Adaptive Wi-Fi channel switching using machine learning and software-defined radio in DFS environments

Dynamic Frequency Selection (DFS) plays a crucial role in ensuring coexistence between Wi-Fi systems and radar operations in the 5 GHz band. This study presents the development of an adaptive DFS system based on Software-Defined Radio (SDR) using HackRF, integrated with radar pulse simulation (via GNU Radio) and signal classification powered by Machine Learning (1D CNN). The system successfully detected FCC Type 4 radar with a classification accuracy of 96.82%, 98.88% precision, and 92.59% recall for the radar class. Experimental results demonstrate that the system can detect radar interference on active DFS channels (e.g., Channel 52), vacate the channel in an average time of 4.2 seconds, switch to a new channel (e.g., Channel 157), and comply with the 30-minute Non- Occupancy Period (NOP) requirement. A VB.NET-based interface enables real-time monitoring and displays RF spectrum activity and DFS event status based on system logs retrieved from a router.