Concluding 50 Years of Research in Wireless Communications: Algorithms for Artificial Intelligence and Optimization in Networks Beyond 5G and Thereafter

点击翻译按钮获取中文摘要

The proposed project will conclude our ongoing research on algorithms for networks beyond 5G and thereafter. Our specific focus will be on the topics of spectrum monitoring, resource optimization in edge networks, and device localization. In networks beyond 5G, we will continue to see a surge in the number of wireless devices. For optimal spectrum utilization, a critical step is to monitor the spectrum actively, detect all the signals present and map them to their transmitters. In recent work, we explored deep learning for this task and proposed a spectrogram-based solution to detect Wi-Fi-type signals. We will now make architectural extensions to detect other heterogeneous signals like Bluetooth, as experienced in a real-life scenario. We will pursue protocol-level and device-level classification using defining features such as the wireless packet inter-arrival times, the number of subcarriers, and hopping periods. We will also develop denoising algorithms to alleviate the performance degradation in the low signal-to-noise ratio region. The number of services offered at the network edge keeps growing. For optimal resource utility in networks beyond 5G, we need fair and efficient resource allocation algorithms at the edge. In recent work, we investigated the joint edge resource management and pricing problem and proposed a game theory-based solution. We will now extend the work to include the service placement cost and the uncertainties due to demand variability, node failures, and fluctuating resource prices. We will tackle the resulting complications using techniques from stochastic, chance-constrained, and risk-constrained optimization. We will also develop a robust optimization framework for the service provider to determine the optimal locations for service placement and the amount of resources to purchase from each location. Accurate localization will be crucial in networks beyond 5G due to the growing need for device sensing and cooperation in low-latency communications. In recent work, we proposed machine learning algorithms to locate the devices transmitting at a fixed power level in a 5G system. We will now extend the work to heterogeneous devices with diverse antenna gains and power control mechanisms. We will seek improvements in the localization accuracy through fusion techniques, which augment the signal strength with angle and time information. Towards developing efficient tracking algorithms, we will model the device mobility using appropriate state-space models and design Bayesian filters that can recursively estimate the time-varying device location. Network operators across Canada strive to deliver various location-aware services while providing quality applications at affordable rates. The proposed research will immensely benefit them in growing their subscription base. Our spectrum monitoring methods will also help detect malicious transmitters and protect our safety-critical airspaces such as airports.