Replication of the HSCA Weiss & Aschkenasy Acoustic Analysis
Michael O'Dell
“Scientific
researchers propose hypotheses as explanations of phenomena, and design
experimental studies to test these hypotheses via predictions which can
be derived from them. These steps must be repeatable, to guard against
mistake or confusion in any particular experimenter.” –
Wikipedia article on “Scientific method”.
History
In 1978 the House Select Committee on Assassinations (HSCA) concluded
that there was probably a conspiracy involved in the assassination of
President Kennedy, a conclusion primarily based on the acoustic
evidence contained on the Dallas Police Department radio recordings.
An unknown motorcycle tuned to channel I had a defective microphone
button that caused it to continuously transmit over a five minute
period during which the assassination took place. If this motorcycle
had been part of the motorcade it might have picked up sounds of the
gunshots. If true, those sounds could be used to determine how many
shots were fired, their timing, and using echo location methods, the
source of the shots.
Working for the HSCA, a team (BRSW) from Bolt, Beranek and Newman Inc.
(BBN) headed by James E. Barger, studied the recordings. BRSW performed
a series of test shots in Dealey Plaza and used recordings of these
shots to compare with signals on the DPD recordings. BRSW concluded
that channel I contained impulses probably caused by the gunshots, with
a 50% probability that one shot came from the grassy knoll in front of
the President. Because of the level of uncertainty in this finding the
HSCA asked another team try to raise the confidence level of the
results. Weiss and Aschkenasy (W&A) used an acoustic modeling
method and concluded there was a 95% probability of a shot from the
grassy knoll.
Science and Reproducibility
The BBN tests compared impulse patterns from the Dictabelt recording to
patterns from live test firings done in Dealey Plaza. Recordings
or other records of those test firings are no longer available.
Without them there is no practical way to attempt a replication of the
BBN procedure. However, it was not the BBN tests that determined
there was a shot from the grassy knoll or a fourth shot. The BBN
tests were inconclusive on these points. Those conclusions came
from the W&A analysis. For this reason, although it may be
enlightening to obtain those test fire recordings and they may be
necessary to replicate the BBN test, they are not necessary for
evaluating the evidence for a grassy knoll shot. It is also why
any falsification of the BBN tests does not automatically falsify the
HSCA acoustic evidence conclusion.
Despite the historic significance of these tests and despite years of
back and forth scientific debate about them since they were done,
nobody has replicated the experiments. It does seem to be a
strange oversight.
Reproducibility is a fundamental principle of the scientific
method. An experiment performed only once has little
significance. Science is a human endeavor and subject to human
frailties. Replication is one of the ways that science controls
for human error.
The Current Effort
In 2007 I acquired a full sized copy of the Drommer & Associates
survey map from Anthony Marsh. This is the same map that Weiss
& Aschkenasy used to perform their analysis. It gave every
appearance of being the same map, to the same scale. It matches
the image of the map that was presented in Weiss &
Aschkenasy’s testimony before the committee. Known
distances of features, like the widths of buildings, were confirmed on
the map. It has the same scale, 1 inch equal to 10 feet.
Since acquiring the map it had been a goal to one day attempt a
replication of the W&A analysis. There has been discussion
for years about additional tests that could be done to help clarify our
understanding of the acoustic evidence. There have been efforts
to acquire the test shot recordings that were made in Dealey Plaza in
1978. There have been suggestions about analyzing the other
suspect shots the same way Weiss & Aschkenasy did for the suspect
Grassy Knoll shot. These could be worthwhile efforts, but
replication of the Weiss & Aschkenasy experiment could have been
done at any time.
Not long after receiving the map I made some preliminary calculations
of some echo paths presented in the W&A report and got results
indicating buildings might be in the wrong place. This was
puzzling, and I wanted to double check the accuracy of the map. I
asked Dale Myers to confirm some long distance measurements of
buildings in Dealey Plaza. Dale made his own extensive surveys of
the plaza while building his virtual recreation so this would be
independent confirmation. Again measurements on the map were
confirmed. The discrepancies with the W&A report remained
unexplained, but I did not pursue them further at the time.
