Finite wordlength effects in discrete time wavelet transform.

Abstract

This thesis investigates the finite wordlength effects in one-dimensional discrete-time wavelet transform (DTWT). A MATLAB DTWT model is written based on an efficient algorithm suggested by the Multiresolution Analysis (MRA). This model allows users to investigate the overall non-ideal effects of a MRA-based DTWT system by specifying round-off method for computation, the number of quantization bits employed in representing filter coefficient and internal computational results. Further, the possibility of discarding the high frequency portion of a signal in DTWT signal reconstruction phase (Selective Subband Reconstruction) is shown practical with the presence of finite wordlength effects. The feasibility of using either direct structure filter or lattice structure filter to realize a MRA-based DTWT system is also considered in this thesis. Simulation results show that the lattice structure filters have superior magnitude and phase responses in most scenarios under investigation to its direct structure-based counterparts. Moreover, lattice structure filter demands relatively less computation than direct structure filter in the same filter length and DTWT configuration. In conclusion, the lattice structure filter is a better alternative than the direct structure filter for MRA-based DTWT system implementation.