Optimal Settings for Measuring Frequency Response Functions withWeighted Overlapped Segment Averaging

This paper investigates the measurement errors involved in estimating frequency response functions - and related quantities such as the coherence function - from weighted overlapped segment averaging, a technique that has become a standard in modern data analysers due to its computational advantages. A particular attention is paid to leakage errors, for which this technique has been frequently criticised. Our main result is that a half-sine or Diff window with 67% overlap achieves the best compromise to reduce leakage errors in the case of stationary random excitations, and this is independently of the system frequency response function. This conclusion is to be contrasted with the customary habit of using a Manning window with 50% overlap. The same reasoning confirms that a rectangular window with 0% overlap minimises measurement errors in the special case of multi-sine excitations