It is sometimes necessary to repeat analyses using different 

 filters before the optimum one is found. Figure 4 shows the 

 effect of different filters operating on the same input data. 



The oscillator motor determines the time it takes the oscillator 

 to scan the entire frequency range, which consequently determines the 

 maximum analysis time. There are two choices: 11 minutes (fast scan) 

 or 22 minutes (slow scan). Slow scan allows the filter to remain 

 longer in its local frequency environment and "permits better resolution. 

 Fast scan may have a relative "smearing" effect because new frequencies 

 are being scanned by the filter at a faster rate and corresponding 

 new amplitudes are being squared and averaged at the same faster rate. 

 The error or "smearing" thus introduced depends on the steepness of the 

 spectrum. A flat spectrum is unaffected by fast scan. Figure 5 shows 

 the same data subjected to slow and fast scan. In either case, it is 

 important for the filter to scan the entire loop, before the frequency 

 band it is examining moves very much, to get the benefit of all the 

 available data. For this purpose the time (T) it takes the data to 

 make one passage through the analyzer is important. A rule of thumb 

 that suggests TAf» 1 may be used as one guide in the selection of 

 an appropriate filter bandwidth (^f). A 20-minute record, for example, 

 which is sped-up 512 times takes 2.34 seconds to pass through the 

 analyzer once (loop time). A 5-cps filter would appear to be adequate 

 in this case. Choice of both filter and scanning speed depends mostly 

 on the shape of the spectrum. A flat spectrum can tolerate a wide filter 

 and fast scan; a steep, rapidly changing spectrum requires a narrow 



13 



