HSCA 1978 Acoustic Study by BBN – Figure 367

[Michael T. Griffith]

It might help to point out that after BBN did the test firing in Dealey Plaza, they were able to perform more tests to identify gunfire sounds on the dictabelt. One of those tests was the matched filter test. Importantly, they applied this test to all the impulses on the dictabelt. No sound impulses that were recorded after Curry’s “to the hospital” transmission passed the matched filter test. Five of the six suspect sound impulses did pass the test. Here is some information about when and how BBN used the matched filter technique on the dictabelt, from Dr. Barger’s HSCA testimony:

Mr. FITHIAN. And one final sort of terminology question. You used the term "match filter technique," matching filter techniques"—

Dr. BARGER. Yes.

Mr. FITHIAN [continuing]. And that means what?

Dr. BARGER. That means that you have in your--you expect to receive one of many kinds of signals. By "many kinds," I mean a signal in this case that has a series of impulses that occur in a definite sequence, like, let's say, the first one occurs at a particular time, the second one perhaps 3 milliseconds later, the third one maybe 15, the fourth one 27, the fifth one 121/z after that, and so on, a definite sequence of impulses.

We went to Dallas to find out what the sequence of impulses would be that would be generated by Dealey Plaza if a gun was fired.

Having found out what that sequence of impulses is, you then go through the tape in question and look for sequences of impulses that match it. When you find one that matches it, you say aha, at that time something occurred that generated a pattern of transient events that just matches what we did in Dealey Plaza, and when that occurs, you judge that you have made a detection. You have identified a similar source of noise. The word "matched filter" is a technically correct or often used form, and the use of the word "match" is fairly self-evident, I believe. (2 HSCA 73-74)

Dr. BARGER. There is in the field of detection theory a favorite approach called matched filtering. The matched filter is a device that is used to detect events that you have some understanding of, even though they are subaudible. Matched filters are used in radar sets commonly to detect the presence of impulsive signals in noise, even though they are not visible or audible in the raw data. There was reason to believe that applying these techniques we might be able to detect the impulsive sounds of gunfire. (2 HSCA 18)

Dr. BARGER. Then we counted the number of impulses in each pattern of impulses that we see in the waveform records of the tape and we saw there were about 10. We realized there was still a possibility that these impulsive sounds that we saw in the record of the tape were in fact caused by gunfire.

Mr. CORNWELL. At this point then you had devised six screening tests, any one of which I take it might have been sufficient to rule out these impulses as being gunshots, and they in fact passed all six tests, is that correct?

Dr. BARGER. Quite so.

Mr. CORNWELL. Now, at this point did you have any conclusions or, on the other hand, did you feel that further testing was required?

Dr. BARGER. At this point we felt we were justified in suggesting to the committee that a matched filter detection trial was warranted on the tape. As I said, the patterns that formed the basis for the match would have to be obtained by an acoustical reconstruction. The reason for suggesting the matched filter procedure for detecting the events was it is the most powerful method we know of with which to do that.

Mr. CORNWELL. How about telling us in just plain, common language what you are referring to when you say an acoustical reconstruction?

Dr. BARGER. The objective is to obtain echo patterns of the sort that I described briefly before, and the purpose for having these patterns is to become the basis of the match in the matched filter detector. In order to get these echo patterns, it was necessary to design a test that would get echo patterns that would in fact match with the events on the tape if in fact there were events on the tape that were gunfire. (2 HSCA 46-47)

And, needless to say, the matched filter test found that five of the six dictabelt impulse patterns that passed the initial screening tests contained echo patterns that matched those of shots from the test firing.

I’d like to note that the HSCA’s use of acoustical analysis to determine the existence and location of gunfire was neither novel nor based on unproven technology. Acoustical analysis was used in the Kent State shooting. As chance would have it, the lead acoustical scientist in that case was James Barger, who was also one of the HSCA acoustical experts. In the Kent State case, Barger used the same procedure that he used for the HSCA, namely, echo location based on an audio record of the incident. Barger was able to identify the physical locations of the first gunshots fired at Kent State to within 10 feet of where they were later determined to have been fired when the National Guardsmen who fired them were arrested and admitted they had fired the shots.

So no one should claim that acoustical analysis technology is unproven. The same technology has been used for many years by the Army for locating enemy gun emplacements and snipers, by the Navy to navigate underwater, and by geologists to find oil deposits.

