Junli Fang,João F. C. Mota,Baoshan Lu,Weicheng Zhang,Xuemin Hong
Abstract:The joint source-channel coding (JSCC) framework leverages deep learning to learn from data the best codes for source and channel coding. When the output signal, rather than being binary, is directly mapped onto the IQ domain (complex-valued), we call the resulting framework joint source coding and modulation (JSCM). We consider a JSCM scenario and show the existence of a strict tradeoff between channel rate, distortion, perception, and classification accuracy, a tradeoff that we name RDPC. We then propose two image compression methods to navigate that tradeoff: the RDPCO algorithm which, under simple assumptions, directly solves the optimization problem characterizing the tradeoff, and an algorithm based on an inverse-domain generative adversarial network (ID-GAN), which is more general and achieves extreme compression. Simulation results corroborate the theoretical findings, showing that both algorithms exhibit the RDPC tradeoff. They also demonstrate that the proposed ID-GAN algorithm effectively balances image distortion, perception, and classification accuracy, and significantly outperforms traditional separation-based methods and recent deep JSCM architectures in terms of one or more of these metrics.
Machine Learning,Artificial Intelligence,Computer Vision and Pattern Recognition,Information Theory,Probability
What problem does this paper attempt to address?
### Problems the paper attempts to solve
This paper attempts to solve the strict trade - off problems among channel rate (Rate), distortion (Distortion), perception quality (Perception) and classification accuracy (Classification) in the joint source coding and modulation (JSCM) framework. Specifically, the paper proposes the following points:
1. **Research background**:
- Traditional communication systems follow Shannon's source - channel coding theorem, that is, source coding and channel coding can be designed separately without losing optimality. However, in extreme scenarios (such as highly bandwidth - limited or rapidly changing channels), this modular design becomes no longer applicable.
- The joint source coding and modulation (JSCM) framework directly maps signals to the IQ domain (complex domain), thus performing joint design among source coding, channel coding and modulation schemes.
2. **Problem definition**:
- In the JSCM framework, the paper introduces the rate - distortion - perception - classification (RDPC) function, which describes the problem of minimizing the channel rate under certain constraint conditions.
- The specific optimization problem can be expressed as:
\[
R(D, P, C)=\min_{p_Y|X, p_{\hat{X}|\hat{Y}}, \Sigma} \sum_{i = 1}^m \log \left(1+\frac{1}{\Sigma_{ii}}\right)
\]
with the constraint conditions:
\[
\begin{aligned}
&\mathbb{E}[\Delta(X, \hat{X})]\leq D \\
&d(p_X, p_{\hat{X}})\leq P \\
&\mathbb{E}[\epsilon_{c_0}(X, \hat{X})]\leq C
\end{aligned}
\]
where \(D\) is the distortion constraint, \(P\) is the perception quality constraint, and \(C\) is the classification error constraint.
3. **Research objectives**:
- The goal of the paper is to design the encoder - decoder pair \((p_Y|X, p_{\hat{X}|\hat{Y}})\) and the noise matrix \(\Sigma\) so that the channel rate \(R(D, P, C)\) is minimized under the above - mentioned constraint conditions.
- Through analysis and experimental verification, show how different values of \(D\), \(P\), and \(C\) affect the achievable channel rate \(R(D, P, C)\).
4. **Main contributions**:
- Prove that in the JSCM framework, there is a strict trade - off relationship among channel rate, distortion, perception quality and classification error.
- Propose two algorithms to solve the JSCM trade - off problem: one is the simple RDPCO algorithm, which is suitable for restrictive assumptions; the other is an algorithm based on the inverse domain generative adversarial network (ID - GAN), which can transmit images at extreme compression rates while maintaining semantic information, perception quality and reconstruction fidelity.
- Through experimental results, show the similarity in performance between these two algorithms, and the ID - GAN algorithm is superior to traditional methods and recent deep - learning methods in terms of perception quality and classification accuracy.
### Summary
By introducing the RDPC function, this paper systematically studies the trade - off problems among channel rate, distortion, perception quality and classification error in the joint source coding and modulation (JSCM) framework. The paper not only theoretically proves the strict trade - off relationship among these indicators, but also proposes two effective algorithms to solve this problem and verifies the effectiveness of these algorithms through experiments.