Cees Binkhorst
2014-04-12 09:46:21 UTC
http://iheartmatlab.blogspot.com.au/2014/04/analysis-of-suspected-mh370-pings.html
Wednesday, April 9, 2014
Analysis of suspected MH370 pings recorded by Ocean Shield
So this is a bit of a deviation from my normal MATLAB oriented posts.
In-fact MATLAB was not used once in this analysis!
I've been following the MH370 disappearance since it happened. With such
a lack of confirmed information there have been a lot of interesting
theories discussed on multiple forums (Whirlpool and PPRUNE are a couple
of interesting ones).
As I have a knowledge in acoustics (I am an Analyst / Software Engineer
for an underwater acoustics company and I specialise in tracking things
with pingers fitted to them), I've been trying to spread facts when I
know them and try and dispel myths.
Today I saw an interesting video from Sky News on a forum of the
suspected MH370 pings recorded by ADV Ocean Shield using the SUBSALV
Towed Pinger Locator TPL-25 (pinger audible 18 seconds in):
I thought I would see if I could glean any information from the audio in
the publicly released video. In particular I was wanting to look at the
inter-ping interval to see how steady it was to try and determine if the
noise is man-made or natural (ie marine mammal noise).
The signal is quite noisy with ambient ship noise and other transients
(scuffing of shoes on deck!) but a repeated click can be heard.
Process
Extracted the audio from the Youtube video using a web-based MP3
extractor
Opened the MP3 in Adobe Audition and trimmed the file to the time
segment corresponding to good audio (start time 41.787s from beginning
of file, duration 22.462 seconds)
Bandpass filtered the audio between 2kHz and 10kHz to remove LF
ambient noise (NOTE: the 'pinger' is audible at about 3.4kHz however it
obviously has been demodulated down from the raw signal observed on the
TPL-25 to bring into the audible range)
Filtered Time Domain Signal - Red ticks added at repeated 'ping' noise
A manual search of the waveform (zooming in tight where necessary)
was performed to measure the start time of each ping transient to within
a couple of wavelengths
Results
The detection times of the pings are listed in the table below along
with some basic statistics.
Ping # Time (s) Interval (s)
1 0.758
2 1.861 1.103
3 2.964 1.103
4 4.067 1.103
5 5.171 1.104
6 6.274 1.103
7 7.38 1.106
8 8.486 1.106
9 9.592 1.106
10 10.696 1.104
11 11.801 1.105
12 12.907 1.106
13 14.014 1.107
14 15.12 1.106
15 16.226 1.106
16 17.333 1.107
18 19.549 1.108
19 20.657 1.108
20 21.764 1.107
Mean 1.10544
Std Dev 0.00172
As you can see, the ping interval is very stable with an average of
1.105s and standard deviation of less than 2ms.
This is incredibly stable over the 20 seconds that I've looked at and
very unlikely to be a marine mammal (ie beaked or sperm whale) which DO
create broadband pulses at regular intervals, but I've never seen them
that regular.
This strongly suggests the recorded signal is man-made. Highly probably
it is from the Underwater Locator Beacon due to the very close match to
the expected ping rate of 0.9Hz or 1 ping per 1.111 seconds.
However it COULD be from other repetitive signal sources such as a stray
echo sounder nearby; although that would be a soul crushing
disappointment to all onboard ADV Ocean Shield and the other involved in
the search. Also.. WHAT CRAZY PERSON WOULD PUT A SOUNDER ON DURING AN
ACOUSTIC SEARCH!?!
One last thing to note about the observed ping intervals is that they
appear to increase slightly over time (order of 4ms). Whilst the sample
set isn't long enough to show consistency, I see patterns like these all
the time when tracking moving items fitted with pingers from stationary
receivers (conceptually the same thing as the stationary pinger and
moving receiver on ADV Ocean Shield).
It could be either:
- The decreasing battery voltage is causing the ping interval to
increase. However it doesn't seem likely this phenomenon would be
observed over such a short time window
- The audio recording segment corresponds to when the towed pinger
locator has performed a Closest Point of Approach (CPA) and is now
opening in range from the source. When this happens a effect similar to
Doppler occurs in that the time between successive pings appears to
increase as because between successive pings the receiver has moved
further away.
For the segment analysed it corresponds to a slant range increase of 6
metres over 20 seconds, or 0.3m/s or 0.6 knots. If the towed pinger
locator is transiting at 2 knots and the depth of the receiver is known,
you could estimate a very rough range to the source by looking at the
vector component in the horizontal direction. (its late where I am but
I'll try and do some example calculations tomorrow)
Posted by Rodney Thomson at 12:45 AM
Wednesday, April 9, 2014
Analysis of suspected MH370 pings recorded by Ocean Shield
So this is a bit of a deviation from my normal MATLAB oriented posts.
