Machine Unlearning for Equalizer Classifiers in 6G Wireless Communication Systems

This letter proposes a new direction for wireless signal processing by applying machine unlearning (MU) to neural-network-based equalizers in 6G receivers. MU enables a trained equalizer to mitigate the influence of corrupted training samples without full model retraining. A new neural equalizer is designed to better capture nonlinear radio-frequency impairments, inter-symbol interference, and channel-memory effects characteristic of 6G systems. We also develop MU-enabled equalization schemes and evaluate their performance under various data corruption scenarios. Simulation results show that MU effectively restores classification accuracy while reducing computational cost, making it a practical and efficient approach for preserving equalizer robustness in dynamic and imperfect 6G environments.