期刊名称：International Journal of Electrical and Computer Engineering
电子版ISSN：2088-8708
出版年度：2015
卷号：5
期号：3
页码：548-561
DOI：10.11591/ijece.v5i3.pp548-561
语种：English
出版社：Institute of Advanced Engineering and Science (IAES)
摘要：Blind algorithms based on the Euclidean distance (ED) between the output distribution function and a set of Dirac delta functions have a heavy computational burden of due to some double summation operations for the sample size and symbol points. In this paper, a recursive approach to the estimation of the ED and its gradient is proposed to reduce the computational complexity for efficient implementation of the algorithm. The ED of the algorithm is comprised of information potentials (IPs), and the IPs at the next iteration can be calculated recursively based on the currently obtained IPs. Utilizing the recursively estimated IPs, the next step gradient for the weight update of the algorithm can be estimated recursively with the present gradient. With this recursive approach, the computational complexity of gradient calculation has only . The simulation results show that the proposed gradient estimation method holds significantly reduced computational complexity keeping the same performance as the block processing method
其他摘要：Blind algorithms based on the Euclidean distance (ED) between the output distribution function and a set of Dirac delta functions have a heavy computational burden of due to some double summation operations for the sample size and symbol points. In this paper, a recursive approach to the estimation of the ED and its gradient is proposed to reduce the computational complexity for efficient implementation of the algorithm. The ED of the algorithm is comprised of information potentials (IPs), and the IPs at the next iteration can be calculated recursively based on the currently obtained IPs. Utilizing the recursively estimated IPs, the next step gradient for the weight update of the algorithm can be estimated recursively with the present gradient. With this recursive approach, the computational complexity of gradient calculation has only . The simulation results show that the proposed gradient estimation method holds significantly reduced computational complexity keeping the same performance as the block processing method
