Broadband spectrum detection method

Abstract

This invention discloses a broadband spectrum detection method, which includes: in response to acquired broadband spectrum raw data, converting the raw data into a one-dimensional salient vector; extracting signal features from the salient vector; reconstructing the salient vector after linear interpolation upsampling of the signal features; calculating the cross-entropy loss between the reconstructed vector and the salient vector; and calculating the parameter information required for broadband spectrum detection based on the cross-entropy loss. This invention can perform broadband spectrum detection in complex electromagnetic environments.

Description

Technical Field

[0001] This invention belongs to the field of signal detection technology, and particularly relates to a broadband spectrum detection method. Background Technology

[0002] With the rapid development of various radio communication systems and the development and application of various electromagnetic communication devices, a large number of electromagnetic signals with complex signal systems and modulation patterns, dense overlap, and dynamic interleaving permeate the electromagnetic environment. Civilian and military signals with different power, bandwidth, modulation methods, coding methods, and time-varying signals are intertwined in the receiving frequency band, making the detected signals exhibit complex characteristics such as massive quantity, high dimension, diversity, and dynamics. These complex electromagnetic environments are manifested in the following aspects: (1) The number of electromagnetic signals in space is increasing, and the background noise is significantly increasing; (2) Various interferences are becoming increasingly serious, and the background noise often exhibits non-Gaussian characteristics; (3) The noise is highly dynamic due to various sudden signals, interferences, and background influences.

[0003] Among traditional spectrum detection algorithms, matched filtering, energy detection, cyclostationary detection, and eigenvalue detection are the most widely used and classic methods. In addition, wavelet analysis, which extracts target signals by performing hierarchical analysis of signals at different scales, has become a research hotspot in the field of signal detection in recent years.

[0004] The matched filter method is often limited in practical applications due to its requirement for prior knowledge. Energy detection methods require accurate estimation of the variance of ambient noise. However, under low signal-to-noise ratio conditions, the signal variance often cannot be accurately estimated, leading to a sharp decline in the performance of energy detection methods. Eigenvalue detection methods are limited by random matrix theory; the precise threshold of this method is difficult to obtain theoretically and often requires extensive practical experiments for correction, which greatly affects its practical application. The difficulty and bottleneck of wavelet analysis lies in the selection of wavelet basis functions. To date, there is no standard or universal method to obtain wavelet basis functions. Furthermore, the resolution of wavelet transform is limited by sampling frequency and length. All of the above methods are model-driven, requiring prior knowledge of signal or noise information. However, in real-world environments, signals are often unknown, and noise conditions are dynamically changing, hindering the performance of traditional spectrum sensing methods. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art by providing a broadband spectrum detection method that can achieve broadband spectrum detection without relying on experts, has strong environmental adaptability, is easy to train, has a fast convergence speed, and can effectively improve the detection probability.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A broadband spectrum detection method, the method comprising:

[0008] In response to the acquired raw broadband spectrum data, the raw data is converted into a one-dimensional salient vector;

[0009] Extract the signal features of the salient vector;

[0010] After performing linear interpolation upsampling on the signal features, the salient vector is reconstructed, and the cross-entropy loss between the reconstructed vector and the salient vector is calculated.

[0011] The parameters required for broadband spectrum detection are calculated based on the cross-entropy loss.

[0012] Furthermore, the step of converting the original data into a one-dimensional salient vector specifically includes:

[0013] The parts of the original data containing signals are marked as salient regions, and the parts without signals are marked as non-salient regions, resulting in a one-dimensional salient vector as the label.

[0014] Furthermore, the extraction of signal features from the salient vector specifically includes:

[0015] The original data is normalized by minimax and normalization, and then input into a deep learning model along with the corresponding labels to extract the low-level detail features and high-level semantic features of the salient vectors.

[0016] Global contextual features are extracted using high-level semantic features;

[0017] The high-level semantic features, the low-level detail features, and the global context features are fused to obtain fused features and a feature map of the fused features;

[0018] The signal features of the salient vector are obtained by refining the feature mapping using the internal characteristics of the feature map.

