Partially blind estimation: ML-based approaches and Cramer-Rao bound

Abstract In digital communication systems, unknown channel response can be estimated using a training signal (nonblind identification) or statistics of the received data (blind identification). Our goal is to form a partially blind identifier that uses both the training signal and received data statistics for improved channel estimation. In this paper, we explore maximum-likelihood-based approaches to partially blind channel estimation, particularly the expectation–maximization (EM) technique. We also present a novel formulation of the Cramer–Rao bound that includes a known training signal and models the nontraining portion of the symbol stream as stochastic rather than simply deterministic and unknown. Using this formulation we study the effects of the length of the training signal and observation vector (at the receiver) on the CR bound. Although CR bound is not always attainable in a given situation, it can be seen (from the simulation results) that the partially blind EM method comes very close to it.