| Summary: | 碩士 === 國立中山大學 === 電機工程研究所 === 84 === The main concerns of this thesis is to improve the
performance of non-integer time delay estimation by the
so-call adaptive γ-LMS algorithm. When noise is present,
in an adaptive TDE system, the FIR weight coefficients
may not converge to the correct values, resulting in TDE
performance degradation. To speed up The convergence rate
of the tap-weight coefficients, an adaptive γ-LMS algorithm
is presented. After the weight coefficients are obtained
the direct delay estimation formula is employed for non-
integer time delay estimation. Computer simulation results
verified the adaptive γ-LMS algorithm over the
conventional adaptive LMS algorithm in term of non-integer
time delay estimation. Moreover, to examine the steady-
state property, the steady-state mean weight vector and the
sum of the mean squared weight errors, the weight-error
variance of the γ-LMS algorithm are derived. Another topic
which will be considered in this thesis is the variable step-
size, The convergence rate of the ada- ptive filtering
algorithm is proportional to step-size, and generally, the
misadjustment value is inversely pro- portional to step-
size. The proper choice of the step- size is the key to
assure good performance. The variable step size of the
adaptive γ-LMS is desgined to reduce the squared
estimation error and to speed up the conver- gence rate.
|
|---|
