A novel jitter separation method based on Gaussian mixture model

Jitter is random variations in bit period of a digital signal. It may be regarded as single most limiting factor in high speed digital links. Separating jitter into its components and identifying their root causes help in improving phase locked loop design. Proposed methodology for separating and estimating total jitter is based on Gaussian mixture model (GMM). Expectation-maximization (EM) algorithm is used to find the maximum likelihood estimation of GMM parameter. The total jitter time series data is directly fit using the EM algorithm. The method eliminates problem of careful initial value selection for EM algorithm and automatically find the unknown number of mixing kernels using Bayesian information criterion (BIC). After finding the fitting parameter dual-Dirac Model can be used to calculate total jitter at the bit error rate (BER) level of interest.