I assumed that replication would not be a problem. The strange
initial results I had gotten that prompted confirmation with Dale I
attributed to some unknown error that would work out once I really got
into it. There are some easily observed errors in the W&A
report as well, but I thought those could be resolved and not be a
barrier to replication. The intention was to build a computer
model to calculate echoes in Dealey Plaza, replicate the W&A
results, and then use the model to test additional scenarios and
locations that W&A were unable to check. However, this
subject is just one of many things in which I’m interested and
there are always projects to be done. The enthusiasm to start on
this just wasn’t there and years went by without doing anything.
In late 2012 I began receiving inquiries from media people and authors
that were working on projects for the 50th anniversary in 2013. I
realized that it was time to get this project done.
Using Excel for the model was always my intent. When Jefferson
Morley came out to interview me in 2006 for his Playboy article I
showed him a crude Excel spreadsheet calculating echo paths. I
doubt he thought much of it, it wasn’t very impressive.
I started by measuring the positions of objects on the map and entering
those into a table on a spreadsheet, then writing formulas to calculate
distances, echo paths, conversions to millimeters and milliseconds, and
the other necessary functions.
Before long I realized the results of runs could be plotted directly on
the map image, and that it would run orders of magnitude faster if all
of the calculation were removed from cell functions and run in
background VBA code. The model gradually evolved into a very
sophisticated tool that could test possibilities Weiss & Aschkenasy
could never have attempted.
The W&A report depends on a single real live fire test shot for
calibration purposes, and the data for that impulse pattern is provided
in the report. The rest of the procedure required the survey map
of Dealey Plaza and a recording of the Dictabelt, both of which are now
readily available. Measurements W&A made using the map are
published in the report.
I have made every effort to reproduce
the process as laid out in the report, using the same materials and
techniques but with more powerful tools. Numerous errors have
been found with the data provided in the report, including basic errors
involving the measurement of delay times, waveform peaks, and object
positions. Some of the errors are necessary to the finding of an
echo correlation to the suspect Dictabelt pattern. The Weiss
& Aschkenasy report does not stand up to even limited scrutiny, and
the results it contains cannot be reproduced.
Materials
1. Drommer &
Associates survey map prepared for the HSCA, provided by Anthony Marsh
(image: http://the-puzzle-palace.com/DP1280.jpg). Scale: 1 inch =
10 feet
2. Copy of the Weiss & Aschkenasy report to the
HSCA
(http://www.history-matters.com/archive/jfk/hsca/reportvols/vol8/pdf/HSCA_Vol8_AS_1_Weiss.pdf)
3. Copies of the Dallas Police Department recordings, provided by Dr. Norman Ramsey and Dr. Paul Horowitz.
4. A copy of the Dictabelt recording used in the
Court TV program in 2003, provided by Robert Berkovitz of
Sensimetrics.
Methods
The Weiss & Aschkenasy procedure involved measuring distances on
the survey map and converting those to time in milliseconds based on
the speed of sound, and comparing those values to delay times derived
from either the Dictabelt or a test fire recording.
In 1978 this was done manually using a ruler and lengths of
string. Locations were measured on a high resolution scan (4923 x
3596) of the survey map. Map locations were identified by two
dimensional coordinates, as number of pixels from the upper left corner
of the map, using Adobe Photoshop. Coordinates and other
variables were entered into a Microsoft Excel 2010 spreadsheet, in
which Excel VBA code was written to handle conversions and the
computations of echo paths, and other functions.
W&A measured distances on the map in millimeters. For this
effort distances were measured as individual pixels from the upper left
corner of the image on the scanned map. A ratio has to be
established to convert pixels to millimeters. That was done by
taking measurements of long distances directly on the paper map and
comparing those with pixel measurements on the image. The
conversion ratio was also later cross-checked with fixed location data
in the W&A report. The ratio was determined to be 3.92
pixels/mm.
Calibration
The W&A analysis was an effort to match a suspect impulse pattern
on the Dictabelt recording to a specific location on the survey map of
Dealey Plaza, using echo location. The assumption was that it
would be very unlikely to be able to pin a specific location to the
pattern if it was not in fact a real gunshot sound, recorded in Dealey
Plaza.
