Age-of-information in wireless networks : theory and implementation

Abstract

Emerging data-driven applications will increasingly rely on sharing time-sensitive information for monitoring and control. Examples are abundant: mobile robots in automated warehouses sharing status information to cooperate with each other and with humans, self-driving cars exchanging safety-related information with other vehicles and infrastructure, and smart-cities analyzing data from Internet-of-Things (IoT) sensors to provide real-time feedback for vehicles and traffic management systems. In such application domains, it is essential to keep state information fresh, as outdated information loses its value and can lead to system failures and safety risks. The Age of Information (AoI) is a recently proposed performance metric that captures the freshness of information from the perspective of the destination.

Optimizing AoI is a challenging objective that goes beyond low latency, it requires that packets with low delay are delivered regularly over time to every destination in the network. In this thesis, we use rigorous theory to gain insight into the AoI optimization problem and to develop practical network control mechanisms, and we leverage system implementation to evaluate the performance of these mechanisms in real operating scenarios. We consider a broadcast single-hop wireless network with a base station and a number of nodes sharing time-sensitive information through unreliable communication links. We formulate a discrete-time decision problem and use tools from mathematical optimization and stochastic control to develop network control mechanisms that optimize AoI. Our first approach is to develop an algorithm that computes the optimal transmission scheduling decision at every time t.

As expected, this optimal solution is impractical due to its high computational complexity - shown to grow exponentially with the size of the network. To overcome this challenge, we propose low-complexity transmission scheduling policies with provable performance guarantees in terms of AoI. For example, we use Lyapunov Optimization to develop an AoI-based Max-Weight policy, show that this policy is optimal for symmetric networks, and show that, for general networks, this policy is guaranteed to be within a factor of two away from the optimal AoI. Numerical results suggest that this Max- Weight policy achieves near-optimal performance in various network settings.

Throughout the thesis, we analyze, optimize, and evaluate important classes of centralized and distributed low-complexity transmission scheduling algorithms, namely Max-Weight, Maximum Age First, Stationary Randomized, Whittle's Index, Slotted-ALOHA and Carrier-Sense Multiple Access, using tools from Dynamic Programming, Lyapunov Optimization, Renewal Theory and the Restless Multi-Armed Bandits framework. Leveraging the theoretical results, we propose WiFresh: an unconventional network architecture that scales gracefully, achieving near optimal information freshness in wireless networks of any size, even when the network is overloaded. We propose and realize two strategies for implementing WiFresh: one at the MAC layer in a network of FPGA-enabled Software Defined Radios using hardware-level programming, and another at the Application layer, without modifications to lower layers of the communication system, in a network of Raspberry Pis using Python 3.

Our experimental results show that the more congested the network, the more prominent is the superiority ofWiFresh when compared to an equivalent WiFi network, with WiFresh achieving two orders of magnitude improvement over standard WiFi. Our measurements suggest that WiFresh is well-suited for large-scale applications that rely on sharing time-sensitive information.