Microprocessor based signal processing techniques for system identification and adaptive control of DC-DC converters

Abstract

Many industrial and consumer devices rely on switch mode power converters (SMPCs) to provide a reliable, well regulated, DC power supply. A poorly performing power supply can potentially compromise the characteristic behaviour, efficiency, and operating range of the device. To ensure accurate regulation of the SMPC, optimal control of the power converter output is required. However, SMPC uncertainties such as component variations and load changes will affect the performance of the controller. To compensate for these time varying problems, there is increasing interest in employing real-time adaptive control techniques in SMPC applications. It is important to note that many adaptive controllers constantly tune and adjust their parameters based upon on-line system identification. In the area of system identification and adaptive control, Recursive Least Square (RLS) method provide promising results in terms of fast convergence rate, small prediction error, accurate parametric estimation, and simple adaptive structure. Despite being popular, RLS methods often have limited application in low cost systems, such as SMPCs, due to the computationally heavy calculations demanding significant hardware resources which, in turn, may require a high specification microprocessor to successfully implement. For this reason, this thesis presents research into lower complexity adaptive signal processing and filtering techniques for on-line system identification and control of SMPCs systems. The thesis presents the novel application of a Dichotomous Coordinate Descent (DCD) algorithm for the system identification of a dc-dc buck converter. Two unique applications of the DCD algorithm are proposed; system identification and self-compensation of a dc-dc SMPC. Firstly, specific attention is given to the parameter estimation of dc-dc buck SMPC. It is computationally efficient, and uses an infinite impulse response (IIR) adaptive filter as a plant model. Importantly, the proposed method is able to identify the parameters quickly and accurately; thus offering an efficient hardware solution which is well suited to real-time applications. Secondly, new alternative adaptive schemes that do not depend entirely on estimating the plant parameters is embedded with DCD algorithm. The proposed technique is based on a simple adaptive filter method and uses a one-tap finite impulse response (FIR) prediction error filter (PEF). Experimental and simulation results clearly show the DCD technique can be optimised to achieve comparable performance to classic RLS algorithms. However, it is computationally superior; thus making it an ideal candidate technique for low cost microprocessor based applications.