Wireless Network Communication Method and Framework System Based on Blockchain and Semantic Communication

Abstract

The present invention proposes a wireless network communication method and framework system based on blockchain and semantic communication, which includes four main parts: a semantic encoder, a wireless channel, semantic verification, and a semantic decoder. The semantic encoder obtains the original image generated by the edge device, encodes the original image and extracts semantic information therefrom; the wireless channel transmits the extracted semantic information to generate noisy semantic information; the state channel is implemented by the smart contract of the blockchain to perform off-chain semantic verification on the noisy semantic information updated to the blockchain; the semantic decoder uses a neural network model to decode the verified semantic information and restore the image. The present invention constructs a semantic sharing mechanism based on the state channel and the task-related information bottleneck method. Considering the resource-limited wireless network, the original image is transmitted through the blockchain semantic framework based on the neural network, which can effectively improve the efficiency of semantic sharing.

Description

Technical Field

[0001] The present invention belongs to the field of semantic communication, and particularly relates to a wireless network communication method and framework system based on blockchain and semantic communication. Background Art

[0002] The fifth generation (5G) introduced a service-based architecture that transforms data-oriented performance metrics into service-oriented performance metrics through low-latency communication. The development from 1G to 5G has focused on how to accurately transmit communication bits, but due to limited physical layer resources, it cannot meet the growing intelligent needs of wireless networks. Semantic communication is a new task-oriented communication architecture that enables edge devices to transmit the semantic information of messages and evaluate the accuracy of the meaning conveyed by the semantic information. Instead of transmitting the original data, the edge device can extract the semantic information from the original data and transmit it to the target recipient. By restoring the data through shared knowledge, communication overhead can be reduced and resource efficiency can be improved. Therefore, semantic communication can be classified as a context-aware and semantic-related paradigm to alleviate the network burden.

[0003] Although semantic communication can effectively process and exchange information, due to the lack of trust among participants, it is difficult for participants to update shared knowledge in a secure manner. In addition, since semantic information is user-centered and user-generated, it is necessary to ensure the data value and protect the semantic information when transmitting among participants. Blockchain is a peer-to-peer network that can protect data from being tampered with in a decentralized, transparent, and secure manner. The integration of blockchain and semantic communication can bring reliable updates of shared knowledge, build trust among participants, and identify the value of semantic information. Blockchain provides characteristics such as decentralization, transparency, and security for the semantic ecosystem. However, since blockchain cannot actively verify the semantic information submitted in the real world, it is difficult to address the challenge of the authenticity of the source of semantic information.

[0004] Currently, there is a lack of integrated solutions for blockchain and semantic communication at home and abroad. Existing work focuses on the integration of blockchain and the semantic web, while paying little attention to blockchain and semantic communication. Semantic blockchains for resource registration, resource discovery, resource selection, and resource payment have been proposed in the prior art, embedding intelligence into pervasive computing. There are also semantic smart contracts based on the semantic network proposed for blockchain to enhance the query of specific terms across multiple ledgers, and a semantic difference transaction mechanism is proposed to minimize information redundancy. In addition, there are some works focusing on semantic communication to provide more effective interactions. For example: federated learning and wave-vector-based autoencoders proposed in the prior art for semantic communication to transmit audio through wireless networks; semantic communication and data adaptation networks for image transmission to convert data into empirical data in a similar form; task-oriented single-model multi-user semantic communication; a deep learning-supported semantic communication system for semantic transmission.