[Joe Elliott]

It might help to point out that after BBN did the test firing in Dealey Plaza, they were able to perform more tests to identify gunfire sounds on the dictabelt. One of those tests was the matched filter test. Importantly, they applied this test to all the impulses on the dictabelt. No sound impulses that were recorded after Curry’s “to the hospital” transmission passed the matched filter test. Five of the six suspect sound impulses did pass the test. Here is some information about when and how BBN used the matched filter technique on the dictabelt, from Dr. Barger’s HSCA testimony:

No, they did not apply the “matched filter test” to all the impulses on the Dictabelt. Specifically, it was only applied to all the impulses on the first sequence of 10.1 seconds, not on the second sequence of 4.0 seconds that was recorded on the Dictabelt 30 seconds later.

The “matched filter test” was comparing the 7 impulses from the first impulse sequence, covering 10.1 seconds, with the 2,408 impulses from the 1978 test with 69 rifle shots recorded on 36 microphones.

“Five out of six did passed the “match filter test”. First of all, it was seven impulses, not six. And only four of the seven passed, that is, had a correlation coefficient of 0.8 or higher. And the four that did pass, barely passed.

But the main point is, the second sequence, the 4-second sequence, was never given the “matched filter test”. For all we know, some of the impulses in that 4.0-second sequence would have passed as well, which would have served to discredit the BBN study.

But of course, the BBN study was discredited anyway, by their own data. Check out the BBN’s Exhibit F-367 table which I have in my initial post of this thread. We don’t have a “match”, for a certain 1978 recreation shot, with both an impulse found in the Dictabelt 10.1-second and the 4.0-second sequence, because this test was never made. But we do have a couple of matches, for both the 137.70 and 145.61 Dictabelt shots, where matches of 0.8 correlation coefficient was found for a shot fired at two different targets. That shouldn’t happen. Each shot should have its own unique fingerprint. So, a correlation of 0.8 is not sufficient to say “We have a perfect match with this shot, fired from this location, at this target, recorded from this location” because they have two different shots that reach 0.8. Clearly a higher threshold than 0.8 was needed to make this determination, but the BBN could not find any.

[Michael T. Griffith]

No, they did not apply the “matched filter test” to all the impulses on the Dictabelt. Specifically, it was only applied to all the impulses on the first sequence of 10.1 seconds, not on the second sequence of 4.0 seconds that was recorded on the Dictabelt 30 seconds later.

I conflated two events in my mind as I wrote. Barger said:

We went to Dallas to find out what the sequence of impulses would be that would be generated by Dealey Plaza if a gun was fired.

Having found out what that sequence of impulses is, you then go through the tape in question and look for sequences of impulses that match it. When you find one that matches it, you say aha, at that time something occurred that generated a pattern of transient events that just matches what we did in Dealey Plaza, and when that occurs, you judge that you have made a detection. You have identified a similar source of noise. The word "matched filter" is a technically correct or often used form, and the use of the word "match" is fairly self-evident, I believe. (2 HSCA 73-74)

They "went through the tape" and looked "for impulses that match it" via the screening tests in the preliminary analysis. The six impulse patterns that passed the screening tests were the ones to which they applied the matched filter test.

The “matched filter test” was comparing the 7 impulses from the first impulse sequence, covering 10.1 seconds, with the 2,408 impulses from the 1978 test with 69 rifle shots recorded on 36 microphones.

“Five out of six did passed the “match filter test”. First of all, it was seven impulses, not six. And only four of the seven passed, that is, had a correlation coefficient of 0.8 or higher. And the four that did pass, barely passed.

Well, that didn't take long: You've quickly wandered back into the Land of Confusion. "Five of the six" refers to impulse patterns, not impulses in the patterns. You are mistakenly referring to impulses in individual impulse patterns as if they were impulse patterns. Six impulse patterns passed the screening tests. From the HSCA report:

Six sequences of impulses that could have been caused by a noise such as gunfire were initially identified as having been transmitted over channel 1.(13) Thus, they warranted further analysis.

These six sequences of impulses, or impulse patterns, were subjected to preliminary screening tests to determine if any could be conclusively determined not to have been caused by gunfire during the assassination. . . .

All six impulse patterns passed the preliminary screening tests. (HSCA Report, p. 68)

Five of those six impulse patterns passed the matched filtering test.

Now, as we've discussed, one of those five impulse patterns was wrongly rejected as a false alarm, and that's why in the published HSCA materials we read that four of the six passed the matched filtering test, but actually five passed. The fifth was excluded based on non-acoustical grounds, as Dr. Thomas has documented.

