A semantic domain multiple access transmission interference cancellation algorithm

Abstract

This invention discloses a semantic domain multiple access transmission interference cancellation algorithm, belonging to the field of wireless communication. The method first introduces a user-independent random scrambling scheme, effectively achieving user differentiation while approximating inter-user interference as Gaussian noise. Users can use the same semantic codec, effectively solving the dependence on user-specific encoding / decoding models in traditional semantic communication. In uplink multiple access transmission scenarios, leveraging the reversibility of random scrambling, a decoding-re-encoding approach is used to denoise features, and the denoised features are pre-removed, effectively improving the signal-to-interference-plus-noise ratio (SNR). Experimental results demonstrate the effectiveness of the proposed scheme. This invention overcomes the shortcomings of previous algorithms in this system, and the algorithm introduces minimal additional cost and has no power constraints, greatly expanding the application scope of asymptotic interference cancellation schemes and effectively exploring interference cancellation mechanisms in the semantic domain, a new resource domain.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication, and specifically relates to a semantic domain multiple access transmission interference cancellation algorithm for semantic communication. Background Technology

[0002] Wireless communication technology has developed rapidly in recent decades, supporting mobile communication systems that have evolved from the first generation (1G) to the fifth generation (5G), with ongoing standardization efforts for the sixth generation (6G). Traditional wireless communication systems typically employ a modular design paradigm, where source data undergoes sequential processing stages such as source coding, channel coding, modulation, beamforming, and radio frequency transmission during transmission. Each module is designed independently, with its own clear optimization goals and theoretical support. This design paradigm has reduced system design complexity to some extent, effectively promoting the development of wireless communication technology. However, with technological evolution, some key modules (such as channel coding) have gradually approached the limits of Shannon's theory. For example, polar codes and low-density parity-check (LDPC) codes have approached their theoretical limits in terms of performance, creating bottlenecks in transmission rate and performance improvement for traditional communication systems. Against this backdrop, researchers have begun to focus on higher-level information transmission mechanisms, namely the semantic-level communication paradigm proposed by Shannon and Vivo. Leveraging the advantages of neural networks in semantic feature extraction and representation, modern semantic communication systems are typically deeply integrated with deep learning models to achieve joint mapping and optimization from source to channel symbols through end-to-end learning. This approach holds the promise of overcoming the performance limitations imposed by traditional modular designs, thereby improving communication efficiency and quality of service at the semantic level.

[0003] However, while neural network-based semantic communication systems achieve end-to-end performance optimization, they also bring new challenges to multi-user transmission. In multi-user scenarios, inter-user interference is unavoidable, and in semantic communication systems, this interference mainly manifests as the superposition of semantic features encoded by multiple users. Since these features are generated by neural networks, they possess strong nonlinearity and "black box" characteristics, making it difficult to accurately model inter-user interference using traditional analytical models. Furthermore, existing research shows that, under the same power conditions, the damage to key semantic information caused by such semantic feature interference is often significantly higher than that caused by Gaussian noise, further exacerbating the design difficulty of multi-user semantic communication systems. To address these issues, existing methods typically employ the design of personalized DeepJSCC (Deep Joint Source-Channel Coding) encoding and decoding models for different users to achieve interference differentiation and suppression. However, such methods require model matching and maintenance during actual deployment, resulting in high system overhead, and also face significant complexity and scalability issues when the user scale expands. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies, consider the multi-user transmission problem in semantic communication, explore user differentiation methods and inter-user interference management under the semantic communication joint source-channel coding paradigm, and provide a semantic domain multiple access transmission interference cancellation algorithm. In this invention, all users share a unified DeepJSCC model, and user differentiation and interference management are achieved through a designed mechanism, thereby reducing system complexity while improving scalability and practicality in multi-user scenarios.

