Atmospheric Radar Signal Processing using Iterative Principal Component Analysis under Spiked Covariance Estimation Structure

Abstract

The signal subspace methods based on eigenvalue decomposition are used for extracting most relevant information from large dimensional data sets. In this regard, we consider the iterative robust covariance matrix estimation under spiked covariance structure by using the Tyler’s estimator principle. Incorporating this structured estimation procedure is beneficial to improving estimation accuracy. We tested the algorithm for spectrum estimation of both narrow band and complex exponential signals which may contain two different frequencies in presence of noise. For simulated signals, it is observed that the proposed algorithm (PALG) works better even for low SNR, i.e., it detects the number of frequency peaks in highly noise corrupted data also. Finally, the algorithm has been used to process MST radar data for estimating the Doppler power spectrum. The zonal U and meridional V and wind speed W have been estimated from Doppler frequency components. Compared with existing methods, the PALG works well at higher range bins and the results have been validated using the GPS radio sonde data.