As I've also mentioned, Dr. Barger admitted to Dr. Thomas that the reasoning behind the rejection of the fifth match was circular and "ad hoc," which is polite verbiage for bogus.

But the main point is, the second sequence, the 4-second sequence, was never given the “matched filter test”.

Sigh. . . .  That's because it failed the initial screening tests in the preliminary analysis! Sheesh, we've been over this already. Hello? That sequence is mostly a bunch of squawks from radios keying in. It was judged to be too dissimilar to the suspect impulse patterns to warrant further analysis.

For all we know, some of the impulses in that 4.0-second sequence would have passed as well, which would have served to discredit the BBN study.

Oh. . . .  Okay. . . .  So all the BBN scientists were involved in a conspiracy to lie about the 4-second impulse. LOL. Got it.

Here's a question to chew on: The NRC panel had unlimited funding and could have easily tested that 4-second sequence with all the tests that the HSCA experts applied to the six gunshot sequence patterns that passed the screening tests. Why didn't they do that? Because they knew it would be a waste of time?

But of course, the BBN study was discredited anyway, by their own data.

Oh, please. You don't know enough to have any business making such a ridiculous claim, as  you make clear when you elaborate on it below.

Check out the BBN’s Exhibit F-367 table which I have in my initial post of this thread. We don’t have a “match”, for a certain 1978 recreation shot, with both an impulse found in the Dictabelt 10.1-second and the 4.0-second sequence, because this test was never made. But we do have a couple of matches, for both the 137.70 and 145.61 Dictabelt shots, where matches of 0.8 correlation coefficient was found for a shot fired at two different targets. That shouldn’t happen. Each shot should have its own unique fingerprint. So, a correlation of 0.8 is not sufficient to say “We have a perfect match with this shot, fired from this location, at this target, recorded from this location” because they have two different shots that reach 0.8. Clearly a higher threshold than 0.8 was needed to make this determination, but the BBN could not find any.

Oh boy. You seem to be a glutton for embarrassment. This is a ball of confusion and a bundle of error. It would take me two or three pages to unravel the confusion, mistaken readings, and erroneous statements packed into your paragraph.

First of all, a correlation of 0.8 is very high, very close to perfect, as Barger explained. 0.5 was the minimum threshold, although Dr. Barger mostly limited his discussion to correlations that were at least 0.6, for reasons that Barger also explained. Dr. Barger said that a correlation of 0.75 meant that two impulse patterns matched "quite admirably." So 0.8 is a very solid correlation.

Now, in order to fully explain exhibit F-367, I'm going to have to quote Dr. Barger's testimony about it, so no one will say "oh, that's just what you say." The quote is going to have to be very long, because F-367 represents a lot of information and was one of several exhibits that Dr. Barger used in this segment of his testimony.

For others who are reading this thread, I think once you read Barger's explanation of F-367 quoted below, you will understand what it does and does not say. Also, Dr. Barger used exhibits F-347 and F-370 to help explain F-367, and vice versa.

Dr. Barger used those three exhibits to explain how BBN distinguished between correct detections and false alarms among the 15 correlations between dictabelt impulse patterns and echo patterns from 432 Dealey Plaza test-firing shots that had a correlation of 0.5 or higher, although Dr. Barger mostly talked about correlations that were 0.6 or higher.

Dr. BARGER. From this time on, I will mostly talk about those matches that
exceeded a correlation coefficient of 0.6.

Mr. CORNWELL. I would like to show you JFK exhibit F-347, and
ask you if you would tell us what that is.

Dr. BARGER. This illustrates two types of data. Here are three
test patterns. These three test patterns were generated by a shot
from the depository with the muzzle 2 feet behind the plane of the
window and fired at the target No. 1, which was located just at the
head of Elm Street in a position previously described, and it was
received by microphones 4, 5, and 6 in the second array position.

Those microphones were on Houston near Elm, and we see in
each of these that the first sound that arrived was the muzzle
blast. There is no shock wave that precedes the muzzle blast, and
that is to be expected because in this case the shot is fired in this
direction, and the microphone is over here, and according to the
first exhibit I showed, the shock wave would not be seen 90°
laterally.

As you look at the arrival of the muzzle blast, you see that in
each channel it occurs progressively later in time, so that if you
connect the peaks, they slant. This is because channel 4 microphone
is farther away from the rifle than is the channel 5 and
channel 6 microphone. However, if you look at these peaks out
here near one second, these are the echoes from the Post Office
Annex.

As the microphone moves away from the location of the rifle, it
is moving toward the Post Office Annex. Therefore, the echo in fact
comes in sooner, so when you connect the dots signifying each of
those echoes, they have a slope in this direction.

