Identification of Frequency Hopping Modulation Mode Based on Reinforcement Learning Feature Selection

‍ ‍Aiming at the difficulty of selecting the optimal feature set for modulation identification of frequency hopping signal， a frequency hopping modulation identification algorithm based on reinforcement learning feature selection was proposed in this paper. Firstly， the time-frequency energy map of the frequency hopping signal was extracted and the adaptive wiener filtering algorithm was used to remove the noise. Then， the algorithm extracted 20-dimensional feature vectors of the time-frequency energy map， including geometric invariant moments， pseudo-Zernike moments and Rayleigh entropy. The algorithm established a Markov reinforcement learning feature selection model. The improved Q-learning algorithm was used to explore and learn the original feature set， so as to find the optimal feature set and output the recognition result. Simulation results showed that the reinforcement learning feature selection algorithm effectively overcomed the problems of single artificial experience standard and poor robustness， reduced the redundancy and dimension of the original feature set. Under the condition of signal-to-noise ratio of 5 dB， the identification accuracy rate of 10 kinds of frequency hopping signal modulation methods can reach 95%.