Pipeline Implementation of Adaptive Filters Enabling Fast Mobile Communication

点击翻译按钮获取中文摘要

In wireless communication systems, many interferences, such as inter-symbol interference (ISI), or inter-access interference (IAI) are occurred due to the noise and to the fading. Interference canceler, that is used for canceling those unwanted interferences, become an important unit in wireless communication systems. Their importance will increase as contents carried in wireless communications will shift from speech-only information to multimedia data.In this research, we considered a method for increasing operable frequency of an interference canceler with maintaining its performance. For that purpose, we developed methods for pipeline implementation of the adaptive filters used in interference cancelers. The pipeline implementation is a technique to increase the throughput of the circuits by shortening its longest path. Note that we use the term throughput to show the number of input samples per unit time that can be processed by a canceler.In interference cancelers, two types of ad … More aptive filters are usually used, that is, the LMS adaptive filter and the RLS adaptive filter. The LMS filter is known to be implemented with rather small amount of computation although it requires long time for leaning. On the other hand, the RLS filter can provide faster learning by paying huge computational costs. We should, therefore, select one of them according to the requirement of the applications and available resources such as the amount of usable hardware, or power consumptions allowed.Because the basic characteristics of the adaptive filters differ, we should prepare different scenario for realizing pipeline implementation for those filters. From this view point, we considered two techniques for pipeline implementation, namely, hardware pipelining and software pipelining according to the characteristics of the adaptive filter. We then derived a hardware pipelining method for the LMS adaptive filters and a software one for RLS adaptive filters. We showed that the proposed method for the LMS filter enables us to pipeling the filter without any processing delays so that no performance loss will be generated. On the other hand, the software method for the RLb filter enables us to reduce the more than half of the required amount of calculation when the method is implemented using DSP processors. Less