Neural network design method for personnel sensing wireless signal generation

Abstract

The invention discloses a neural network design method for personnel perception wireless signal generation, and belongs to the field of wireless signal processing deep learning, and the method comprises the steps: S1, converting a wireless signal inputted into a neural network into a spectrogram format, and obtaining a signal; s2, extracting local frequency characteristics in the wireless signals by using a one-dimensional Fourier transform network; s3, further extracting time domain features of the wireless signals by using a ViT network; s4, performing two-dimensional frequency domain feature extraction, and extracting global frequency features in the wireless signals by using a two-dimensional Fourier transform network; and S5, inputting the extracted time-frequency characteristics of the wireless signals into the signal generation network for finally outputting the signals. According to the neural network design method for the personnel to perceive the generation of the wireless signals, the generated wireless signals have richer and more real frequency information, and the quality of the generated signals is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning for wireless signal processing, and in particular to a neural network design method for generating human perception wireless signals. Background Art

[0002] The human wireless perception technology is a technology that uses wireless signals (such as Wi-Fi, millimeter wave, Bluetooth, ultra-wideband, etc.) to realize the perception of human activities, behaviors and physiological states. It processes wireless signals, analyzes the changes in their propagation characteristics near the human body, and extracts information related to human behaviors, and is widely used in the fields of smart home, health monitoring, security monitoring, etc.

[0003] In recent years, with the rapid development of Artificial Intelligence (AI) technology, AI-based wireless signal processing methods have been widely used. However, the training of AI models requires a large amount of wireless signal data, and the acquisition cost of high-quality wireless signals is extremely high. Therefore, generating simulated wireless signals has become a very promising solution. This method can not only significantly reduce the cost of signal acquisition, but also provide rich training data for the model, thereby improving the efficiency and accuracy of wireless signal processing.

[0004] Benefiting from the powerful ability of generative artificial intelligence in generating content, a large amount of existing work has generated wireless signals based on this technology. The existing wireless signal generation methods mainly use Convolutional Neural Network (CNN) to extract features from the original signals, and then generate new simulated wireless signals based on the extracted features. However, CNN has limitations in the feature extraction and generation process of wireless signals, mainly reflected in the insufficient processing of the temporal and frequency features of the signals. Wireless signals are essentially time-series data containing rich dynamic information, but CNN is difficult to capture these long-term dependencies, resulting in the inability to fully reproduce the time-domain features. In addition, the frequency features in wireless signals (such as Doppler frequency shift) are crucial for detecting the target speed, while the local convolution operation of CNN is difficult to extract the global spectrum information of the signals, resulting in frequency-domain distortion of the generated signals. These deficiencies make the wireless signals generated by CNN less than ideal in terms of accuracy and authenticity.

[0005] In summary, in order to generate high-quality wireless signals, there is an urgent need for a neural network that can effectively extract the temporal and frequency features of wireless signals at the same time, improve the accuracy and authenticity of signal generation, so as to meet the needs of high-precision applications. Summary of the Invention

[0006] The object of the present invention is to provide a neural network design method for generating human perception wireless signals, which can effectively solve the problems in the prior art that the rich physical features in wireless signals are not fully considered, it is difficult to extract the time-frequency features in wireless signals, and it is difficult to generate wireless signals with high precision and high authenticity.

[0007] To achieve the above object, the present invention provides a neural network design method for generating human perception wireless signals, including the following steps: S1. Convert the wireless signal input into the neural network into a spectrogram format to obtain a signal , indicating that the input wireless signal format is a three-dimensional matrix vector with a length of M, a width of N, and a channel number of C; S2. Perform one-dimensional frequency domain feature extraction, and use a one-dimensional Fourier transform network to extract the local frequency features in the wireless signal; S3. Perform time domain feature extraction, and use a ViT network to further extract the time domain features of the wireless signal; including: S31. Extract basic time domain features; S32. Extract enhanced time domain features; S4. Perform two-dimensional frequency domain feature extraction, and use a two-dimensional Fourier transform network to extract the global frequency features in the wireless signal; S5. Input the extracted time-frequency features of the wireless signal into a signal generation network for final output of the signal.

[0008] Preferably, the content of S2 is as follows: S21. Perform one-dimensional Fourier transform and extract features; S22. Perform adaptive feature extraction.

