A New Approach to Parameter Estimation of Time-Varying Systems, Applied to Fading Digital Mobile Radio Channels

Mikael Sternad, Anders Ahlén and Lars Lindbom

Report UPTEC 90052R, May 1990.

Report In Pdf.

Abstract:

Parameters of time-varying dynamic systems may sometimes be modelled as deterministic functions of time. This is shown to be feasible, and to improve the estimation accuracy, in the estimation of fading digital mobile radio channels. Such systems may be described as finite impulse responses, with coefficients varying, on a short time-scale, as sinusoids.

We propose to estimate the phase, amplitude and frequency of these sinusoids, in each data burst in a TDMA system.

A prediction error identification algorithm has been developed, assuming each data burst to begin with a known training sequence. The parameter estimation is separated from the equalizer design and the equalization phase.

The parameter estimation was tested on simulated data, generated by a two ray channel model, with 10dB SNR. The proposed algorithm provided very accurate channel models, also in time intervals with severe fading. In contrast, the alternative approach of using recursive identification with a sliding data window was found not to behave well.

Related publications:

Conference paper, NUTEK 1991, which presented the results.
Licentiate Thesis by L Lindbom, 1992, which develops the algorithm further.
Conference paper in IEEE ICASSP'93 on using stochastic models of time-variations .
PhD Thesis by L Lindbom, presenting an improved general design methodology.