Performance comparison of several contemporary equalizer structures applied to selected field test data

The application of adaptive equalization techniques for the purpose of receiving underwater acoustic digital data has been well established by many researchers. However, the choice and exact implementation of the adaptive equalizer has been debated and still remains an open research area. Typically, the adaptive equalization algorithm is fed by a feedforward and feedback tap placement algorithm. The purpose of the tap placement algorithm is to determine the placement and support for the feedforward filter section and each of the feedback filter sections. The adaptive equalizer usually employs a decision feedback equalizer with a second order digital phase-locked loop that is structured around a filter coefficient update algorithm. Some examples of currently used equalizer update algorithms are the Adaptive step size Least Mean Squares (A-LMS) and Modular Stabilized Fast Transversal Filters (MOD-SFTF) algorithms. The MOD-SFTF algorithm is a faster implementation of the Recursive Least Squares (RLS) algorithm. The MOD-SFTF algorithm makes use of linear prediction theory, i.e. time and order updating coupled with transversal filter techniques to achieve the speed increase. Both the A-LMS and the MOD-SFTF algorithms can be modified to allow for sparsing of feedback filter sections. The author of this paper has implemented an n-section sparse feedback version of the MOD-SFTF algorithm that has been shown to remove sparse intersymbol interference (ISI) from field test data. Both the A-LMS and the MOD-SFTF algorithms require roughly the same order of magnitude number of complex operations for N filter taps. A more recent channel estimation based equalizer algorithm has been presented by Stojanovic et al.. This algorithm may use the A-LMS, normal RLS, or other similar format update equations. However, in its current form, the channel estimation based equalizer cannot be used with the MOD-SFTF algorithm. A benefit that the channel estimation based equalizer has is that it finds and places the appropriate sparse filter taps at multipath locations that have a magnitude above a given truncation threshold through the use of the training sequence