[0005] However, the above-mentioned existing technologies face the following three challenges: 1) how to construct a unified framework based on blockchain and semantic communication, 2) how to verify the semantics of information circulation to prevent spam information from flowing in the blockchain framework, and 3) how to extract task-related semantic information and verify the semantic correctness in an effective manner. Summary of the Invention

[0006] In view of the deficiencies in the prior art, the present invention provides a wireless network communication method and framework system based on blockchain and semantic communication. Among them, a wireless network framework based on blockchain and semantic ecosystem is proposed for the first time. The framework includes a state channel and a semantic sharing mechanism based on the task-related information bottleneck (IB) method to effectively verify the correctness of semantic information.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A wireless network communication method based on blockchain and semantic communication, characterized by including:

[0009] Obtain the original images generated by edge devices, encode the original images and extract semantic information therefrom;

[0010] Transmit the extracted semantic information through a wireless channel to generate noisy semantic information;

[0011] Update the noisy semantic information to the blockchain, and perform off-chain semantic verification through a state channel implemented by the smart contract of the blockchain;

[0012] Use a neural network model to decode the verified semantic information and restore the image.

[0013] To optimize the above technical solutions, the specific measures taken also include:

[0014] Further, the obtaining the original images generated by edge devices, encoding the original images and extracting semantic information therefrom includes:

[0015] Obtain a set of original images r = [r1, r2,..., r N , where N represents the number of samples of the original images;

[0016] Use a neural network model to extract semantic information x = [x1, x2,..., x S from r, where S represents the number of elements mapped to r. Further, the noisy semantic information is: y = c·x + σ, where c is the channel coefficient, x is the semantic information, represents an independent and identically distributed zero-mean Gaussian noise channel, and the noise variance is σ 2, where \(I\) is the identity matrix.

[0017] Furthermore, updating the noisy semantic information to the blockchain and performing off-chain semantic verification through the state channel implemented by the smart contract of the blockchain includes:

[0018] Send the edge device \(d\) i Sign the noisy semantic information \(y\) and pass through the state channel Share the channel message \(\{y, l, m\) j with the target edge device \(d\) r , \(\delta\) i \}, where the state channel \(d\) is the set of edge devices, \(s_0\) is the initial state, \(s\) p is the current state proof, \(s\) c is the current state, \(l\) is the sequence for off-chain semantic sharing to prevent replay attacks, \(m\) r is the cumulative channel information of the Merkle root proof, \(\delta\) i is the signature of the sending edge device and the target edge device;

[0019] The target edge device \(d\) j Signs the message to verify the off-chain state until the cumulative semantic information is updated to the blockchain, and the blockchain records the semantic information in a decentralized manner;

[0020] Verify the signature of the cumulative semantic information through the smart contract and update the current state proof \(s\) and the current state \(s\) p c .

[0021] Furthermore, the goal of the neural network model is to maximize the mutual information of the noisy semantic information while minimizing the coding complexity between the noisy semantic information and the original image. The objective function is:

[0022]

[0023] where \(I(Y, Z)\) represents the mutual information between the noisy semantic information \(Y\) and the classification output \(Z\) of the neural network model, \(I(Y, R)\) represents the mutual information between the noisy semantic information \(Y\) and the original image \(R\), and \(\beta\) is a trade-off hyperparameter;

[0024] The lower bound formula of

[0025]

[0026] where represents the expectation under the condition of \(p(z, r)\), represents under​ The expectation under the condition of D KL () represents the KL divergence, p channel (y|x) represents the probability of output y under the condition of input x in the wireless channel, represents the probability of output x given input r under the condition that the neural network parameters are θ e The probability of output x given input r under the condition that the neural network parameters are θ, p(z, r) represents the joint probability distribution of z and r, q(z|y) is a variational distribution approximating p(z|y), p(z|y) represents the conditional probability distribution of z and y, r(y) represents the probability distribution of y, x is a sample of the semantic information X, y is a sample of the noisy semantic information Y, z is a sample of the classification output Z of the neural network model, and r is a sample of the original image R;

[0027] According to The lower bound formula of, perform Monte Carlo sampling, given a batch of inputs Calculate to obtain:

[0028]

[0029] Among them, N represents the number of samples of the original image, x n represents the nth semantic information, y n represents the nth noisy semantic information, z n represents the nth classification output of the neural network model, r n represents the nth original image.