[0009] Preferably, the content of S21 is as follows: S211. Perform a discrete Fourier transform along the length dimension of the wireless signal, and the specific formula is as follows: ; where represents the signal of the th column, represents the imaginary unit, represents at the spectrum; S212. Apply the operation in S211 to each channel to obtain a spectrogram .

[0010] Preferably, the content of S22 is as follows: S221. Randomly initialize a learnable parameter vector , replicate and extend it along the width direction and the channel direction respectively to obtain a parameter vector ; S222. Randomly initialize a learnable parameter vector , replicate and extend it along the length direction and the channel direction respectively to obtain a parameter vector ; S223. Multiply the parameter vector in S221 by the parameter vector in S222 to obtain the final learnable parameter vector ; S224. Element-wise multiply the learnable parameter vector in S223 by the spectrogram in S22 to obtain the filtered spectrogram , and the specific formula is as follows: ; S225. Perform the inverse discrete Fourier transform on the filtered spectrogram in S224 to obtain the one-dimensional frequency-domain feature ; Among them, represents the -th element in the -th column of the filtered signal , and represents the -th element in the -th column of the spectrogram ; S226. Apply the operation in S225 to each channel to obtain the filtered signal .

[0011] Preferably, the calculation formula of S225 is as follows: .

[0012] Preferably, S31. Extract the basic time-domain feature content as follows: S311. Initialize a two-dimensional convolution module , set its kernel size parameter to , set its stride parameter to , and set its number of channels parameter to ; S312. Apply the two-dimensional convolution module in S311 to the filtered signal in S226 to obtain the feature vector , and the specific formula is as follows: ; Among them, ; Preferably, the enhanced time-domain feature extraction in S32 is as follows: S321. Initialize the two-dimensional convolution module , and set its kernel size parameter to , and set its stride parameter to , and set its padding 0 parameter to , and set its number of channels parameter to ; S322. Apply the two-dimensional convolution module in S321 to the filtered signal in S226 to obtain a feature vector , and the specific formula is as follows: ; where, .

[0013] Preferably, S4 is as follows: S41. Shape reshaping: S411. Reshape the feature vector in S312 to obtain a reshaped feature vector ; S412. Reshape the feature vector in S322 to obtain a reshaped feature vector ; S42. Layer normalization: S421. Calculate the mean for each row of the reshaped feature vector in S411 along the feature dimension . Taking the th row as an example, the specific formula is as follows: ; where, represents the mean of the th row, represents the th element of the rd column of the th row of the feature vector S422. Calculate the variance for each row of the reshaped feature vector in S411 along the feature dimension . Taking the th row as an example, the specific formula is as follows: ; where, represents the variance of the th row; S423. Normalize each element of the reshaped feature vector in S411 to obtain a normalized feature vector . Taking the element in the -th row and -th column as an example, the specific formula is as follows: ; S424. Perform the same operations as S421 to S423 on the reshaped feature vector in S412 to obtain a normalized feature vector ; S43. Two-dimensional Fourier transform and feature extraction: Perform two-dimensional Fourier transform processing on the normalized feature vector in S423 to obtain a spectrogram ; S44. Adaptive feature extraction: S441. Randomly initialize a learnable parameter vector ; S442. Element-wise multiply the learnable parameter vector in S441 with the spectrogram in S43 to obtain a filtered spectrogram , and the specific formula is as follows: ; S443. Perform two-dimensional inverse discrete Fourier transform on the filtered spectrogram in S442 to obtain two-dimensional frequency domain features . Taking the element in the -th row and -th column as an example, the specific formula is as follows: ; S444. Perform the same operations as S441 to S443 on the reshaped feature vector in S424 to obtain two-dimensional frequency domain features ; S45. Shape reshaping: S451. Reshape the reshaped feature vector in S443 to obtain a feature vector ; S452. Reshape the reshaped feature vector in S444 to obtain a feature vector ; S46. Layer normalization: S461. For each row of the feature vector in S451 along the feature dimension Calculate the mean value , taking the th row as an example, the specific formula is as follows: ; where, represents the mean value of the th row, represents the th row and the th element of the feature vector; S462. Calculate the variance of each row of the feature vector in S451 along the feature dimension , taking the th row as an example, the specific formula is as follows: ; where, represents the variance of the th row; S463. Normalize each element of the feature vector in S451 to obtain the feature vector , taking the element in the th row and the th column as an example, the specific formula is as follows: ; S464. Perform the same operations as S461 to S463 on the feature vector in S452 to obtain the feature vector .

