Semantic Channel Equalization Strategies for Deep Joint Source-Channel Coding

TL;DR

DeepJSCC enables end-to-end semantic transmission but assumes matched latent spaces, an assumption violated in multi-vendor deployments leading to semantic noise. The authors propose semantic channel equalization with three aligners—linear, neural, and Parseval-frame—that use semantic pilots to align TX/RX latent spaces under both semantic and physical impairments. Through CIFAR-10 image reconstruction experiments over AWGN and fading channels, they quantify trade-offs in complexity, data efficiency, and fidelity, showing that convolutional, neural-based aligners reach high quality with few pilots, while Parseval-frame methods offer robust zero-shot deployment. The work provides practical deployment guidelines for heterogeneous AI-native wireless networks and lays groundwork for extending semantic equalization to broader tasks and channel conditions.

Abstract

Deep joint source-channel coding (DeepJSCC) has emerged as a powerful paradigm for end-to-end semantic communications, jointly learning to compress and protect task-relevant features over noisy channels. However, existing DeepJSCC schemes assume a shared latent space at transmitter (TX) and receiver (RX) - an assumption that fails in multi-vendor deployments where encoders and decoders cannot be co-trained. This mismatch introduces "semantic noise", degrading reconstruction quality and downstream task performance. In this paper, we systematize and evaluate methods for semantic channel equalization for DeepJSCC, introducing an additional processing stage that aligns heterogeneous latent spaces under both physical and semantic impairments. We investigate three classes of aligners: (i) linear maps, which admit closed-form solutions; (ii) lightweight neural networks, offering greater expressiveness; and (iii) a Parseval-frame equalizer, which operates in zero-shot mode without the need for training. Through extensive experiments on image reconstruction over AWGN and fading channels, we quantify trade-offs among complexity, data efficiency, and fidelity, providing guidelines for deploying DeepJSCC in heterogeneous AI-native wireless networks.

Figures (7)

• Figure 1: Scheme of the proposed DeepJSCC model with semantic equalization.
• Figure 2: Reconstruction of a Kodak patch at DeepJSCC-trained SNRs {$-10$, $0$, $10$, $20$, $30$} dB, evaluated only under the presence of AWGN at the same SNR. Results are shown for no mismatch, mismatched without equalizer, and results with Linear, MLP, 2-layer CNN, 1-layer CNN, and PFE equalizers (all with $|{\mathcal{P}}|= 10\,000$ pilots).
• Figure 3: PSNR versus $|{\mathcal{P}}|$, with both ${\text{SNR}_\text{Align}}$ and ${\text{SNR}_\text{DeepJSCC}}$ fixed at $-10$ dB, considering only the AWGN component.
• Figure 4: PSNR versus $|{\mathcal{P}}|$, with both ${\text{SNR}_\text{Align}}$ and ${\text{SNR}_\text{DeepJSCC}}$ fixed at $20$ dB, considering only the AWGN component.
• Figure 5: PSNR versus ${\text{SNR}_\text{Align}}$, with ${\text{SNR}_\text{DeepJSCC}}\!=\!-10\,$dB and $|{\mathcal{P}}|\!=\!10\,000$.