[0005] The specific technical solution adopted in this invention is as follows: In a first aspect, the present invention provides a semantic domain multiple access transmission interference cancellation algorithm, as follows: S1: After each user obtains a semantic feature vector through the same semantic encoder, each user independently generates a scrambling mode to distinguish users. Based on the generated scrambling mode, the semantic feature vector is scrambled at the position index. At the receiving end, an inverse scrambling mode is maintained for each user to distinguish target signals and interference signals at the semantic feature level. S2: For the aliased signal at the receiving end, each user is first decoded independently, and noise reduction is performed according to a feature-based noise reduction method based on the codec. Then, multi-level elimination of interference between users is achieved through iteration. The specific feature denoising method based on the encoder / decoder is as follows: First, the corresponding inverse scrambling is performed via the inverse scrambling mode described in S1, converting the structured interference from other users into Gaussian noise. Then, the noisy features containing the Gaussian noise are reconstructed and restored using the denoising capability of the semantic codec itself. Finally, the reconstructed and restored image is re-encoded to achieve nonlinear denoising of the original noisy features.

[0006] Preferably, the semantic encoder is the DeepJSCC model.

[0007] Preferably, S1 is as follows: Considering the uplink communication image transmission task, each user uses the same codec to encode their respective images. The encoding process is represented as follows: ; in, On behalf of users The semantic feature vector to be transmitted N Indicates the length of the semantic feature vector; Indicates user The image to be transmitted H and W These represent the height and width of the image, respectively. This represents an encoder composed of neural networks; For the semantic feature vector to be transmitted, the power limit is as follows: ; in, This indicates a demand for expectation; Subsequently, the semantic feature vectors of each user are randomly scrambled, and the definition is... The scrambled matrix is ​​a matrix whose row and column vectors each contain only one non-zero element (1), with the rest being zero. The scrambled semantic feature vectors are further mapped to the channel symbol sequence. The expression is as follows: ; Wherein, mapping function This is used to convert real-valued vectors into corresponding complex-valued channel symbol sequences. The specific conversion method is as follows: ;in, f k Represents the real number characteristics to be transmitted, [ f k ] i express f k The i 1 element, [ f k ] i+N express f k The i+N Each element represents the imaginary unit; For symbol transmission, consider an uplink multi-user single-input single-output system; the system includes a base station equipped with one receiving antenna and There are multiple users, each equipped with one transmit antenna, transmitting information simultaneously within a resource block on the same frequency. The signal received by the base station is a superposition of the signals transmitted by multiple users, as expressed below: ; in, Indicates user The channel fading coefficient at the receiving end; This represents the received signal at the base station. Represents a Gaussian noise vector; The base station decodes the data for each user; for each user The corresponding received signal is equivalent to the original characteristic signal after additive Gaussian noise, as expressed below: ; in, Represents noise power; Channel fading coefficient The phase, i.e. ; The unit of Gaussian noise follows a standard complex Gaussian distribution. Finally, the DeepJSCC decoder produces the final reconstructed image, expressed as follows: ; in, Represents the decoding function. Parameters representing the decoder, On behalf of users k The reconstructed image.

[0008] Preferably, in S2, the method for reconstructing and restoring noisy features containing Gaussian noise is to use the DeepJSCC model itself to learn and recover the original image from the noisy semantic feature vector.

[0009] Preferably, in step S2, the specific method for achieving multi-level elimination of interference between users through iteration is as follows: S21: Received signal vector Channel status information for each user Orthogonal scrambling matrix for each user DeepJSCC decoder With encoder Maximum number of iterations ;in, K Indicates the number of users; S22: Iteration Rounds Initialize the estimated semantic feature vectors of all users to empty, i.e. ; S23: Repeat S231-S232 until... ; S231: Iterate through all users : Based on the latest estimates of other user characteristics, reconstruct and eliminate interference: ; Recover the target user using the DeepJSCC decoder. Image: ; The reconstructed image is then re-input into the DeepJSCC encoder to obtain the denoised semantic feature vector: ; S232: Update iteration steps ; S24: Output the final reconstructed images generated by each user: .

[0010] In a second aspect, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement a semantic domain multiple access transmission interference cancellation algorithm as described in any of the first aspects.

[0011] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a semantic domain multiple access transmission interference cancellation algorithm as described in any of the first aspects.

