amplified and made to drive a vibration galvanometer. It is 
clear that if there is a component in the record having N 
complete cycles in the peripheral length of the wheel, this 
will produce a resonance of the galvanometer at its natural 
freouency of p cyc./sec. when the wheel is rotating at a 
speed of p/N rev./sec. The wheel is made to revolve at a 
speed which gradually decreases from a high value and the 
vibration galvanometer performs a series of transient 
resonances, one for each periodicity in the record. The 
resonances of the vibration galvanometer are converted to 
an electrical signal which drives a pen recorder, and the 
curve drawn by this pen is a series of peaks which constitute 
a Fourier amplitude spectrum on the curve on the record. ...." 
The envelope of the individual spikes in the record would seem 
to be related to the power spectrum of the record. The width and 
shape of the spike would therefore be related to the band pass filter 
of the analysis and the figure suggests that the resonant galvano- 
meter is very sharply peaked and responds to an extremely narrow band 
of the power in the wave record. Note how the amplitude of the record 
falis down to very low values on each side of each peak. 
Now note how extremely irregular the envelope of the peak appears 
to be. From 1700 to 1900 in the first two spectra marked gaps appear 
inside of the range of # where one would expect only minor variations 
from the theories contained in this paper. If the irregularities 
were to reflect actual physical changes in the record, this would 
be most disconcerting, but they really do not. 
The irregularities from record to record and from point to point 
ir tse same record are simply due to too great a resolution for too 
small a record length. The wave records were 20 minutes long and 
there are about 15 spikes between 27/15 and 27/12 in the spectra 
shown. This suggests a band width of the analysis given byApw equal 
to 27/4-15-15. From equation (10.39), and since 20 minutes times 
118 
