In order to estimate the full spectrum, It Is necessary to evaluate 

 the amplitude at several frequencies, spanning the range of the desired 

 spectrum. This Is accomplished by convolving a bank of band pass fil- 

 ters, centered at different frequencies, with the data and evaluating 

 the amplitude of the band-limited signals discretely. Knowing the 

 "power law" functional form of the spectrum In advance, one can fit the 

 several amplitude versus frequency estimates at discrete points In 

 space, using the Iterative regression technique described In Appendix A. 

 The regression coefficients a and b, now available at every point along 

 the profile, are often highly variable and must be smoothed. Also, 

 because the two parameters are statistically correlated, It Is prefer- 

 able to use the exponent of frequency (b) and total band-limited RMS as 

 detection parameters. Just as the presence of white noise at high fre- 

 quencies must not be Included In the regression analysis of the ampli- 

 tude spectrum, amplitude estimates at the noise level In the spatial 

 domain are also Ignored. The method Is described In detail In Appendix 

 B, along with the results of various performance tests on signals of 

 known properties used to calibrate the sensitivity of the detector. 



Generation of Amplitude Spectra 



Having delineated statistically homogeneous segments of data on the 

 basis of their estimated frequency spectra, the next step Is to generate 

 amplitude spectra from these segments. Were the spatial domain esti- 

 mates adequate, it would not be necessary to generate the spectra at 

 all. However, due largely to Instabilities In the slope (b) parameter, 

 those estimates are not adequate and true FFT' s must be run to estimate 



48 



