Spectrum occupancy detection method and related device, electronic equipment and storage medium

Abstract

The invention discloses a spectrum occupancy detection method, a related device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a one-dimensional vector with the same length as the total number of frequency points based on the energy value of each frequency point in the power spectrum density of a to-be-detected wireless signal; remodeling based on the one-dimensional vector to obtain input features of image segmentation network support processing; processing the input features based on an image segmentation network to obtain a prediction mask; wherein the prediction mask is the same as the input feature in size; linear mapping is carried out based on a one-dimensional mask obtained through prediction mask remodeling, and a target mask is obtained; wherein the one-dimensional mask, the target mask and the total number of the frequency points are equal in length, and the target mask comprises a probability value that each frequency point is occupied; and determining the occupied frequency band interval based on the target mask. According to the scheme, the detection precision of spectrum occupancy can be improved, especially under the conditions of large energy dynamic range, low signal-to-noise ratio and non-stationary noise.

Description

Technical Field

[0001] This application relates to the field of spectrum analysis technology, and in particular to a spectrum occupancy detection method and related devices, electronic equipment and storage media. Background Technology

[0002] With the development of wireless communication technology, wireless spectrum resources are becoming increasingly scarce and exhibit characteristics such as dynamic, asymmetric, and sparse occupation. Therefore, it is necessary to detect the occupation of the spectrum.

[0003] Currently, existing spectrum occupancy detection technologies still have shortcomings. For example, in situations with a large energy dynamic range, low signal-to-noise ratio, and non-stationary noise, the detection capability decreases due to the use of a single threshold for occupancy detection. Therefore, improving the detection accuracy of spectrum occupancy, especially in situations with a large energy dynamic range, low signal-to-noise ratio, and non-stationary noise, has become an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem addressed by this application is to provide a spectrum occupancy detection method and related devices, electronic equipment, and storage media, which can improve the detection accuracy of spectrum occupancy, especially in the case of large energy dynamic range, low signal-to-noise ratio, and non-stationary noise.

[0005] To address the aforementioned technical problems, the first aspect of this application provides a spectrum occupancy detection method, comprising: obtaining a one-dimensional vector of equal length to the total number of frequency points based on the energy values ​​of each frequency point in the power spectral density of the wireless signal under test; reshaping the one-dimensional vector to obtain input features that can be processed by an image segmentation network; processing the input features based on the image segmentation network to obtain a prediction mask; wherein the prediction mask has the same size as the input features; performing a linear mapping based on the one-dimensional mask reshaped from the prediction mask to obtain a target mask; wherein the one-dimensional mask and the target mask have the same length as the total number of frequency points, and the target mask contains the probability value of each frequency point being occupied; and determining the occupied frequency band interval based on the target mask.

[0006] To address the aforementioned technical problems, a second aspect of this application provides a spectrum occupancy detection device, comprising: a vector extraction module, a vector reshaping module, a mask prediction module, a linear mapping module, and an interval determination module. The vector extraction module is used to obtain a one-dimensional vector of equal length to the total number of frequency points based on the energy values ​​of each frequency point in the power spectral density of the wireless signal under test. The vector reshaping module is used to reshape the one-dimensional vector to obtain input features that can be processed by an image segmentation network. The mask prediction module is used to process the input features based on the image segmentation network to obtain a prediction mask; wherein the prediction mask has the same size as the input features. The linear mapping module is used to perform linear mapping based on the one-dimensional mask obtained by reshaping the prediction mask to obtain a target mask; wherein the one-dimensional mask and the target mask have the same length as the total number of frequency points, and the target mask contains the probability value of each frequency point being occupied. The interval determination module is used to determine the occupied frequency band interval based on the target mask.

[0007] To address the aforementioned technical problems, a third aspect of this application provides an electronic device comprising at least a memory and a processor coupled to each other, wherein the memory stores at least program instructions, and the processor executes the program instructions to implement the spectrum occupancy detection method of the first aspect described above.

[0008] To address the aforementioned technical problems, a fourth aspect of this application provides a computer-readable storage medium storing program instructions executable by a processor, the program instructions being used to implement the spectrum occupancy detection method of the first aspect described above.

