A USRP-Based Testbed of Multi-agent Reinforcement Learning for Dynamic Spectrum Anti-Jamming

In this article, we develop a demonstrated multi-agent dynamic spectrum anti-jamming (MDSA) system using LabVIEW software and USRP-based soft defined radio platform. In the system, we design four subsystems, i.e., wireless transmission subsystem, wideband spectrum sensing subsystem, autonomous decision subsystem, and jamming subsystem. A multi-agent collaborative Q-learning (MACQL) algorithm is adopted in the autonomous decision subsystem to avoid the jamming and the co-channel interference between the agents. The dynamic process of the experiment is illustrated by the screenshots of the software. By showing that the data are successfully received and the performance of the MACQL algorithm is better than the sensing-based method, the MDSA system is realized and the effectiveness of the MACQL algorithm is demonstrated.