Abstract: We propose a novel modulation classifier based on the Markov chain Monte Carlo (MCMC) methods for amplitude-phase modulated signals over the frequency-selective fading channel with multiple unknown parameters such as noise power,carrier frequency and phase offset. The framework for an optimal maximum posterier (MAP) classifier is developed. MCMC methods are employed to generate ergodic random samples from the posterior conditional distributions of the unknown parameters and transmitted symbols, which are derived from the prior distributions of the received signals. Since a close-form expression of the integration of high-dimensional function in the posterior distribution of the modulation can rarely be obtained in the proposed classifier, the Monte Carlo integration is then used to approximate it with these samples. The convergence property and the robust performance of the proposed classifier are then verified via extensive simulations and comparisons with existing approaches.

Key words: modulation classification, Bayesian methods, MCMC, Gibbs sampler