Some observations on implementing various recursive least squares adaptive filtering algorithms

Schutze and Ren (1992) have investigated the behaviour of an extensive range of covariance domain algorithms, but restricted their simulations to single-precision arithmetic with a forgetting factor (λ) of 1.0. Yang and Bohme (1992) concentrate mainly on square-root information domain algorithms, performing limited precision simulations to determine the minimum number of bits required for stability. This work combines the approaches of both papers to enable a comparison to be made between covariance and square-root information domain algorithms in a limited precision environment. The problem considered is one of adaptive system identification, with a pseudo-white centered Gaussian noise of unit variance as the input signal, and the unidentified system´s output plus uncorrelated Gaussian noise at -30 dB as the desired response