Towards Intelligent Next Generation Wireless Systems: Multi-Dimensional Intelligent Sensing and Signal Identification
By the year 2024, mobile data traffic in the Middle East and Africa region will grow nine times than what is today. Therefore, next generation wireless networks (NGWNs), starting with fifth generation (5G), are expected to immensely improve in terms of data-rates, latency, and reliability. Furthermore, with the advent of direct communication technologies such as machine-to-machine and vehicle-to-everything, wireless communications nodes will communicate directly in local premises, independent of core networks. However, these opportunities mean new challenges in terms of communications interoperability, coordination, interference management, and multi-dimensional (i.e., frequency, time, code, and space) resource sharing since the communications opportunities can be exploited jointly in all dimensions, thanks to the introduction of massive multiple-input and multiple-output, adaptive beamforming and tracking techniques. Ongoing extensive research indicate that the answer to these challenges will come from the artificial intelligence (AI) domain through advanced techniques that most probably will exploit deep learning (DL) and deep reinforcement learning (DRL) since DL learns from the features and DRL learns from the processes. On the other hand, until this day resource allocation has been handled in the upper layers and limitations in the physical layer were not addressed properly even though the bottleneck is the radio frequency (RF) condition. Thus, most of the research and development (R&D) efforts are not felt or directly valued by end users, and this has led to the stall or decline of average revenue per-user for service providers. This phenomenon, if not resolved timely, will hamper the investments for R&D and lead to catastrophic results. The solution is to grow the revenue by getting the services closer to end users as done by the Internet companies. It seems that this can only be achieved by making decisions in physical layer intelligently. The nature of the physical layer demands incorporation of intelligence in some form. A literature survey indicates that machine learning (ML) has been extensively used in cognitive radio networks in the upper communication layers, but application of ML at the physical layer is challenging owing to the complex and vastly varying nature of the channel environments. To address this, DL and DRL have lately been applied in many fields, and their potential application to physical layer problems was quickly recognized. However, so far, no effective DL or DRL based solution to the core issues of interoperability, coordination, interference management, and multi-dimensional resource sharing for 5G has been proposed..........