Wireless Networks Design in the Era of Deep Learning: Model-Based, AI-Based, or Both?
This work addresses the use of emerging data-driven techniques based on deep learning and artificial neural networks in future wireless communication networks. In particular, a key point that will be made and supported throughout the work is that data-driven approaches should not replace traditional design techniques based on mathematical models. On the contrary, despite being seemingly mutually exclusive, there is much to be gained by merging data-driven and model-based approaches. To begin with, a detailed presentation is given for the reasons why deep learning based on artificial neural networks will be an indispensable tool for the design and operation of future wireless communications networks, as well as a description of the recent technological advances that make deep learning practically viable for wireless applications. Our vision of how artificial neural networks should be integrated into the architecture of future wireless communication networks is presented, explaining the main areas where deep learning provides a decisive advantage over traditional approaches. Afterwards, a thorough description of deep learning methodologies is provided, starting with presenting the general machine learning paradigm, followed by a more in-depth discussion about deep learning. Artificial neural networks are introduced as the peculiar feature that makes deep learning different and more performing than other machine learning techniques. The most widely-used artificial neural network architectures and their training methods will be analyzed in detail. Moreover, bridges will be drawn between deep learning and other major learning frameworks such as reinforcement learning and transfer learning. After introducing the deep learning framework, its application
to wireless communication is addressed. This part of the work first provides the state-of-the-art of deep learning for wireless communication networks, and then moves on to address several novel case-studies wherein the use of deep learning proves extremely useful for network design. In particular, the connection between deep learning and model-based approaches is emphasized, proposing several novel techniques for cross-fertilization between these two paradigms. For each case-study, it will be shown how the use of (even approximate) mathematical models can significantly reduce the amount of live data that needs to be acquired/measured to implement data-driven approaches. For each application, the merits of the proposed approaches will be demonstrated by a numerical analysis in which the implementation and training of the artificial neural network used to solve the problem is discussed. Finally, concluding remarks describe those that in our opinion are the major directions for future research in this field.