| The objective of this study is to design a reliable and comprehensive identification algorithm for symbol constellations in digital communications, using finite-length sample sequences. Such algorithms are known as modulation recognizers and are expected to play an important role in many communication applications such as the realization of "universal" receivers, signal confirmation and interference identification, multi-drop networks, surveillance, electronic warfare and military threat analysis.; Despite the fact that several studies on modulation recognition have appeared in the last fifteen years, many problems remain. Most existing algorithms cannot tackle large sets of constellations, multipath-fading channels, and constrained channels (in power, bandwidth and signal shape, etc.). The present investigation attempts to resolve some of these issues.; Two classifiers are proposed here for QAM constellation identification in noise (chapter 2). They are the optimal maximum-likelihood (ML) classifier (which maximizes the average probability of correct decision) and a (non-parametric) feature-based classifier which is based on Fourier-domain analysis. The performance of these classifiers is analyzed, and trade-offs are studied between performance and the amount of information required for reliable decision-making. The non-parametric classifier is then extended to operate in multipath-fading channels (chapter 3).; Then a blind equalization technique is presented. It provides preprocessing for the modulation recognizers and can also be used in improving demodulation and information retrieval. In devising this blind equalization technique, we noted that most existing blind equalization algorithms assume little about the multipath channel which they equalize. They do, however, require significant amount of information about the communicated constellation. In some applications the situation is reversed--the communicated constellation is unknown, but a multi-ray fading model can be confidently assumed. For this situation an iterative algorithm based on a non-parametric multipath channel model was developed (chapter 4). It jointly estimates the information sequence and the channel parameters over a fixed block of data in a "bootstrapping" manner. The algorithm is shown to be very effective for modulation recognition and information retrieval from the constellation set {dollar}{lcub}{dollar}BPSK, QPSK, 8PSK, l6QAM, 64QAM, 256QAM, 9QPR, 49QPR{dollar}{rcub},{dollar} even if the order of the channel model is underestimated. |