To do this they first had to establish which objects on the map should
be used to calculate echoes. They took a test shot recording made
in Dealey Plaza by BBN, and matched up the echo peaks on that impulse
pattern with objects on the map. This generated the list of echo
producing objects and echo paths that are used in the succeeding steps.
The calibration procedure involved the following steps:
1. Establish the position of the rifle and microphone used for the test shot.
2. Measure echo times of peaks on the test shot recording.
3. Identify echo objects, using pins at the rifle and
microphone locations on the map and strings cut to the distance each
echo would travel, to draw an ellipse. Each ellipse should
intersect the object that produced the echo peak.
4. Generate a numbered list of echo paths using
objects identified on the map. Some peaks may be produced by
compound echoes, and their paths will involve two objects.
Rifle and Microphone positions
Figure 1: W&A Fig. 5, depicting microphone location.
W&A placed the test rifle north of the corner of the wooden fence
on the grassy knoll, where the shooter fired during the reconstruction
tests. The microphone was the fourth in the third array, as
depicted in their Figure 4 diagram. The report documents these
positions in text accompanying their Figure 5.
Page 22 of the report cites the path of echo 2 in this diagram and gives the following measurements:
Rifle to diffraction point: 499 mm.
Rifle to microphone: 213 ft. or 185.2 ms.
Diffraction point to microphone: 92 mm.
These are fixed measurements for three sides of a triangle, with one
apex on the diffraction point and another touching the fence
line. This is enough information to geometrically determine the
positions of the rifle and microphone.
One problem I noticed with the report
early on is that the diagram does not match the description. Echo
path 2 as depicted in the illustration diffracts off of the northwest
corner of the Dallas County Records Building. The distance
between the fence and that building is easily checked on the map.
The DCRB is about 930 mm from the fence, not 499 mm.
There is however only one object on the list of objects that is 499 mm
away from the fence. It is not labeled on the illustration.
It is “Wall A, Corner 2” on the object list. That is
very near “Wall A, Corner 1”, which is the path depicted by
path 1 on the illustration.
If we ignore the W&A illustration and use the provided dimensions,
the rifle is along the fence line 499 mm from “Wall A, Corner
2”. That position is 892 pixels right and 1750 pixels down
from the corner of the map. This also corresponds to 15 mm north
of the corner of the fence.
Figure 2: Corrected diagram for microphone location.
The microphone position for the test would be 92 mm from “Wall A,
Corner 2”, and 185.2 ms. from the rifle. Given the rifle
position, the microphone coordinates are 2924 pixels right and 1139
pixels down.
Echo Times of Test Shot Peaks
The test shot recordings are unavailable. The only evidence of
the test shot used by W&A is Figure 6 in their report, a poor
quality copy of the waveform without a scale. However W&A
provide measurements from the waveform in their Table 3. That
data is used here.
Identify Echo Objects and Paths
Table 2 is a list of echo paths. Each path is a line connecting
the rifle position, to an object position, and then to the microphone
position. Some paths use two object positions making a compound
echo. Each path is made using objects listed in Table 1.
Table 1 contains another of the
obvious errors. Both object 16 and 20 are described as the same
object, “DCRB : Southwest corner.” Yet the echo times
for these two paths are different. One or both must be labeled
incorrectly.
Table 2 contains a third obvious error. Path 26 is said to bounce
off of object 23. There are only 22 objects in the object list.
Each of the 26 numbered paths is listed in Table 3, along with the
relevant measurements. Since each echo path was supposed to
correspond to one of the peaks on the test shot pattern, each path has
an associated echo travel time and delay times. Weiss &
Aschkenasy used the echo travel times to draw ellipses on the map,
using pins and string. If an ellipse intersects a potential echo
producing object on the map then it is an indication that the object
produced the echo.
We can do the same and use the model to plot the ellipses for each path
to check the object selections and record their map coordinates.
Most echo time ellipses do in fact
match with an object recorded in the object list, confirming those
objects and also the accuracy of the model. But there are
problems with some of the identifications used by W&A.
