Automatic Modulation Identification Based on the Probability Density Function of Signal Phase

Automatic modulation recognition is advantageous for wireless communication systems employing adaptive modulation, software-defined radio, and cognitive radio. In this paper, we consider a phase based maximum likelihood (ML) approach for identifying the modulation format of a linearly modulated signal. Since the optimal ML scheme is computationally intensive, we propose two approximate ML alternatives, which can offer close-to-optimal performance with reduced complexity. We then present a general performance analysis for classification of K types of modulation constellations. For K<;=5, probability of correct classification (Pcc) can be evaluated via simplified integration. In the case of K>;5, we obtain a set of upper bounds on Pcc, which provide a tradeoff between accuracy and complexity in calculating the Pcc. In addition, asymptotic behavior of phase based ML classification schemes is investigated.

I. Introduction

Automatic modulation recognition [1], originally used in military and civilian applications like spectrum monitoring, signal surveillance, and interference identification, is now a desirable feature for emerging wireless communication systems such as software-defined radio [2], [3] and cognitive radio [4], since it can facilitate more flexible and less expensive system designs. Methods for identifying the modulation format of a given signal generally fall into two categories. in the first category, statistical properties of the signal are used to determine its modulation format [5]–[7]. in the second category, the maximum likelihood (ML) principle is applied to choose the most possible modulation type in use [8]–[11]. A comprehensive review of modulation classification techniques can be found in [16]. Some recent advances on the two approaches are reported in [12], [13] and [14], [15], respectively.