This paper focuses on overcoming the high Peak to Average Power Ratio challenging within Orthogonal Frequency Division Multiplexing Asynchronous (OFDMA) communication system. This signal is modulated using differential phase-shift keying and quadrature amplitude modulation techniques. It is degraded under different noise attacks. Signal recovery algorithms are supposed for overcoming the high peak power problem. This issue is resolved through the detection of maximum peaks within threshold peak values of the power amplifier output signal. Additionally, these peaks are identified depending on the extracted power density and order statistics features. This system is trained using an artificial neural network under different channel impairments. Besides, the error rate of this network is computed in presence of different channel fadings. Rayleigh fading imposes the best performance for the transmitted OFDMA signal. Moreover, recognition of 100% is attained using the high order statistics compared to the power density features in presence of noise attacks.