[0014] Preferably, the calculation formula of S43 is as follows: Taking the element in the th row and the th column as an example, ; where the spectrogram is .

[0015] Preferably, the content of S5 is as follows: S51. Input dimension unification: S511. Use the fully connected layer of the neural network to map the feature vector in S463 to a unified dimension to obtain the unified feature vector , and the specific formula is as follows: ; where, ; S512, using the neural network fully connected layer The feature vector in S464 Map to a unified dimension to obtain a unified feature vector , the specific formula is as follows: ; in, ; S52, Feature Recovery: S521, unify the feature vector in S511 Perform all the same operations as S4 , and the restored feature vector is obtained , the specific formula is as follows: ; S522, unify the feature vector in S512 Perform all the same operations as S4 , and the restored feature vector is obtained , the specific formula is as follows: ; S53, output dimension unification: S531. Using the fully connected layer of the neural network The restored feature vector in S521 Map to a unified dimension to obtain a unified feature vector , the specific formula is as follows: ; in, ; S532, using the neural network fully connected layer The restored feature vector in S522 Map to a unified dimension to obtain a unified feature vector , the specific formula is as follows: ; in, ; S54, output generation signal: S541, unify the feature vector in S531 Reshape to get feature vector ; S542, unify the feature vector in S532 Reshape to get feature vector ; S543, the feature vector in S541 and the eigenvector in S542 Add bit by bit along the center point to finally generate the wireless signal .

[0016] Therefore, the present invention adopts the above-mentioned neural network design method for generating wireless signals for personnel perception, and has the following beneficial effects: (1) It can specifically extract effective frequency domain features for wireless signal generation from the original wireless signal. Compared with the traditional signal generation method, by integrating one-dimensional Fourier transform and two-dimensional Fourier transform into the neural network, and at the same time enabling the neural network to learn and extract effective frequency features in an adaptive manner, the generated wireless signal has richer and more real frequency information; (2) It can specifically extract effective time domain features for wireless signal generation from the original wireless signal. Based on the ViT architecture, a global receptive field is established through the self-attention mechanism, enabling the neural network to capture long-distance time relationships, and at the same time adaptively learning and extracting effective time domain features, thereby improving the quality of the generated signal.

[0017] Next, through the drawings and embodiments, the technical solutions of the present invention will be further described in detail. Description of the Drawings

[0018] Figure 1 It is a flowchart of wireless signal feature extraction and generation proposed in an embodiment of the present invention; Figure 2 It is a schematic diagram of the input wireless signal in an embodiment of the present invention; Figure 3 It is a schematic diagram of the wireless signal generated by the wireless signal generation method proposed in an embodiment of the present invention and the real wireless signal; among them, (a) is the real wireless signal , (b) is the generated wireless signal . Specific Embodiments

[0019] The following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the present invention claimed, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0020] Please refer to Figure 1 , a neural network design method for generating wireless signals for personnel perception, including the following steps: S1. As Figure 2 , convert the wireless signal input into the neural network into a spectrogram format to obtain the signal , indicating that the input wireless signal format is a three-dimensional matrix vector with a length of M, a width of N, and a channel number of C.

[0021] S2. Perform one-dimensional frequency-domain feature extraction, and use a one-dimensional Fourier transform network to extract local frequency features in the wireless signal.

[0022] S21. One-dimensional Fourier transform and feature extraction. The content is as follows: S211. Perform a discrete Fourier transform along the length dimension of the wireless signal. The specific formula is as follows: ; where represents the signal 's th column, represents the imaginary unit, represents at the spectrum.

[0023] S212. Apply the operation in S211 to each channel to obtain a spectrogram .

[0024] S22. Perform adaptive feature extraction.

[0025] S221. Randomly initialize a learnable parameter vector , and replicate and extend it along the width direction and the channel direction respectively to obtain a parameter vector ; S222. Randomly initialize a learnable parameter vector , and replicate and extend it along the length direction and the channel direction respectively to obtain a parameter vector ; S223. Multiply the parameter vector in S221 with the parameter vector in S222 to obtain a final learnable parameter vector ; S224. Element-wise multiply the learnable parameter vector in S223 with the spectrogram in S22 to obtain a filtered spectrogram , and the specific formula is as follows: ; S225. Perform an inverse discrete Fourier transform on the filtered spectrogram in S224 to obtain one-dimensional frequency-domain features , and the specific formula is as follows: ; where, represents the th column of the filtered signal th row of element representing the spectrogram the th column of the element; S226. Apply the operation in S225 to each channel to obtain the filtered signal .

