Two new adaptive filter structures: Performance analysis, implementation and applications

Adaptive filters have become standard components of many current digital signal processing systems. With the continuing development of new digital technologies such as digital high-definition television and digital portable telecommunications, interest will continue to be placed on fast, low complexity, easily implementable adaptive filtering algorithms. This thesis considers two such algorithms, their performance, implementation, and some applications for which they can be used.; The LMS filter has been proposed for sinusoidal interference cancellation in direct-sequence spread spectrum communications. It has been previously shown that the smoothing (or two-sided transversal) filter outperforms the prediction filter in terms of bit error rate. The optimal smoothing filter has symmetric coefficients, and this symmetry is exploited to produce an adaptation algorithm with one-half as many multiplications as the equivalent order predictor. For an input of a real sinusoid in white Gaussian noise, it is proved that the real LMS smoother has faster MSE convergence than the predictor when the filters are designed for equal steady-state misadjustment. Although the modal decomposition techniques of the proof do not carry over to the case of a complex sinusoid input, simulation results indicate the same performance advantage for the complex smoother over the complex LMS predictor.; One alternative to the LMS tap-delay line filter is the Gradient Adaptive Lattice (GAL). Unlike the LMS tap-delay line, the convergence rate of the GAL is not strongly dependent upon the statistics of the input signal and under some common signal conditions can outperform the LMS algorithm. The GAL, though, requires more arithmetic operations to implement than the LMS algorithm. Consequently, a GAL algorithm using power-of-two quantization to replace multiplications and divisions with shifting operations in the filter implementation is proposed. This filter is analyzed through a new, second order method that correctly models the filter coefficient bias and steady-state misadjustment that previous first order analyses ignored.; Finally, the application of estimating the ionospheric refraction multipath of a HF transmission is considered. An adaptive smoothing method and a maximum likelihood algorithm are presented and compared.