Figure 3: Echo time ellipses for three echo times.
 |
|
Object 1, used in path 1, is labeled as, “South shelter : South
door, east post.”. In fact, the ellipse for path 1 crosses
the west post. |
Figure 4: For object 1, path 1
|
 |
|
Object 2, used in path 2, is labeled as, “South shelter : East
door, south post”. In fact, the ellipse identifies the
north post. |
Figure 5: For object 2, path 2
|
 |
|
Object
4, used in path 7, is labeled as, “North shelter : South door,
west post.”. In fact, the ellipse for path 7 misses all the
posts in both shelters. The only potential object with which it
intersects is the end of a wall just north of Wall “A” that
does not appear on the object list at all.
Object 6, used in path 8, is labeled as, “North shelter : East
door, south post.”. However, it uses the same echo time as
object #4 and therefore draws the same ellipse, which intersects no
posts. |
| Figure 6: For objects 4 and 6, paths 7 and 8 |
|
|
An interesting pattern emerges when checking objects numbered higher
than 10. When a path cites an object higher than 10, the ellipse
for that path will intersect the object that is numbered one
less. That is, when a path cites object 13, it really intersects
at object 12. A path cites object 18 but really intersects object
17.
This observation helps explain the obvious mistake in the path list
regarding object 23. Path 26 cites object 23 when there is no
object 23. No path cites object 11, when the only reason an
object appears on the list is because it is used in a path.
The objects in the path list are numbered incorrectly. They can
be corrected by subtracting one from every object number higher than
10, wherever the object number appears. For example, instead of
path 19 using object 18, it should use object 17.
The duplicate labeling of objects 16 and 20 can now be corrected.
Object 16 is now cited by path 25. The echo time ellipse for path
25 intersects only one potential object, the southwest corner of the
Dal-Tex building.
 |
|
There
are two additional ambiguities on the object list. First, object
19 is, “DCRB : Roof edge on west wall.”. The west
wall of the DCRB is already used as object 18, with a different echo
time. The “roof edge” would be at the top of this
tall building. This introduces the third dimension into
W&A’s work for the only time, and there are no notations on
the map regarding the height or position of the “roof
edge”. It is possible they used another source for that
information. Ignoring the doubts raised by choosing to introduce
the third dimension for this one and only echo, we can resolve the
ambiguity by giving them the benefit of the doubt. A position for
the unknown roof edge was used that would give the result cited by
W&A. |
| Figure 7: For object 16, path 25 |
|
|
 |
|
Another
ambiguity involves object 22. The table calls this,
“DCRB—New DCCCB : Alley wall between
buildings”. No such alley wall is depicted on the survey
map. We might assume they used other sources for that information
as well, but there is a problem with that. The ellipse drawn for
this object would place it significantly off the right edge of the
map. Even if there is a wall that far back in the alley the walls
of the alley would prevent any echoes coming from it from reaching most
of the plaza, including the part of the street we are interested
in. That cannot be the real object. However the ellipse for
path 26, object 22 does identify one potential object, the southwest
corner of the New DCCCB. |
Figure 8: For object 22, path 26
|
After corrections to objects 1, 2, 16, 22, and fixing the incorrect
numbering in the path list, only objects 4 and 6 are unresolved.
Those are both based on the same reported echo time of 215 ms. which
does not map to either of the objects reported in the list but does at
the end of an unidentified wall.
One resolution would be to assign this wall as the echo producing
object instead of the two shelter posts they were supposed to be.
The conservative option would be to use a resolution that favors
W&A if one is available. Their Table 3 also lists other echo
times for the paths. The “Predicted” column is the
number they calculated from the map after assigning an object
position. The “predicted” echo time for path 7 is two
milliseconds less than the value reported for the waveform
measurement. That in itself is strange since two milliseconds
would qualify it to be a different echo. However, since we
don’t have good data for the test shot recording, it’s
possible the error is in recording the “measured”
value. The conservative option is therefore to plot the ellipse
with the “predicted” value. When that’s done it
does intersect with the object described in the list for object 4,
“North shelter : South door, west post”.
This does not help resolve object 6. The difference between the
“Predicted” and “Measured” column values is
only 0.4 milliseconds, and it does not move the ellipse enough to make
a difference. The only object that intersects this line is the
south end of the unidentified wall. The choice is to use that
object, or to strike the path from the list for possible matching.