[0012] Fourthly, the present invention provides a computer electronic device, including a memory and a processor; The memory is used to store computer programs; The processor is configured to, when executing the computer program, implement a semantic domain multiple access transmission interference cancellation algorithm as described in any of the first aspects.

[0013] Compared with the prior art, the present invention has the following advantages: First, by introducing a user-independent random scrambling scheme, user differentiation is effectively achieved, while inter-user interference is approximately transformed into Gaussian noise. Under this invention, users can use the same semantic codec, effectively solving the dependency problem of user-specific encoding / decoding models in traditional semantic communication. Furthermore, in uplink multiple access transmission scenarios, utilizing the reversibility of random scrambling, a decoding-re-encoding approach is used to denoise features, and the denoised features are pre-removed, effectively improving the signal-to-interference-plus-noise ratio (SNR). Experimental results demonstrate the effectiveness of this invention. Overall, this invention proposes an asymptotic interference cancellation algorithm for uplink semantic communication multiple access transmission, overcoming the shortcomings of previous algorithms in this system. Moreover, the algorithm introduces minimal additional cost. Compared to traditional asymptotic interference cancellation schemes based on power allocation, the proposed scheme has no power constraints, greatly expanding the application scope of asymptotic interference cancellation schemes and effectively exploring interference cancellation mechanisms in the new resource domain of the semantic domain. Attached Figure Description

[0014] Figure 1 This is a topology diagram of the experimental platform in this invention; Figure 2 The diagram shows the transmission PSNR performance of the proposed solution under different numbers of users. Figure 3 The diagram shows the SSIM transmission performance of the proposed solution under different numbers of users. Detailed Implementation

[0015] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. Technical features in various embodiments of the present invention can be combined accordingly without mutual conflict.

[0016] This invention provides a semantic domain multiple access transmission interference cancellation algorithm. The problem it addresses is that when users employ the same DeepJSCC model, the algorithm randomly scrambles the inter-user interference to Gaussianize, then independently decodes each user's data by descrambling, and further maximizes the reconstruction performance of each user through semantic asymptotic interference cancellation. The method includes steps such as user identification based on random scrambling, reversible Gaussianization of inter-user interference, codec-based feature denoising, and asymptotic interference cancellation without power constraints, as detailed below: S1: After each user obtains a semantic feature vector through the same semantic encoder, each user independently generates a scrambling mode to distinguish between users. Based on the generated scrambling mode, the obtained semantic feature vector is scrambled at the position index. At the receiving end, an inverse scrambling mode is maintained for each user to distinguish between target signals and interference signals at the semantic feature level.

[0017] This step involves assigning independent scrambling modes to users, performing statistical analysis and performance impact analysis on the scrambled semantic feature vectors, and observing that the scrambled features are close to Gaussian noise at the same power, thus enabling consistent handling of inter-user interference and channel noise.

[0018] In a preferred embodiment of the present invention, the semantic encoder is the DeepJSCC model. For the semantic feature vectors generated by the DeepJSCC model, users are distinguished by assigning each user an independent scrambling pattern. Furthermore, through statistical analysis and ablation experiments, random scrambling can approximate user-interaction interference into Gaussian noise, effectively solving the problem of difficult user-interaction processing caused by the black-box nature of neural networks.

[0019] As a preferred embodiment of the present invention, step S1 is specifically as follows: 1) Transmission process Considering the uplink communication image transmission task, each user uses the same codec to encode their respective images. The encoding process can be represented as follows: (1); in, On behalf of users The semantic feature vector to be transmitted N Indicates the length of the semantic feature vector; Indicates user The image to be transmitted H and W These represent the height and width of the image, respectively. This represents an encoder composed of neural networks; For the semantic feature vector to be transmitted, the power limit is as follows: (2); in, This indicates a demand for expectation; Subsequently, the semantic feature vectors of each user are randomly scrambled, and the definition is... The scrambled matrix is ​​a matrix whose row and column vectors each contain only one non-zero element (1), with the rest being zero. , For example, there are That is, to achieve the original Random shuffling.