[0030] In addition, the present invention also proposes a wireless network framework system based on blockchain and semantic communication, which is characterized by including: a semantic encoder, a wireless channel, a state channel, and a semantic decoder;

[0031] The semantic encoder obtains the original image generated by the edge device, encodes the original image and extracts semantic information from it;

[0032] The wireless channel transmits the extracted semantic information to generate noisy semantic information;

[0033] The state channel is implemented by the smart contract of the blockchain, and the state channel performs off-chain semantic verification on the noisy semantic information updated to the blockchain;

[0034] The semantic decoder uses a neural network model to decode the verified semantic information and restore the image.

[0035] Further, the semantic encoder obtains a set of original images r = [r1, r2,..., r N , N represents the number of samples of the original image; use a neural network model to extract semantic information x = [x1, x2,..., xS , where S represents the number of elements mapped to r.

[0036] Furthermore, the noisy semantic information generated by the wireless channel is: y = c·x + σ, where c is the channel coefficient, x is the semantic information, represents an independent and identically distributed zero-mean Gaussian noise channel with noise variance σ 2 , and I is the identity matrix.

[0037] Furthermore, the state channel performs off-chain semantic verification on the noisy semantic information updated to the blockchain, including:

[0038] Send the edge device d i to sign the noisy semantic information y and pass it through the state channel to share the channel message {y, l, m j , δ r} with the target edge device d i , where the state channel d is a set of edge devices, s0 is the initial state, s p is the current state proof, s c is the current state, l is the sequence for off-chain semantic sharing to prevent replay attacks, m r is the cumulative channel information of the Merkle root proof, δ i is the signature of the sending edge device and the target edge device;

[0039] The target edge device d j signs the message to verify the off-chain state until the cumulative semantic information is updated to the blockchain, and the blockchain records the semantic information in a decentralized manner;

[0040] Verify the signature of the cumulative semantic information through the smart contract and update the current state proof s and the current state s p of the state channel c .

[0041] Furthermore, the goal of the neural network model adopted by the semantic decoder is to maximize the mutual information of the noisy semantic information while minimizing the coding complexity between the noisy semantic information and the original image. The objective function is:

[0042]

[0043] where I(Y, Z) represents the mutual information between the noisy semantic information Y and the classification output Z of the neural network model, I(Y, R) represents the mutual information between the noisy semantic information Y and the original image R, and β is a trade-off hyperparameter;

[0044] The lower bound formula of

[0045]

[0046] where denotes the expectation under the condition of p(z, r), denotes the expectation under the condition of and D KL () represents the KL divergence, and p channel (y|x) represents the probability of output y under the condition of input x in the wireless channel, denotes the probability of output x given input r under the condition that the neural network parameters are θ e p(z, r) represents the joint probability distribution of z and r, q(z|y) is a variational distribution approximating p(z|y), p(z|y) represents the conditional probability distribution of z and y, r(y) represents the probability distribution of y, x is a sample of the semantic information X, y is a sample of the noisy semantic information Y, z is a sample of the classification output Z of the neural network model, and r is a sample of the original image R;

[0047] According to the lower bound formula of Monte Carlo sampling is performed, and given the batch input

[0048]

[0049] we calculate and obtain: n where N represents the number of samples of the original image, and x n denotes the nth semantic information, y n denotes the nth noisy semantic information, z n denotes the nth classification output of the neural network model, and r

[0050] The beneficial effects of the present invention are as follows: The present invention constructs a semantic sharing mechanism based on the state channel and the task-related information bottleneck method. Considering the resource-limited wireless network, the original image is transmitted through the blockchain semantic framework based on the neural network, which can effectively improve the efficiency of semantic sharing and prevent garbage information from flowing in the blockchain framework. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 FIG. is a schematic diagram of the wireless network framework system based on blockchain and semantic communication proposed by the present invention.

[0052] Figure 2 FIG. is a schematic diagram of the semantic proof mechanism proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] The present invention will now be described in further detail with reference to the accompanying drawings.