[0019] Furthermore, the step of extracting global contextual features through high-level semantic features also includes:

[0020] By utilizing spatial and channel attention mechanisms to enhance spatial regions and features that have a high response to salient objects, enhanced high-level semantic features are obtained.

[0021] Furthermore, the step of reconstructing the salient vector after linear interpolation upsampling of the signal features, and calculating the cross-entropy loss between the reconstructed vector and the salient vector specifically includes:

[0022] The signal features are linearly interpolated and upsampled to make the feature dimension the same as the input vector;

[0023] The deconvolutioned data is mapped to real numbers between 0 and 1 to reconstruct the input vector, and the cross-entropy loss between the reconstructed vector and the input salient vector is calculated.

[0024] Furthermore, the step of calculating the required parameter information for broadband spectrum detection based on the cross-entropy loss and completing the broadband spectrum detection specifically includes:

[0025] The significance of the original broadband spectrum data is obtained based on the cross-entropy loss.

[0026] The start and end points of the broadband spectrum are calculated based on the significance results, and the parameter information contained in the broadband spectrum is calculated based on the start frequency and frequency resolution of the original broadband spectrum data.

[0027] Furthermore, the step of normalizing the original data by maximum and minimum values ​​specifically includes:

[0028] The cost function is used for calculation.

[0029]

[0030] Among them, X norm Here are the normalized sample data, and X is the original data. min X is the minimum value of the original data. max This represents the maximum value of the original data.

[0031] Furthermore, the enhancement of spatial regions and features with high response to salient objects using spatial and channel attention mechanisms to obtain enhanced high-level semantic features specifically includes:

[0032] The high-level semantic features are enhanced using the following formula:

[0033] y=sigmod(fc2(RELU(fc1(F))));

[0034] F out =F·y;

[0035] Where fc1 and fc2 represent fully connected layer 1 and fully connected layer 2, F represents the input high-level semantic features, f is the channel weight feature vector obtained by downsampling the high-level semantic features, ReLU and sigmoid represent activation functions, · represents dot product, y represents the final weight vector, and F out This represents the selected feature.

[0036] Furthermore, the extraction of global context features through high-level semantic features specifically includes:

[0037]

[0038] yt = sigmod(fc4(RELU(fc3(f gap ))))

[0039]

[0040] Where F represents the high-level semantic features of the input, conv represents the convolution operation, and f gap This represents the feature obtained by global average pooling of high-level semantic features, which includes global contextual information. The dot (·) represents the dot product. This represents the final output of global context feature extraction.

[0041] Furthermore, the process of fusing the high-level semantic features, the low-level detail features, and the global context features to obtain the fused features specifically includes:

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050] in, Represents characteristics of high-level personnel. Represents low-level characteristics. Representing global features, `upsample` represents bilinear interpolation upsampling, `t` represents the stage number, `W` represents the generated mask, and `concat` represents concatenation by channel. The feature fusion module first combines low-level features... The input is fed into the convolutional layer conv1 to obtain the same number of channels as the high-level features. Then, high-level characteristics The mask is obtained by bilinear interpolation upsampling after inputting it into the convolutional layer conv2. mask and After performing a dot product operation and then applying the ReLU activation function, the first type of fused feature from the high and low layers is obtained. The same method can be used to obtain another fused feature of high and low layer features. and global low-level fusion features Finally, the three preliminary fusion features are concatenated by channel and then passed through a convolutional layer to obtain the final fusion feature.

[0051] Furthermore, the specific steps of refining the feature mapping to obtain the salient vector using the internal characteristics of the feature map include:

[0052]

[0053]

[0054] Among them, f in The input features are represented by f, and conv represents the convolution operation. First, f is... in The input is fed into a 3x3 convolutional layer to obtain... Then The inputs are fed into two convolutional layers to obtain the mask W and the bias b, respectively. Based on W, b, ... The final output f is obtained out .