[0009] The above scheme obtains a one-dimensional vector of equal length to the total number of frequency points based on the energy values ​​of each frequency point in the power spectral density of the wireless signal under test. This vector is then reshaped to obtain the input features that the image segmentation network can process. The input features are then processed by the image segmentation network to obtain a prediction mask, which has the same size as the input features. A linear mapping is then performed on the one-dimensional mask obtained from the reshaped prediction mask to obtain the target mask. Both the one-dimensional mask and the target mask have the same length as the total number of frequency points. The target mask contains the probability values ​​of each frequency point being occupied. Based on the target mask, the occupied frequency band intervals are determined. On the one hand, because the spectrum occupancy detection problem is cleverly transformed into an image segmentation problem with input features constructed from the energy values ​​at frequency points, compared to single threshold detection of spectrum occupancy, even under conditions of large energy dynamic range, low signal-to-noise ratio, and non-stationary noise, masked modeling of spectrum occupancy detection can be achieved. Furthermore, the image segmentation model can also rely on its own modeling of depth features during processing. On the other hand, because masked modeling can predict the occupancy probability at the frequency point level, the precision of spectrum occupancy detection can be improved as much as possible. Therefore, it can improve the detection accuracy of spectrum occupancy, especially in situations with a large energy dynamic range, low signal-to-noise ratio, and non-stationary noise. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating an embodiment of the spectrum occupancy detection method of this application; Figure 2 This is a schematic diagram of the framework of an embodiment of the spectrum mask prediction model of this application; Figure 3 This is a schematic diagram of the framework of an embodiment of the spectrum occupancy detection device of this application; Figure 4 This is a schematic diagram of the framework of an embodiment of the electronic device of this application; Figure 5 This is a schematic diagram of a framework of an embodiment of the computer-readable storage medium of this application. Detailed Implementation

[0011] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0012] In the following description, specific details such as particular system architectures, interfaces, and technologies are presented for illustrative purposes rather than for limiting purposes, in order to provide a thorough understanding of this application.

[0013] In this paper, the terms "system" and "network" are often used interchangeably. The term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the slash " / " generally indicates that the preceding and following related objects have an "or" relationship. Furthermore, "many" in this paper indicates two or more objects.

[0014] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the spectrum occupancy detection method of this application. Specifically, it may include the following steps: Step S11: Based on the energy values ​​of each frequency point in the power spectral density of the wireless signal under test, obtain a one-dimensional vector with the same length as the total number of frequency points.

[0015] In one implementation scenario, the wireless signal to be tested can be a wireless signal collected in the target area. It should be noted that the target area can be the location area where spectrum occupancy detection needs to be performed. For example, when spectrum occupancy detection is needed for an airport, the target area can be the airport itself; or, when spectrum occupancy detection is needed for a residential area, the target area can be the residential area. Of course, the above examples are only a few possible examples of the target area in practical applications, and other possible scenarios for the target area will not be listed here.

[0016] In one implementation scenario, in order to obtain the power spectral density of the wireless signal under test, IQ signal sampling can be performed in the target area at a certain sampling rate for a preset duration to obtain a complex discrete signal. Based on this, the power spectral density of the wireless signal under test can be obtained by power spectral density estimation methods such as Welch power spectral density estimation.

[0017] In a specific implementation scenario, the sampling rate can be set according to the actual application needs, such as including but not limited to 100MHz, 150MHz, 200MHz, etc. The specific value of the sampling rate is not limited here.

[0018] In a specific implementation scenario, the preset duration of sampling maintenance can also be set according to the actual application needs, such as including but not limited to 0.1 seconds, 0.2 seconds, 0.5 seconds, etc. There is no limit to the preset duration of sampling maintenance here.

[0019] In a specific implementation scenario, let's take the power spectral density estimation method using Welch power spectral density estimation to obtain the power spectral density of the wireless signal under test as an example. For ease of description, the aforementioned complex discrete signal can be denoted as x[n] = I[n] + j*Q[n]. Based on this, the complex discrete signal can be divided into K potentially overlapping sub-segments (each segment can have a length of L, and the overlap rate can be set to 50%, etc.), and the k-th segment can be denoted as x. k [n]. For example, the length L of each segment can be set to 10000, etc., and other possible settings are not limited here. Next, each signal x segment can be... k Multiplying [n] by the windowing function w[n] yields the windowed sequence. After obtaining the windowed sequence of each signal segment, the periodogram estimate P can be calculated for each segment. k (f):

[0020] In the above formula, f represents the frequency point, and U represents the window function energy. It should be noted that the window function energy can be obtained by taking the average of the window function, as shown below:

[0021] Finally, the power spectral density can be obtained by averaging all periodograms to reduce the estimation variance:

[0022] In the above formula, This represents the power spectral density of the wireless signal under test. It should be noted that the above example is merely one possible implementation example when using Welch power spectral density estimation to measure the power spectral density of the wireless signal under test. Other possible estimation methods are not limited here, nor will they be listed one by one.

[0023] In one implementation scenario, as a possible example, after obtaining the power spectral density of the wireless signal under test, the energy value of each frequency point in the power spectral density can be directly used as the element value of the corresponding frequency point in a one-dimensional vector of the same length as the total number of frequency points. For example, the energy value of the i-th frequency point in the power spectral density is P. i Then, the value of the i-th element in a one-dimensional vector of the same length as the total number of frequency points can also be P. i .

