Performance Testing of Cellular Communication System with Change-point Detection
Examensarbete för masterexamen
Complex adaptive systems (MPCAS), MSc
This thesis discusses the use of change-point detection in an automated performance testing environment for WCDMA base stations. It reviews three existing methods for change-point detection namely least-squares importance fitting, Bayesian online change-point detection and wavelet footprints. It also introduces a novel method based on a penalty formulation of average variance which is optimized with evolutionary optimization. Benchmarking of the algorithms is performed by using binomial data and first performing a parameter sweep to set the false detection rate to a constant value. Secondly all algorithms are tested to determine the detection rate given a specific false detection rate. Wavelet footprints exhibited the best performance in terms of the detection rate for a specific change-point size. It had a 90% detection rate for change-points of 12%. Average variance with penalty performed the worst and had approximately half the relative performance than wavelet footprints. Bayesian change-point detection and least-squares importance fitting had really similar performance and did not exhibit any position dependence in terms of false alarms whereas the other methods did. The algorithms have unique strengths and weaknesses but can all be used in an automated testing environment. The use of such algorithms seem to be a promising way of analyzing large time series in performance testing environments.
Signalbehandling , Transport , Signal Processing , Transport