Prompt-Assisted Semantic Interference Cancellation on Moderate Interference Channels

TL;DR

This work tackles interference management in moderate interference channels by rethinking transmission through semantic communication. It introduces DeepPASIC, a deep-learning framework that partitions each user's semantic content into a common part and a private prompt, using the private part to guide a GAN-based separator that cancels interference at the semantic level. The approach combines a trainable semantic encoder/decoder with a GAN separator and a three-stage training strategy, achieving notable PSNR gains over orthogonal, TIN, and SIC baselines for image transmission under moderate interference. While effective in moderate regimes, the method shows performance degradation as interference strengthens, highlighting avenues for future improvements in robust semantic interference cancellation. This framework has potential practical impact on wireless systems requiring reliable semantic recovery under interference while preserving end-to-end learning benefits.

Abstract

The performance of conventional interference management strategies degrades when interference power is comparable to signal power. We consider a new perspective on interference management using semantic communication. Specifically, a multi-user semantic communication system is considered on moderate interference channels (ICs), for which a novel framework of deep learning-based prompt-assisted semantic interference cancellation (DeepPASIC) is proposed. Each transmitted signal is partitioned into common and private parts. The common parts of different users are transmitted simultaneously in a shared medium, resulting in superposition. The private part, on the other hand, serves as a prompt to assist in canceling the interference suffered by the common part at the semantic level. Simulation results demonstrate that the proposed DeepPASIC outperforms conventional interference management strategies under moderate interference conditions.

Figures (5)

• Figure 1: Two-stage transmission strategy.
• Figure 2: The framework and data pipeline of proposed DeepPASIC at a typical user $k$.
• Figure 3: Some visualized results for two-user DeepPASIC under moderate IC, where $|h_{ij}|=1$, $C=12$, and $P=4$. (a), (b) represent original images encoded and transmitted by the two users, respectively. (c) displays the reconstruction results of both users without interference separation, which are identical due to the symmetric channel state. (d), (e) show the reconstruction results of received signals after applying the proposed DeepPASIC.
• Figure 4: PSNR comparisons between different interference management methods with a similar symbol rate.
• Figure 5: PSNR comparisons between DeepPASIC ($E=16$) and the orthogonal bit transmission under varying $|h|$.