[0024] In another implementation scenario, differing from the aforementioned implementation, as another possible example, after obtaining the power spectral density of the wireless signal under test, the energy value at each frequency point in the power spectral density can be added to a preset constant and then its logarithm taken. This logarithm is then used as the element value in the reference vector corresponding to each frequency point. It should be noted that the preset constant can also be a small constant to minimize numerical overflow during logarithmic calculations. For example, the preset constant can be set to 10. -12 The specific value of the preset constant is not limited here. For ease of description, the above process of adding the preset constant and then taking the logarithm can be represented as:

[0025] In the above formula, log represents logarithmic operation, ε represents a preset constant, and PSD dB (f) represents the power spectral density energy value obtained after adding a preset constant and taking the logarithm. Based on this, normalization can be performed using a reference vector to obtain a one-dimensional vector. For example, normalization operations can include, but are not limited to, sigmoid, min-max, etc. The specific algorithms for normalization are not limited here, nor will they be listed individually. In the above method, the energy value at each frequency point in the power spectral density is added to a preset constant and then the logarithm is taken as the element value in the reference vector corresponding to each frequency point. Normalization is then performed based on the reference vector to obtain a one-dimensional vector, which facilitates subsequent processing by the network as much as possible.

[0026] Step S12: Reshape the image based on a one-dimensional vector to obtain the input features that the image segmentation network can support processing.

[0027] In one implementation scenario, an image segmentation network can incorporate one-dimensional convolutions, thus enabling direct processing of one-dimensional feature vectors. In this case, during the reshaping process based on a one-dimensional vector, the one-dimensional vector itself can be directly used as the reshaped one-dimensional vector, i.e., as the input feature that the image segmentation network can process. For example, if the length of the one-dimensional vector is 10000, then during the reshaping process, the one-dimensional vector (1*10000) itself can be directly used as the reshaped one-dimensional vector.

[0028] In another implementation scenario, image segmentation networks can also incorporate two-dimensional convolutions, thus enabling the processing of two-dimensional feature vectors. In this case, during the reshaping process based on a one-dimensional vector, the one-dimensional vector can be reshaped into a two-dimensional vector. Taking a one-dimensional vector with a length of 10000 as an example, the one-dimensional vector can be reshaped into a 100*100 two-dimensional vector. That is, the one-dimensional vector can be restored by concatenating each row of the two-dimensional vector sequentially.

[0029] It should be noted that the examples above are merely possible examples of reshaping one-dimensional vectors when image segmentation networks support processing one-dimensional and two-dimensional feature vectors respectively. Other possible scenarios are not limited here, nor will they be listed individually. Furthermore, image segmentation networks can include, but are not limited to, U-Net, SegNet, etc. The network structure of image segmentation networks is not limited here. For example, an image segmentation network can include an encoder and a decoder, and there can be skip connections between the encoder and decoder. For better understanding, please refer to [the relevant documentation / references]. Figure 2 , Figure 2 This is a schematic diagram of the framework of an embodiment of the mask prediction model of this application. Figure 2 As shown, in practical applications, the image segmentation network can specifically use U-Net. The left half of U-Net is the encoder, and the right half is the decoder. The red arrows between them indicate skip connections. It should be noted that... Figure 2 In the image segmentation network shown, the numbers labeled above the graphical elements (representing feature maps) represent the number of channels. For more details, please refer to the relevant documentation. Figure 2 Table 1 is a hierarchical schematic table of an embodiment of the mask prediction model of this application.

[0030] Table 1. Hierarchical diagram of an embodiment of the mask prediction model of this application.

[0031] like Figure 2As shown in Table 1, the encoder includes multiple operations of "3*3 convolution + max pooling" (e.g., K (e.g., 5) layers of convolution + pooling can be used to extract semantic features), so that the number of channels gradually increases from 1 to 64, 128, 256, 512, and 1024. The decoder can include multiple operations of "upsampling + skip connections" (e.g., K (e.g., 5) layers of deconvolution + concatenation can be used to gradually restore resolution), so that the number of channels gradually decreases from 1024 to 512, 256, 128, and 64. The output layer can contain 1*1 convolutions, mainly used to restore the number of channels from 64 back to 1. For more detailed processing procedures, please refer to the technical details of U-Net, which will not be elaborated here. Of course, Figure 2 The examples shown in Table 1 are merely one possible case when using U-Net as an image segmentation network; no specific structure of the image segmentation network is specified here.

[0032] Step S13: Process the input features based on the image segmentation network to obtain the prediction mask.

[0033] In this embodiment of the disclosure, the prediction mask can have the same size as the input feature. For example, if the input feature is a one-dimensional vector of size 1*10000, the size of the prediction mask can also be 1*10000; or, as another example, if the input feature is a two-dimensional vector of size 100*100, the size of the prediction mask can also be 100*100. It should be noted that the above examples are only two possible examples of different sizes of the prediction mask, and other possible cases will not be listed here.