[0055] Furthermore, the calculation of the cross-entropy loss between the reconstructed vector and the salient vector includes the calculation of the final cross-entropy loss and the calculation of the upsampled cross-entropy loss, specifically calculated as follows:

[0056]

[0057]

[0058] Among them, G ij S represents the true labels of points i and j. ij Represents the predicted label, l dom The loss representing the final predicted result. The auxiliary loss representing the intermediate result, γ i This represents the weighting coefficient.

[0059] Furthermore, the parameter information, specifically the parameter information included in the calculation of the broadband spectrum based on the starting frequency and frequency resolution of the original broadband spectrum data, includes:

[0060] f start =f startin +Start index *f step

[0061] f end =f startin +End index *f step

[0062] f c =fstart +(f end -f start ) / 2

[0063] bw = (end index -Start index )*f step

[0064] Among them, f startin f represents the starting frequency of the original broadband spectrum data. step Indicates frequency resolution, Start index Indicates the start point of the signal, End index Indicates the termination point of the signal, f start f represents the starting frequency of the signal. end fc represents the signal's termination frequency, fc represents the signal's center frequency, and bw represents the signal's bandwidth.

[0065] The beneficial effects of this invention are as follows:

[0066] (1) This invention adopts data-driven deep learning technology, which can mine the potential patterns and information inside the data through complex network structure and weight connection. By continuously adapting the distribution of the data through network weights, it can automatically complete the learning of data features. No threshold and parameter settings are required during the analysis process, which greatly reduces the dependence on expert knowledge and reduces the analysis manpower and time costs.

[0067] (2) This invention is based on training with real-world data. During the training process, it deeply mines the low-level features, high-level features, and global context features of the real-world data, and fully integrates these three types of features through a feature fusion module. This allows the fused features to possess the advantages of these three types of features, enabling more effective extraction of salient information from the input data, suppression of noise effects, and recovery of more structural and detailed information about the target. It can ensure the integrity of salient target detection in complex environments, reduce noise interference, and is more environmentally adaptable than traditional methods that set thresholds based on background noise. Furthermore, it can guarantee the signal detection effect even when background noise changes.

[0068] (3) This invention establishes a saliency-based deep learning model for broadband signal detection tasks; the deep learning model is trained under supervision using the actual sampled signal training set, thus transforming the broadband detection problem into pixel-level detection, where one pixel corresponds to one label, salient regions are considered positive samples, and non-salient regions are considered negative samples. Since one pixel corresponds to one label, the saliency-based broadband detection method does not require a large amount of data for training, has a smaller computational load compared to other deep learning models, is easier to train, and has a faster convergence speed. Simultaneously, the deep neural network model can be deployed on high-performance GPUs or dedicated AI processing chips for rapid parallel inference, enabling fast computation of the deep neural network model.

[0069] (4) This invention targets real-world broadband spectrum data, employing high-precision floating-point operators to construct a deep neural network model. The model is trained using high-precision labeled data and trained in a supervised manner. Multiple features are extracted and fused using a backbone network. Linear interpolation is performed at different feature fusion stages, followed by cross-entropy calculation with the input saliency map for multiple reconstructions. This ensures that the features obtained during multi-layer feature extraction are beneficial for saliency map generation. The trained deep neural network model is used to process the broadband spectrum data, automatically generating a saliency map of the signal. Based on the saliency map, parameters such as the signal's center frequency and bandwidth are calculated. This achieves highly stable and accurate detection of real-world broadband spectrum signals. It not only boasts high detection stability and accuracy but also high parameter estimation accuracy. Furthermore, simulations using computational methods can be used to verify aspects such as frequency offset acquisition range, signal-to-noise ratio working threshold, and level reception range. Compared to traditional methods, it exhibits smaller detection errors and a higher detection probability. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of the broadband spectrum detection method provided in an embodiment of the present invention;

[0071] Figure 2 This is a schematic diagram of the signal detection model structure in an embodiment of the present invention. Detailed Implementation

[0072] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0073] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] The matched filter method is often limited in practical applications due to its requirement for prior knowledge. Energy detection methods require accurate estimation of the variance of ambient noise. However, under low signal-to-noise ratio conditions, the signal variance often cannot be accurately estimated, leading to a sharp decline in the performance of energy detection methods. Eigenvalue detection methods are limited by random matrix theory; the precise threshold of this method is difficult to obtain theoretically and often requires extensive practical experiments for correction, which greatly affects its practical application. The difficulty and bottleneck of wavelet analysis lies in the selection of wavelet basis functions. To date, there is no standard or universal method to obtain wavelet basis functions. Furthermore, the resolution of wavelet transform is limited by sampling frequency and length. All of the above methods are model-driven, requiring prior knowledge of signal or noise information. However, in real-world environments, signals are often unknown, and noise conditions are dynamically changing, hindering the performance of traditional spectrum sensing methods.