[0020] Next, the scrambled semantic feature vectors are further mapped to the channel symbol sequence. The expression is as follows: (3); Wherein, mapping function This is used to convert real-valued vectors into corresponding complex-valued channel symbol sequences. The specific conversion method is as follows: ;in, f k Represents the real number characteristics to be transmitted, [ f k ] i express f k The i 1 element, [ f k ] i+N express f k The i+N One element, Represents the imaginary unit; For symbol transmission, consider an uplink multi-user single-input single-output (MU-MISO) system. This system consists of a base station equipped with one receive antenna and... There are multiple users, each equipped with one transmit antenna, transmitting information simultaneously within a resource block on the same frequency. The signal received by the base station is a superposition of the signals transmitted by multiple users, as expressed below: (4); in, Indicates user The channel fading coefficient at the receiving end, considering slow fading channels, i.e., during transmission... During the process of one channel symbol Remain unchanged; This represents the received signal at the base station. Represents a Gaussian noise vector; Next, the base station decodes the data for each user; for each user... The corresponding receiver feature vector can be obtained in the following way: (5); in, Represents noise power; Channel fading coefficient The phase, i.e. ; Finally, the DeepJSCC decoder produces the final reconstructed image, expressed as follows: (6); in, Represents the decoding function. Parameters representing the decoder, On behalf of users k The reconstructed image.

[0021] 2) Received signal analysis The transmission scheme of this invention differs from traditional semantic communication systems based on the DeepJSCC paradigm primarily in that it incorporates a random scrambling operation. The purpose of this random scrambling operation is as follows: 1) When multiple users transmit simultaneously on the same frequency in the uplink, user signals will overlap; this random scrambling mode can be used to distinguish user feature vectors; 2) When random scrambling is used to distinguish users, the user end can use the same DeepJSCC encoder, eliminating the need for user-specific codecs, effectively improving flexibility and scalability for the number of users; 3) At the receiving end, by using the user's corresponding descrambling mode, the corresponding semantic features can be effectively extracted. The interference caused by the semantic features of other users is replaced by scrambled semantic features, which have less impact on the recovery of key semantics; 4) By assigning the corresponding scrambling mode, at the base station decoding end, the user's semantic information can be processed and used as prior information to eliminate interference when decoding other user data, thereby effectively suppressing interference between users.

[0022] As shown in formula (3), after the transmitter performs feature encoding, the features are scrambled using a random scrambling matrix. When the base station decodes, formula (5) can be further expressed as follows: (7); in, The unit of Gaussian noise follows a standard complex Gaussian distribution. Represents noise power; As can be seen from the above formula, the inter-user interference term is: When users With users When randomly sampling the shuffled pattern, Equivalent to Random sampling, compared using t-SNE feature analysis. , And from the standard Gaussian noise distribution, we can know There is a significant difference between the distribution of and the Gaussian distribution, and after random shuffling, The distribution almost perfectly overlaps with the Gaussian distribution on the t-SNE characteristic distribution, which verifies the effectiveness of random scrambling in Gaussianizing inter-user interference. Furthermore, analyzing their respective impacts on reconstruction performance also reveals... The impact on reconstruction performance is similar to that of Gaussian noise at the same power, while Its impact on reconstruction performance is far greater than that of Gaussian noise. Based on the above analysis, Therefore, the above formula can be simplified to: (8); As can be seen from the above formula, the received signal can be equivalent to the original feature signal processed by additive Gaussian noise. This form removes the dependence on the user-personalized DeepJSCC model on the one hand, and simplifies the training difficulty of the DeepJSCC model itself on the other hand, that is, it only needs to be trained offline at the base station and then broadcast to each user.

[0023] S2: For the aliased signal at the receiving end, each user is first decoded independently, and noise reduction is performed according to a feature-based noise reduction method based on the codec. Then, multi-level elimination of interference between users is achieved through iteration.

