Design and fault classification of charge pump PLLs using statistical learning

Phase-locked loops (PLLs) form an essential component of mixed-signal systems shared by communication and control systems as they commonly provide functionally related features such as frequency synthesis, phase alignment, and modulation/demodulation as well as numerous other uses. Because of their continued use in such applications, the demand for efficient methods of testing PLLs is becoming increasingly important both during prototype phase and in high volume manufacturing processes. All of these factors compound the effect of testing costs and complexity. A statistical learning-based fault diagnosis approach is proposed for charge pump (CP) PLLs, where Back Propagation Neural Network (BPNN) and Probabilistic Neural Network (PNN) are employed to distinguish different fault conditions . Simulation results indicate an overall 99.77% percent accuracy of fault classification as well as an average precision of 0.998, recall of 0.998, and F1 score of 0.998 for PLL fault classes indicating a high reliability of the proposed classification method. The results confirm that the developed PNN-based fault classification method is successful at classifying nonlinear patterns of fault characteristics as well as successfully accounting for variations in the input parameters to the CP PLL and to the PNN classifier. Since this proposed classification method for CP PLL faults is able to classify faults quickly, accurately, and automatically, it will improve the overall efficiency and reduce cost of testing PLL devices.