One selects all of the significant impulses on these test patterns.
We have placed dots on them. Some of the dots are obscured in
these dark areas where the photographer has overexposed them,
but nevertheless, they are there. We have connected all of those
that we think are caused by the same echo-generating device by
lines, to show how the time that that echo arrives is changing
continuously as you move the position of the receiver.

Up here is shown a portion, a segment, of the Dallas police tape
that was also prepared at the same time scale, 16 inches equals a
second with intensity vertical on the scale in decibels. The threshold
has been made, and all of those impulses that exceeded have
been identified and numbered, and the plus or minus 6 milliseconds
acceptance regions have been marked, these to accommodate
the uncertainty of the exact position of the motorcycle.

I am prepared to show how this echo pattern matches the test
pattern-and I knew I would probably forget which one it is that
matches with it, but it is quite evident. If you tried to match this
pattern with this shot, the significant impulses at this point would
not in fact match with the significant impulses in this pattern,
even though with this setting the echo from the Post Office Annex
does.

However, if you match it with the test impulse obtained at
channel 5, which is a different place, then they match quite admirably
in fact. If you count the dots signifying significant echoes in
the echo pattern with the marks signifying the significant impulses
in the Dallas tape, you find there are 12 matches out of 17 possible
impulses, and if you count these, 15 possible echoes. The cross correlation
coefficient for that match is 0.75, above our threshold value of 0.6.

Mr. CORNWELL. Given the amount of noise in the Dallas Police
Department tape, would you expect that you would ever get a
complete match, all 17 out of 17 in this case?

Dr. BARGER. Many of the impulses on the tape, on the Dallas
police tape, this segment of it in particular, that correspond to the
total number that were above the threshold value of 17 are caused
undoubtedly by nonacoustical events. Examples are the key transients
that I described when I was showing the results of the
spectrographic analysis .

However, none of those impulses in this particular segment of
the tape have been conclusively identified as being any of those.
The noise from whatever its origin that is present in the police
recording tape, there is demonstrably noise there, in addition to
any impulses that may be caused by gunfire, those would rise up
and compete with the impulses caused by gunfire and reduce the
value of the correlation coefficient to some number less than one.

Mr. CORNWELL. So in spite of the fact that the correlation coefficient
was not one, the match was not perfect, your words were that
this was a quite adequate match. In other words, it had a correlation
coefficient which approximated one; is that correct?

Dr. BARGER. Well, it was not possible to reach that judgment by
looking at one alone. We looked at 2,600 of them, and reached our
conclusions from that. This was to illustrate just one. . . .

Mr. CORNWELL. I would now like to direct your attention to JFK
exhibit F-367, and for your assistance ask that F-337 and F-344 be
placed up there simultaneously.

Dr. BARGER. Yes. This one and this one have been introduced as
evidence. This is new.

Mr. CORNWELL. Would you tell us what F-367 is?

Dr. BARGER. It is a list of those 15 matches that-of the 2,600
approximate matches we attempted--that that did in fact exhibit a
correlation coefficient higher than 0.6. . . .

Dr. BARGER. Very well. There are 15 descriptors here. Each one
describes a case where an acoustical test pattern matched better
than the threshold value of 0.6 with a segment of the Dallas tape.

The first situation where this occurred I will label with blue.
There were four test patterns that corresponded with the segment
of the tape that began at 137.7 seconds after the stuck button, with
coefficient, correlation coefficient, larger than 0.6, and these are
the four. I will note with a 1 that that is the first time in the tape
that any of the test patterns correlated with any of the impulse
patterns in the police tape with a score better than 0.6, and it
occurred four times.

Mr. CORNWELL. So at that point you are telling us that there is a
segment of the Dallas police tape which very closely approximates
or at least has a correlation coefficient of over 0.6 with respect to
the various test shots?

Dr. BARGER. Yes. This section may contain the sound of gunfire.
Then going on down in the list, we have what I will label the
second time, the second place on the Dallas tape where correlations
or matches were achieved that were good enough to exceed the
threshold value, and I will label that with red brackets to highlight
it, and there were five of them.

Then in the same way at a later time, around 145.15 seconds, in
green, I will label and highlight the three test shot patterns that
correlated with that part of the tape better than 0.6, and, finally,
at 145 seconds-yellow is not the best, is it-well, the fourth part of
the tape at 145.61 seconds had three different test patterns that
achieved the correlation score greater than 0.6.

