Anti-Jamming Communication Using Imitation Learning.

Area of Science:

• Wireless Communication
• Cybersecurity
• Machine Learning

Background:

• Wireless communication systems face significant reliability threats from malicious jamming.
• Existing anti-jamming methods like Reinforcement Learning (RL) and Deep Q Network (DQN) have limitations in utilizing historical data.
• Current strategies often focus on immediate state changes, neglecting valuable historical experience.

Purpose of the Study:

• To propose a novel anti-jamming communication scheme using imitation learning to enhance reliability under malicious jamming.
• To address the limitations of existing methods in leveraging historical data for anti-jamming decisions.
• To develop an efficient and sequential decision-making process for anti-jamming in dynamic environments.

Main Methods:

• Proposed an Expert Trajectory Generation Algorithm to derive expert strategies from historical samples.
• Employed an imitation learning neural network to train a user strategy by mimicking the expert strategy.
• Implemented a functional user strategy for sequential and efficient anti-jamming decisions.

Main Results:

• The proposed imitation learning method demonstrated superior performance compared to RL and DQN-based anti-jamming techniques.
• Effectively solved continuous-state spectrum anti-jamming problems without the 'curse of dimensionality'.
• Showcased enhanced robustness against channel fading, noise, and variations in jamming patterns.

Conclusions:

• Imitation learning offers a promising approach for developing resilient anti-jamming communication systems.
• The proposed method provides a more effective way to utilize historical data for adaptive anti-jamming.
• This technique enhances wireless communication reliability and security in the presence of sophisticated jamming threats.