[0075] To address the aforementioned technical problems, the following embodiments of the broadband spectrum detection method of the present invention are proposed.

[0076] Reference Figure 1 ,like Figure 1 The diagram shown is a schematic flowchart of the broadband spectrum detection method provided in this embodiment. The method specifically includes the following steps:

[0077] Step 1: In response to the acquired broadband spectrum raw data, the raw data is converted into a one-dimensional salient vector.

[0078] This step marks the parts of the original data containing signals as salient regions (represented by 1) and the parts without signals as non-salient regions (represented by 0), resulting in a binary one-dimensional 0-1 salient vector as the label. Simultaneously, the original data undergoes max-min normalization and is input into the deep learning model along with the corresponding labels.

[0079] As one implementation method, this embodiment normalizes the input data by maximum and minimum values, and the cost function is calculated as follows:

[0080]

[0081] Among them, X norm Here, X represents the normalized sample data, and X represents the original sample data. min X is the minimum value of the sample data. max This represents the maximum value of the sample data.

[0082] Step 2: Extract the signal features of the salient vector.

[0083] Reference Figure 2 ,like Figure 2 The diagram shown is a schematic diagram of the signal detection model structure in this embodiment.

[0084] This embodiment uses a deep residual network as the backbone network to extract various features from the input data, including low-level detail features and high-level semantic features.

[0085] The extracted high-level feature maps are input into the channel selection module, which utilizes spatial and channel attention mechanisms to enhance spatial regions and features that have a high response to salient objects, resulting in more representative high-level semantic features. Simultaneously, these high-level semantic features are input into the global feature extraction module to capture the relationships between multiple salient objects or different parts of salient objects from a global perspective, thus obtaining global contextual features. Finally, the enhanced high-level semantic features, low-level detail features, and global contextual features are input into the feature fusion module for fusion, integrating the functions of the three features to achieve the effect of suppressing noise while recovering more structural and detailed information.

[0086] As one implementation method, this embodiment uses a 50-layer deep residual network to extract four low-level features and one high-level feature.

[0087] As one implementation method, the feature selection module in this embodiment selects high-level features from the output using the following formula:

[0088] y=sigmod(fc2(RELU(fc1(F)))) (2)

[0089] F out =F·y (3)

[0090] Where fc1 and fc2 represent fully connected layer 1 and fully connected layer 2, F represents the input high-level features, f is the channel weight feature vector obtained by downsampling the high-level features, ReLU and sigmoid represent activation functions, · represents dot product, y represents the final weight vector, and F out This represents the selected feature.

[0091] The global feature extraction module extracts contextual information from high-level data, using the following formula:

[0092]

[0093] y t = sigmod(fc4(RELU(fc3(f gap (5)

[0094]

[0095] Where F represents the high-level features of the input, conv represents the convolution operation, and f gap This represents the feature obtained by global average pooling of high-level features, which includes global contextual information. The dot (·) represents the dot product. This represents the final output of the global feature extraction module.