Now a feature of a detection by a receiver that was designed to
detect the possibility of otherwise subaudible events by using the
threshold correlation procedure is that it can give threshold exceedences,
the threshold having been 0.6, under two circumstances.

One, it exceeds the threshold when it has correctly detected the
event, and the other is, it exceeds the threshold when it has incorrectly
detected the event. The latter circumstance is called a false
alarm.

It is the purpose of the rest of my testimony now to examine the
question: Which, if any, are false alarms?

Mr. CORNWELL. Before you do that, I take it that you took each
of the four segments of the Dallas Police Department tape, which
you have indicated with the numbers 1, 2, 3, and 4, and compared
them with all of the test patterns, and what you have simply
illustrated on the chart is a match very similar to the one that you
showed us physically how you performed earlier with respect to a
shot in the first time frame. Is that correct?

Dr. BARGER. That is correct.

Mr. CORNWELL. Then would you use the exhibits which are presently
in place and tell us what that means in terms of the other
diagrams as to the location of the microphones and the direction
and location of the shots.

Dr. BARGER. I can say a few preliminary things about that with
these exhibits, a few preliminary things. The results suggest that
there are detections at four different times of day.
If the motorcycle were in Dealey Plaza, it would only be at one
place at each of those times of day and would either be standing
still or moving in some reasonable pattern.

The correlations achieved, or the matches achieved, at the first
time when any matches were achieved are either at microphone 5
or 6 in the second-array position. There are four correlations there,
so that at this time on the tape, we would tentatively estimate that
the motorcycle was there.

Mr. CORNWELL. Let me show you at this point then JFK exhibit
F-370, and ask you if you would tell us what that is. . . .

Dr. BARGER. We want to examine now the meaning of these detections that
passed the threshold level to see if there is any reason to believe
that they are not all false alarms, possibly. I will attempt it in this
way. . . .

Dr. BARGER. All right, now I have explained where those 15 dots came from.
Those 15 dots represent these 15 correlations that passed the
threshold of 0.6, and they are illustrated as a function of the time
when they occurred and the position down the street where the
microphone was that picked up the test pattern that gave the
correlations.

Mr. CORNWELL. May we have JFK exhibit F-370 entered into the
record?

Chairman STOKES. Without objection, so ordered.

Dr. BARGER. Now, we look at these and immediately see the
motorcycle can't be at all these places, but there is a high degree of
order in this diagram.

The negative hypothesis would be that the motorcycle was not in
Dealey Plaza. If that were true, then this scale that describes the
distance down the street of the motorcade would be meaningless in
the data, and the data would occur in time and in distance down
the street at random.

But the eye can see that they tend to follow a sloping line. It can
particularly see that because of these prior lines that I drew in.
There is a lot of order in the occurrence of these 15 correlations.

Now, how much order? Well, if one segments the position of
microphones along the street into four bins, or four compartments,
and segments the time at which they occurred into the four compartments
that are naturally the four compartments into which
the data are segmented, then one can question what is the likelihood
that this ordered pattern could have occurred by chance. In
other words, was it likely this pattern would have occurred if the
motorcycle wasn't there.

There is a test for that sort of thing, and it is called the Chi
square test. If you segment the data into four times and four
places, as I have done, it is a test done with nine degrees of
freedom. The Chi squared, value, which is a measure of orderliness,
is 17 1/z . For those of you that have tables of the Chi square distribution,
the meaning of that number is this much order would occur
only 5 times out of 100 if this was caused by chance.

In other words, if the motorcycle was not there and so the data
were distributed at random, there is only a 5-percent chance that
that would have occurred. This much order in the data suggests
there is a 95-percent likelihood that the motorcycle was moving in
the motorcade.

That is just about at the level of statistical significance that gives
a person confidence that there are correct detections in the data.
On the other hand, there are demonstrably also false alarms.
This can be seen by observing that if some of those correlations,
in fact, indicate the position of the motorcycle, then some of them
must be wrong because the motorcycle can't be in two places at
once.

Mr. Cornwell, I could proceed with what I am doing now or we
could put up those other three. I think it might be easier if I
proceed.

Mr. CORNWELL. Go right ahead and proceed.

Dr. BARGER. It is now the task of the committee and me to try to
identify the best we can which of these detections are false alarms
and which ones are not. We have a good deal of confidence that
many of them are not.

Now, in order that the motorcycle could achieve this position 130
feet down the street from the blue position in the 1.6 seconds, it
would have to go 55 miles an hour.

