Machine Learning for Physical Layer Wireless Communications
Wireless communication technologies are characterized by a massive number of sophisticated interconnected devices, virtual/augmented reality, and internet-of-things. Modelling such a system to achieve high capacity, low latency, and massive connectivity has thereby become a challenging task.
Technologies such as massive multi-input multi-output (MIMO) have been proposed to satisfy the high data rate communication requirements. However, when the system operates in complex propagation environments with unknown channel properties, or when there are hardware impairments, traditional approaches, that rely on certain assumptions in order to maintain tractable mathematical models, exhibit some limitations in capturing the actual system properties. In such cases, data-driven approaches can be used to model and learn the complex properties of the environments and neural networks can be trained in order to complete a precise communication task, by capturing system properties, which might be modeled inaccurately, or not at all, by classical approaches. Moreover, conventional model-driven approaches often lead to highly complex algorithms in terms of computational complexity, and/or run-time. In this case, machine learning techniques can be more effective in modelling relationships between the inputs and the outputs of latency-critical applications.
Our research activities include machine learning for channel estimation, for channel prediction and for channel clustering.