[0096] In this embodiment, the feature fusion module has three inputs: low-level features, high-level features, and global features. High-level features come from the output of the previous level, low-level features come from the output of the backbone network, and global features come from the global information extraction module. The feature fusion module fuses the two types of features through upsampling and dot multiplication, which enhances the target's noise suppression. The features after the first fusion are then fused again using convolution to obtain the final result. The specific fusion strategy is as follows:

[0097]

[0098]

[0099]

[0100]

[0101]

[0102]

[0103]

[0104]

[0105] in, Representing high-level characteristics, f l t Represents low-level characteristics. Representing global features, upsample represents bilinear interpolation upsampling, t represents the stage number, W represents the generated mask, and concat represents concatenation by channel. The feature fusion module first combines the low-level features f l t The input is fed into the convolutional layer conv1 to obtain the same number of channels as the high-level features. Then, high-level characteristics The mask is obtained by bilinear interpolation upsampling after inputting it into the convolutional layer conv2. mask and After performing a dot product operation and then applying the ReLU activation function, the first type of fused feature from the high and low layers is obtained. The same method can be used to obtain another fused feature of high and low layer features. and global low-level fusion features Finally, the three preliminary fusion features are concatenated by channel and then passed through a convolutional layer to obtain the final fusion feature.

[0106] In this embodiment, the feature enhancement module uses multiplication and addition modules to further refine and enhance the feature map that has passed through the feature selection module and the feature fusion module. The specific operations are as follows:

[0107]

[0108]

[0109] Among them, f in The input features are represented by f, and conv represents the convolution operation. First, f is... in The input is fed into a 3x3 convolutional layer to obtain... Then The inputs are fed into two convolutional layers to obtain the mask W and the bias b, respectively. The obtained W and b are then compared... The final output f is obtained by performing the operation according to formula (16). out .

[0110] Step 3: After performing linear interpolation upsampling on the signal features, reconstruct the salient vector and calculate the cross-entropy loss between the reconstructed vector and the salient vector.

[0111] This step uses the SoftMax function to reconstruct the input vector by mapping the deconvolutioned data to real numbers between 0 and 1, and then calculates the cross-entropy loss between the reconstructed vector and the input salient vector.

[0112] The cross-entropy loss function is used to calculate the reconstruction error. In addition to the final result, the loss function is also calculated for the intermediate three-layer upsampling results.

[0113]

[0114]

[0115] Among them, G ij S represents the true labels of points i and j. ij Represents the predicted label, l dom The loss representing the final predicted result. The auxiliary loss representing the intermediate result, γ i This represents the weighting coefficient.

[0116] Step 4: Calculate the parameters required for broadband spectrum detection based on the cross-entropy loss.

[0117] After inference through the deep saliency detection network model, the saliency results of the input data are obtained. Once the final saliency results are obtained, the portion exceeding a threshold and continuously ending at 1 is considered a signal, and its start and end points are considered the signal's start and end points. Based on the saliency results, the signal's start and end points are calculated. Furthermore, based on the start frequency and frequency resolution of the original broadband spectrum data, parameters such as the number of signals contained in the broadband spectrum, the signal center frequency, and the signal bandwidth are calculated, thus completing signal detection and parameter extraction in complex electromagnetic environments.

[0118] After obtaining the start and end points of the signal, the signal center frequency and signal bandwidth parameters are calculated based on the start frequency and frequency resolution of the original broadband spectrum data. The specific calculation method is as follows:

[0119] f start =f startin +Start index *f step (19)

[0120] f end =f startin +end index *f step (20)

[0121] f c =f start +(f end -f start ) / 2 (21)

[0122] bw = (end index -Start index )*f step (twenty two)

[0123] Among them, f startin f represents the starting frequency of the original broadband spectrum data. step Indicates frequency resolution, Start index Indicates the start point of the signal, End index Indicates the termination point of the signal, f start f represents the starting frequency of the signal. end fc represents the signal's termination frequency, fc represents the signal's center frequency, and bw represents the signal's bandwidth.