There is no evidence to indicate that it did that, and so this
particular detection is labeled a false alarm. It couldn't be true. It
leaked through because we lowered our threshold of detection to
the point where we had enough correlations so we could be reasonably
certain that the true answers would emerge. We wouldn't
want to shut them out. . . .

If we assume that one of these last two occurrences represents
the so-called head shot, then we know at that time where the
limousine was. It was at frame 313. Frame 313 is 250 feet down the
street from the blue dot, so 250 feet at that time of occurrence is
here, so this must be where the limousine was at that time.

It was going at about 11 miles an hour as determined by photographic
evidence. If one plots back at 11 miles an hour, one finds at
the time of the first occurrence the limousine was somewhere 120
feet ahead of the motorcycle, which would have put it right there.

Now, again, I am examining the question about whether these
three or these three are candidates for false alarms. If these three
are truth, then the motorcycle was going 18 miles an hour, catching
up with the limousine, and, in fact, having achieved a position
only 40 or 50 feet behind it at the time of the head shot.

Now, if you recall the first thing we noticed on the tape was that
there was a diminution of the sound due to the motorcycle 3
seconds prior to the first impulsive pattern that we originally
suspected could be caused by gunfire.

There was no obvious explanation for that, until one sees that at
that time the motorcycle was just beginning a 110° turn, and on
the inside track apparently, and he would therefore have to slow
down to execute the turn.

Now, it was further observed that the motorcycle sound stayed
diminished after the turn. It did not increase to the level that it
had formerly had. Therefore, it would seem that it couldn't have
increased speed, which it would have had to do to achieve this
position in 8 seconds.

If, on the other hand, it had continued at the same speed of the
motorcade, it would have achieved this position in that time.

There is, therefore, the diminished sound of the motorcycle that
indicates that these are false alarms. Now, that is an example of
the kind of corroborating or disqualifying evidence that is of nonacoustical
origin. We are inferring that the motorcycle didn't speed
up because the noise didn't increase, this allows us to identify as
false alarms some of these correlations we have accepted by lowering
the threshold sufficiently to catch the correct detections.

In other words, indications of detection that were accepted by the
test, but that were shown by other reasons not to be possible, are
therefore, found to be false alarms.

As a result of that judgment, the estimate of the motorcycle
position at the time of the second impulse, the red one, would be
there, which is right there, and the estimated position then of the
motorcycle at the time of the third occurrence, which is here, is
right there.

I lost my graphical symbolism a little, and that is right there,
and at the time of the last segment labeled No. 4, which at this
time we would estimate it to be halfway between those two right
there, and that is there, 120 feet behind the limousine at the time
of the head shot, if in fact these impulses represent the sound of
the head shot.

There is the possibility of labeling one of these four threshold
crossings as a potential false alarm because it involves firing from
this place at this target at the time that the limousine was here.

That is almost 180° out. It is inconceivable that anyone would do
that, and on that basis one of these can be judged a false alarm.
The fact that some of those are thought to be correct detections
was illustrated by all of the order in the data, as I explained
earlier. (2 HSCA 61, 63-69)

[Michael T. Griffith]

No matter which pattern matching methods where applied to locate the shots on Ch-1, the location of "Hold everything" cross talk captured on Ch-1 destroys any hope of the pulses actually being shots, according to O'Dell.

The argument is surprisingly simple and is explained by O'Dell in his article: http://mcadams.posc.mu.edu/odell/index.htm

You must be kidding. I already destroyed this argument in the "Poor Scholarship on Display" thread. In fact, I destroyed that argument in reply to your repetition of it in that thread. See https://www.jfkassassinationforum.com/index.php/topic,2704.msg98676.html#msg98676.

The "hold everything" crosstalk is a bogus time indicator and is refuted by the 12:28 time notation on Channel 1, the 12:30 time notation on Channel 2, the simultaneous Fisher "I'll check" crosstalk, and Curry's two assassination-period transmissions on Channel 2. So it is worthless as evidence against the intricate correlations between the dictabelt impulse patterns and the test-firing impulse patterns.

