non-orthogonal multiple access (NOMA) can satisfy the fifth-generation (5G) wireless communication requirements. For a traditional NOMA detection strategy, successive interference cancellation (SIC) at the receiver side is necessary for both uplink and downlink broadcasts. Because of the complex multipath channel environment and error propagation concerns, the traditional SIC technique has limited performance. In this paper a novel Deep Learning-Based Channel Estimation and Signal Recovery for OFDM Systems Over Rician Fading Channels. The transmitter performs traditional steps such as pilot insertion, IDFT, and cyclic prefix addition, followed by signal transmission. At the receiver, after standard preprocessing steps, the received signal is passed through a hybrid neural network combining convolutional and recurrent layers. The convolutional layers extract spatial features, while the recurrent layers capture temporal dependencies, enhancing signal detection performance in complex channel conditions. This RNN outperforms conventional detection techniques by improving robustness to interference and fading. The proposed model demonstrates the potential of integrating deep learning in advanced wireless communication systems for efficient and accurate signal recovery.