SVR-based RPD approach for complex processes and its application in circuit optimization

Abstract

During the lifespan of electronic products, the output voltage and current fluctuate due to the random fluctuations of parameter values of circuit components and environmental noise. Extant methods of circuit designs, such as parameter sweep and sensitivity analysis, are hard to obtain global robust optimization of output characteristics. This paper proposes a SVR-based robust parameter design approach to reach global circuit optimization. First, the approach fits an empirical model of process responses by using SVR. Next, it introduces the fluctuations of controllable factor variations and noise factors into response model by probability density functions, and calculates process means and variances by integration. Finally, it obtains optimal parameter combination by model optimization. An empirical study of the robust design of an inductor-resistor series circuit is conducted. The results show that the proposed approach not only avoids the disadvantage of ignoring interactions between factors when using parameter sweep and sensitivity analysis, but also overcomes the shortcoming of only achieving non-continuous optimization by Taguchi method and the limitation of obtaining local optimization by DRSM, and therefore, enhances the robustness of the circuit outputs.

DOI: http://dx.doi.org/10.11591/telkomnika.v11i3.2228