Now, let's get back to the five impulse patterns that matched echo patterns from the test-firing shots. Under pressure from Blakey, BBN came up with a circular, non-acoustic excuse for rejecting the shot at 140.3 as a false alarm, namely, that it came too close to the previous shot to have been fired by the alleged murder weapon! This shot was selected for rejection to appease Blakey because it had a correlation coefficient of 0.6 and had only one echo pattern match, but 0.6, though lower than the other gunshots' scores, was 20% higher than the 0.5 threshold and represented a significant degree of correlation. Dr. Thomas on the questionable reject of the 140.3 impulse pattern:

Reference to Table 2 shows that five patterns passed the echo delay matching test. The five patterns are identified by their chronological position: 137.70, 139.27, 140.32, 145.15, and 145.61 seconds after the beginning of the motorcycle segment. One of the five, the pattern at 140.32 sec was judged to be a false alarm and discarded. This was the one glaring error in the acoustical analysis. The BBN Report states,

"The entry in Table II that occurred at 140.32 sec is a false alarm, because it occurred only 1.05 sec later than earlier correlations also obtained from the TSBD. The rifle cannot be fired that rapidly. Since there are three correlations plausibly indicating the earlier shot, the one occurring 1.05 sec later must be a false alarm."

The logic of this statement is that if it didn’t come from Lee Harvey Oswald’s rifle then, it was not a shot. But of course, the whole purpose of the inquiry was to test the Warren Commission’s single assassin theory against the facts, not the other way around. The fifth shot was dismissed because five shots were less palatable to the committee members than four shots. Palatability is not, however, a scientific criterion for judging the validity of evidence. Moreover, it was illogical to dismiss the pattern at 140.32 as a false positive because it was too close to the previous shot. The first two putative shots are only 1.7 sec apart, also too close together to have been fired from Lee Harvey Oswald’s rifle.

If the subject sounds are the assassination gunfire, and if three of the shots are attributable to Oswald’s rifle, then the second pattern is the rogue shot, not the third. But, the second pattern, at 139.27 sec, could not be dismissed as a false alarm because it was supported by multiple correlations, including a robust correlation coefficient of 0.8. The weakest supported pattern was the pattern at 140.32, with a score of only 0.6, and only one match, and was thus selected as the “false alarm.” The fact remains that five candidate patterns passed the initial screening tests, and all five matched to a significant degree with the test shot patterns. The time intervals between these putative shots, corrected for tape speed, were: 1.7, 1.1, 4.6 and 0.7 sec. (https://www.maryferrell.org/pages/Essay_-_Acoustics_Overview_and_History_-_part_2.html)

[Joe Elliott]

Five of those six impulse patterns passed the matched filtering test.

Look again at my initial post, which contains a copy of Dr. Barger’s Exhibit F-367, which he presented on September 11, 1978 to the HSCA:

It contains the following impulses at the following times:

Impulse # 1 – 136.20 – 0.5 - Rejected
Impulse # 2 – 137.70 – 0.8
Impulse # 3 – 139.27 – 0.8
Impulse # 4 – 140.32 – 0.6 - Rejected
Impulse # 5 – 145.15 – 0.8
Impulse # 6 – 136.20 – 0.8
Impulse # 7 – 136.20 - 0.5 - Rejected


Question 1:

You see 7 impulses, Correct?

Question 2:

The four impulses that were judged good all had a correlation coefficient of 0.8, correct?

Question 3:

So, this means that not 5 out of 6, but 4 out of 7 passed the BBN tests as real gunshots in 1978, correct?

Question 4:

The three rejected impulses had an inferior correlation coefficient of 0.5 or 0.6, correct?

Question 5:

One of the rejected impulses, at 140.32, corresponding to z216 (BBN’s estimate) and z224 (Dr. Thomas’s estimate), had a correlation coefficient of 0.6. Inferior to all the other accepted shots, correct?

So, it appears to me, that this “z224” was not rejected because of pressure from Robert Blakey but because it did not correlate well enough with any of the 1978 shooting tests.

Question 6:

Are you saying that Robert Blakey put pressure on Dr. Barger, and as a result Dr. Barger not only removed the shot at 140.32 as a possible shot (at either z216 or z224) but forged his data, to show a correlation coefficient of 0.6 instead of 0.8?

Question 7:

On what basis, do you claim this shot at 140.32 has just as good support from the evidence as the four impulse patterns the BBN accepted as shots back in 1978?

Oh. . . .  Okay. . . .  So all the BBN scientists were involved in a conspiracy to lie about the 4-second impulse. LOL. Got it.

Well, you seem to think that all the BBN scientists were involved in a conspiracy to suppress the evidence for a shot at 140.32. Apparently even forging some of there data, to make it appear that the best correlation for this time was only 0.6.

But, no, I don’t support a BBN conspiracy. Only a reluctance to investigate the 4-second impulse sequence, which could undermine their conclusions about the 10-second impulse sequence.

