Coherence of Transients
Abstract
This technical report analyzes the performance of various methods of Squared Magnitude Coherency (SMC) estimation when the available data is very short. Three methods of Squared Magnitude Coherency estimation were examined. The first, Coherency estimation using the periodogram and a FFT (the frequency domain). The second, Fitting an AR model to the process, estimating its coefficients, and calculating the SMC from the coefficients (the time domain analysis). the third method is a direct coherency estimation via constrained least squares linear prediction. For a short data the SMC estimation using the time domain method has a large bias error, and a large confidence interval. The other two methods have reasonable performance for short data. Most of the previous work that calculated the SMC estimation confidence interval, assumed that to the estimated SMC is Gaussian distributed (Bendat & Piersol, Brockwell & Davis). Simulation results show that this assumption holds only for SMC values between 0.3-0.8, for long data as well as short data. Simulation results reveal that as soon as the number of samples is such that the law of large numbers holds (n greater then =40), all the expressions of the estimated SMC density function developed by Carter et al. and Sakai et al. still hold.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1990
- Accession Number
- ADA230168
Entities
People
- Amos Dotan
- William Hodgkiss
Organizations
- Scripps Institution of Oceanography