The procedure being used here is the same one prescribed by W&A for
selecting all of the objects, and it does in fact point to an
object. What reason would we have for not using it, other than
the fact that W&A didn’t?
One reason might be that W&A believed this object would be occluded
by structures between it and the rifle. That’s very
possible, Wall “A” runs just south of this object.
However, object 10 appears to be equally occluded by the wall and it
made the list. Such objections might be raised about the several
post objects in the two shelters. To get to the position reported
for object 7 sound from the rifle would have to pass through the south
shelter (through what appears to be solid wall) and barely pass through
the south doorway of the north shelter. This analysis is not
intended to dispute the physical plausibility of these echo paths
however, so such potential occlusion is not grounds for dismissing an
echo path. According to the procedure defined by W&A we
should use the object indicated by the ellipse.
After replicating the W&A
procedure for selecting echo producing objects and echo paths, 5
objects have been found to be incorrect, and 17 of the path definitions
have to be altered. In addition the diagram depicting the
microphone placement is found to be in error.
Some of these mistakes may just be errors in compiling the
report. Renumbering the object numbers in the paths, or changing
a label on an object would have no material effect if W&A used the
correct values when they performed the experiment.
However, if W&A selected the wrong objects and used those positions
going forward, then they weren’t comparing the right values to
the echo peaks on the Dictabelt. Testing of the echoes for object
1 does indicate that they used the incorrect post.
 |
|
Figure
9 depicts results from the model when the parameters were set up for
the Dictabelt test. The two colored areas are plots of the
possible microphone locations where Dictabelt peak 1 would match the
predicted value (+/- 1 ms.) for echo path 1. The only difference
between them is which post is selected for the echo path. The
green area uses the correct post shown by plotting the echo time
ellipse. The lower purple area uses the incorrect post cited by
Weiss & Aschkenasy. |
| Figure 9: Possible microphone positions with alternative selections for object 1 |
Using the incorrect post is required to get the reported Dictabelt correlation.
Dictabelt Pattern
The ultimate purpose of this entire calibration procedure was to create
sets of predicted echoes for comparison to the Dictabelt
recording. Using the echo paths that were defined, the microphone
and rifle locations could be altered and a new pattern of echo peaks
generated. These could then be compared to the Dictabelt
pattern. It is just as important to measure the Dictabelt pattern
accurately as it is to generate a proper set of comparison peaks.
Figure 10 depicts the W&A Dictabelt pattern exhibit with labels for
each of the 26 echoes that were assigned to paths. The echo time
measurements for these peaks were recorded in W&A Table 4.
The “Impulse delay time” column of Table 4 would be the
time elapsed from the muzzle blast to each labeled peak.
Figure 10 has no scale of any kind so the measurements can’t be
completely confirmed by referring to that illustration. One check
that can be made with Figure 10 is to see if the measurements are
accurately made regardless of scale. To do that we can assume
peak 26 represents the 369.2 ms. mark as the table says, and compare
all the other peaks relative to that. When I did, many of the
peaks were found to produce measurements that were significantly
different than those presented in the table.
Other peaks were tried as the anchor with similar results. Even a
sliding scale was used to find the best possible match, which minimized
the sum of all the differences. However no such adjustments could
reconcile all of the peak measurements. Even when using a scale
with minimum differences many of the peaks differed by more than the 1
ms. tolerance that W&A claimed to be allowing.
Considering the possibility that the printed Dictabelt pattern may have
been distorted in the production process, I needed to check the
measurements against the actual recording. Figure 11 is the
suspect pattern taken from a digital copy of the Dictabelt recording.
Visually the two patterns confirm each other. Once they are
scaled to the same size the two images could be placed on transparent
backgrounds and laid over each other and the peaks will line up.
There is no significant distortion in the printed pattern.
The digital recording has the advantage of being directly
measureable. Each peak in Figure 8 can be measured very precisely
in the audio software. When measurements of all 26 peaks were
made very large differences from the Table 4 values were
observed. This is best seen by looking at the longest echo delay,
the one for peak 26.
Figure 10: W&A Dictabelt pattern of suspected grassy knoll shot.