[0124] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A wideband spectrum detection method, characterized in that, The method comprises: in response to the acquired wideband spectrum raw data, converting the raw data into a one-dimensional saliency vector; extracting signal features of the saliency vector; reconstructing the saliency vector after linear interpolation upsampling of the signal features, and calculating cross-entropy loss of the reconstructed vector and the saliency vector; calculating parameter information required for wideband spectrum detection according to the cross-entropy loss; the conversion of the raw data into a one-dimensional saliency vector specifically comprises: marking the signal part in the raw data as a saliency region, and marking the non-signal part as a non-saliency region to obtain a one-dimensional saliency vector as a label; the extraction of the signal features of the saliency vector specifically comprises: performing maximum and minimum value normalization on the raw data, and inputting the normalized data and the corresponding label into a deep learning model to extract low-level detail features and high-level semantic features of the saliency vector; enhancing the spatial region and features of the saliency object with high response by using spatial and channel attention mechanisms to obtain enhanced high-level semantic features; extracting global context features through the enhanced high-level semantic features; fusing the enhanced high-level semantic features, the low-level detail features and the global context features to obtain fused features and a feature map of the fused features; refining the characteristic mapping by using the internal characteristics of the feature map to obtain the signal features of the saliency vector.

2. The wideband spectrum detection method of claim 1, wherein, the reconstruction of the saliency vector after linear interpolation upsampling of the signal features, and the calculation of the cross-entropy loss of the reconstructed vector and the saliency vector specifically comprise: linear interpolation upsampling of the signal features to make the feature dimension the same as the input vector; mapping the data after upsampling to a real number between 0 and 1 to reconstruct the input vector, and calculating the cross-entropy loss of the reconstructed vector and the input saliency vector.

3. The wideband spectrum detection method of claim 1, wherein, the calculation of the parameter information required for wideband spectrum detection according to the cross-entropy loss, and the completion of the wideband spectrum detection specifically comprise: obtaining saliency results of the wideband spectrum raw data according to the cross-entropy loss; calculating the start point and the end point of the wideband spectrum according to the saliency results, and calculating the parameter information contained in the wideband spectrum according to the start frequency and the frequency resolution of the original wideband spectrum data.

4. The wideband spectrum detection method of claim 1, wherein, the maximum and minimum value normalization of the raw data specifically comprises: using the calculation method of a cost function: ； wherein, is the normalized sample data, is the original data, is the minimum value of the original data, is the maximum value of the original data.

5. The wideband spectrum detection method of claim 1, wherein, the enhancement of the spatial region and the features of the saliency object with high response by using spatial and channel attention mechanisms to obtain enhanced high-level semantic features specifically comprises: enhancing the high-level semantic features by using the following formula: ； ； wherein, and denote the fully connected layer 1 and the fully connected layer 2, denote the input high-level semantic feature, f is a channel weight feature vector obtained by down-sampling the high-level semantic feature, RELU and sigmod represent activation functions, denote dot product, y represents the final weight vector, denote the selected features.

6. The wideband spectrum detection method of claim 1, wherein, the extraction of global context features through the high-level semantic features specifically comprises: wherein, represents an input high-level semantic feature, represents a convolution operation, represents a feature including global context information obtained by global average pooling of the high-level semantic feature, represents a dot product, represents a final output of global context feature extraction.

7. The wideband spectrum detection method of claim 1, wherein, the fusion of the high-level semantic features, the low-level detail features and the global context features to obtain fused features specifically comprises: wherein, represents high-level features, represents low-level features, represents global features, represents bilinear interpolation up-sampling, t represents the stage number, and W represents the generated mask, represents channel concatenation; the feature fusion module first inputs the low-level features into the convolutional layer conv1 to obtain the same number of channels as the high-level features , then inputs the high-level features into the convolutional layer conv2 to perform bilinear interpolation up-sampling to obtain the mask , performs point multiplication between the mask and , and then passes through the RELU activation function to obtain the first fusion feature of the high-level and low-level features ; the same method can be used to obtain another fusion feature of the high-level and low-level features and a global low-level fusion feature , and finally, the three preliminary fusion features are concatenated by channels and then pass through a convolutional layer to obtain the final fusion feature .

8. The wideband spectrum detection method of claim 1, wherein, the parameter information, and the calculation of the parameter information contained in the wideband spectrum according to the start frequency and the frequency resolution of the original wideband spectrum data specifically comprises: wherein, represents a start frequency of the original wideband spectrum data, represents a frequency resolution, represents a start point of the signal, represents an end point of the signal, represents a start frequency of the signal, represents an end frequency of the signal, represents a center frequency of the signal, represents a bandwidth of the signal.

Images