Lars Lindbom

Licentiate Thesis, Report UPTEC 92124R, 125pp, November. 1992.

Outline:

The receiver in a digital mobile radio system has to detect and adjust for time-variations in the dispersive channel. In some standards, such as D-AMPS in North America, the time-variability is rapid and provides severe challenges for existing adaptation algorithms, such as LMS and RLS. The thesis presents novel algorithms, one of which has subsequently been patented.

Abstract:

The development of indirect methods for adaptive equalization, applied to fading digital mobile radio channels encountered in the proposed North American Digital Cellular system, is the main subject of this thesis. New algorithms for channel estimation in severe Rayleigh fading environments are presented, They are based on the principles of stochastic and deterministic internal modelling of time-varying coefficients of a FIR channel model.

In the stochastic case, channel estimators are based on simplified second order autoregressive models and low-complexity approximation of a Kalman estimator. A novel averaging approach is used to replace the on-line update of a Riccati equation with the determination of a constant matrix. When using a simplified second order AR model of the time-varying parameters, the resulting adaptation gain filters can be expressed in analytical form. This results in algorithms with high performance at LMS computational complexity.

In the deterministic case, channel coefficients are parameterized by first order Fourier series expansions, with unknown fundamental frequencies. A simplified prediction error identification algorithm is derived to estimate the Fourier coefficients and the fundamental frequencies, simultaneously.

Several combinations of equalizers/Viterbi schemes with different channel estimators were studied on simulated data, generated from a Rayleigh fading channel model. As compared to channel estimation with LMS and RLS algorithms, the new channel estimators provide higher performance. Channel estimators based on second order internal stochastic models, used in conjunction with decision feedback equalizers (DFE's), provide adaptive equalization with high performance, at a low computational complexity.

Indirect adaptation of equalizer parameters, based on channel estimation, is also compared with the conventional direct approach to adaptive equalization. The indirect method offers superior performance, mainly since Rayleigh fading channel coefficients change in a regular way. Direct adjustment of the equalizer coefficients would instead have to track optimal adjusted values which drift with strongly time-varying rates of change.

Related publications:

A series of four papers outlining the later development of a complete design methodology, based on stochastic models of time-varying parameters:

Design of general constant-gain adaptation algorithms.

Part II: Analysis of stability and performance, for slow and fast variations.

The Wiener LMS adaptation algorithm, a special case with low complexity.

A Case Study on IS-136 1900MHz channels.

PhD Thesis by L Lindbom 1995, presenting the general design methodology.

Conference paper (IEEE ICASSP'93) summarizing the proposed KLMS algorithm.
Sinusoid modelling of time-varying channel coefficients in IS-136 800 MHz systems.