[0034] Specifically, as in the previous example, taking an image segmentation network using an encoder-decoder architecture as an example, the input features are first processed layer by layer by the encoder, with the feature resolution gradually decreasing and the number of channels gradually increasing. Then, they are processed layer by layer by the decoder, with the feature resolution gradually increasing and the number of channels gradually decreasing. During the layer-by-layer processing of the decoder, each layer also makes skip connections with the corresponding layer in the encoder to combine the output features of the corresponding layer. Finally, the output layer restores the original resolution (i.e., the same size as the input features) and the number of channels to a prediction mask. For the specific process, please refer to the technical details of image segmentation networks such as U-Net, which will not be elaborated here. It should be noted that the element value at any position in the prediction mask can represent the probability that the corresponding frequency point in the input features is occupied. The higher the element value, the higher the probability, and vice versa. In this way, even when facing extremely wide bandwidths and huge scale differences (e.g., within the same observation bandwidth, signals from different communication devices may have bandwidth spans that differ by more than a hundred times, i.e., extremely narrowband to relatively wideband coexist), compared to using fixed windows or fixed anchor frames, the spectrum occupancy detection using Welch power spectral density and the U-Net image segmentation network can solve the detection problem under extremely wide bandwidths and huge scale differences through spectrum smoothing estimation. In addition, the combination of power spectral density and U-Net multi-scale feature fusion modeling also helps to solve the problem of target sparsity (i.e., "mostly idle, few occupied" under large bandwidth).

[0035] Step S14: Perform a linear mapping based on the one-dimensional mask reshaped from the predicted mask to obtain the target mask.

[0036] In this embodiment, the one-dimensional mask and the target mask can be of the same length as the total number of frequency points, and the target mask can contain the probability value of each frequency point being occupied. It should be noted that since the input features of the image segmentation network are obtained by reshaping a one-dimensional vector, in order to minimize the possible impact of vector reshaping on mask prediction, a linear mapping can be performed on the prediction mask after obtaining it.

[0037] In one implementation scenario, as a possible example, as mentioned earlier, the image segmentation network can include one-dimensional convolutions, thus supporting the direct processing of one-dimensional feature vectors. In this case, the one-dimensional vector itself can be directly used as the input feature of the image segmentation network, and the prediction mask itself can be directly used as a one-dimensional mask. Alternatively, as another possible example, as mentioned earlier, the image segmentation network can also include two-dimensional convolutions, thus supporting the processing of two-dimensional feature vectors. In this case, the one-dimensional vector can be reshaped into a two-dimensional feature vector and used as the input feature that the image segmentation network can process. In this case, the prediction mask itself, since it is essentially a two-dimensional matrix, can be reshaped into a one-dimensional matrix (e.g., each row of a two-dimensional matrix is ​​concatenated sequentially to obtain a one-dimensional matrix) as a one-dimensional mask. It should be noted that the above examples are only a few possible examples of obtaining a one-dimensional mask from the prediction mask in practical applications; other possible cases will not be listed here.

[0038] In one implementation scenario, please continue to refer to [the relevant documentation]. Figure 2 According to Table 1, in order to perform linear mapping on the one-dimensional mask, the one-dimensional mask can be processed based on a fully connected layer to obtain an output logits of the same length as the total number of frequency points. Then, the output logits can be processed based on a normalization layer to obtain the target mask. Taking a one-dimensional vector of size 1*10000 as an example, in the case of an image segmentation network containing two-dimensional convolution, the one-dimensional vector can be reshaped into a 100*100 input feature. After processing by the image segmentation network, a prediction mask of size 100*100 can be obtained. This mask can then be reshaped back into a one-dimensional mask of size 1*10000. The one-dimensional mask of size 1*10000 can then be processed by a fully connected layer to obtain an output logits of size 1*10000 (e.g., it can be processed by two fully connected layers sequentially; the first processing can obtain an output logits of size 1*512, and the second processing can obtain an output logits of size 1*10000). Finally, a normalization layer such as sigmoid can be used to normalize the output logits to obtain the target mask (e.g., a target mask of size 1*10000 contains the probability values ​​of 10000 frequency points being occupied).

[0039] In one implementation scenario, as mentioned earlier, the target mask can be obtained by a mask prediction model predicting the input features. This mask prediction model can include an image segmentation network and a linear mapping network (such as one containing fully connected layers, normalization layers, etc.) to implement the linear mapping. Before applying the mask prediction model, it can be trained first. Specifically, based on the sample energy values ​​of each sample frequency point in the sample power spectral density of the sample wireless signal, the sample input features that the image segmentation network can process can be obtained. Based on the frequency band occupancy label of the sample wireless signal, a true sample mask is generated. Then, the sample input features are processed based on the mask prediction model to obtain the sample target mask. Both the true sample mask and the sample target mask are of equal length to the total number of sample frequency points. Finally, the network parameters of the mask prediction model can be adjusted based on the difference between the true sample mask and the sample target mask. The above method generates a sample real mask by using the frequency band occupancy label of the sample wireless signal. During the training process, the network parameters of the mask prediction model are adjusted according to the difference between the sample real mask and the sample target mask. This forces the mask prediction model to learn whether the predicted frequency point is occupied by combining the power spectral density.

