Very Large Scale Integrated High Throughput Rate Architectures for Systolic Array Processors

The Lease-Squares(L.S.)算法是许多现代信号处理应用的基础，包括频谱分析、波束形成、测向、自适应天线阵、均衡、卡尔曼滤波和参数估计。对于高性能的实时应用，三角形和线性脉动阵列是获得高计算吞吐速率的最有效方法。该公司正在对算术复杂性(乘数、除数等)进行详细比较。在Givens变换、FAST Givens变换、修正的FAST Givens变换、修正的Gram-Schmidt变换和Household变换下，三角和线性内部和边界的脉动处理单元的有限精度计算效果。基于QR分解方法的估计。通过选择计算复杂度最小的处理算法，可以确定用于VLSI制造的高通量脉动阵列处理器的体系结构。可以为各种信号处理应用开发通用(可编程)脉动阵列处理芯片。评审者给出了16分、17分和18分的一致高分。他们说，这个研究问题定义得很准确，影响很大。该提案知识渊博，具有时效性，而且内容全面。成功将带来具有非常高吞吐速率的信号处理架构。这些人员的资质都很好。这项研究将有助于设计未来需要高速度数据处理的仪器。

The Lease-Squares (L.S.) algorithm is the basis for many modern signal processing applications including spectral analysis, beamformation, direction finding, adaptive antenna array, equalization, Kalman filtering and parameter estimation. Triangular and linear systolic arrays are the most efficient ways to achieve high computational throughput rates for high performance real-time applications. The company is making a detailed comparison of the arithmetical complexity (number of multiples, divides, etc.) and finite precision computational effects of the triangular and linear internal and boundary systolic processing cells under Givens, fast Givens, modified fast Givens, modified Gram-Schmidt, and Householder transformations for l.s. estimation based on the QR decomposition approach. By selecting the processing algorithm with the least computational complexity, it can determine the architecture of the high throughput rate systolic array processor for VLSI fabrication. A generic (programmable) systolic array processing chip can be developed for a variety of signal processing applications. Reviewers awarded this uniformly high scores of 16, 17 and 18. They said that the research problem was precisely defined and was one of high impact. The proposal is knowledgeable, current and comprehensive. Success will lead to signal processing architectures with very high throughput rates. The personnel have excellent qualifications. This research will contribute to the design of future instruments that require a high rate of data manipulation.