Research On Partial Updating RLS Algorithm

Adaptive algorithm is an important branch of the field of signal processing. For its efficiency and practical, it was widely used in the field of Antenna Array, Pre-distortion, Echo Cancellation and so on. Adaptive algorithm was mainly composed of RLS algorithm and LMS algorithm. The LMS algorithm's complexity is low but its convergence is very slow, on the other hand The RLS algorithm's complexity is high but its convergence is very fast.This paper's discussion was based on the features of data invariant. Base on this situation, some people have proposed Split RLS algorithm, HRLS algorithm and PU-RLS algorithm respectively. The best one was the PU-RLS algorithm (Partial Updating RLS Algorithm). But its alternately updating influences the convergence of PU-RLS algorithm.Based on this problem, this paper dose a convergence analysis for the PU-RLS algorithm with correlative input, and obtains the Ensemble Average Learning Curve of PU-RLS algorithm. And then get some parameters about the PU-RLS algorithm through the analysis; What's more this paper proposed a kind of Improved Partial Updating RLS Algorithm (IPU-RLS) in which the power weight factor plays an important role in the adaptive process. Firstly, we do a few of iterations with the PU-RLS algorithms to initially determine the relationship among the coefficients of the adaptive filter. Then the coefficients are listed in descending order and separated into two parts with sub-filters of the same order in each part. In the next step, only to update the first sub-filter until it reaches a relative stable state, then alternately update two sub-filters. The simulation proves that the algorithm can accelerate the convergence with complexity nearly to the PU-RLS algorithm. Its performance is superior to the PU-RLS algorithm.Finally, some improved was proposed for IPU-RLS which used the NLMS algorithm as the pretreatment. The cause to use the NLMS algorithm was that its convergence was faster than the LMS, and its complexity was similar to the LMS algorithm. This method can finished the pretreatment with small complexity and more faster. So this method can decrease the complexity of IPU-RLS efficiently.