[0026] S3. Perform time-domain feature extraction and use the ViT network to further extract the time-domain features of the wireless signal

[0027] S31. Basic time-domain feature extraction: S311. Initialize the two-dimensional convolutional module , set its kernel size parameter to , set its stride parameter to , set its number of channels parameter to ; S312. Apply the two-dimensional convolutional module in S311 to the S226 filtered signal to obtain the feature vector , and the specific formula is as follows: ; where, ; S32. Enhanced time-domain feature extraction: S321. Initialize the two-dimensional convolutional module , set its kernel size parameter to , set its stride parameter to , set its padding 0 parameter to , and set its number of channels parameter to ; S322. Apply the two-dimensional convolutional module in S321 to the S226 filtered signal to obtain the feature vector , and the specific formula is as follows: ; where, .

[0028] S4. Perform two-dimensional frequency-domain feature extraction and use the two-dimensional Fourier transform network to extract the global frequency features in the wireless signal

[0029] S41. Shape reshaping: S411. Reshape the feature vector in S312 to obtain the reshaped feature vector ; S412, the feature vector in S322 Reshape to get the reshaped feature vector ; S42, layer normalization: S421, reshape the feature vector in S411 Each row of Calculate the mean , with the Behavior example, the specific formula is as follows: ; in, Indicates The mean of the rows, Represents the feature vector No. The first elements; S422, reshape the feature vector in S411 Each row of Calculating variance , with the Behavior example, the specific formula is as follows: ; in, express No. Variance of rows; S423, reshape the feature vector in S411 Each element of is normalized to obtain the normalized feature vector , with the Line Taking column elements as an example, the specific formula is as follows: ; S424, reshape the feature vector in S412 Perform the same operations from S421 to S423 to obtain the normalized feature vector ; S43, two-dimensional Fourier transform and feature extraction: Normalized feature vector in S423 Perform two-dimensional Fourier transform processing to obtain the spectrum diagram , with the Line Taking column elements as an example, the specific formula is as follows: ; Among them, the spectrum ; S44, Adaptive feature extraction: S441, Randomly initialize the learnable parameter vector ; S442, the learnable parameter vector in S441 Spectrum diagram with S43 Element-by-element product to get the filtered spectrum , the specific formula is as follows: ; S443, the spectrum after filtering in S442 Perform a two-dimensional discrete Fourier inverse transform to obtain two-dimensional frequency domain features , with the Line Taking column elements as an example, the specific formula is as follows: ; S444, reshape the feature vector in S424 Perform the same operations from S441 to S443 to obtain the two-dimensional frequency domain features ; S45, Shape Reshaping: S451, reshape the feature vector in S443 Reshape to get feature vector ; S452, reshape the feature vector in S444 Reshape to get feature vector ; S46, layer normalization: S461, for the eigenvector in S451 Each row of Calculate the mean , with the Behavior example, the specific formula is as follows: ; in, Indicates The mean of the rows, Represents the feature vector No. The first elements; S462, for the characteristic vector in S451 Each row of Calculating variance , with the Behavior example, the specific formula is as follows: ; in, express The variance of the row; S463. Normalize each element of the eigenvector in S451 to obtain the eigenvector . Taking the element in the row and column as an example, the specific formula is as follows: ; S464. Perform the same operations as S461 to S463 on the eigenvector in S452 to obtain the eigenvector .

[0030] S5. Input the time-frequency features of the extracted wireless signal into the signal generation network for the final output signal.

[0031] S51. Unified input dimension: S511. Use the fully connected layer of the neural network to map the eigenvector in S463 to a unified dimension to obtain the unified eigenvector , and the specific formula is as follows: ; where ; S512. Use the fully connected layer of the neural network to map the eigenvector in S464 to a unified dimension to obtain the unified eigenvector , and the specific formula is as follows: ; where ; S52. Feature restoration: S521: Perform all the same operations as S4 on the unified eigenvector in S511 to obtain the restored eigenvector , and the specific formula is as follows: ; S522. Perform all the same operations as S4 on the unified eigenvector in S512 to obtain the restored eigenvector , and the specific formula is as follows: ; S53. Unified output dimension: S531. Use the fully connected layer of the neural network Map the recovery feature vector in S521 to a unified dimension to obtain a unified feature vector , and the specific formula is as follows: ; wherein, ; S532. Use the fully connected layer of the neural network to map the recovery feature vector in S522 to a unified dimension to obtain a unified feature vector , and the specific formula is as follows: ; wherein, ; S54. Output the generated signal: S541. Reshape the shape of the unified feature vector in S531 to obtain a feature vector ; S542. Reshape the shape of the unified feature vector in S532 to obtain a feature vector ; S543. Add the feature vector in S541 and the feature vector in S542 bit by bit along the center point to obtain the final generated wireless signal , as Figure 3 shown.