[0024] One encoder-decoder-based feature denoising method uses the feature distribution of the data itself to perform nonlinear denoising on the received noisy features. The proposed scheme involves first reconstructing the received noisy data using a decoder, and then re-encoding it using an encoder to obtain the corresponding denoised features, as detailed below: First, the inverse scrambling pattern obtained from S1 is used to perform corresponding inverse scrambling, converting the structured interference from other users into Gaussian noise. Then, the denoising capability of the semantic codec itself is used to reconstruct and restore the noisy features containing the Gaussian noise. Finally, the reconstructed and restored image is re-encoded to achieve nonlinear denoising of the original noisy features.

[0025] In S2, while reconstructing images for each user, a queue of denoised semantic features is maintained. This queue is dynamically updated during iteration. Based on random scrambling, a bidirectional conversion between features and Gaussian noise can be effectively achieved. When the denoised features approach the original semantic features, the interference caused by these features to other semantic features can be eliminated, improving the reconstruction performance of other users. Simultaneously, the improved reconstruction performance of other users corresponds to their denoised semantic features becoming closer to the original features. Through this iterative update and elimination method, the signal-to-interference-plus-noise ratio (SNR) of all users can be effectively improved, thus significantly enhancing reconstruction performance.

[0026] In S2, based on the reversibility of the random scrambling process, during uplink transmission, each user's data is decoded sequentially according to a specific descrambling pattern. Furthermore, the decoded data yields corresponding feature vectors. When decoding other users' data, the interference corresponding to that user's feature vector is stripped away again through scrambling, thus reducing it and achieving asymptotic interference cancellation in the semantic domain. This cancellation strategy has many advantages over traditional asymptotic interference cancellation strategies in non-orthogonal multiple access, most notably that it does not require constraining user power, explicitly distinguishing between primary and secondary users, and provides good support for multi-user expansion.

[0027] In a preferred embodiment of the present invention, the method for reconstructing and restoring noisy features containing Gaussian noise includes steps to achieve user... Feature estimation requires estimating the original features from the noisy features, i.e. (10); in, For users m The original characteristics, The power representing the useful signal. Represents noise power. Represents standard Gaussian noise; This problem is a denoising problem, which utilizes the denoising capabilities of the DeepJSCC model itself. Specifically, the DeepJSCC model itself learns to recover the original image from noisy semantic features, thus possessing a certain denoising ability. To obtain the semantic features corresponding to the recovered image, a method of decoding followed by encoding can be used.

[0028] As shown in Equation (8), random scrambling can approximate inter-user interference into Gaussian white noise of the same power, which can effectively suppress the structured interference caused by the semantic feature distribution itself. However, it cannot directly improve the signal-to-interference-plus-noise ratio of the target signal at the base station. In the uplink system, the base station needs to reconstruct the feature information from multiple users. During the decoding process, the feature information of different users is known to each other. Taking advantage of this characteristic, inter-user interference from other users can be further reduced.

[0029] Specifically, assuming the base station has already obtained the user's information... semantic feature estimation In the context of users When decoding the features, you can first determine the features from the user based on (7). The semantic feature interference is reduced as follows: (1); As can be seen from formula (1), if the relevant user The feature estimation is accurate enough, that is This comes from the user. Interference items Approaching zero, at which point for the user In this regard, its signal-to-interference-plus-noise ratio can be effectively improved.

[0030] As a preferred embodiment of the present invention, the specific method for achieving multi-level elimination of interference between users through iteration is as follows: S21, Input: Received signal vector Channel status information for each user Orthogonal scrambling matrix for each user DeepJSCC decoder With encoder Maximum number of iterations ;in, K Indicates the number of users; S22, Initialization: Iteration Rounds Initialize the estimated semantic feature vectors of all users to empty, i.e. ; S23: Repeat S231-S232 until... ; S231: Iterate through all users : Multiple access interference cancellation: Based on the latest estimates of other user characteristics (using the Gauss-Seidel update mode), reconstruct and eliminate interference. ; Image Reconstruction: Recovering the Target User Image via DeepJSCC Decoder Image: ; Encoding Denoising and Feature Update: The reconstructed image is re-inputted into the DeepJSCC encoder to obtain the denoised semantic feature vector. ; S232: Number of update iterations: ; S24: Output the final reconstructed images generated by each user: .