[0040] In a specific implementation scenario, the sample wireless signal can be referred to in the aforementioned description of the wireless signal under test, the sample power spectral density can be referred to in the aforementioned description of the power spectral density, the sample frequency point can be referred to in the aforementioned description of the frequency point, the sample input features can be referred to in the aforementioned description of the input features, and the sample target mask can be referred to in the aforementioned description of the target mask. All of these will not be repeated here.

[0041] In a specific implementation scenario, as a possible example, the sample wireless signal can be labeled with [[f] start1 ,f end1 ],[f start2 ,f end2 Frequency band occupancy tags such as ],……] indicate that the sample wireless signal is in [f start1 ,f end1 The frequency band is occupied, in [f start2 ,f end2 The frequency band is occupied. Of course, the above example is only one possible example of segment occupancy tags in practical applications, and other possible situations will not be listed here.

[0042] In a specific implementation scenario, as mentioned earlier, similar to power spectral density, the sample power spectral density can contain the sample energy values ​​of each sample frequency point. For any sample frequency point, it can be determined whether the sample frequency point is within the frequency band marked by the frequency band occupancy label. If so, the element corresponding to the sample frequency point in the sample true mask can be set to 1; otherwise, it can be set to 0. For ease of description, the sample true mask M[f] can be represented as:

[0043] In a specific implementation scenario, after obtaining the sample target mask, the difference between the sample true mask and the sample target mask can be measured based on the cross-entropy loss function to obtain the training loss. For ease of description, taking the cross-entropy loss function as an example, the training loss Loss can be expressed as:

[0044] In the above formula, σ represents the sigmoid function, and M[f] represents the true mask of the sample. This represents the sample target mask, and N represents the batch size. After obtaining the training loss, the network parameters of the mask prediction model can be adjusted based on the training loss.

[0045] Step S15: Based on the target mask, determine the frequency band intervals that have been occupied.

[0046] In one implementation scenario, after obtaining the target mask, the probability values ​​in the target mask can be binarized based on a probability threshold to obtain a binary mask. That is, the probability threshold is used to binarize the probability values ​​corresponding to each frequency point in the target mask to either 1 (indicating the frequency point is occupied) or 0 (indicating the frequency point is not occupied), thus obtaining the binary mask. Based on this, during the sequential scanning of the binary mask, the starting frequency point f of the frequency band interval can be determined based on the rising transition point (i.e., the transition from 0 to 1). start (For example, the frequency point corresponding to "1" in the rising transition point can be determined as the starting frequency point), and based on the falling transition point (i.e., 1 transitions to 0), the ending frequency point f of the frequency band interval can be determined. end (For example, the frequency point corresponding to "0" in the descent transition point can be determined as the end frequency point), and based on the starting frequency point f start and the end frequency f end Determine the frequency band interval [f start ,f end For example, each pair of adjacent start and end frequencies can determine a frequency band interval. The above method, by binarizing the target mask and then determining the start and end frequencies by combining rising and falling transition points during sequential scanning, can alleviate the target sparsity problem as much as possible.

[0047] In a specific implementation scenario, the probability threshold can be set according to the actual application needs. For example, if it is necessary to minimize the error rate, the probability threshold can be set appropriately high; or, if it is necessary to minimize the omission rate, the probability threshold can be set appropriately low; or, if it is necessary to maintain a balance between the error rate and the omission rate as much as possible, the probability threshold can be set appropriately moderate. As a possible implementation example, the probability threshold can be set to 0.5. Of course, the above example is only one possible example of the probability threshold, and the specific value of the probability threshold is not limited here, nor will it be listed in detail.

[0048] In a specific implementation scenario, when the probability value is not lower than (or higher than) the probability threshold, the probability value can be reset to the first value representing that the frequency point is occupied, such as 1. Conversely, when the probability value is lower than (or not higher than) the probability threshold, the probability value can be reset to the second value representing that the frequency point is not occupied, such as 0. For ease of description, the probability threshold can be denoted as T, and the binary mask B[f] can be expressed as:

[0049] In another implementation scenario, unlike the aforementioned implementation, after obtaining each group of adjacent start and end frequencies through binarization and sequential scanning, each candidate interval can be determined separately based on each group of adjacent start and end frequencies (as mentioned earlier, each group of adjacent start and end frequencies determines one candidate interval). Based on this, it can be determined whether to merge adjacent candidate intervals into a frequency band interval based on the frequency interval between them. It should be noted that the frequency interval between adjacent candidate intervals can be calculated by the difference between the end frequency of the preceding and following candidate intervals. Furthermore, when the frequency interval is less than (or not greater than) the interval threshold, adjacent candidate intervals can be merged into a frequency band interval; conversely, when the frequency interval is not less than (or greater than) the interval threshold, adjacent candidate intervals do not need to be merged. For example, the interval threshold can be set according to actual application needs, such as 1kHz, i.e., 0.001MHz. Of course, the above example is merely one possible implementation example in practical applications, and the specific value of the interval threshold is not limited here. Through the above processing, the occupied frequency band intervals can finally be obtained, such as The above method, after obtaining candidate intervals defined by start and end frequencies, further determines whether to merge adjacent candidate intervals based on the frequency interval between them, which can alleviate short-term breaks in mask prediction.

