WEINSTEIN: EXPLOSIVE SOUND-SOURCE STANDARDS 



of propagation loss with good S/N ratio was surprisingly good; 

 essentially no bias and a standard deviation of 0.14 dB. 



The automated systems used to process a large quantity of data 

 are computer controlled but can employ either analog filtering or FFT 

 processing. In the latter case, the FFT spectral levels are summed 

 in the time domain to cover the total signal interval, and are 

 summed in the frequency domain to construct square 1/3-octave or 

 one-octave bands. A comparison of analog and FFT processing for 

 noise, with everything else the same, yielded a bias of about 0.3 dB 

 and a standard deviation of 0.8 dB in the 50-Hz 1/3-octave band. 

 For propagation loss, again with good S/N ratio, there was essentially 

 zero bias, and a standard deviation of about 0.5 dB. 



Christian (these Proceedings) has pointed out that in processing 

 one set of short-range recordings, the level changed by as much as 

 0.8 dB as the FFT bandwidth was changed. This observation is unex- 

 plained. There are a number of difficulties when we seek high 

 accuracy. Concerning FFT processing, it appears that most of our 

 knowledge is derived from consideration of long-duration Gaussian 

 signals. Explosive signals do not satisfy this criterion. They 

 consist of a series of short transients with deterministic spectral 

 characteristics. Specifically, we must know how the broadband FFT 

 levels depend upon: 



• The bandwidth selected for processing 



• The digitization rate 



• The number of bits 



• Whether Manning is or is not used 



• Whether coherent or incoherent summation in the 

 time domain should be used. 



