Short-Data-Record Adaptive Receivers for Rapidly Changing Communications Environments
Abstract
We defined and pursued a novel line of research that lies in a multidisciplinary intersection of Estimation Theory, Communications Theory, and Mean-Square optimum linear filtering. Consider an arbitrary input signal vector space and a given information bearing signal vector to be protected or recovered in the presence of multiuser or other forms of heavy interference. Based strictly on statistical conditional optimization principles, we developed an iterative algorithm that starts from the conventional matched-filter correlator and generates a sequence of linear filters ("auxiliary-vecto?" filters) that converges to the exact MS-optimum solution. At each iteration step, the filter is given as a direct function of the input autocorrelation matrix, the signal vector waveform to be protected, and the filter at the previous iteration. When the autocorrelation matrix is sample-average estimated from a short data record, this procedure offers the means for effective control over the filter estimator bias versus (co-)variance trade-off. For a fixed data record size the filter estimators in this sequence have rapidly decreasing bias and gracefully increasing variance. They outperform other known estimators such as Sample-Matrix-lnversion (SMI), Diagonal-Loading (DL) SMI, RLS, LMS, reduced-rank eigenvector decomposition, and "multistage" nested Wiener filter. While all of the above estimators converge to the optimum MMSE/MVDR solution for infinitely long data records, for any given finite data set there is at least one AV filter estimator in the sequence that outperforms all SMI, DL-SMI, RLS, LMS, reduced-rank eigenvector and multistage nested Wiener filter estimators. The theoretical and practical implications of these results are far reaching. Biased estimators and algorithms that offer fall control over the biasivariance balance are rarely reported in the literature, if any in a communications applicable context.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2003
- Accession Number
- ADA417694
Entities
People
- Dimitris A. Pados
- Stella N. Batalama
Organizations
- University at Buffalo