Abstract

At the CCSL, we are engaged in research and teaching on wireless communication methods for future wireless communication systems. In future wireless communication 5G and beyond, an extremely high number of heterogeneous devices, such as smartphones, sensors, robots, and vehicles, will communicate with each other. Consequently, the need for higher data rates and lower latency will increase significantly, posing major challenges for resource allocation. Deep learning and data-driven algorithm approximation schemes have recently received significant attention as means to perform resource allocation with reduced complexity in 5G and beyond networks. This project considers the challenging case of Reflective Intelligent System (RIS) assisted, mmWave, frequency selective, massive MIMO systems with hybrid architecture and develops deep-learning based resource allocation frameworks. In these frameworks, prior data-set observations and deep neural network models will be leveraged to learn the mapping from received measurements to channels, beamformer, and power allocations. Furthermore, deep neural networks will be used to approximate the optimization problems by selecting the suitable parameters that minimize the approximation error. The usage of a deep neural network framework reduces the computational complexity and processing overhead since it only requires a limited number of layers of matrix-vector multiplications which can reduce processing time substantially.

Deliverables

1. This project aims to improve the overall performance on emerging and beyond 5G wireless systems in boosting system capacity with improved robustness and high data rates, via a low-cost, low-latency, and green implementation. For this purpose, deep learning or machine learning methods shall be applied, which can adapt to the dynamic changes of the wireless communication system and its environment and exploit the past experience to improve the future performance of the system.
2. A paper-style report documenting survey literature, simulation setup, and results are expected at end of the program.