In-fact MATLAB was not used once in this analysis!
I've been following the MH370 disappearance since it happened. With such
a lack of confirmed information there have been a lot of interesting
theories discussed on multiple forums (Whirlpool and PPRUNE are a couple
of interesting ones).
As I have a knowledge in acoustics (I am an Analyst / Software Engineer
for an underwater acoustics company and I specialise in tracking things
with pingers fitted to them), I've been trying to spread facts when I
know them and try and dispel myths.
Today I saw an interesting video from Sky News on a forum of the
suspected MH370 pings recorded by ADV Ocean Shield using the SUBSALV
Towed Pinger Locator TPL-25 (pinger audible 18 seconds in):
I thought I would see if I could glean any information from the audio in
the publicly released video. In particular I was wanting to look at the
inter-ping interval to see how steady it was to try and determine if the
noise is man-made or natural (ie marine mammal noise).
The signal is quite noisy with ambient ship noise and other transients
(scuffing of shoes on deck!) but a repeated click can be heard.
Process
Extracted the audio from the Youtube video using a web-based MP3
extractor
Opened the MP3 in Adobe Audition and trimmed the file to the time
segment corresponding to good audio (start time 41.787s from beginning
of file, duration 22.462 seconds)
Bandpass filtered the audio between 2kHz and 10kHz to remove LF
ambient noise (NOTE: the 'pinger' is audible at about 3.4kHz however it
obviously has been demodulated down from the raw signal observed on the
TPL-25 to bring into the audible range)
Filtered Time Domain Signal - Red ticks added at repeated 'ping' noise
A manual search of the waveform (zooming in tight where necessary)
was performed to measure the start time of each ping transient to within
a couple of wavelengths
Results
The detection times of the pings are listed in the table below along
with some basic statistics.
Ping # Time (s) Interval (s)
1 0.758
2 1.861 1.103
3 2.964 1.103
4 4.067 1.103
5 5.171 1.104
6 6.274 1.103
7 7.38 1.106
8 8.486 1.106
9 9.592 1.106
10 10.696 1.104
11 11.801 1.105
12 12.907 1.106
13 14.014 1.107
14 15.12 1.106
15 16.226 1.106
16 17.333 1.107
18 19.549 1.108
19 20.657 1.108
20 21.764 1.107
Mean 1.10544
Std Dev 0.00172
As you can see, the ping interval is very stable with an average of
1.105s and standard deviation of less than 2ms.
This is incredibly stable over the 20 seconds that I've looked at and
very unlikely to be a marine mammal (ie beaked or sperm whale) which DO
create broadband pulses at regular intervals, but I've never seen them
that regular.
This strongly suggests the recorded signal is man-made. Highly probably
it is from the Underwater Locator Beacon due to the very close match to
the expected ping rate of 0.9Hz or 1 ping per 1.111 seconds.
However it COULD be from other repetitive signal sources such as a stray
echo sounder nearby; although that would be a soul crushing
disappointment to all onboard ADV Ocean Shield and the other involved in
the search. Also.. WHAT CRAZY PERSON WOULD PUT A SOUNDER ON DURING AN
ACOUSTIC SEARCH!?!
One last thing to note about the observed ping intervals is that they
appear to increase slightly over time (order of 4ms). Whilst the sample
set isn't long enough to show consistency, I see patterns like these all
the time when tracking moving items fitted with pingers from stationary
receivers (conceptually the same thing as the stationary pinger and
moving receiver on ADV Ocean Shield).
It could be either:
- The decreasing battery voltage is causing the ping interval to
increase. However it doesn't seem likely this phenomenon would be
observed over such a short time window
- The audio recording segment corresponds to when the towed pinger
locator has performed a Closest Point of Approach (CPA) and is now
opening in range from the source. When this happens a effect similar to
Doppler occurs in that the time between successive pings appears to
increase as because between successive pings the receiver has moved
further away.
For the segment analysed it corresponds to a slant range increase of 6
metres over 20 seconds, or 0.3m/s or 0.6 knots. If the towed pinger
locator is transiting at 2 knots and the depth of the receiver is known,
you could estimate a very rough range to the source by looking at the
vector component in the horizontal direction. (its late where I am but
I'll try and do some example calculations tomorrow)
Posted by Rodney Thomson at 12:45 AM