Likelihood‐based Classifiers

Likelihood-based (LB) modulation classifiers are by far the most popular modulation classification approaches. This chapter presents the maximum likelihood (ML) classifier. It discusses the alternatives of average likelihood ratio test (ALRT), generalized likelihood ratio test (GLRT) and hybrid likelihood ratio test (HLRT). The chapter focuses on deriving the likelihood function in the AWGN channel while modification of the likelihood function in fading channels and non-Gaussian channels are also mentioned briefly. Different from the ML likelihood function, the ALRT likelihood function replaces unknown parameters with the integral of all their possible values and their corresponding probabilities. The chapter describes the complexity reduction of the likelihood-based classifiers. The complexity and accuracy analysis provided suggests that the classification performance is largely associated with the level of quantization.