the last part of the spectrum was smoothed by a three point running 

 weighted mean (0.25, 0.50, and 0.25) and then the last ten values 

 were averaged.* This average was treated as white noise and sub- 

 tracted from all spectral estimates. When the reduced values were 

 multiplied by the appropriate calibration curves, the usual result 

 was a fairly smooth spectrum that decreased regularly toward zero 

 values at high frequency. By such a procedure some of the spectral 

 values at high frequency will be negative. These values were auto- 

 matically set equal to zero in the rest of the computations. 



Inadvertently, Table 1 was applied where Table 2 ought 

 to have been applied to some of the spectra. The result was ex- 

 ceptionally high values at high frequencies. Thus the calibrations 

 given in these two tables do distinguish between the characteristics 

 of the two ships as, when done correctly, reasonable results were 

 obtained. 



Even with these corrections, there were a few spectra that 

 still became exceptionally large for frequencies greater than about 

 0.25 cycles per second. This behavior was apparently caused by the 

 original quality of the record and not by the digitization procedure. 

 These spectra were further modified by arbitrarily setting the cali- 

 bration curve equal to one above a certain frequency that was 

 selected by inspection of each spectrum. 



The result of such a sequence of computations should yield 

 fairly reliable spectral estimates for frequencies ranging from zero 

 to 0.25 cycles per second, but the values at high frequencies should 

 not be used to decide on any features of the high frequency end. 



*See p. 6 for details. 



