Digital signal processing methods of global nonparametric frequency domain audio testing

Abstract

In this paper we propose advanced digital signal processing methods of nonparametric frequency domain audio system distortion testing. The traditional frequency domain audio tests are based on the theory of electrical networks and are suitable for local distortion characterization only. Locality points at the dependency on simple distortion models and test signal characteristics. Local tests are of only low significance, if digital audio systems have to be tested. Iterative test methods are known which are based on representative models of weakly nonlinear systems. These tests can be extended to result in global characterizations that are obtained under typical operating conditions. Global tests estimate the frequency response and, based on this linear characterization, the power spectral density of the nonlinear distortion contributions. In addition to the already known high efficiency frequency response test signal types of multitone signals or chirp-signals, we propose the use of maximum length sequences which give superior results w.r.t. accuracy and convergence speed not only for measuring the frequency response but also for measuring the distortion power spectral density. If desired, the distortion power spectral density can be also measured as part of an offline or an online test setup with any appropriate signals like music signals.