Paper Info
Title
The recursive batch least squares filter: An efficient RLS filter for floating-point hardware
Abstract
Conventional Recursive Least Squares (RLS) filters have a complexity of 1.5L <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> products per sample, where L is the number of parameters in the least squares model. The recently published FWL RLS algorithm has a complexity of L <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> , about 33% lower. We present an algorithm which has a complexity between 5L <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> /6 and L <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> /2. The algorithm is in theory as fast and accurate as the other RLS ones, but employs a batch approach, waiting for K≥L consecutive samples and processing them together. When K = L, complexity is highest, but still lower than in the conventional and FWL RLS algorithms. When K >> L complexity converges to one third of conventional RLS algorithms, or one half of the FWL RLS one. The algorithm may have stability problems in fixed-point because of accumulation of numerical errors, and it can only be effectively implemented in floating-point arithmetic. Some DSP processors and advanced FPGAs are capable of using floating-point arithmetic: the algorithm may thus be employed in many advanced DSP hardware. We test it in a C++ implementation.
Year
DOI
Venue
2017
10.1109/ECCTD.2017.8093223
2017 European Conference on Circuit Theory and Design (ECCTD)
Keywords
Field
DocType
Adaptive filtering,Recursive least squares,RLS
Least squares,Digital signal processing,Computer science,Floating point,Matrix decomposition,Field-programmable gate array,Algorithm,Filtering theory,Computer hardware,Recursive least squares filter,Recursion
Conference
ISBN
Citations 
PageRank 
978-1-5386-3975-7
0
0.34
References 
Authors
3
2
Name
Order
Citations
PageRank
Pietro Monsurrò14812.78
A. Trifiletti243363.29