[0054] In one embodiment, the present invention proposes a wireless network framework system based on blockchain and semantic communication as shown in Figure 1 and its semantic proof mechanism is as shown in Figure 2 . The present invention considers a resource - limited wireless network and transmits the original image through a blockchain semantic framework based on neural networks. The framework includes four main parts: a semantic encoder, a wireless channel, semantic verification, and a semantic decoder.

[0055] 1. Semantic encoder: The semantic encoder is used to encode the original image and extract semantic information from the original data. The sending edge device generates the original data and uses the semantic encoder to extract semantic information. The input of the semantic encoder is a set of images r = [r1, r2,..., r N , where r n represents the nth item in N samples. The classification output of r is z = [z1, z2,..., z S , where z S represents the corresponding classification output of r S . The semantic encoder uses a neural network (NN) model with neural network parameters θ e to extract semantic information from r. The output of the semantic encoder is semantic information x = [x1, x2,..., x S , where there are S elements that are mapped to the input r. Therefore, the semantic information can be expressed as where represents the semantic encoder.

[0056] 2. Wireless channel: Since the semantic information is transmitted in a resource - limited time - varying wireless communication environment, it is affected by noise. The semantic information affected by the wireless channel can be given by y = c·x + σ, where y is the noisy semantic information output by the wireless channel encoder, c is the channel coefficient, represents an independent and identically - distributed zero - mean Gaussian noise channel with noise variance σ 2 , and I is the identity matrix. Since the semantic encoder and the wireless channel are parameterized by neural networks, they can be jointly trained in the framework. The noise - damaged semantic information y is transmitted through the noisy wireless channel to the target edge device. The compression ratio (CR) is used to represent the ratio of the image resolution to the original data r and can be defined as CR = log(y) / log(r). Since traditional communication methods must rely on centralized entities to achieve information sharing, the semantic information is underestimated, insecure, and uncontrolled. Therefore, the generated semantic information should be added to the blockchain to maintain decentralization, transparency, and security.

[0057] 3. Semantic Verification: Different from wireless channels, state channels do not have miners verify semantic information before adding it to the blockchain. Instead, complex on-chain semantic verification is transferred to off-chain channels. Semantic sharing between edge devices can be regarded as an off-chain state transition, which can be submitted to the blockchain for consensus after accumulating multiple states. It can also open multiple state channels to achieve efficient semantic sharing given multiple semantic tasks.

[0058] State channels are implemented by smart contracts of the blockchain. Edge devices should deposit and lock tokens in the smart contract and open the state channel by triggering corresponding transactions. This can prevent edge devices from submitting malicious states. Each state channel can provide extensive semantic sharing between edge devices for different semantic tasks. The state channel can be represented as a tuple where d = {d0, d1,..., d n} is a set of edge devices, s0 is the state when the on-chain state opens the channel, s p is the current state proof submitted for off-chain semantic sharing between edge devices, and s c is the current on-chain state verified by miners. Instead of submitting semantic information y immediately after receiving it, edge devices can share semantic information y multiple times and submit the off-chain state proof s p to the smart contract for verification, which can save on-chain verification overhead. Therefore, s p can be represented as s p = {Δs, l, m r , δ ij}, where Δs represents the cumulative state, l represents the sequence for off-chain semantic sharing to prevent replay attacks, m r represents the cumulative state of the Merkle proof, and δ ij represents the signatures of the producer and consumer of the semantic information. The off-chain semantic verification mechanism is detailed as follows.

[0059] The producer d i (i.e., the sending edge device) and the consumer d j (i.e., the target edge device) open the state channel with the initial state s0, the state proof s p , and the current state s c . After successfully opening , the producer d i extracts semantic information x from the source image r and obtains the semantically corrupted information y through the wireless network. The producer d i should sign y and share the channel message {y, l, m } with the consumer d j through r, δ i}。Consumer d j Signs the message to verify the off-chain status. The above process can be repeated multiple times until d i or d j Updates the accumulated semantic information Δs to the blockchain, recording the semantic information in a decentralized manner. The smart contract verifies the signature of the status proof Δs and updates the current status proof s of the status channel p and the current status s c . Since d i or d j may submit an old status for profit, miners can form a committee to judge the accumulated status proofs in the dispute between d i and d j instead of verifying all state transitions to reduce the computational overhead.

