PhD thesis Defense: Estimation and Control in Energy Harvesting Wireless Communication Networks
In the context of this PhD thesis defense, Miguel Calvo-Fullana presents his work realized at CTTC.
Long thought to be an unattainable ambition, self-sustainable and green-powered wireless networks are rapidly becoming a reality. This is driven by recent hardware improvements in what is known as Energy Harvesting (EH). This technology makes it possible for devices to scavenge energy from the environment, be it from solar, thermal, kinetic or other sources. One area where this idea has shown considerable promise is in Wireless Sensor Networks (WSNs). These networks consist of inexpensive, small and low-power sensors, making them a prime candidate for the deployment of energy harvesting technologies. However, when devices are equipped with these new technologies, the intermittent and random nature of the energy supply makes it necessary to take a new approach to the design of communication policies.
It is the main objective of this dissertation to study, evaluate and solve problems that arise in wireless sensor networks with energy harvesting capabilities. The nature of the problems studied can be grouped into two categories. On one hand, we address problems of estimation, which arise when EH sensors collect measurements of some physical phenomenon. On the other hand, we study problems of control, which emerge when EH sensors are part of a dynamical system.
First, we address the estimation problem in EH-powered wireless sensor networks. We approach the problem in a coded manner, where sensors transit their measurements to a Fusion Center (FC) digitally. We consider a point-to-point sensor-to-FC communication scenario, where the measured sources are time-correlated. We derive the transmission policies minimizing the average reconstruction distortion for both delay-constrained and delay-tolerant scenarios. Next, we study the case in which multiple sensors collaborate in the estimation of a source. In this problem, only a limited number of sensors can transmit simultaneously, due to the reduced number of sensor-to-FC channels. The goal is to jointly design the power allocation and sensor selection policies that minimize the average reconstruction distortion. However, this problem is not convex. To overcome this, we propose two policies, an iterative joint policy that finds a stationary solution of the original problem; and a heuristic separate policy in which the optimal power allocation is given by a convex optimization problem. Both policies are related to each other in the fact that the latter can be used as an initialization point of the former iterative policy. Further, as an alternative approach to the problem, we have also proposed the use of sparsity-promoting techniques.
Then, we turn our attention to problems of a control nature in energy harvesting communication networks. First, we study the problem of jointly routing and scheduling traffic in a communication network with EH-powered nodes. The routing-scheduling policies proposed act as a generalization the stochastic backpressure policies to energy harvesting communication networks. Specifically, we provide two policies, an easy to compute policy and a randomized policy with improved stability guarantees. Furthermore, we ensure that given sustainable data and energy arrivals, the proposed policies stabilize the data queues over all the network. Finally, we study a more general control problem in which energy harvesting sensors share a wireless medium over which they transmit measurements to their respective controllers. Since the medium is shared, simultaneous transmissions might lead to packet collisions. To overcome this issue, we propose the use of a random access scheduling policy. Furthermore, we show that given sustainable energy and stability requirements, the policies stabilize all the control systems while satisfying the energy harvesting constraints.