Block-update parallel processing QRD-RLS algorithm for throughput improvement with low power consumption

Lijun Gao, Keshab K Parhi

Research output: Contribution to conferencePaperpeer-review

3 Scopus citations

Abstract

In this paper, a block-update parallel processing algorithm is proposed for increasing the throughput of the CORDIC-based QRD-RLS filtering with low power consumption. The proposed algorithm employs single-state-update parallel processing, and with this algorithm, the throughput of a block-by-block weight-update QRD-RLS filter can be increased at the cost of linear increase in hardware resource, which is the same as in [1]. However, the proposed algorithm does not change the iteration bounds and clock frequency of the QRD-RLS filters. As a result, the functional units need not be pipelined and the power consumption only increases linearly instead of quadratically as compared to the approach in [1] for throughput increase. Due to non-pipelining and less power consumption, a higher folding factor can be used for a folding transformation [3][4] and a great reduction in hardware resource can be achieved without exceeding the physical limitation on pipelining level and power density. Therefore, the proposed algorithm can serve as an important stage in designing and mapping a QRD-RLS filter onto physical hardware or computing resources, and thus is better for both ASIC chip design and parallel computing when block-by-block weight-update is applicable.

Original languageEnglish (US)
Pages225-234
Number of pages10
StatePublished - Jan 1 2000
Event2000 IEEE International Conference on Application-Specific Systems, Architectures, and Processors - Boston, MA, USA
Duration: Jul 10 2000Jul 12 2000

Other

Other2000 IEEE International Conference on Application-Specific Systems, Architectures, and Processors
CityBoston, MA, USA
Period7/10/007/12/00

Fingerprint Dive into the research topics of 'Block-update parallel processing QRD-RLS algorithm for throughput improvement with low power consumption'. Together they form a unique fingerprint.

Cite this