W911NF-17-S-0002 Learning Based Policies for Mobile Relaying (related to i.(2) Information Processing and Fusion, Active and Collaborative Sensing)
Abstract
Relaying is a well-established technique for improving quality of service in wireless communications, combating blockage and shadowing effects, increasing communication range and achieving efficient use of resources. Relay beamforming can achieve the relaying advantages with a reduced number of hops, which implies reduced signaling and scheduling overhead, reduced power consumption and lower latency. It can also be used to increase the rate of reliable information delivery (secrecy rate) to the intended receiver. Optimal positioning of relays is key in achieving good performance. In spatio-temporally varying channel environments, such as those experienced by mmWave signals in dense urban cities, relay mobility offers additional degrees of freedom to improve system performance. However, to reap the benefits of mobility, the relay control mechanism needs to be able to predict the evolution of the channel in time and space. Fortunately, in certain scenarios, the channel exhibits a spatio-temporal correlation structure, arising due to shadowing, which can be exploited to predictively estimate the best relay positions, or learn the best policy for selecting the best relay positions. The proposed project makes foundational contributions to the problem of cooperative beamforming and predictive spatial relay control in dynamic multiuser environments. Prior approaches have assumed complete knowledge of channel models, and/or were developed for single source-destination pairs. Compared to the existing literature, the proposed project will use partial information, or no information at all on the channel statistics, and will study more general scenarios, involving multiple source destination pairs. The specific objectives are (i) novel schemes for joint beamforming and predictive relay control by treating the channel as a spatio-temporal random process that follows a certain model with unknown parameters, (ii) novel schemes for predictive relay control that do not rely on channel models but rather use reinforcement learning (RL); (iii) novel RL-based relay control policies for cooperative jamming, aiming to achieve secret communications; (iv) feasibility study of the applicability of the proposed ideas to urban mmWave scenarios on Rutgers COSMOS testbed. The project will have significant impact on several applications of interest to the military. The proposed policies will enable relays, in the form of UAVs, to be quickly deployed in the battlefield allowing the fighters to remain connected even in difficult channel environments, and protecting their communications from eavesdroppers. The proposed project will enable mmWave technology to meet its goal of high rates in urban environments, which will be important in combat operations in cities. The contribution will be demonstrated via simulations corresponding to realistic channel scenarios, obtained via experiments at the Rutgers COSMOS testbed.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 25, 2021
- Source ID
- W911NF2110071
Entities
People
- Athina P. Petropulu
Organizations
- Army Contracting Command
- Rutgers University
- United States Army