Initial Analysis of Acoustic Data from the Narcine Array

Initial Analysis of Acoustic Data
from the Narcine Array
Paul White
Institute of Sound and Vibration Research (ISVR),
University of Southampton
• Array design
• Introduction to the acoustic data
• Spectral Analysis
– Short-term
– Long-term
• Frequency-wavenumber analysis
• Beamformer results
Array Design
Sampling frequency = 10 kHz,
Bit depth = 24
Array design frequency ~ 1.7 kHz (Frequency such that d=0.45=l/2)
Data collected for 40 mins (most of the analysis here is based on 30 mins)
Time Series
• Example time series
Channels 1 and 16 unusual.
Channels 5 and 10 are typical of the
12 channels not shown.
Spectra (next slide)
• Computed using the standard segment averaging approach
(Welch’s method) with ~5 Hz resolution (2048 point FFT)
• Green lines: show the average of all the spectra across all the
• Blue lines: show the spectra of each channel
Tonal Noise
Effect of “rare” electrical
Median Spectra
• Computed as in a typical power spectra, with the difference
that the median of the segment’s FFTs are computed instead
of the mean.
• This means transient components do not significantly affect
the spectral estimates.
• (The for the median estimates are not computed to represent
a PSD – the values on the y-axis are not comparable to those
on the last slide).
Long-term Spectra
PSDs of 5 s blocks from channel 8 shown over 30 mins.
Electrical noise
Frequency-Wavenumber Analysis
• The wavenumber vector (k) defines spatial frequency.
• Temporal frequency, f, and the magnitude of the wavenumber
are related to each other via the sound speed c:
k 
• The component of the wave number in the direction of the
array is
2f sin  q 
karray 
where q is the angle of arrive of the acoustic wave.
Constraints in the f-k Analysis
• For any given frequency the wavenumber must satisfy:
 karray 
• Energy outside this region cannot be due to acoustic waves in
the medium.
– e.g. flow noise.
• Energy inside the region are from wave propagating at a
speed of c or greater.
Tonal electrical noise
Alias of own “ship”
f-k spectrum
Own “ship” noise
Low frequency
non-acoustic noise
Energy-Bearing Plots
Beamformer outputs.
Energy measured over 10 s segment
Tonal electrical noise
Example of Beamoutputs
Long-Term Spectrograms of Beam Outputs
Closing comments
• Some noise issues on the array:
– Low frequency noise, below 250 Hz (flow noise?)
– Some narrowband electrical noise.
• Lowest tone just below 1.5 kHz
• Narrowband filters may be adequate to remove this, if necessary.
– Impulsive noise.
• Seems to emanate from the autonaut.
• Effects can be reduced, in some circumstances, using median processing.
– Beamformers can be used to mitigate these noise sources.
• Some evidence of ambient noise changes can be seen in this