System Level Study of the Performance Losses Caused by Inaccurate Interference Modeling for Indirect Linear Equalizers Utilized in Cellular Downlinks

Adaptive equalization of time variant and frequency selective channels is a crucial task of advanced mobile radio receivers. Channel-estimate based methods and direct methods are the two main strategies for automatic equalization. Even though several performance comparisons between the two implementation choices are available (see the full text for references), they have all been performed at the "link level" and their results cannot be extrapolated to the "system level" because the noise models typically used in link level simulations do not faithfully reflect the actual interference characteristics observed in cellular downlinks. This is of redoubled significance since most channel-estimate based methods make use of some of those same model simplifications during the calculation of the equalizer coefficients. Traditional square error minimization algorithms (referred to as direct methods) make no assumptions about noise characteristics and are thus more general. Consequently, only a full system level experiment can appropriately evaluate possible performance differences between these two classes of algorithms. This paper is an experimental evaluation of the losses that typical channel-estimate based equalization methods would experience in cellular downlinks due to their modeling assumptions. The results are relevant when choosing a filter adaptation strategy since direct methods do not suffer from this effect