Orthogonal Frequency Division Modulation (OFDM) is the modulation employed by the LTE downlink physical layer. Recent studies on pilot-aided channel estimation, typical of such modulation, show that assuming a parametric channel model can effectively improve the estimate accuracy. The algorithms based on this model perform the frequency response estimation of a multipath environment through the knowledge of the tap-delay parameters, i.e., the delays’ values and their number. This thesis focuses on the estimation of the number of delays, required by the tap-delay estimators, e.g., ESPRIT. For this purpose the state-of-the-art MDL detection scheme is investigated over one of the proposed LTE channels and over a dynamic channel. Furthermore, it is compared with an approach based on a simple threshold operation. The study has carried out jointly with the preprocessing scheme named spatial smoothing. It’s shown that over the static channel, MDL exhibits a certain robustness against the time observation reduction, while the Threshold algorithm suffers from a significant increase in the estimate’s error. Tests over the dynamic channel show a general performances decrement. However MDL average error keeps constant within a wide window size range and it doesn’t drastically get worse as the Threshold one. The state-of-the-art performances then, are not exceeded, but the noticed problems still leave this topic open.