[0050] The above scheme obtains a one-dimensional vector of equal length to the total number of frequency points based on the energy values ​​of each frequency point in the power spectral density of the wireless signal under test. This vector is then reshaped to obtain the input features that the image segmentation network can process. The input features are then processed by the image segmentation network to obtain a prediction mask, which has the same size as the input features. A linear mapping is then performed on the one-dimensional mask obtained from the reshaped prediction mask to obtain the target mask. Both the one-dimensional mask and the target mask have the same length as the total number of frequency points. The target mask contains the probability values ​​of each frequency point being occupied. Based on the target mask, the occupied frequency band intervals are determined. On the one hand, because the spectrum occupancy detection problem is cleverly transformed into an image segmentation problem with input features constructed from the energy values ​​at frequency points, compared to single threshold detection of spectrum occupancy, even under conditions of large energy dynamic range, low signal-to-noise ratio, and non-stationary noise, masked modeling of spectrum occupancy detection can be achieved. Furthermore, the image segmentation model can also rely on its own modeling of depth features during processing. On the other hand, because masked modeling can predict the occupancy probability at the frequency point level, the precision of spectrum occupancy detection can be improved as much as possible. Therefore, it can improve the detection accuracy of spectrum occupancy, especially in situations with a large energy dynamic range, low signal-to-noise ratio, and non-stationary noise.

[0051] Please see Figure 3 , Figure 3 This is a schematic diagram of the framework of an embodiment of the spectrum occupancy detection device of this application. The spectrum occupancy detection device 30 includes: a vector extraction module 31, a vector reshaping module 32, a mask prediction module 33, a linear mapping module 34, and an interval determination module 35. The vector extraction module 31 is used to obtain a one-dimensional vector of the same length as the total number of frequency points based on the energy values ​​of each frequency point in the power spectral density of the wireless signal under test. The vector reshaping module 32 is used to reshape the one-dimensional vector to obtain the input features that the image segmentation network can process. The mask prediction module 33 is used to obtain a prediction mask based on the input features processed by the image segmentation network. The prediction mask has the same size as the input features. The linear mapping module 34 is used to perform linear mapping based on the one-dimensional mask obtained by reshaping the prediction mask to obtain a target mask. The one-dimensional mask and the target mask have the same length as the total number of frequency points, and the target mask contains the probability value of each frequency point being occupied. The interval determination module 35 is used to determine the occupied frequency band intervals based on the target mask.

[0052] In the above scheme, the spectrum occupancy detection device 30 obtains a one-dimensional vector of the same length as the total number of frequency points based on the energy values ​​of each frequency point in the power spectral density of the wireless signal under test. This one-dimensional vector is then reconstructed to obtain input features that the image segmentation network can process. The input features are then processed by the image segmentation network to obtain a prediction mask, which has the same size as the input features. A linear mapping is then performed based on the one-dimensional mask reconstructed from the prediction mask to obtain a target mask. Both the one-dimensional mask and the target mask have the same length as the total number of frequency points. The target mask contains the probability value of each frequency point being occupied. Furthermore, based on... Target masking identifies occupied frequency bands. On one hand, by cleverly transforming the spectrum occupancy detection problem into an image segmentation problem using energy values ​​at frequency points as input features, it achieves masked modeling of spectrum occupancy detection, even under conditions of large energy dynamic range, low signal-to-noise ratio, and non-stationary noise, compared to single-threshold detection. Furthermore, the image segmentation model can leverage its own deep feature modeling capabilities during processing. On the other hand, masked modeling can predict frequency-level occupancy probabilities, thus maximizing the precision of spectrum occupancy detection. Therefore, it improves the accuracy of spectrum occupancy detection, especially under conditions of large energy dynamic range, low signal-to-noise ratio, and non-stationary noise.

[0053] In some disclosed embodiments, the vector extraction module 31 includes an energy processing submodule, which is used to add a preset constant to the energy value of each frequency point in the power spectral density and then take the logarithm, so as to obtain the element value in the reference vector corresponding to each frequency point; the vector extraction module 31 includes a first normalization submodule, which is used to normalize based on the reference vector to obtain a one-dimensional vector.

[0054] In some disclosed embodiments, the linear mapping module 34 includes a fully connected submodule for processing a one-dimensional mask based on the fully connected layer to obtain an output logits of the same length as the total number of frequency points; the linear mapping module 34 includes a second normalization submodule for processing the output logits based on the normalization layer to obtain a target mask.

[0055] In some disclosed embodiments, the interval determination module 35 includes a binarization submodule, used to binarize the probability values ​​in the target mask based on a probability threshold to obtain a binary mask; the interval determination module 35 includes a mask scanning submodule, used to determine the starting frequency point of the frequency band interval based on the rising transition point and the ending frequency point of the frequency band interval based on the falling transition point during the sequential scanning of the binary mask, and to determine the frequency band interval based on the starting frequency point and the ending frequency point.

