In this article, we applied Deep Learning on LTE-A uplink channel estimation system. The work involved creating of two SC-FDMA databases for training and for test, based on three types of channel propagation models. The first section of this work consists of applying an Artificial Neural Network to estimate the channel of SC-FDMA link. Neural Network training is an iterative process which consists on adapting the values of its parameters: weights and bias. After training, the Neural Network was tested and implemented on the receiver. The second section of this work deals with the same experimentation but by using Deep Learning instead of classic Neural Networks. The simulation results showed a strong improvement given by Deep learning compared to classic method concerning the Bit Error Rate and the process speed. The third part of this work had been reserved to the complexity study. We had proved that Deep Learning gives better performance than MMSE estimator with low complexity.