Optimizing Wireless Communication with Deep Learning in Pilot-Free Systems

In the presently used wireless communication systems, pilot signals are utilized for channel estimation and synchronization. However, such pilots introduce a lot of overhead, which in turn limits the spectral efficiency as well as the system performance. In this research paper, a new framework based on deep learning methodologies is introduced to use new wireless communication end-to-end systems that do not require pilot signals. The use of CNNs and RNNs are proposed within the methodology to co-design the transmitter and receiver elements of the system. This model is trained with extensive datasets in representing different channel environments and the model learns the channels' features and synchronizations implicitly. This end-to-end learning framework highly contributes to the flexibility and robustness of the communication system in terms of the discussed dynamic and heterogeneous nature of various applications. The work of the performance evaluation prove the efficiency increases BER and thru-put than traditional pilot based system, especially in high mobility and interference environment. Furthermore, the offered system has minimum latency compared to other existing systems and thus, makes it as a potential candidate for future generation wireless communication system like 5G and beyond. Exploring such opportunities, this research opens the way toward definable improvements in the nature of wireless communication paradigms and their applicative development both from methodological and technological viewpoints.