ANDERSON: VERTICAL NOISE DISTRIBUTION 



and the resulting spectra averaged with a three-point weight, simulating 

 Hamming shading on the basic time series. The resulting sidelobes in 

 frequency are down 25 dB, as shown for the 23 Hz passband spectrum in 

 Figure 7. This still leaves the possibility for out-of-band noise con- 

 taminating the filter output since the total noise band is 500 Hz and 

 the passband is 0.8 Hz. This factor must be considered in the sub- 

 sequent data interpretation. 



The next data problem was encountered during an attempt to con- 

 struct noise-power histograms. Figure 8a illustrates the type of 

 histogram obtained initially. Because the original 1000-point data 

 set was reduced to a single data point at each frequency of interest, 

 there was a total of only 300 such points for each histogram (corres- 

 ponding to 10 minutes of data) . The combination of quantization errors 

 and the logarithmic tabulation base resulted in some level-bins having 

 no possibility of being occupied. Basically, the level quantization 

 was taxing the available level accuracy, resulting in "noisy" histo- 

 grams like Figure 8a. This problem was corrected (Figure 8b) by 

 weighting the points and interpolating over the empty bins. These 

 corrected histograms were used to compute the mean power shown in 

 subsequent figures. 



Figure 9 displays noise power as a function of steered angle for 

 the three frequency bands of 23, 36 and 100 Hz. While at first these 

 data appear to show highly directional noise, peaked toward the hori- 

 zontal,, closer inspection showed that they were contaminated by 

 common-mode noise: a broadband noise with the same phase at each 

 element so that it appears as a horizontally arriving signal. In 

 fact, these noise patterns, especially at the two lower frequencies, 

 are quite similar to the beam-response patterns shown in Figures 4, 

 5, and 6 for a horizontal signal. 



867 



