detectors disagree. All of the pertinent software Is listed la Appendix 

 B.2. The program runs with a core size of approximately 540 K bytes, 

 and was written In ANSI Standard FORTRAN (1977) for a ONIVAC 1180/2 com- 

 puter. The maximum profile length Is 4100 Interpolated data points and 

 145 points are lost from the end of the signal due to filtering. 



Just as an electrical engineer Investigates the performance of an 

 Instrument by operating on signals of known properties, the same tech- 

 nique can be used here to test the performance of the province picker 

 algorithm. While an engineer might use a step, ramp, or Impulse func- 

 tion as Input, random signals with known spectral forms are used In this 

 analysis. Many of the seemingly arbitrary choices of filters, averaging 

 procedures, and other design decisions Incorporated Into the present 

 province picker, were selected through feedback from performance tests 

 with known signals. 



An obvious choice of a signal with a known amplitude spectrum Is 

 random "white noise," with a continuous spectrum of zero slope. There 

 are several means of producing such a signal. The simplest Is to gener- 

 ate pseudo-random noise series of either a normal or uniform distribu- 

 tion. Figures B-2 and B-3 Illustrate such signals with their spectra, 

 which are Indeed relatively flat. Notice the large amount of scatter In 

 the amplitude estimates. An alternative method of generating "white 

 noise" Is actually to use a constant amplitude spectrum and uniformly 

 distributed random phase spectrum, separate their real and Imaginary 

 parts, and Inverse-transform the signals using the FFT. In this manner 

 a perfectly flat, non-varying spectrum Is assured, as Is Illustrated In 

 Figure B-4. This Is the signal source used In testing for this study, 

 and the algorithm Is presented In SUBROUTINE WHINOI. 



139 



