Reducing Pattern Noise in Blind Energy Detection of Nonnegative PAM Symbols

In this paper, we consider blind initialization of decision-directed (BDD) parameter estimation for energy detection of nonnegative pulse-amplitude-modulated (PAM) symbols. Short pulses and guard intervals are used to mitigate intersymbol interference. In the previous work, no analytical model has been given to predict how the BDD approach converges, hindering the usage of the approach. We present analytical mean square error expressions to infer a sufficient observation interval that makes the effect of the pattern noise negligible in comparison with the noise component originated from the thermal noise of the receiver. Using a proper observation interval, the symbol error rate of the BDD method is demonstrated to be comparable with that of a corresponding data-aided estimator which, unlike the BDD approach, needs a periodical training signal.