Trustworthy Semantic-Enabled 6G Communication: A Task-oriented and Privacy-preserving Perspective

Shuaishuai Guo,Anbang Zhang,Yanhu Wang,Chenyuan Feng,Tony Q. S. Quek

2024-08-08

Abstract:Trustworthy task-oriented semantic communication (ToSC) emerges as an innovative approach in the 6G landscape, characterized by the transmission of only vital information that is directly pertinent to a specific task. While ToSC offers an efficient mode of communication, it concurrently raises concerns regarding privacy, as sophisticated adversaries might possess the capability to reconstruct the original data from the transmitted features. This article provides an in-depth analysis of privacy-preserving strategies specifically designed for ToSC relying on deep neural network-based joint source and channel coding (DeepJSCC). The study encompasses a detailed comparative assessment of trustworthy feature perturbation methods such as differential privacy and encryption, alongside intrinsic security incorporation approaches like adversarial learning to train the JSCC and learning-based vector quantization (LBVQ). This comparative analysis underscores the integration of advanced explainable learning algorithms into communication systems, positing a new benchmark for privacy standards in the forthcoming 6G era.

Signal Processing,Systems and Control

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is how to ensure the privacy of data transmission while guaranteeing the efficiency of task - oriented semantic communication (ToSC) in 6G communication. Specifically, the paper focuses on the following points:

1. **Trade - off between efficient communication and privacy protection**:
- ToSC improves communication efficiency and reduces latency by transmitting only the key information directly related to a specific task. However, this selective transmission may cause privacy issues because the transmitted feature information may be reconstructed by malicious adversaries, thus leaking sensitive data.

2. **Selection and evaluation of privacy protection methods**:
- The paper explores multiple privacy protection strategies, including Differential Privacy, Encryption, Adversarial Learning, and Learning - based Vector Quantization (LBVQ). These methods aim to prevent potential privacy leakage while minimizing the impact on the performance of the communication system.

3. **Challenges in dealing with complex attacks**:
- With the improvement of machine learning and data analysis capabilities, malicious adversaries can extract valuable information from a small amount of data. Therefore, it is necessary to develop privacy protection mechanisms that can resist advanced attacks.

### Specific research content

- **Differential Privacy**: Protect privacy by adding noise to the DeepJSCC output. This method can provide strict privacy protection, but too much noise will affect the performance of task inference.

\[
L_{\text{DP}}=\frac{1}{N}\sum_{i = 1}^{N}\|x_i-\tilde{x}_i+\epsilon\|_2^2
\]
where \(\epsilon\) is the added Laplace noise.

- **Encryption**: Encrypt the output of DeepJSCC to ensure the security of data during transmission. However, the encryption and decryption processes will significantly increase the computational overhead and are vulnerable to attacks without channel coding.

- **Adversarial Learning**: By introducing adversarial loss, make the model more robust during the training stage and be able to effectively resist unknown or intentional attacks. Although this method requires continuous updating and improvement of the model when facing new attacks.

- **LBVQ**: Map continuous features into discrete codebooks, reducing the possibility of accurately reconstructing the original input from the latent representation, thereby enhancing privacy protection.

### Experimental evaluation

The paper evaluates the performance and privacy protection degree of different privacy protection schemes through two experiments, image classification and face recognition. The experimental results show that:

- **Classification accuracy**: Under different signal - to - noise ratio (SNR) conditions, the classification accuracy of each scheme varies. For example, DeepJSCC - DP performs well under a lower privacy budget, while DeepJSCC - LBVQ performs excellently under high SNR conditions.

- **PSNR value of the attacker**: A lower PSNR value means better privacy protection because it means that the quality of the image reconstructed by the attacker is poor and it is difficult to obtain useful information from it.

In summary, this paper is committed to exploring and evaluating multiple privacy protection methods in the context of 6G communication to achieve efficient task - oriented semantic communication while ensuring the security and privacy of user data.