Performance Analysis of Adaptive Filtering Algorithms for System Identification

The paper presents a comparative study of NLMS (Normalized Least Mean Square), NVSS (New Variable Step Size) LMS (Least Mean Square), RVSS (Robust Variable Step Size) LMS, TVLMS (Time Varying Least Mean Square) and IVSS (Improved Variable Step Size) LMS adaptive filter algorithms. Four performances criterion are utilized in this study: Minimum Mean Square Error (MSE), Convergence Speed, Algorithm Execution Time, and Tracking Capability. The comparisons of all algorithms are demonstrated using uncorrelated and correlated input data in both stationary and nonstationary environments. The Step Size Parameter (μ) in all algorithms is chosen to obtain the same exact value of Misadjustment (M) equal to 2% for white Gaussian input and 6% for correlated input in stationary environment. Simulation Plots are obtained by ensemble averaging of 200 independent simulation runs. The simulation results show that RVSS algorithm has fastest convergence speed and superior tracking capability. The algorithm execution time is lowest in case of IVSS algorithm in stationary environment. For nonstationary environment the performance of all algorithms is equivalent.