A Parallel Adaptive Filtering Algorithm Based on the Mean-Square Deviation Analysis for Large-Scale Data

This paper proposes a parallel adaptive filtering algorithm via analysis of the mean-square deviation for largescale data. In this algorithm, large-scale data is divided into several sub-blocks to reduce the computational cost. Based on each data sub-block, a normalized least-mean-square algorithm estimates the parameters of interest at each sub-filter. Furthermore, the mean-square deviation analysis of the estimation result at each sub-filter leads to a variable step-size method and an intermittent-update method. These methods provide not only fast convergence rate and small steady-state error but also high computational efficiency. Finally, the estimation results of each sub-filter are combined through a combination method by determining the weights for each estimation result based on their error variance. The proposed combination method provides robustness to abnormalities of the data. Simulation results show that the proposed algorithm performs well for estimation with large-scale data.