th decoder; This indicates a spectrum decoding operation.

7. The distributed compressed sensing system for broadband communication anti-interference as described in claim 6, characterized in that, The broadband spectrum obtained by the fusion center The expression is: ； in, Let m represent the reconstructed short-time spectrum of the subband obtained by the m-th decoder, where m = 1, 2, ..., M.

8. A distributed compressed sensing method for broadband communication interference immunity, applicable to the distributed compressed sensing system for broadband communication interference immunity as described in any one of claims 1-7, characterized in that, The method includes: Based on the sensor of the distributed compressed sensing system, the sub-band signal of the broadband signal to be sensed is sensed, and the short-time spectrum of the sub-band is obtained. The encoder of the distributed compressed sensing system performs compression coding operations based on the short-time spectrum of the sub-band to obtain the coding vector. Based on the binarization module of the distributed compressed sensing system, a binarization operation is performed according to the encoding vector to obtain a binary spectrum information compressed frame. The decoder based on the distributed compressed sensing system performs spectrum decoding operation on the compressed frame according to the binary spectrum information to obtain the reconstructed short-time spectrum of the sub-band. Based on the fusion center of the distributed compressed sensing system, the broadband spectrum is obtained according to the reconstructed short-time spectrum of the sub-band, so as to realize the distributed sensing and compressed reconstruction of the broadband signal to be sensed.

Images