Skip to Main Content
Conservative selection of cyclic prefix (CP) length to completely eliminate interference results in a waste of resources, while fast adaptation of CP to given values of the root mean square (RMS) delay spread incurs increased complexity. We propose a method to slowly adapt CP to the underlying lognormal statistics of the RMS delay spread in a multiuser OFDMA downlink where different users experience location dependent statistics of RMS delay spread. Employing the approximate average rate as a performance measure and incorporating the approximate average rate requirements of different users, the CP optimization is shown to involve a set of linear programming problems which are solved analytically by duality analysis. With the additional assumption of an exponential decaying power delay profile, a closed form approximation of the original problem is formulated and solved in a computationally simple manner. Finally, an online adaptation is presented to estimate the statistics of RMS delay spread from feedback whenever available.