Algorithm for Adaptive Processing of High-speed Integrated Filters

Abstract

The most popular adaptive algorithm for high-speed integrated filters today is the LMS algorithm, due primarily to its straightforward and robust digital hardware implementation. However, in analog adaptive filters, implementation of the LMS algorithm is neither straightforward nor robust. The hardware required to generate gradient information is cumbersome and power-hungry. The LMS algorithm’s accuracy is also hindered by the presence of dc offsets on the state and error signals. These considerations are the primary factors which presently limit the use of analog adaptive filters. This paper seeks to combine the advantages of digital adaptation algorithms with a high-speed analog signal path. The aim is to perform digital adaptation of analog filters and also several potential applications for digitally adaptive analog filters are presented. All are mixed-signal digital communications systems often including considerable dedicated digital signal processing hardware, usually integrated on the same die as the analog front end. In such systems analog circuit design is particularly challenging, so it is especially desirable to implement the adaptation algorithm digitally.