This paper deals with a Maximum Likelihood (ML) receiver for a Single Carrier (SC) signal distorted by a nonlinear Power Amplifier (PA). The well known optimal Matched Filter (MF) loses its optimality when a nonlinear PA is considered. We first propose an optimal receiver based on the ML criterion taking into account the PA nonlinearity. In case of a Additive White Gaussian Noise (AWGN), the ML receiver has to find the minimum Euclidean distance between the received vector and all nonlinearly distorted vectors that the transmitter may send. The exhaustive search solution of this problem has exponential complexity. To reduce the complexity, we propose a sub-optimal receiver that minimizes the Euclidean distance, seen as a cost function, using the Gradient Descent (GD) algorithm. We also present a low complexity version of the GD based receiver which is suitable for a parallel processing implementation. Comparative study of Symbol Error Rate (SER) performance of the proposed receiver against the optimal SC MF based receiver is given.