[0031] The following experimental results demonstrate the performance of the semantic domain multiple access transmission interference cancellation algorithm of the present invention. The technical effects achieved by the embodiments are as follows: Figure 2 and Figure 3 To compare the semantic domain multiple access transmission interference cancellation algorithm proposed in this invention with existing algorithms, the experimental conditions were as follows: transmission center frequency of 1.9 GHz, sampling rate of 1 MHz, transmit gain of 10 dB, OFDM carrier number of 64, and cyclic prefix length of 16. The experimental platform is as follows: Figure 1 As shown, a scenario simulating simultaneous transmission by four devices is simulated by aliasing up to four randomly scrambled signals at the transmitting end using a single device. At the receiving end, the four signals are extracted from the received signals. Both the transmitting and receiving ends use a single-antenna configuration. The Deep JSCC model is an offline-trained model in a dynamic AWGN channel (i.e., the signal-to-noise ratio dynamically changes from -15dB to 15dB). The model architecture is a hybrid of convolutional networks and visual Transformers. It should be noted that the theoretical conclusions should be similar when using other Deep JSCC models. A comparison is made between interference-free cancellation schemes and the asymptotic interference cancellation scheme presented in this paper. Figure 2 and Figure 3 To compare reconstruction performance under different user numbers, PSNR and SSIM were used as reconstruction metrics. It can be seen that when using interference cancellation mechanisms, although the random scrambling method can approximate inter-user interference into Gaussian noise, it fails to effectively improve the signal-to-interference-plus-noise ratio (SNR), especially when the number of users is large. When the number of iterations is set to 2, the performance degrades significantly. However, after introducing an asymptotic interference cancellation mechanism, the signal-to-interference-plus-noise ratio (SNR) is effectively improved. Compared to the scheme without interference cancellation, the PSNR performance is improved by 1 dB when the number of iterations is set to 2. Furthermore, comparing different iteration numbers reveals that the optimal number of iterations varies for different performance metrics. For the PSNR performance metric, 2 iterations result in the best performance, with performance decreasing as the number of iterations increases. For the SSIM metric, 4 iterations result in the best performance. In practical applications, the number of iterations needs to be specifically set and optimized according to the specific reconstruction metrics.

[0032] In the method of this invention, firstly, by introducing a user-independent random scrambling scheme, user differentiation is effectively achieved, while inter-user interference is approximately transformed into Gaussian noise. Under this scheme, users can use the same semantic codec, effectively solving the problem of dependence on user-specific encoding and decoding models in traditional semantic communication. Furthermore, in the uplink multiple access transmission scenario, the reversibility of random scrambling is utilized to denoise features using a decode-and-encode approach, and the denoised features are pre-removed, effectively improving the signal-to-interference-plus-noise ratio. Experimental results demonstrate the effectiveness of the proposed scheme. Overall, this invention proposes an asymptotic interference cancellation algorithm for uplink semantic communication multiple access transmission, making up for the shortcomings of previous algorithms in this system. Moreover, the algorithm introduces a small additional cost. Compared with the traditional asymptotic interference cancellation scheme based on power allocation, the proposed scheme has no power constraint requirements, greatly expanding the application scope of the asymptotic interference cancellation scheme and effectively exploring interference cancellation mechanisms in the new resource domain of the semantic domain.

[0033] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A semantic domain multiple access transmission interference cancellation algorithm, characterized in that, Specifically as follows: S1: After each user obtains a semantic feature vector through the same semantic encoder, each user independently generates a scrambling mode to distinguish users. Based on the generated scrambling mode, the semantic feature vector is scrambled at the position index. At the receiving end, an inverse scrambling mode is maintained for each user to distinguish target signals and interference signals at the semantic feature level. S2: For the aliased signal at the receiving end, each user is first decoded independently, and noise reduction is performed according to a feature-based noise reduction method based on the codec. Then, multi-level elimination of interference between users is achieved through iteration. The specific feature denoising method based on the encoder / decoder is as follows: First, the corresponding inverse scrambling is performed via the inverse scrambling mode described in S1, converting the structured interference from other users into Gaussian noise. Then, the noisy features containing the Gaussian noise are reconstructed and restored using the denoising capability of the semantic codec itself. Finally, the reconstructed and restored image is re-encoded to achieve nonlinear denoising of the original noisy features.