Figure 11: Pattern seen in digital recording of Dictabelt.
Table 4 records the value of the echo delay for peak 26 as 369.2
ms. The same peak measured directly in the recording is 337.1 ms
after the muzzle blast. That’s a difference of almost 10%.
 |
|
On
average W&A Table 4 gives delay times that are 13.4% greater than
those measured off the recording. That’s only the
average. Some peaks have a much greater difference, as much as
34% for peak 1. If it were only a matter of using different
scales, or playback speeds, the percentage of difference would be the
same for every peak.
Playback speed is an issue in the reports, but cannot explain the
discrepancies I observed. W&A explained that they adjusted
Dictabelt times by a factor of 1.043 because that gave the best
match. This was said to be justified because BBN found that the
recorder was running at, “0.95 of correct speed.”
However the recording I used to measure the peaks had already been
established as very close to real time. The very same digital
recording was used in the 2003 Court TV analysis performed by Robert
Berkovitz. Berkovitz created pattern detection software to
evaluate possible matches between W&A’s echo pattern and the
Dictabelt recording. He reported that there was no match of any
significance. For his analysis he made the 5% correction to the
playback speed called for by BBN. Don Thomas and I both found
that this was an error because the recording he used had already been
speed corrected, and his software produced a better correlation with a
0% speed correction. |
Figure 12: Dictabelt peak variances from W&A data
|
I performed another check on the playback speed by comparing signals
between recordings. Both channel 2 and channel 1 contain a
simulcast announcement that is almost 10 seconds long. The speed
of channel 2 can be calculated with some certainty because of the
prominent power hum that was unmistakably deposited at record time
(see: http://mcadams.posc.mu.edu/odell). This test showed that
perhaps a correction to Dictabelt times as much as 1.5% may be
justified, but in the opposite direction. That adjustment would
only increase the differences with Table 4.
There is also a conflict between W&A and BBN regarding the total
span of the Dictabelt impulses. In the relevant test BBN says
W&A found, “14 impulses in a 320 msec time span”, and
that value forms a part of the probability calculation upon which the
95% chance of “gunfire from the knoll” is based.
Table 4 contains peaks out to 369.2 ms.
Because the range of differences runs from 8% to 34%, the values in
Table 4 cannot be completely reconciled by any adjustment for speed.
The bottom line is that the values
presented in Table 4 for the Dictabelt pattern do not appear to be
valid measurements of the peaks on the recording. A test that
supposedly identifies a gunshot on the Dictabelt recording must, at a
minimum, correctly measure the sound being tested on the
Dictabelt. If it does not then it is not a real test.
Conclusions About the Attempt to Reproduce Weiss & Aschkenasy
The Weiss & Aschkenasy report to the HSCA formed the basis for the
scientific finding a gunshot from the grassy knoll. The analysis
depended upon measuring distances on a survey map, identifying echo
producing structures on that map, predicting echoes that would be made
at various locations on the map, and measuring the Dictabelt pattern to
compare those projected echoes.
This attempt to reproduce that experiment has demonstrated significant
errors in Weiss & Aschkenasy’s results at every step.
An attempt to completely reproduce the process without any dependence
on data from the report may require additional sources, like the
calibration shot recording, that are not currently available.
Someone with a religious devotion to belief in the acoustic evidence
might say that falsifying the report does not prove there was no shot
from the grassy knoll on the Dictabelt recording. Maybe Weiss
& Aschkenasy did the test correctly and everything got all messed
up when the report was written. True, maybe that happened.
Maybe it didn’t. We also can’t prove there was no
shot from the Dal-Tex building, or the overpass, or the storm
drain. We are not often required to prove negatives such as
that. The basis for believing the Weiss & Aschkenasy result
is the report. A hypothetical experiment conducted in private and
not written down does not count.
Falsifying the Weiss & Aschkenasy report basically puts us back
where things were before the HSCA hired them to do the analysis.
We have a report from BBN that is unable to say whether there was a
shot from the grassy knoll, or even a fourth shot. That report
also has never been reproduced and relied on data we can no longer
examine.
Weiss, Aschkenasy, and Barger were all contacted about these results. All declined to comment.
© 2013, Michael O'Dell