Increase in communication devices demand for more intelligent and robust communication systems. In this work, we present our initial research on various aspects of combining deep learning in communication systems majorly for the task of signal detection, recovery and de-noising. In part one of the paper, we provide a comprehensive view on various aspects of applying deep learning in implicit channel estimation and signal detection: like the effect of different neural net complexity, activation functions, length of cyclic prefix, and the number of pilot symbols on the Symbol Error Rate (SER) with respect to increasing Signal to Noise Ratio (SNR). We have implemented a Recurrent Neural Network for the task of signal detection. In the second part, we introduce a new concept of implementing deep learning for signal recovery and de-noising in a communication system using auto-encoders.