A New Data Extrapolation Approach Based on Spectral Partitioning

Abstract

In this letter, we propose a new linear predictive (LP) data extrapolation approach. It involves partitioning the spectrum into multiple spectral subbands and using a different autoregressive (AR) process to model each subband. The new extrapolation approach is then combined with the classical discrete Fourier transform (DFT) to produce a new hybrid LP-DFT spectral estimator to address the detection and estimation problem of multiple sinusoids in a discrete data sequence. Simulation results demonstrate the superiority of the proposed hybrid technique over an existing popular hybrid LP-DFT technique, where a single AR process is used to model the entire spectrum of the data sequence.