[0032] Test on two publicly available wireless signal datasets, the Dop-NET dataset and the IURHA2023 dataset. Run 3 times with different random seeds for the experiment, and report the final average value and standard deviation results, which are specifically reflected as follows: As Figure 2 shown, on the IURHA2023 dataset, the wireless signal generated by S533 is extremely visually similar to the real signal; As shown in Table 1, the wireless signal generated in this embodiment shows an improvement in both the average peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) metrics compared to the traditional method.

[0033] Table 1 Comparison of the quality of the neural network-generated signal in this embodiment with the traditional neural network ;

[0034] Therefore, the present invention adopts the above-mentioned neural network design method for personnel perception wireless signal generation. By embedding a learnable frequency domain feature extraction module in the neural network, features are adaptively extracted from the wireless signal for generating refined wireless signals. The frequency domain feature extraction module includes one-dimensional frequency domain feature extraction and two-dimensional frequency domain feature extraction. The adaptively extracted features mean that the neural network can change and gradually optimize the extracted features by continuously iteratively updating the neural network parameters, so as to improve the quality of the generated wireless signals.

[0035] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A neural network design method for personnel perception of wireless signal generation, characterized in that: The following steps are involved: S1. Convert the wireless signal input into the neural network into a spectrogram format to obtain the signal , indicating that the input wireless signal format is a three-dimensional matrix vector with a length of M, a width of N, and a number of channels of C; S2, performing one-dimensional frequency domain feature extraction, using a one-dimensional Fourier transform network to extract local frequency features in the wireless signal; S3. Perform time domain feature extraction and use the Vision Transformer network to further extract the time domain features of the wireless signal; including: S31, extracting basic time domain features; S32, extracting enhanced time domain features; S4, performing two-dimensional frequency domain feature extraction, using a two-dimensional Fourier transform network to extract global frequency features in the wireless signal; S5. Input the extracted time-frequency features of the wireless signal into a signal generation network for a final output signal.

2. A neural network design method for generating a human-perceived wireless signal according to claim 1, characterized in that: The contents of S2 are as follows: S21, one-dimensional Fourier transform and feature extraction; S22. Perform adaptive feature extraction.

3. A neural network design method for generating a human-perceived wireless signal according to claim 2, characterized in that: S21 content is as follows: S211. Perform discrete Fourier transform along the length latitude of the wireless signal. The specific formula is as follows: ； in Indicates signal No. List, represents the imaginary unit, express exist The spectrum at S212, apply the operation in S211 to each channel to obtain a spectrum diagram .

4. A neural network design method for generating a human-perceived wireless signal according to claim 3, characterized in that: S22 content is as follows: S221, Randomly initialize the learnable parameter vector , copy and extend it along the width direction and channel direction to obtain the parameter vector ; S222, Randomly initialize the learnable parameter vector , and copy and extend it along the length direction and channel direction to obtain the parameter vector ; S223, the parameter vector in S221 and the parameter vector in S222 Multiply to get the final learnable parameter vector ; S224, the learnable parameter vector in S223 Spectrum diagram with S22 Element-by-element product to get the filtered spectrum , the specific formula is as follows: ； S225, the spectrum after filtering in S224 Perform inverse discrete Fourier transform to obtain one-dimensional frequency domain features ; in, Represents the filtered signal No. Column No. elements, Representation of the spectrum No. Column No. elements; S226, apply the operation in S225 to each channel to obtain a filtered signal .

5. A neural network design method for generating a human-perceived wireless signal according to claim 4, characterized in that: The S225 calculation formula is as follows: 。 6. A neural network design method for generating a human-perceived wireless signal according to claim 5, characterized in that: The content of basic time domain features extracted by S31 is as follows: S311, initialize the two-dimensional convolution module , setting its kernel size parameter to , setting its step size parameter to , set its channel number parameter to ; S312, the two-dimensional convolution module in S311 Applied to S226 filter signal Get the feature vector , the specific formula is as follows: ； in, .