[0060] 4. Semantic decoder: The semantic decoder is used to decode semantic information and restore the image. The target edge device receives semantic information from the blockchain and uses the semantic decoder for decoding. The input of the semantic decoder is the transmitted semantic information y. The semantic decoder uses a neural network model parameterized by θ d to reconstruct a distorted version of the original image r The classification output is This operation can be given as where represents the semantic decoder.

[0061] The goal of the proposed mechanism is to accurately convey the desired meaning of the transmitted symbol while maintaining decentralization, which means minimizing the error between r and . The relationship between the semantic encoder and the semantic decoder can be represented by a Markov process Capital letters represent random variables, mapping the semantic information above. z represents the classification output, R represents the input image, X represents the extracted semantic information, Y represents the noisy semantic information, represents the reconstructed image, represents the classification output of the reconstructed image. According to the probabilistic Markov process, it should satisfy:

[0062]

[0063] where, represents the probability of outputting semantic information x for the input image r under the condition that the neural network parameters are θ e , p channel (y|x) represents the probability of outputting the noisy semantic information y under the condition that the input semantic information x is in the wireless channel channel, Indicates that the neural network parameter is θ d Under the condition of inputting noisy semantic information y, the reconstructed image is output probability.

[0064] (a) Objective: The objective of the proposed method is to maximize the mutual information of the semantic information of the channel output Y while minimizing the encoding complexity between the semantic information of the channel output Y and the source data R. Therefore, referring to the Information Bottleneck (IB) principle, the mutual information-based objective function of the proposed method can be expressed as Where I(Y, Z) and I(Y, R) represent the mutual information between random variables, and β is a trade-off hyperparameter. The maximization of the objective function can be used to balance the task-related information and encoding complexity carried by the semantic information, thereby reducing the semantic information that is not related to the task.

[0065] (b) I(Y, Z): First, we use the definition of mutual information to derive the lower bound of I(Y, Z), which can be expressed as Where p(y, z) represents the joint probability distribution of y and z, p(y) represents the probability distribution of y, p(z) represents the probability distribution of z, p(z|y) represents the conditional probability distribution of y and z, y is a sample of noisy semantic information Y, and z is a sample of the classification output Z of the neural network model. Due to the Kullback-Leibler (KL) divergence D with two probability distributions KL is always non-negative, so the following relationship can be obtained:

[0066]

[0067] Among them, q(z|y) is a variational distribution similar to p(z|y), which is different from the semantic encoder. Therefore, the lower bound of I(Y, Z) can be expressed as:

[0068]

[0069] Among them, H(Z) = ∫p(y, z)logp(z)dydz = ∫p(z)logp(z)dz is the entropy of Z. Since H(Z) represents the distribution of Z and is a constant, it can be ignored. In addition, because is related to R, so I(Y,Z) is transformed to:

[0070] I(Y, Z)≥∫p(r)p(y|r)p(z|r)logq(z|y)drdydz,

[0071] Among them, p(y, z) = ∫p(r, y, z)dr = ∫p(r)p(y|r)p(z|r)drdydz, where r is a sample of the original image R.

[0072] (c) I(Y, R): I(Y, R) can also derive an upper bound using the definition of mutual information and can be expressed as The KL divergence is also used to approximate I(Y, R) and can be expressed as:

[0073]

[0074] Among them, r(y) is a variational distribution approximating p(y). Therefore, the upper bound of I(Y, R) can be derived as:

[0075]

[0076] Among them, p(r) represents the probability distribution of r, p(y|r) represents the conditional probability distribution of y and r, r(y) represents the probability distribution of y, p(z|r) represents the conditional probability distribution of z and r, q(z|y) is a variational distribution approximating p(z|y), and p(z|y) represents the conditional probability distribution of z and y.

