Deep Reinforcement Learning-Based Adaptive Bandpass Filter With Reconfigurable Frequency and Bandwidth

Abstract

This letter proposes a novel adaptive bandpass filter with reconfigurable center frequency and bandwidth (BW) based on a deep Q-network (DQN). To the best of our knowledge, this is the first implementation of a DQN-based tunable filter capable of simultaneously adjusting both frequency and BW. To improve filter-control efficiency, a method using only positive voltage actions is introduced. Additionally, a reward equation is formulated to enable the DQN to effectively control both the frequency and the BW. The proposed approach is validated through the design of a two-pole adaptive filter.

Cheolsoo Park

Professor

His research interests include machine learning, adaptive signal processing, computational neuroscience, and wearable technology.