Analysis of and techniques for adaptive equalization for underwater acoustic communication

Abstract

Underwater wireless communication is quickly becoming a necessity for applications in ocean science, defense, and homeland security. Acoustics remains the only practical means of accomplishing long-range communication in the ocean. The acoustic communication channel is fraught with difficulties including limited available bandwidth, long delay-spread, time-variability, and Doppler spreading. These difficulties reduce the reliability of the communication system and make high data-rate communication challenging. Adaptive decision feedback equalization is a common method to compensate for distortions introduced by the underwater acoustic channel. Limited work has been done thus far to introduce the physics of the underwater channel into improving and better understanding the operation of a decision feedback equalizer. This thesis examines how to use physical models to improve the reliability and reduce the computational complexity of the decision feedback equalizer. The specific topics covered by this work are: how to handle channel estimation errors for the time varying channel, how to use angular constraints imposed by the environment into an array receiver, what happens when there is a mismatch between the true channel order and the estimated channel order, and why there is a performance difference between the direct adaptation and channel estimation based methods for computing the equalizer coefficients. For each of these topics, algorithms are provided that help create a more robust equalizer with lower computational complexity for the underwater channel.

Graphical model driven methods in adaptive system identification

“…The update of the LMS algorithm is particularly simple: 22) where µ lms is the step-size of the LMS algorithm.…”

“…For instance, it was observed by Blair [22,Chapter 3] that simply trying to directly exploit Toeplitz structure by constraining the inverse covariance matrix of the data does not lead to the performance gains that may be expected while estimating adaptive equalizer coefficients. In the context of this work as well, it is discussed in Section 5.3.4 that enforcing graphical model constraints on the input to the system without suitably constraining the output leads to a deterioration in performance.…”

“…It appears to be a non-trivial and interesting problem as to how other kinds of input structure could be exploited for system identification, and what makes it challenging is the effect of input structure on the output of the system. The effect of not accounting for that effect may be seen in, e.g., the work of Blair [22].…”