Concept Drift Detection for Deep Learning Aided Receivers in Dynamic Channels

Deep learning-aided receivers can operate in settings where classical non-trainable receivers struggle. However, a core challenge associated with deep receivers arises from the dynamic nature of wireless communications. This often results in a train/test distributions mismatch, requiring one to retrain the receiver using newly transmitted samples that capture the new environment. However, frequent retraining is costly and ineffective, while in practice, not every channel variation necessitates adaptation of deep receivers. In this paper, we study concept drift detection for identifying when does a deep receiver no longer match the channel, to allow re-training only when necessary. We adapt existing drift detection mechanisms from the machine learning literature for deep receivers in dynamic channels, and propose a novel soft-output detection mechanism tailored to the communication domain. The provided numerical studies show that even in a rapidly time-varying scenario, concept drift detection dramatically reduces the number of re-training times with little compromise on performance.