Research

Machine Learning Approaches for Communications

Data traffic of wireless mobile networks is dramatically increasing, but available resources, such as time, bandwidth, and transmission power are limited. Therefore, efficient resource allocation schemes are becoming increasingly important.
However, since wireless mobile networks have highly complicated features, such as numerous nodes, nodes’ mobility, channel variation, and channel interference, etc., the optimal resource allocation is a challenging problem.
Recently, machine learning techniques are achieving remarkable success in many application domains. Machine learning does not require accurate network models because it can estimate the hidden relationship between voluminous input data and complicated system outputs. Moreover, some machine learning techniques, e.g., reinforcement learning, can automatically adapt the learning results to the new environments. These features enable machine learning approach to efficiently solve the resource allocation problems in complex, dynamic and time-varying networks.

Recently, applications of vehicular networks for future intelligent transportation, such as advanced driver assistance on active safety and traffic efficiency, are emerging.
To support these applications, vehicle-to-everything (V2X) communication has been proposed, which includes: 1) vehicle-to-vehicle (V2V), 2) vehicle-to-pedestrian (V2P), and 3) vehicle-to-infrastructure (V2I) communications.
To deal with high mobility of vehicles and support safety-critical applications, V2X communications need to satisfy strigent quality-of-service (QoS) requirements: ultra low latency and high reliability.

Recently, mobile users' demands for computation-intensive applications such as VR/AR and speach recognition are more and more increasing.
Those applications usually require huge energy consumption and rigorous delay constraints.
However, due to mobile devices' limited resources, e.g., battery energy and computation speed, it is challenging for mobile devices to support the strict requirements on energy and delay.
Mobile edge computing (MEC), which enables the mobile users to offload part or all of their computation works to the nearby edge servers with powerful computation capability, has been considered as a promising way to address this issue.
By offloading their computation works to edge servers, mobile users can reduce the energy consumption and delay for executing the computation-intensive applications.

The amount of energy consumed by mobile users is increasing rapidly, but the development of battery capacity has not kept up with their energy-supply demands.
Wireless energy transfer (WET) technology can significantly extend the battery lifetime of wireless devices by supplying them with energy.
Combined with wireless information transmission (WIT) in communication systems, WET has newly introduced system architectures, e.g., simultaneous wirless information and power transfer (SWIPT) and wireless powered communication network (WPCN).
In a SWIPT system, a transmitter broadcasts radio frequency (RF) signals that carry both information and energy at the same time, and then wireless devices split the received signal into two parts, one for information decoding and the other one for energy harvesting.
In a WPCN, wireless devices first harvest energy from the RF signals broadcasted by a transmitter and then use the harvested energy to transmit data signals to the transmitter.

The increasing demand of mobile Internet poses challenging requirements for wireless communications, such as high spectral efficiency and massive connectivity.
In conventional orthogonal multiple access (OMA), where different users are allocated orthogonal resources in either the time or frequency domain, the number of supported users is limited by the number of available orthogonal resources, which may hinder the efficient use of the limited resources.
Thus, we consider non-orthogonal multiple access (NOMA), which uses superposition coding at the transmitter and successive interference cancellation (SIC) at the receiver.
NOMA can support more connections by letting more than one user simultaneously access the same frequency or time resources, which leads to higher spectral efficiency.