2. The semantic domain multiple access transmission interference cancellation algorithm according to claim 1, characterized in that, The semantic encoder is the DeepJSCC model.

3. The semantic domain multiple access transmission interference cancellation algorithm according to claim 1, characterized in that, S1 is specifically as follows: Considering the uplink communication image transmission task, each user uses the same codec to encode their respective images. The encoding process is represented as follows: ； in, On behalf of users The semantic feature vector to be transmitted N Indicates the length of the semantic feature vector; Indicates user The image to be transmitted H and W These represent the height and width of the image, respectively. This represents an encoder composed of neural networks; For the semantic feature vector to be transmitted, the power limit is as follows: ； in, This indicates a demand for expectation; Subsequently, the semantic feature vectors of each user are randomly scrambled, and the definition is... The scrambled matrix is ​​a matrix whose row and column vectors each contain only one non-zero element (1), with the rest being zero. The scrambled semantic feature vectors are further mapped to the channel symbol sequence. The expression is as follows: ； Wherein, mapping function This is used to convert real-valued vectors into corresponding complex-valued channel symbol sequences. The specific conversion method is as follows: ;in, f k Represents the real number characteristics to be transmitted, [ f k ] i express f k The i 1 element, [ f k ] i+N express f k The i+N One element, Represents the imaginary unit; For symbol transmission, consider an uplink multi-user single-input single-output system; the system includes a base station equipped with one receiving antenna and There are multiple users, each equipped with one transmit antenna, transmitting information simultaneously within a resource block on the same frequency. The signal received by the base station is a superposition of the signals transmitted by multiple users, as expressed below: ； in, Indicates user The channel fading coefficient at the receiving end; This represents the received signal at the base station. Represents a Gaussian noise vector; The base station decodes the data for each user; for each user The corresponding received signal is equivalent to the original characteristic signal after additive Gaussian noise, as expressed below: ； in, Represents noise power; Channel fading coefficient The phase, i.e. ; The unit of Gaussian noise follows a standard complex Gaussian distribution. Finally, the DeepJSCC decoder produces the final reconstructed image, expressed as follows: ； in, Represents the decoding function. Parameters representing the decoder, On behalf of users k The reconstructed image.

4. The semantic domain multiple access transmission interference cancellation algorithm according to claim 2, characterized in that, In S2, the method for reconstructing and restoring noisy features containing Gaussian noise is to use the DeepJSCC model itself to recover the original image from the noisy semantic feature vector by learning.

5. The semantic domain multiple access transmission interference cancellation algorithm according to claim 1, characterized in that, In step S2, the specific method for achieving multi-level elimination of interference between users through iteration is as follows: S21: Received signal vector Channel status information for each user Orthogonal scrambling matrix for each user DeepJSCC decoder With encoder Maximum number of iterations ;in, K Indicates the number of users; S22: Iteration Rounds Initialize the estimated semantic feature vectors of all users to empty, i.e. ; S23: Repeat S231-S232 until... ; S231: Iterate through all users : Based on the latest estimates of other user characteristics, reconstruct and eliminate interference: ； Recover the target user using the DeepJSCC decoder. Image: ； The reconstructed image is then re-input into the DeepJSCC encoder to obtain the denoised semantic feature vector: ； S232: Update iteration steps ; S24: Output the final reconstructed images generated by each user: .

6. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it can implement a semantic domain multiple access transmission interference cancellation algorithm as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements a semantic domain multiple access transmission interference cancellation algorithm as described in any one of claims 1 to 5.

8. A computer electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to implement, when executing the computer program, a semantic domain multiple access transmission interference cancellation algorithm as described in any one of claims 1 to 5.

Images