BCH (Bose-Chaudhuri-Hocquenghem) codes over complex fields in the frequency
domain have been proposed for image coding applications as robust channel
coding methods. The embedded problem of estimation of complex sinusoids cor
rupted by white additive noise may be solved in different ways. A standard
approach for modelling such signals employs low-order nearly nonstationary
autoregressive (AR) models with complex parameters. In the paper linear pre
diction/least squares based methods are used for parameter estimation inclu
ding forward linear prediction (FLP) and forward-backward linear prediction
(FBLP). The two parameter estimation methods have been applied to the firs
t- and the second-order AR models with minimal number of samples to obtain
efficient frequency estimators for decoding of BCH codes. Computer simulati
on has been carried out to compare different frequency estimation algorithm
s with regards to three significant criteria: frequency resolution, computi
ng complexity and coding redundancy. The results show that different algori
thms exhibit optimal combinations of characteristics for different constrai
nts and Importance of criteria.