Towards Breaking the Gridlock: Delay, Convergence, and Complexity in Highly Dynamic Tactical Networks
Abstract
The tactical wireless environment presents a number of key challenges: (i) A network environment that is highly dynamic and constantly changing, which means that by the time traditional resource allocation and control algorithms converge, the network is already operating in a new state, thus rendering these algorithms ineffective. (ii) Multi-hop and multi-path traffic with a diverse set of quality of service (QoS) requirements. While the research community has made substantial strides in developing throughput optimal algorithms, these approaches do not typically work well for other performance metrics, e.g., often resulting in large latencies. (iii) Low Overhead and Asynchronous requirements. These networks operate in environments where message exchanges between nodes, which corresponds to each iteration of a resource allocation algorithm, need to be small. This means that traditional network utility maximization algorithms that take a long time to converge, also require a substantial amount of overhead. Further, these algorithms typically require synchronization among the various nodes, again not comporting well with the military network requirement. (iv) Limited feedback overhead so that the protocols may need to operate without explicit channel or environmental information. To address these issues, our goal in this proposed effort will be to build the analytical foundations that are conceptually unifying, mathematically rigorous, and lead to the development of low-complexity and practically-implementable distributed control and learning algorithms that will provide a tactical advantage to the U.S. military networks. Toward that end, our proposed efforts are organized around the following two research thrusts: 1. Optimization-based Network Control: We will develop a new optimization algorithmic framework for distributed control of highly dynamic tactical networks, which will be built upon an inexact Uzawa Alternating Direction Method of Multipliers (ADMM) network optimization method that can simultaneously offer utility optimality, linear convergence rate, low latency, and low complexity. Our research will focus on exploiting the insights behind the inexact Uzawa ADMM method and fully harness the potential performance gains in non-stationary dynamic networks. 2. Learning-based Control in Highly Dynamic Networks: We will design network-centric learning techniques in which multiple network entities jointly learn and track the variations, correlations, and inter-dependence of the network state and environment, and leverage timely learned information to improve multi-dimensional performance metrics (e.g., throughput, delay, convergence, reliability) of network control algorithms.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 25, 2021
- Source ID
- W911NF2110244
Entities
People
- Ness Shroff
Organizations
- Army Contracting Command
- Defense Advanced Research Projects Agency
- Ohio State University