Stochastic MV-PURE estimator: robust reduced-rank estimator for stochastic linear model
IEEE Transactions on Signal Processing
Generic weighted filtering of stochastic signals
IEEE Transactions on Signal Processing
| Hi-index | 35.69 |
In time-slotted mobile communication systems with antenna array at the receiver, the space-time channel matrix is conventionally estimated by transmitting pilot symbols within each data packet (or block). This work is focused on reduced rank (RR) estimation methods that exploit the low-rank property of the space-time channel matrix to estimate single or multiple user channels from the observation of single or multiple training blocks. The proposed RR methods allow to improve the estimate accuracy by reducing the set of unknown parameters (rank reduction) and extending the training set (multiblock processing). The maximum likelihood RR estimate is obtained as the projection of the prewhitened full-rank (FR) estimate onto the spatial or temporal signal subspace. The paper shows that, even for time varying channels, these subspaces can be considered to be slowly varying, and therefore, they can be estimated with increased accuracy by properly exploiting training signals from several blocks. The analytical and numerical performance in terms of mean square error for the RR estimate shows that the main advantage of the proposed method with respect to the conventional FR one can be ascribed to the reduced complexity of the channel parameterization.