[0077] (d) Solution: Therefore, based on the above analysis, the lower bound of can be derived as:

[0078]

[0079] can be converted to:

[0080]

[0081] Among them, represents the expectation under the condition that the variable is p(z, r), represents the expectation under the condition that the variable is of.

[0082] Given a mini-batch input z n , r n , y n respectively represent the nth classification output, image input, and noisy semantic information. Monte Carlo sampling and the reparameterization trick are used to derive the empirical estimate:

[0083]

[0084] In another embodiment, the present invention also proposes a communication method using the above wireless network framework system, that is, a wireless network communication method based on blockchain and semantic communication, including the following steps:

[0085] Obtain the original image generated by the edge device, encode the original image and extract semantic information from it;

[0086] Transmit the extracted semantic information through a wireless channel to generate noisy semantic information;

[0087] Update the noisy semantic information to the blockchain and perform off-chain semantic verification through the state channel implemented by the smart contract of the blockchain;

[0088] Use a neural network model to decode the verified semantic information and restore the image.

[0089] The specific processes of each step have been described in the first embodiment, so they will not be elaborated here.

[0090] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should be regarded as within the protection scope of the present invention.

Claims

1. A wireless network communication method based on blockchain and semantic communication, characterized in that Including: Obtain the original image generated by the edge device, encode the original image, and extract semantic information therefrom; Transmit the extracted semantic information through a wireless channel to generate noisy semantic information; Update the noisy semantic information to the blockchain and perform off-chain semantic verification through the state channel implemented by the smart contract of the blockchain; Use a neural network model to decode and restore the image for the semantic information that has passed the verification; the goal of the neural network model is to maximize the mutual information of the noisy semantic information while minimizing the coding complexity between the noisy semantic information and the original image, and the objective function is: Wherein, I(Y,Z) represents the mutual information between the noisy semantic information Y and the classification output Z of the neural network model, I(Y,R) represents the mutual information between the noisy semantic information Y and the original image R, and β is a trade-off hyperparameter; The lower bound formula is as follows: wherein, denotes the expectation under the condition of p(z,r), denotes the expectation under the condition of , D KL () represents the KL divergence, p channel (y|x) represents the probability of output y under the condition of input x in the wireless channel, denotes the probability of outputting x for input r under the condition that the neural network parameter is θ e ; p(z,r) represents the joint probability distribution of z and r, q(z|y) is a variational distribution approximating p(z|y), p(z|y) represents the conditional probability distribution of z and y, r(y) represents the probability distribution of y, x is a sample of semantic information X, y is a sample of noisy semantic information Y, z is a sample of the classification output Z of the neural network model, and r is a sample of the original image R; According to the lower bound formula for Monte Carlo sampling, given a batch of inputs it is calculated that: Among them, N represents the number of samples of the original image, x n represents the nth semantic information, y n represents the nth noisy semantic information, z n represents the classification output of the nth neural network model, r n represents the nth original image.

2. The wireless network communication method based on blockchain and semantic communication according to claim 1, characterized in that: The obtaining the original image generated by the edge device, encoding the original image, and extracting semantic information therefrom includes: Obtain a set of original images \(r = [r_1, r_2, \ldots, r N \), where \(N\) represents the number of samples of the original images; Extract semantic information x = [x1, x2,..., x from r using a neural network model, where S represents the number of elements mapped to r. S ​ 3. The wireless network communication method based on blockchain and semantic communication according to claim 1, characterized in that: The semantic information with noise is: y = c·x + σ, where c is the channel coefficient, x is the semantic information, represents an independent and identically distributed zero-mean Gaussian noise channel with a noise variance of σ 2 , and I is the identity matrix.