[0056] In some disclosed embodiments, the mask scanning submodule is specifically used to determine each candidate interval based on each group of adjacent start and end frequency points; and to determine whether to merge adjacent candidate intervals as frequency band intervals based on the frequency interval between adjacent candidate intervals.

[0057] In some disclosed embodiments, the target mask is obtained by predicting input features using a mask prediction model. The mask prediction model includes an image segmentation network and a linear mapping network for implementing linear mapping. The spectrum occupancy detection device 30 includes a sample extraction module, which obtains the sample input features that the image segmentation network can process based on the sample energy values ​​of each sample frequency point in the sample power spectral density of the sample wireless signal, and generates a sample true mask based on the frequency band occupancy label of the sample wireless signal. The spectrum occupancy detection device 30 includes a sample prediction module, which processes the sample input features based on the mask prediction model to obtain a sample target mask. The sample true mask and the sample target mask are each equal in length to the total number of sample frequency points. The spectrum occupancy detection device 30 includes a parameter adjustment module, which adjusts the network parameters of the mask prediction model based on the difference between the sample true mask and the sample target mask.

[0058] In some disclosed embodiments, the image segmentation network includes an encoder and a decoder, with skip connections between the encoder and the decoder.

[0059] In some disclosed embodiments, the power spectral density is obtained by Welch power spectral density estimation.

[0060] In some publicly available embodiments, the input feature is either a one-dimensional feature or a two-dimensional feature.

[0061] Please see Figure 4 , Figure 4 This is a schematic diagram of a framework of an embodiment of the electronic device of this application. The electronic device 40 includes at least a memory 41 and a processor 42 coupled to each other. The memory 41 stores at least program instructions, and the processor 42 is used to execute the program instructions to implement the steps in any of the above-described embodiments of the spectrum occupancy detection method. For details, please refer to the foregoing disclosed embodiments, which will not be repeated here.

[0062] Specifically, processor 42 controls itself and memory 41 to implement the steps in any of the above-described embodiments of the spectrum occupancy detection method. Processor 42 may also be referred to as a CPU (Central Processing Unit). Processor 42 may be an integrated circuit chip with signal processing capabilities. Processor 42 may also be a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor. Furthermore, processor 42 may be implemented using integrated circuit chips.

[0063] In the above scheme, the electronic device 40 obtains a one-dimensional vector of the same length as the total number of frequency points based on the energy values ​​of each frequency point in the power spectral density of the wireless signal under test. This one-dimensional vector is then reshaped to obtain input features that the image segmentation network can process. The image segmentation network then processes these input features to obtain a prediction mask, which has the same size as the input features. A linear mapping is then performed based on the one-dimensional mask reshaped from the prediction mask to obtain the target mask. Both the one-dimensional mask and the target mask have the same length as the total number of frequency points. The target mask contains the probability value of each frequency point being occupied. Finally, based on the target... Masking is used to determine the occupied frequency band intervals. On one hand, by cleverly transforming the spectrum occupancy detection problem into an image segmentation problem using energy values ​​at frequency points as input features, compared to single-threshold spectrum occupancy detection, masked modeling of spectrum occupancy detection can be achieved even under conditions of large energy dynamic range, low signal-to-noise ratio, and non-stationary noise. Furthermore, the image segmentation model can rely on its own modeling of depth features during processing. On the other hand, because masked modeling can predict the occupancy probability at the frequency point level, the fineness of spectrum occupancy detection can be improved as much as possible. Therefore, the detection accuracy of spectrum occupancy can be improved, especially under conditions of large energy dynamic range, low signal-to-noise ratio, and non-stationary noise.

[0064] Please see Figure 5 , Figure 5 This is a schematic diagram of a computer-readable storage medium according to an embodiment of the present application. The computer-readable storage medium 50 stores program instructions 51 that can be executed by a processor. The program instructions 51 are used to implement the steps in any of the above-described embodiments of the spectrum occupancy detection method.

[0065] In the above scheme, the computer-readable storage medium 50 obtains a one-dimensional vector of the same length as the total number of frequency points based on the energy values ​​of each frequency point in the power spectral density of the wireless signal under test. This one-dimensional vector is then reshaped to obtain input features that the image segmentation network can process. The image segmentation network then processes these input features to obtain a prediction mask, which has the same size as the input features. A linear mapping is then performed based on the one-dimensional mask reshaped from the prediction mask to obtain the target mask. Both the one-dimensional mask and the target mask have the same length as the total number of frequency points. The target mask contains the probability value of each frequency point being occupied. Furthermore, based on... Target masking identifies occupied frequency bands. On one hand, by cleverly transforming the spectrum occupancy detection problem into an image segmentation problem using energy values ​​at frequency points as input features, it achieves masked modeling of spectrum occupancy detection, even under conditions of large energy dynamic range, low signal-to-noise ratio, and non-stationary noise, compared to single-threshold detection. Furthermore, the image segmentation model can leverage its own deep feature modeling capabilities during processing. On the other hand, masked modeling can predict frequency-level occupancy probabilities, thus maximizing the precision of spectrum occupancy detection. Therefore, it improves the accuracy of spectrum occupancy detection, especially under conditions of large energy dynamic range, low signal-to-noise ratio, and non-stationary noise.

