A closer look at the LANL report candidate event

After months with no CTBTO access (and other pressing projects) , the acoustic search is back on the front burner.

Using Rottnest hydrophone recordings provided by IMOS, a possible candidate is being examined that may match the LANL report timing.

The LANL study was looking for impact events only, focusing on Leeuwin HA01 hydrophones. The HA08 recordings from Diego Garcia were dismissed due to seismic survey noise. LANL looked at recordings from 00:36:40 to 00:57:05 and analyzed one event in particular that arrived HA01 at 00:51:59 GMT. The event appeared to be partially masked by an Antarctic ice event at bearing 190.5 degrees. The preamble to that event did not correlate well, and was visually aligned to derive a bearing of 246.9 degrees, which pointed to the far Southern portion of the 7th Arc.

The Rottnest 3376 recorder captured a faint signal at 0054:21 that matches the signature of the LANL signal just 5 seconds shy of the predicted arrival time there. Rottnest was experiencing what appears to be surface noise from breaking waves at that time, but noise reduction provided a weak broadband sound.

LANL used a group velocity wave speed of 1.46 km/s, which it turns out is rather low for the region. Using an revised wavespeed of 1.477 provides an estimated time difference between the signals of 140.8 seconds. Using the 1.477 group velocity again turns that into a path difference of 207.96 km between HA01 and Rottnest for the event. The path difference defines a shallow search path arc that crosses the 7th Arc at a more Northerly bearing of 250.0 degrees from HA01. The estimated coordinate of S 39.52 E 85.40 is about half the distance to the active search area than the original LANL estimate. This location coincides closely with the Dr Ulich 2015 contrail result.

Looking more closely at the arrival timings, it appears that the plane would have gone down just before the LANL candidate event. Even with generous wave speed estimates, the plane would have been down before the 7th arc for anywhere South from the LANL crossing. This is important because it the plane would have been exactly on the 7th arc when it pinged. That narrows the search area there to within the accuracy of the satellite timing calculations that define the 7th Arc. Moving North along the arc expands the search area by the distance the plane could have gone after the 7th Arc timing. At the revised estimate for the candidate (if it indeed matched a Rottnest sound), the search area expands to a maximum of about 50 km from the 7th Arc. The 250.0 bearing would be the search path direction, starting about 50km out and crossing the 7th arc. The width of that path is relatively narrow and bounded by the wave speed estimate, which could be further refined. With calibration by a detonation, the search zone could be narrower than the depth of the water.

Another interesting conclusion from the acoustic arrival timing is that given the closest distance from HA01 perpendicular to the 7th Arc of 1553 km, the sound travel time is about 17:23, which added to the 7th arc time of 00:19:37 gives an earliest possible HA01 time of 0037 GMT.

Some novel techniques for sifting through the sometimes noisy spectrograms have been explored. While exploring the mapping of the triad of hydrophone signals into RGB images to use color fringing for flagging the bearing, it was realized that the time delays form a left-right shift in the sonograms that is very similar to a parallax shift when viewing a 3D scene. Mapping pairs of sonograms into a stereo viewer (or going walleyed) produces an intriguing display where the left-right arrival time becomes a depth effect. The effect can be amplified by zooming into the scene. The brain and visual system are naturally adept at sorting out a mixture of overlapping signals. Overlaying a grid of colored dots helps to provide a depth reference.

July 2016 Update:

Work has been ongoing with development of two visualization tools in Octave/MATLAB for examining the signal components. Below is a plot of the the LANL 00:52 signal showing bearing detections over time. Good detail can be seen in the preamble portion of the signal. No significant coherence is seen at the proposed 246.8 degree bearing, but there is a strong component at 30 degrees, with weaker elements at 67 and 190 degrees.

LANL 00:52:00 7th arc candidate showing bearing detection over time

The three hydrophone pairs are shown in Red, Green, and Blue color channels. Where they arrive at the same time with a good coherence and intensity match, the result is monochrome. Each pair can detect a bearing, but with a mirror image in another quadrant. Ambiguity in the plot can be usually be resolved by tracing the arrival signal in each color channel to see if there is a stronger component at a different bearing.

Here are correlation plots aligned at the 190 degree bearing with different window sizes for comparison:


August 3 Update:

A Hydrophone Data and MH370 meta analysis by Richard Godfrey dated Jul 28 was published on Aug 1. It is based on the later 2014 LANL report LA-UR-14-28179 on 7th Arc Noise Sources. The Godfrey report essentially narrows the 7th arc timing and assigns bearings with lat/lon ranges to the arrival times of LANL cluster 3.