7. A neural network design method for generating a human-perceived wireless signal according to claim 6, characterized in that: The contents of S32 extraction and enhancement of time domain features are as follows: S321, initialize the two-dimensional convolution module , setting its kernel size parameter to , setting its step size parameter to , set its expansion 0 parameter to , set its channel number parameter to ; S322, the two-dimensional convolution module in S321 Applied to S226 filter signal Get the feature vector , the specific formula is as follows: ； in, , .

8. A neural network design method for generating a human-perceived wireless signal according to claim 7, characterized in that: S4 content is as follows: S41, Shape Reshaping: S411, the feature vector in S312 Reshape to get the reshaped feature vector ; S412, the feature vector in S322 Reshape to get the reshaped feature vector ; S42, layer normalization: S421, reshape the feature vector in S411 Each row of Calculate the mean , with the Behavior example, the specific formula is as follows: ； in, Indicates The mean of the rows, Represents the feature vector No. The first elements; S422, reshape the feature vector in S411 Each row of Calculating variance , with the Behavior example, the specific formula is as follows: ； in, express No. Variance of rows; S423, reshape the feature vector in S411 Each element of is normalized to obtain the normalized feature vector , with the Line Taking column elements as an example, the specific formula is as follows: ； S424, reshape the feature vector in S412 Perform the same operations from S421 to S423 to obtain the normalized feature vector ; S43, two-dimensional Fourier transform and feature extraction: Normalized feature vector in S423 Perform two-dimensional Fourier transform processing to obtain the spectrum diagram ; Among them, the spectrum ; S44, Adaptive feature extraction: S441, Randomly initialize the learnable parameter vector ; S442, the learnable parameter vector in S441 Spectrum diagram with S43 Element-by-element product to get the filtered spectrum , the specific formula is as follows: ； S443, the spectrum after filtering in S442 Perform a two-dimensional discrete Fourier inverse transform to obtain two-dimensional frequency domain features , with the Line Taking column elements as an example, the specific formula is as follows: ； S444, reshape the feature vector in S424 Perform the same operations from S441 to S443 to obtain the two-dimensional frequency domain features ; S45, Shape Reshaping: S451, reshape the feature vector in S443 Reshape to get feature vector ; S452, reshape the feature vector in S444 Reshape to get feature vector ; S46, layer normalization: S461, for the eigenvector in S451 Each row of Calculate the mean , with the Behavior example, the specific formula is as follows: ； in, Indicates The mean of the rows, Represents the feature vector No. The first elements; S462, for the characteristic vector in S451 Each row of Calculating variance , with the Behavior example, the specific formula is as follows: ； in, express No. Variance of rows; S463, for the characteristic vector in S451 Each element of is normalized to obtain the feature vector , with the Line Taking column elements as an example, the specific formula is as follows: ； S464, for the eigenvector in S452 Perform the same operations from S461 to S463 to obtain the feature vector .

9. A neural network design method for generating a human-perceived wireless signal according to claim 8, characterized in that: The S43 calculation formula is as follows: First Line Take the column element as an example, ； Among them, the spectrum .

10. A neural network design method for generating a human-perceived wireless signal according to claim 9, characterized in that: S5 content is as follows: S51. Unify input dimensions: S511. Using the fully connected layer of the neural network The eigenvector in S463 Map to a unified dimension to obtain a unified feature vector , the specific formula is as follows: ； in, ; S512, using the neural network fully connected layer The feature vector in S464 Map to a unified dimension to obtain a unified feature vector , the specific formula is as follows: ； in, ; S52, Feature Recovery: S521: Unify the feature vectors in S511 Perform all the same operations as S4 , and the restored feature vector is obtained , the specific formula is as follows: ； S522, unify the feature vector in S512 Perform all the same operations as S4 , and the restored feature vector is obtained , the specific formula is as follows: ； S53, output dimension unification: S531. Using the fully connected layer of the neural network The restored feature vector in S521 Map to a unified dimension to obtain a unified feature vector , the specific formula is as follows: ； in, ; S532, using the neural network fully connected layer The restored feature vector in S522 Map to a unified dimension to obtain a unified feature vector , the specific formula is as follows: ； in, ; S54, output generation signal: S541, unify the feature vector in S531 Reshape to get feature vector ; S542, unify the feature vector in S532 Reshape to get feature vector ; S543, the feature vector in S541 and the eigenvector in S542 The final wireless signal is generated by adding bit by bit along the center point .

Images