[0066] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0067] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities between them can be referred to, and for the sake of brevity, they will not be repeated here.

[0068] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus implementations described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0069] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0070] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0071] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0072] If the technical solution of this application involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

Claims

1. A method for detecting spectrum occupancy, characterized in that, include: Based on the energy values ​​of each frequency point in the power spectral density of the wireless signal under test, a one-dimensional vector with the same length as the total number of frequency points is obtained. Based on the one-dimensional vector, the image segmentation network can be reshaped to obtain the input features that it can process. The image segmentation network processes the input features to obtain a prediction mask; wherein the prediction mask has the same size as the input features. A target mask is obtained by linear mapping based on the one-dimensional mask reshaped from the predicted mask; wherein the one-dimensional mask, the target mask and the total number of frequency points are of equal length, and the target mask contains the probability value of each frequency point being occupied. Based on the target mask, the frequency band intervals that have been occupied are determined.

2. The method according to claim 1, characterized in that, The energy values ​​of each frequency point in the power spectral density of the wireless signal under test are used to obtain a one-dimensional vector of the same length as the total number of frequency points, including: For the energy value of each frequency point in the power spectral density, a preset constant is added to each and then the logarithm is taken as the element value in the reference vector corresponding to each frequency point. The one-dimensional vector is obtained by normalizing the reference vector.

3. The method according to claim 1, characterized in that, The step of obtaining the target mask by linear mapping based on the one-dimensional mask reconstructed from the predicted mask includes: The one-dimensional mask is processed by a fully connected layer to obtain an output logits of the same length as the total number of frequency points. The target mask is obtained by processing the output logits using a normalization layer.

4. The method according to claim 1, characterized in that, The step of determining the occupied frequency band interval based on the target mask includes: Binarize the probability values ​​in the target mask based on the probability threshold to obtain a binary mask; During the sequential scanning of the binary mask, the starting frequency of the frequency band interval is determined based on the rising transition point, and the ending frequency of the frequency band interval is determined based on the falling transition point. The frequency band interval is then determined based on the starting frequency and the ending frequency.

5. The method according to claim 4, characterized in that, Determining the frequency band interval based on the starting frequency point and the ending frequency point includes: Based on each group of adjacent starting and ending frequency points, each candidate interval is determined. Based on the frequency interval between adjacent candidate intervals, it is determined whether to merge adjacent candidate intervals as the frequency band interval.

6. The method according to claim 1, characterized in that, The target mask is obtained by a mask prediction model predicting the input features. The mask prediction model includes the image segmentation network and a linear mapping network for implementing the linear mapping. The training steps of the mask prediction model include: Based on the sample energy value of each sample frequency point in the sample power spectral density of the sample wireless signal, the sample input features that the image segmentation network can process are obtained, and the sample true mask is generated based on the frequency band occupancy label of the sample wireless signal. The sample input features are processed based on the mask prediction model to obtain the sample target mask; wherein the sample true mask and the sample target mask are each of equal length to the total number of sample frequency points; Based on the difference between the sample's true mask and the sample's target mask, the network parameters of the mask prediction model are adjusted.

7. The method according to any one of claims 1 to 6, characterized in that, The image segmentation network includes an encoder and a decoder, and there is a skip connection between the encoder and the decoder; And / or, the power spectral density is obtained by Welch power spectral density estimation; And / or, the input feature is either a one-dimensional feature or a two-dimensional feature.

8. A spectrum occupancy detection device, characterized in that, include: The vector extraction module is used to obtain a one-dimensional vector of the same length as the total number of frequency points based on the energy value of each frequency point in the power spectral density of the wireless signal under test. The vector reshaping module is used to reshape the vector based on the one-dimensional vector to obtain the input features that the image segmentation network can process. A mask prediction module is used to process the input features based on the image segmentation network to obtain a predicted mask; wherein the predicted mask has the same size as the input features. A linear mapping module is used to perform linear mapping based on a one-dimensional mask reshaped from the predicted mask to obtain a target mask; wherein the one-dimensional mask, the target mask and the total number of frequency points are of equal length, and the target mask contains the probability value of each frequency point being occupied; The interval determination module is used to determine the occupied frequency band intervals based on the target mask.

9. An electronic device, characterized in that, The method includes at least a memory and a processor coupled to each other, wherein the memory stores at least program instructions, and the processor executes the program instructions to implement the spectrum occupancy detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The device stores program instructions that can be executed by a processor, the program instructions being used to implement the spectrum occupancy detection method according to any one of claims 1 to 7.

Images