A Statistical Approach to Probe Chaos from Noise in Analog and Mixed Signal Designs

Chaotic circuits have gained increasing attention in many engineering applications. Qualitative measures such as Lyapunov Exponent (LE) are the most common methods for identifying chaotic behavior. However, the use of these measures is limited due to the short output signal length and its contamination by noise. In this paper, we propose a novel methodology for modeling and detecting chaotic vs stochastic behavior in AMS designs. First, the design is modeled using a system of recurrence equations for analog and digital parts. Second, a surrogate generation method is performed. The obtained surrogates are a typical realization of the circuit output under the hypothesis that the circuits exhibits noise. Next, hypothesis testing with Gaussian Kernel measure as test statistic is conducted over these surrogates and the original circuit output to statistically assess the circuit behavior. The effectiveness of the proposed methodology is illustrated on several AMS circuits such as PLL or Colpitts oscillator. The obtained results show sufficient improvements over the existing methods. For instance, comparing with the LE method, our approach is an order of magnitude faster and provides a more accurate detection of the chaotic circuit behavior.