Pipeline Implementation of Adaptive Filters Enabling Fast Mobile Communication

在无线通信系统中，由于噪声和衰落的影响，会产生许多干扰，如码间干扰（ISI）或接入间干扰（IAI）。干扰消除器是无线通信系统中的一个重要部件，用于消除不必要的干扰。随着无线通信传输的内容从纯语音信息转向多媒体数据，它们的重要性将会增加。在本研究中，我们考虑了一种在保持干扰消除器性能的前提下提高其工作频率的方法。为此，我们开发了用于干扰消除器的自适应滤波器的流水线实现方法。流水线实现是一种通过缩短其最长路径来提高电路吞吐量的技术。注意，我们使用“吞吐量”一词来表示单位时间内可以由消去器处理的输入样本数。在干扰消除器中，常用的自适应滤波器有两种，即LMS自适应滤波器和RLS自适应滤波器。众所周知，LMS滤波器虽然需要较长的学习时间，但其实现的计算量相当小。另一方面，RLS滤波器可以通过付出巨大的计算成本来提供更快的学习速度。因此，我们应该根据应用程序的需求和可用资源（如可用硬件的数量或允许的功耗）选择其中的一个。由于自适应滤波器的基本特性不同，我们应该准备不同的场景来实现这些滤波器的流水线实现。从这个角度出发，根据自适应滤波器的特点，我们考虑了两种流水线实现技术，即硬件流水线和软件流水线。然后，我们推导了LMS自适应滤波器的硬件流水线方法和RLS自适应滤波器的软件流水线方法。我们证明了LMS滤波器的建议方法使我们能够在没有任何处理延迟的情况下对滤波器进行流水线处理，从而不会产生性能损失。另一方面，RLb滤波器的软件方法使我们能够在使用DSP处理器实现该方法时减少所需计算量的一半以上。少

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