Research on Partial Updating RLS Algorithm
|School||Dalian University of Technology|
|Course||Communication and Information System|
|Keywords||Communications technology Adaptive Algorithm RLS algorithm Section is adapted to Convergence Analysis|
Adaptive algorithm is an adaptive signal processing is an important component of its efficiency, practicality it is widely used in the antenna array, the pre-distortion technology, echo cancellation and other fields. LMS adaptive algorithm is divided into two major categories of algorithms and RLS algorithm, in which the LMS algorithm with low computational complexity, but the slow convergence; while the RLS algorithm converges faster, but the computational complexity is very high. This algorithm is based on the data matrix does not have shift invariance case, based on this case, people were raised Split RLS algorithm, HRLS algorithm and PU-RLS algorithm (Partial Updating RLS Algorithm, abbreviated PU-RLS algorithm). Where PU-RLS algorithm performance is better, but because of its time-interleaved update mechanism, resulting in a number of cross-interference between sub-filters, affecting the convergence speed. Based on the above issues, this paper based on the relevant input environment PU-RLS algorithm is to make a detailed analysis of the convergence, the algorithm gives the ensemble average learning curve expressions that summed affect system performance Several parameters; secondly on the basis of previous work in this paper, an improved partial update RLS algorithm (Improved Partial Updating RLS Algorithm, referred IPU-RLS algorithm), the algorithm uses the weighting coefficients in the adaptive power system plays a major feature role, first to PU-RLS algorithm for the iterative determination of the filter coefficients a small size relationship, and then the filter coefficients are rearranged in descending, the order of all the filter coefficients are divided into a plurality of equal parts, then only the the largest part of the weights to be updated in order to achieve the power to focus on weight coefficient updating purposes. When the weight reaches the largest part is relatively stable, to the various parts of the time-updated alternately. The simulation results show that, IPU-RLS algorithm can not increase in the basic PU-RLS algorithm based on computational complexity, improve the convergence speed. Finally, the paper IPU-RLS algorithm made some improvements, namely preprocessing stage pretreatment NLMS algorithm instead. Because the NLMS algorithm used, its convergence rate faster than the LMS algorithm, while the algorithm is similar to the complexity and the LMS algorithm, with a smaller amount of computation, a faster speed to determine the magnitude relationship between the respective filter coefficients, so as to effectively IPU-RLS algorithm reduces the computational complexity.