1. INTRODUCTION
Signal to noise ratio (SNR) plays an important role in many wireless communication systems. Thus, there is a need to obtain an estimate of the SNR from received signal [1], [2]. The CRB is a well-known lower bound on the variance of any unbiased estimate and, as such, serves as a useful benchmark for practical estimators. Several expressions of the CRB and techniques algorithms for SNR estimation are available in the literature in data-aided (DA) and non-data aided (NDA) context (e.g., [3], [4], [5], [6]) for the additive white Gaussian noise (AWGN) channel, as well as for the constant, frequency selective channels [7], [8]. Among the SNR estimation techniques used in the above references are the ML estimation method, the expectation maximization algorithm (EM), the decision-directed (DD) method, and the method of moments (MM). However, in many applications requiring SNR estimation (e.g., mobile communication), the assumption that the channel is constant throughout the observation period is not valid. In recent years a few papers on the SNR estimation assumed the channel to be time-variant. In [10] the CRB and the DA ML estimator based on the expectation maximization algorithm for polynomial time selective slow-fading channels are derived. In [11] a DA subspace-based method for orthogonal frequency division multiplexing (OFDM) systems is derived on the basis of the eigenvector decomposition of the estimated channel correlation matrix. A CRB work on SNR estimation of noncoherent binary frequency-shift-keying (NBFSK) signals transmitted over uncorrelated Rayleigh fading channels appears in [12].