Ambulatory wireless sensor network power management -- a model-based predictive control approach

The characterisation of power control principles that underpin the performance dynamics of communication is an important feature in the user level in Ambulatory Wireless Sensor Networks. This thesis uses Model Predictive Control (MPC) framework to solve optimisation problems associated with constrained Received Signal Strength Indicator (RSSI) dynamics. Using local RSSI leads to decentralised power control. Sensor data transmitted to the remote base station arrives subject to some uncertainties due to channel fading and interference that frequently cause packet errors. Power is allocated optimally to the $n$ channels in a decentralised fashion by minimising a cost criterion subject to system constraints for ensuring a balance between power consumption and Quality of Service requirements. A Generalised Predictive Control (GPC) strategy is applied and the power control process is modeled based on a Controlled Auto-Regressive Integrated Moving Average form that takes into account the influence of the aforementioned uncertainties as an unknown output disturbance. For the optimisation of such a system, the minimisation of RSSI tracking error and the rate of power control update command that results in the lowest packet error rate and minimum transmit power level under specific sets of constraints is considered. The analysis is primarily validated via experiments using Tmote Sky sensor nodes. In addition, the resource constraint issue of wireless sensors due to memory usage and micro-controller capabilities is addressed, in which an on-line optimisation cannot be efficiently implemented. The optimisation problem is recast as a multi-parametric optimisation problem so that an explicit solution to the constrained GPC problem can be computed by solving a multi-parametric Quadratic Program (mpQP). The methodology is extended to incorporate a min-max MPC framework based on an uncertain linear state-space model of the constrained RSSI tracking error dynamics of each sensor. The resulting min-max optimisation problem is also posed as an mpQP and the solution of the mpQP is computed explicitly as a function of the initial state. The solution to the GPC mpQP and the min-max MPC mpQP is cast as a piecewise-affine function that is evaluated at each power control update. The resulting control algorithm reduces to an on-line evaluation of a lookup table that can be readily implemented on a wireless sensor. The control law is validated experimentally and its performance is shown to compare very favorably with alternative algorithms that have appeared in the literature.