In this paper, we present two new noise variance estimation methods of OFDMA signals transmitted through an unknown multipath fading channel. We focus on blind estimation as it does not require any pilot sequences and is therefore applicable to context, such as cognitive radio for instance, where few prior signal knowledge is available. The two methods are respectively based on the sparse nature of OFDMA signals and on the redundancy induced by the cyclic prefix. Numerical simulations compare the two algorithms and highlight their good performance.