Adversarial poisoning attacks for online deep receivers

Recently, the design of wireless receivers using deep neural networks (DNNs), known as deep receivers, has attracted extensive attention for ensuring reliable communication in complex channel environments. To adapt quickly to dynamic channels, online learning has been adopted to update the weights of deep receivers with over-the-air data (e.g., pilots). However, the fragility of neural models and the openness of wireless channels expose these systems to malicious attacks. To this end, understanding these attack methods is essential for robust receiver design. In this paper, we propose an adversarial poisoning attack method for online receivers. Without knowledge of the attack target, adversarial perturbations are injected to the pilots, poisoning the online deep receiver and impairing its ability to adapt to dynamic channels and nonlinear effects. The attack specifically targets Deep Soft Interference Cancellation (DeepSIC) [1] with online meta-learning, a classic model-driven deep receiver integrating wireless domain knowledge, which adapts to time-varying channels efficiently with few pilots and performs optimally in MIMO systems. Simulations on linear, nonlinear, static and COST 2100 channels verify the attack efficacy: it drastically degrades the performance of online receivers in fast-changing scenarios and remains effective under CSI estimation errors, synchronization drift, and baseline defense mechanisms.