It is well known that transformation of correlated input signal improves the convergence performance in transform domain least mean square (TDLMS) algorithm as compared to normalized least mean square (NLMS) algorithm. But this increase in convergence comes at the cost of increased complexity. Therefore, it is pragmatic to compare real-time performances of the schemes using different transformations, before applying them to solve the real-life problems. In this paper, implementation of transform domain predictor has been carried out using TMS320C6713 DSP starter kit (DSK). The transforms chosen are discrete Fourier transform (DFT), discrete cosine transform (DCT) and discrete wavelet transform (DWT). The models were created in MATLAB-SIMULINK environment and then implemented in real time using the DSK. The performances of different transform domain schemes were compared on the basis of consumed memory and execution time taken by them. It is concluded from the observations that transform domain LMS with DWT transformation, though converges fastest, takes highest memory as well as execution time while NLMS takes the least.

LMS algorithm, transform domain predictor, real-time implementation, TMS320C6713 DSK.