4. The wireless network communication method based on blockchain and semantic communication according to claim 1, characterized in that: The updating the noisy semantic information to the blockchain and performing off-chain semantic verification through the state channel implemented by the smart contract of the blockchain includes: Send edge device d i Sign the noisy semantic information y and pass it through the state channel To the target edge device d j Share the channel message {y, l, m r , δ i}, where the state channel d is the set of edge devices, s0 is the initial state, s p is the current state proof, s c is the current state, l is the sequence for off-chain semantic sharing to prevent replay attacks, m r is the cumulative channel information of the Merkle root proof, δ i is the signature of the sending edge device and the target edge device; Target edge device d j Sign messages to verify the off-chain status until the accumulated semantic information is updated to the blockchain, which records the semantic information in a decentralized manner; Verify the signature of the accumulated semantic information through the smart contract and update the current state proof s of the state channel p and the current state s c .

5. A wireless network framework system based on blockchain and semantic communication, characterized in that Including: A semantic encoder, a wireless channel, a state channel, and a semantic decoder; The semantic encoder obtains the original image generated by the edge device, encodes the original image, and extracts semantic information therefrom; The wireless channel transmits the extracted semantic information to generate noisy semantic information; The state channel is implemented by the smart contract of the blockchain, and the state channel performs off-chain semantic verification on the noisy semantic information updated to the blockchain; The semantic decoder uses a neural network model to decode and restore the image for the semantic information that has passed the verification; the goal of the neural network model used by the semantic decoder is to maximize the mutual information of the noisy semantic information while minimizing the coding complexity between the noisy semantic information and the original image, and the objective function is: Wherein, I(Y,Z) represents the mutual information between the noisy semantic information Y and the classification output Z of the neural network model, I(Y,R) represents the mutual information between the noisy semantic information Y and the original image R, and β is a trade-off hyperparameter; The lower bound formula is as follows: Among them, represents the expectation under the condition of p(x,r), represents the expectation under the condition of , and D KL () represents the KL divergence, and p channel (y|x) represents the probability of output y under the condition of input x in the wireless channel, represents the probability of output x for input r under the condition that the neural network parameters are θ e , p(z,r) represents the joint probability distribution of z and r, q(z|y) is a variational distribution approximating p(z|y), p(z|y) represents the conditional probability distribution of z and y, r(y) represents the probability distribution of y, x is a sample of the semantic information X, y is a sample of the noisy semantic information Y, z is a sample of the classification output Z of the neural network model, and r is a sample of the original image R; According to the lower bound formula for Monte Carlo sampling, given a batch input it is calculated that: Among them, N represents the number of samples of the original image, x n represents the nth semantic information, y n represents the nth noisy semantic information, z n represents the classification output of the nth neural network model, r n represents the nth original image.

6. The wireless network framework system based on blockchain and semantic communication according to claim 5, characterized in that: The semantic encoder obtains a set of original images \(r = [r_1, r_2, \ldots, r N \), where \(N\) represents the number of samples of the original images; the neural network model is used to extract semantic information \(x = [x_1, x_2, \ldots, x S \) from \(r\), where \(S\) represents the number of elements mapped to \(r\).

7. The wireless network framework system based on blockchain and semantic communication according to claim 5, characterized in that: The semantic information with noise generated by the wireless channel is: y = c·x + σ, where c is the channel coefficient, x is the semantic information, and σ ~ represents an independent and identically distributed zero-mean Gaussian noise channel with a noise variance of σ 2 , and I is the identity matrix.

8. The wireless network framework system based on blockchain and semantic communication according to claim 5, characterized in that: The state channel performs off-chain semantic verification on the noisy semantic information updated to the blockchain, including: Send edge device d i Sign the noisy semantic information y and pass it through the state channel to the target edge device d j Share the channel message {y, l, m r , δ i}, where the state channel d is the set of edge devices, s0 is the initial state, s p is the current state proof, s c is the current state, l is the sequence for off-chain semantic sharing to prevent replay attacks, m r is the cumulative channel information of the Merkle root proof, δ i is the signature of the sending edge device and the target edge device; Target edge device d j Sign the message to verify the off-chain status until the accumulated semantic information is updated to the blockchain, which records the semantic information in a decentralized manner; Verify the signature of the accumulated semantic information through the smart contract and update the state channel of the current state proof s p and the current state s c .

Images