One problem with the Godfrey report is that the reported H01 distances for three latitudes on the 7th arc do not match with the known locations of the 7th arc and the H01 array, by 21-22 km at each spot. An error in the 7th arc or H01 placement alone would not account for that, giving distorted reverse triangulations. It is most likely that Godfrey described distances measured using a method different than the great circles used by Google Earth.

Checking further, other sources give distances close to Google Earth. The online geo.javawa.nl distance tool using GeographicLib is within 2 km of Google Earth (which would increase the error to 24 km).

The Godfrey report rounds the 1.466 km/s LANL wavespeed to 1.47 km/s, but the correct value from WOA speed profiling is currently estimated at 1.484 km/s. The LANL ranging error is estimated at 30 km but they described taking this into account. The error from using 1.47 km/s in the Godfrey analysis would shift those results 35-55 km southward along the 7th arc.

Three signal clusters as shown in Figure 4 of LANL report LA-UR-14-28179

LANL Clusters

The LANL noise cluster 1 was the 00:50:00 00:52:00 ice event analyzed above on bearing 190. [ n.b. typo found Aug 2020 ]

Carefully examining LANL clusters 2 and 3 also excludes them from 7th arc bearings. Taking a close look at the LANL Figure 4 timing referred to by Godfrey, cluster 2 is at 00:49:41 on the iceberg bearing 205.5, with another signal following 3.5 sec later at bearing 208-210. The arrival time implies an expected 7th arc bearing of 249-250. No signal components in that range are visible.

A plot showing bearing components of LANL cluster 2
A plot showing signal components by bearing of LANL cluster 2

LANL cluster 3 is for a wider range from 00:39:28 to 00:39:49 centered approximately on a strong signal at 00:39:40 from bearing 196 degrees which would be another ice event. Arrival timing from the 7th arc would have a bearing of 269-270 degrees. The ice signal dominates the cluster. A weaker signal visible at 00:39:44.5 on bearing 267.2 might be of interest but the impact time on that bearing would put the plane in the water 30 seconds before the last ping.

Bearing components of LANL cluster 3 over time
Bearing components of LANL cluster 3 over time

This is all based on reasonably accurate estimates for sound speed and hydrophone placement, of course.

Mar 2019: Significantly improved bearing plot for LANL Cluster 2. Note that the axes are swapped. (click for a larger .pdf file):

Eigenvalue filtered covariance plot of Bearing vs Time for the LANL Cluster 2 event at 00:49:41 UTC.

There are still no components of any signal from bearing 248-250. The new plot reveals LANL Cluster 2 as two separate events from bearings 205.5 and 210.4 degrees.

Mar 2020: Improved bearing plot for LANL Cluster 1. (click for a larger .pdf file):

Improved Bearing vs Time plot for LANL Cluster 1 at 00:50 UTC

This enhanced plot uses covariance of the phase differential, which approximates instantaneous frequency correlation. With this plot it becomes clearer that there is no signal energy from bearing 248 with more than a half second of coherence. The first portion of the signal that was manually aligned has a bearing of 57 degrees, then the overlapping main ice event at 190.6 degrees.

Jun 2020 update: A cleaner bearing plot for LANL Cluster 1 (click for a larger .pdf file):

A cleaner Bearing vs Time plot for LANL Cluster 1 at 00:52 UTC

A new coherence algorithm produces a much cleaner bearing plot of the LANL Cluster 1 signal at 00:52:00 UTC. The colored alias artifacts of previous methods are gone, and lower noise levels reveal weaker events. The arrowhead shape is caused by the frequency dispersion on ice events, where lower frequencies have wider secondary correlation peaks. There are still no significant energy components at other bearings mixed with the main bearing 190.6 signal.

August 2020 Update: A plot of the LANL clusters time period.

Refined Aug 2020 plot matcing the LANL clusters time period.

This new plot shows the same time period covered by the LANL report LA-UR-14-28179 describing three clusters of signal arrivals. (Click the image for a higher resolution .pdf file.) The coherence algorithm used to generate the image looks for covariance of STFT (Short Term Fourier Transform) with a broad window of about 16 seconds. This has significantly reduced false aliasing compared to time domain covariance, with the added advantage of allowing an extended frequency range. A gamma brightening function has been applied to reveal the background noise level.

Noteworthy in the plot is that there are no apparent signal components shown coming from bearing 246.8 at 00:52:00 as reported in 2014 by LANL, or bearing 234.6 at 00:50:00 as reported by Usama Kadri in 2017 and 2019. All of the signals and clusters reported by others in this time frame can be seen to arrive from directions south of the Cape Leeuwin hydrophone array, which classifies them all as typical ice events.

The sensitivity of the plot does show far weaker signals coming from west of H01, which could be analyzed in more detail.