Sensitivity of L-1 regularization on subspace-based SIMO blind channel identification in sample channel measurements
Kareem Bonna, Predrag Spasojević · 2016
Single Input Multiple Output (SIMO) Blind Channel Identification (BCI) using Subspace-based Methods, notably the Cross-Relations (CR) and Noise-Subspace (SS) Methods, are sensitive to selection of the channel order. Previous work on model order selection shows that at low SNR, accurate selection is not always possible, and in some cases, there is difficulty obtaining good estimates using any order. This makes these methods difficult to use in practice. Later works proposed using linear constraints and/or l-1 regularization in order to avoid channel order selection issues and improve performance for sparse channels, though they still depend heavily on parameter selection. In this work, the sensitivity of channel estimates using the CR and SS methods with and without l-1 regularization under a linear constraint, on a BPSK signal, is examined through simulation on real sample channel measurements that had been used in previous work on model order selection.