Here's a question to chew on: The NRC panel had unlimited funding and could have easily tested that 4-second sequence with all the tests that the HSCA experts applied to the six gunshot sequence patterns that passed the screening tests. Why didn't they do that? Because they knew it would be a waste of time?

I don’t think they did have unlimited funding. Or at least it was not made available to the BBN. And more importantly, not unlimited time. The BBN had only 10 days between the August 20, 1978 firing tests at Dealey Plaza to August 30, 1978, to study the data and to make their initial report to the HSCA.

First of all, a correlation of 0.8 is very high, very close to perfect, as Barger explained. 0.5 was the minimum threshold, although Dr. Barger mostly limited his discussion to correlations that were at least 0.6, for reasons that Barger also explained. Dr. Barger said that a correlation of 0.75 meant that two impulse patterns matched "quite admirably." So 0.8 is a very solid correlation.

A correlation coefficient of 0.8 is very high, is it? Well let’s look at a couple of comparisons that met this “very high” rating, from BBN’s Exhibit F-367:

Test	Beginning Time of	Zap.	Zap.	Microphone Array	Rifle	Target	Correlation	Strong	Fluke
ID	First impulse on	Frame	Frame	and	Location	Location	Coefficient**
	Tape Segments (sec)	BBN	Thomas	(Channel Numbers)
B	137.70	168	176	2 ( 5 )	TSBD*	1	0.8	Strong
D	137.70	168	176	2 ( 6 )	TSBD	3	0.8	Strong	Fluke

* Indicates Muzzle Withdrawn 2 ft from Plane of Window
* Correlation coefficient = number of experienced Matches with Impulses divided by the square root of the number of echoes X Number of impulses is Less than or equal to 1.0


Now, each shot, depending on:

•   Where it is fired from.
•   Where the microphone is that records this shot.
•   Where the rifle was aimed at

One gets a unique “fingerprint” for that shot. No other shot should match that fingerprint. Change any one of the three factors and one will get a different waveform.

So, at 137.70, there was a shot from the TSBD, at both Target 1 and Target 3. So, which is it? Was it aimed at Target 1 or Target 3? It was impossible for a bullet to hit near both targets.

So, it seems that, at best, one of these correlations if good. And the other is bogus. So, it is quite evident that a correlation coefficient of 0.8 is not good enough to get a valid result we can be confident in. In truth, some correlations of 0.8 correspond to events that never took place. Perhaps all of them.

[Joe Elliott]

No matter which pattern matching methods where applied to locate the shots on Ch-1, the location of "Hold everything" cross talk captured on Ch-1 destroys any hope of the pulses actually being shots, according to O'Dell.

The argument is surprisingly simple and is explained by O'Dell in his article: http://mcadams.posc.mu.edu/odell/index.htm

From the article, section named "Synchronization", this is what it boils down to:
(My emphasis, click on Fig. 2 for a better image)

Any pair of events on Ch-2 can never have been spaced closer in real time that on the tape, that's why the 61 second is the best one can hope for. IMO, the only way to push the time of the shots back towards 12:30 is to show the recording was paused between shots and "HOLD...".

The "ALL right Chaney" event is not marked on Fig. 2. From what I've seen this is supposed to be the alleged Fisher cross talk "I'll check it" which is disputed by others. To be looked further into....

I found the O’Dell article to be very informative. I didn’t realize that the “Hold Everything Secure” was recorded right on top of some of the “gunshots”. The “K” in “Secure” corresponded to the “shot” at 145.15.

I got the impression from Mr. Griffith that only the shock wave of a bullet could form an N-wave. But it appears that just saying the word “Secure” can also form a similar N-wave.

What an amazing coincidence, that these “shots” all appear all at the same time speech is recorded on the Dictabelt from crosstalk.

Be careful whenever you say the word “Secure”. You don’t want to damage anyone’s eardrums from the resulting shock wave. Choose your words carefully.

[Joe Elliott]

Sigh. . . .  That's because it failed the initial screening tests in the preliminary analysis! Sheesh, we've been over this already. Hello? That sequence is mostly a bunch of squawks from radios keying in. It was judged to be too dissimilar to the suspect impulse patterns to warrant further analysis.

Is “mostly a bunch of squawks” Dr. Barger’s description or yours?

Where does Dr. Barger say, in some many words, that this particular impulse sequence, of 4-seconds, was a bunch of “squawks”, or words to that effect.

From what I have read, Dr. Barger rejects this 4-second sequence of impulses, because it was shorter than 5-seconds.

And yes, he does give the opinion that this 4-second impulse sequence was caused by someone trying to key in, but where does he give evidence to support this opinion?