Two-Stage Message-Driven Frequency Hopping OFDM in Federated Learning for Autonomous Vehicles

Federated learning (FL) has received widespread attention in the field of autonomous driving due to its advantages in preventing privacy leakage. However, there are still hidden dangers in privacy leakage and data transmission security. This paper proposes a two-stage message-driven frequency hopping orthogonal-frequency division multiplexing (TMDFH-OFDM) strategy and make it serve the uplink transmission process of FL for autonomous vehicles (AVs). In order to avoid inaccurate training results which caused by targeted jamming in the communication channel, the method of using scrambled transmission data to select communication frequency twice is proposed. The aperiodic nature of the transmitted data makes the AVs communication frequency unpredictable, so the attacker cannot send effective malicious jamming. In order to keep the privacy information from being leaked, the AVs’ transmission data is divided into three parts, two of which are hidden into the frequency hopping pattern. This strategy makes it impossible for attackers to obtain complete user data from the channel. Theoretical analysis and simulation results show the proposed method can effectively improve the concealment and security of vehicles’ transmitted signals in the process of FL and possesses significant research value.