Communication noise reduction and fidelity method based on semantics

Abstract

The invention relates to a semantic-based communication noise reduction and fidelity method, which comprises the following steps: inputting an original communication statement into a trained FLOW model for vectorization to obtain vector representation of the original communication statement, and then inputting the vector representation V into a special encoder for training to obtain high-quality vector representation. And finally, inputting the semantic vector into a decoder to recover the semantic vector into a high-quality semantic vector. According to the invention, invalid information in the communication process can be obviously reduced, effective information is reserved, and the transmission efficiency is improved.

Description

Technical Field

[0001] The present application relates to the field of communication engineering, and in particular to a semantic-based communication noise reduction and fidelity preservation method. Background Art

[0002] Traditional communication methods mainly rely on the accuracy and efficiency of transmitted signals. With the rapid development of intelligent technology, semantic-based communication methods have received widespread attention. Semantic communication focuses on the meaning and content of data, which can ensure the effectiveness and efficiency of information and has important applications in bandwidth-constrained environments.

[0003] Noise data is inevitably introduced into the traditional semantic-based communication process. Noise will cause the transmission efficiency to decrease, resulting in unnecessary information being mixed in the transmitted information content, which seriously affects the accuracy and fidelity of the semantic communication content. Summary of the invention

[0004] The present invention provides a communication noise reduction and fidelity preservation method based on semantics, which can significantly reduce invalid information in the communication process, retain valid information, and improve transmission efficiency.

[0005] The technical solution of this application is as follows: A communication noise reduction and fidelity preservation method based on semantics comprises the following steps: S01) Pre-train the large language model FLOW, input the original communication sentence into the trained FLOW model for vectorization, and obtain the vector representation of the original communication sentence V ; S02) represents the vector V Input into a dedicated encoder for training to obtain high-quality vector representation V T In the dedicated encoder, the noise effect in the channel transmission is simulated by gradually increasing the noise, and the vector representation is gradually V The formula for adding noise is as follows: ; In the formula, is the time step t The noise proportionality coefficient, is the noise sampled from the normal distribution N(0,I), where I is the identity matrix; The analog channel is realized by adding Gaussian noise or Rayleigh noise. Suppose the added noise is N , then the output vector of the analog channel is: ; In the formula, represents the output vector of the analog channel; S03) Input to the decoder to restore it to V , the decoder gradually removes the noise during the recovery process N , the formula for this process is: In the formula, represents the process vector during the recovery process; Indicates that based on the current vector and the conditional vector The calculated attention weights; represents the dynamic learning rate; represents the noise correction coefficient; C is a vector containing keywords about the original semantic vector; represents noise randomly sampled from the normal distribution N(0,I).

[0006] Further, in step S02), Step by time t The increase gradually decreases.

[0007] Due to the adoption of the above technical solution, the beneficial effects of this application are as follows: Steps S01) and S02) are forward processes, and S03) is a reverse process: 1. In the reverse process of introducing dynamic parameter adjustment, conditional vectors and attention mechanism into the diffusion model, the present application significantly enhances the performance of the semantic communication system in a noisy environment. The above-mentioned improvements enable the model to adapt to the noise level at different stages more flexibly and pay more attention to the key semantic information retained during the denoising process.

[0008] 2. Dynamic learning rate and noise correction coefficient are introduced in this application, which provides the model with the ability to dynamically adjust its behavior according to the actual situation of noise during the entire decoding process, thereby optimizing the denoising effect.

[0009] 3. The application of conditional vectors and attention mechanisms in this application further enhances the model's sensitivity and ability to retain important semantic information in the input data, ensuring high fidelity when restoring the original semantic vectors.

[0010] 4. On the whole, this application not only improves the accuracy and reliability of communication content, but also enhances the adaptability and robustness of the system in complex communication environments, and significantly improves the practical value and efficiency of semantic communication technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flowchart of a semantic-based communication noise reduction and fidelity method provided in this application; Figure 2 This is a diagram of the algorithm framework for communication noise reduction and fidelity preservation in this application. DETAILED DESCRIPTION

[0012] Based on the background technology, the traditional semantic-based communication process inevitably introduces noise data, which will cause the transmission efficiency to decrease, resulting in unnecessary information being mixed in the transmitted information content, seriously affecting the accuracy and fidelity of the semantic communication content. Figure 1 and attached Figure 2 As shown, the present application provides a semantic-based communication noise reduction and fidelity method, comprising the following steps: S01) Pre-train the large language model FLOW, input the original communication sentence into the trained FLOW model for vectorization, and obtain the vector representation of the original communication sentence V .

[0013] Through deep bidirectional training, the FLOW model can capture rich contextual information, thereby generating high-quality semantic vector representations for the input original communication sentences. Let X be the input original text, then: V=FLOW ( X ); V It is the vector representation after FLOW conversion.

[0014] S02) represents the vector V Input into a dedicated encoder for training to obtain high-quality vector representation V T In the dedicated encoder, the noise effect in the channel transmission is simulated by gradually increasing the noise, and the vector representation is gradually V The formula for adding noise is as follows: ; In the formula, is the time step t The noise proportionality coefficient, is the noise sampled from the normal distribution N(0,I), where I is the identity matrix.

[0015] In this embodiment, the network is trained by adding and gradually removing noise, with the aim of improving the robustness of the training network to noise. In the forward process, the vector representation of FLOW output is gradually adjusted by simulating the influence of noise in channel transmission. V Increase the noise. Things that need to be explained, It gradually decreases with the increase of time step t, which means that the noise level gradually increases.

[0016] The analog channel is realized by adding Gaussian noise or Rayleigh noise. Suppose the added noise is N , then the output vector of the analog channel is: ; In the formula, Represents the output vector of the analog channel.

[0017] The output vector of the encoder The simulated communication channel is the non-trainable part of the channel. After adding Gaussian noise or Rayleigh noise through a computer, it is used to simulate the real communication environment.

[0018] S03) Input to the decoder to restore it to V , the decoder gradually removes the noise during the recovery process N , the formula for this process is: In the formula, represents the process vector during the recovery process; Indicates that based on the current vector and the conditional vector The calculated attention weights; represents the dynamic learning rate; represents the noise correction coefficient; C is a vector containing keywords about the original semantic vector; represents noise randomly sampled from the normal distribution N(0,I).

[0019] In step S03), the task of the decoder is to reverse the noise added during the encoding process and restore high-quality semantic vectors. After iterative training according to steps S01) to S03), the final high-quality semantic vector can be obtained. In this application, the dynamic learning rate changes automatically to adjust the parameter amplitude during the training process. By changing the dynamic learning rate, it adapts to different data distributions and model complexities. The noise correction coefficient is derived during the entire algorithm training process. The dynamic learning rate and noise correction coefficient provide the model with the ability to dynamically adjust its behavior according to the actual noise conditions during the entire decoding process, thereby optimizing the denoising effect.

[0020] In the reverse process of introducing dynamic parameter adjustment, conditional vectors and attention mechanism into the diffusion model, the present application significantly enhances the performance of the semantic communication system in a noisy environment. The above-mentioned improvements enable the model to adapt to the noise level at different stages more flexibly and pay more attention to the key semantic information retained during the denoising process.

[0021] In step S03), the application of conditional vectors and attention mechanisms further enhances the model's sensitivity and ability to retain important semantic information in the input data, ensuring high fidelity when restoring the original semantic vectors.

[0022] On the whole, this application not only improves the accuracy and reliability of communication content, but also enhances the adaptability and robustness of the system in complex communication environments, significantly improving the practical value and efficiency of semantic communication technology.

[0023] Anything not described in this application can be achieved by adopting or drawing on existing technologies. The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A semantic-based communication noise reduction and fidelity method, characterized in that: The following steps are involved: S01) Pre-train the large language model FLOW, input the original communication sentence into the trained FLOW model for vectorization, and obtain the vector representation of the original communication sentence V ; S02) represents the vector V Input into a dedicated encoder for training to obtain high-quality vector representation V T In the dedicated encoder, the noise effect in the channel transmission is simulated by gradually increasing the noise, and the vector representation is gradually V The formula for adding noise is as follows: ； In the formula, is the time step t The noise proportionality coefficient, is the noise sampled from the normal distribution N(0,I), where I is the identity matrix; The analog channel is realized by adding Gaussian noise or Rayleigh noise. Suppose the added noise is N , then the output vector of the analog channel is: ； In the formula, represents the output vector of the analog channel; S03) Input to the decoder to restore it to V , the decoder gradually removes the noise during the recovery process N , the formula for this process is: In the formula, represents the process vector during the recovery process; Indicates that based on the current vector and the conditional vector The calculated attention weights; represents the dynamic learning rate; represents the noise correction coefficient; C is a vector containing keywords about the original semantic vector; represents noise randomly sampled from the normal distribution N(0,I).

2. The method for communication noise reduction and fidelity preservation based on semantics according to claim 1, characterized in that: In step S02), Step by time t The increase gradually decreases.

Images