Cardinal Pieces of the Puzzle

In the initial phase of their work, Bolt Beranek and Newman Inc. (BB&N) filtered out periodic signals from their tape of the Dictabelt. They demonstrated the effectiveness of autocorrelation and an adaptive filter to recognize and remove periodic signals from a high-fidelity tape recording of a motorcycle. These proven methods failed to filter the loud interfering signal from the recording of the Dictabelt. (1)

After failing to detect periodic engine sounds, BB&N continued to reenforce the unsupported idea that a motorcycle engine was the source of the loud interfering noise. Primarily they promoted this dubious concept by frequent usage of motorcycle, engine and noise in the same sentence. This repetitious practice diminished the impact of their failure to detect and filter out engine sounds. Finally BB&N assumed that loudness of the interfering noise was proportional to the speed of the motorcycle as it carried the open microphone from one acoustically predicated position to another. Clearly, without evidence of engine sounds on the Dictabelt, this assumption did not justify naming H. B. McLain as the probable driver of the motorcycle with the stuck open microphone. These actions are the first piece of the puzzle.

Spectrographs produced by BB&N showed that the interfering noise had temporal persistence, which eliminated intermittent sources. This consideration excluded a chattering relay contact or faulty muffler and allowed a malfunctioning electronic component as the source. Under these conditions, the persistent noise would be stationary.

A stationary noise arises from an astronomically huge number of random events. The hugeness of this number renders the fluctuations of statistical measures of a stationary noise imperceptible. For practical purposes these measures do not change with time. Hence any probabilistic measurement made now will produce essentially the same result as an earlier or later measurement. This characteristic enables the spectral subtraction technique to anticipate future from present behavior of stationary noise.

In reporting the failure of spectral subtraction to effect a net reduction of the interference, BB&N cleverly disclosed the non-stationary nature of the noise. (2) This information excluded most natural sources of persistent noise. The failure of BB&N to explicitly describe their evidence of a synthetic source of the loud interference is the second piece of the puzzle.

The HSCA evaded these problems by reporting that BB&N filtered out repetitive noise such as repeated firings of the pistons of the motorcycle engine. (3)

In 1982, the Committee on Ballistic Acoustics, commonly known as the Ramsey Panel (RP), and the Watson Research Center of IBM (WRC) published their reviews of the work done by BB&N and Weiss and Aschkenasy. (W&A)

BB&N and W&A based their findings of four shots fired during the assassination of President Kennedy upon less than perfect matches between pulse patterns on a tape recording of the Dictabelt and patterns produced by test shots. Although the limiting characteristics of the selected patterns provided evidence of four gunshots, these studies were subjective. They used highly restricted positions for the locations of the shooter and the open microphone.

Neither the RP nor the WRC used their computers to systematically search for better matches between the contents of a noisy Dictabelt and simulated shots either fired from or recorded at absurd positions. Clearly one better match would have discredited the pulse matching technique of BB&N and W&A and nullified their findings.

Alternately these signal processing experts could have mimicked mechanics who analyzed the performance of engines from their sounds. By cross-correlating the periodic sounds of various engines recorded by the Channel-I radio system with the contents of the historic Dictabelt, the experts could have detected periodic signals well below the noise level and have identified the type of engine, which carried the open microphone. Both organizations failed to report cross-correlation between known engine sounds and contents of the Dictabelt.

The decisions of the RP and the WRC to ignore both weaknesses in works of their opponents are the third piece of the puzzle.

Instead of using their talents to explore these weaknesses, the RP and the WRC followed an independent path. They intended to show interactions between the crosstalk and other signals in the receiver prove that the crosstalk was present during reception. This method had merit.

During an instance of crosstalk with a simultaneous reception, an unmodulated carrier from a competing transmitter beats with the existing carrier and decreases the frequency deviation of the composite signal. Since the amplitude of the detected audio is proportional to the frequency deviation, keying in of a competing transmitter abruptly decreases amplitude of the crosstalk. Similarly when the competing station keys out, the amplitude of the detected audio increases with comparable rapidity. In both cases the narrow bandwidth of audio circuits stretches response times to hundreds of microseconds.

Alternately if the crosstalk were louder than the interference then a competing station of nearly equal carrier strength would produce a heterodyne whose frequency varied with instantaneous amplitude of the crosstalk. These considerations form the basis for showing the by-radio nature of the crosstalk.

For example, at 60.9 seconds after the start of the Bowles tape of Channel I, a station broadcasted a brief message, that said seventy-five place. A heterodyne began after the second word. This event marked the start of a simultaneous broadcast. The presence of the unmodulated carrier reduced the frequency deviation of the composite signal and decreased the average magnitude of the third word. An oscillograph clearly shows the interaction between simultaneously received signals.

Oscillograph of the Brief Message

The dense black portions of this graph are indicative of average magnitudes and the thin vertical lines show peak magnitudes. Normally the peaks of an uncompressed voice extend three times higher than the average. On this brief message the peaks of the first two words are only slightly more than the average. This shows a high degree of audio compression. However the interaction, which produced the heterodyne decreased the average magnitude of the third word and reduced audio compression. These considerations show simultaneous reception of the brief message and the heterodyne by the Channel-I receiver.

A spectral analysis reveals that the third word did not interrupt the heterodyne. This word increased the frequency deviation of the transmitted carrier and imparted rapid and less audible changes to the frequency of the heterodyne. By both standards, this brief message demonstrates the merits of the interaction of received signals as a test of their by-radio nature.

In implementing this meritorious method, the RP and the WRC misapplied principles of AM receiver operation to the FM radio system of the DPD. They showed that AGC action required tens of milliseconds to reduce amplitude of the crosstalk. Both organizations noted recovery times were much longer and did not specify a number to contradict the report of a one second time constant by BB&N. These published figures on the attenuation and recovery times provide clear and convincing evidence that AGC acted on audio.

These fallacious demonstrations of the by-radio nature of the crosstalk by the intelligent and knowledgeable people of the RP and the WRC represent the fourth piece of the puzzle.

The over driving audio and the heterodyne of the brief message show that the audio stages, which connected the receiver to the Dictaphone operated without AGC. However, an oscillograph of signals during the crucial interval of the reported shots shows a heterodyne with indisputable evidence of AGC action. An unmistakable envelope shapes the heterodyne into a decreasing sinusoid. Following termination of this heterodyne a gradual asymptotic increase in the magnitude of the interference shows the slow AGC recovery time. This interval, 233.3 to 234.7 second from the start of the Bowles tape, contains both characteristics of AGC action and is time stamped by crosstalk. Clearly these results show itinerant behavior of the AGC.

Evidence of AGC Action on Audio

An unwise decision by James Bowles to turn on AGC during the recording of the Dictabelt would account for this itinerant behavior. However, in additional to using the Bowles tape, the RP obtained magnetic recordings of the Dictabelt from James Barger of BB&N and the Department of Justice. Hence dismissal of the itinerant behavior of the AGC as accidental requires several similar and foolish actions while making as many independent tapes of the Dictabelt. Further BB&N and the RP compared their tapes with the Dictabelt and found no significant differences. For these reasons the itinerant behavior of the AGC challenges the authenticity of the Dictabelt and becomes the first cardinal piece of the puzzle.

The reported time constants of AGC actions are further evidence of an altered Dictabelt. These numbers coincide with specifications for equipment used by studios to emphasize and fade audio. The RCA-BA-25A AGC Program Amplifier (4) typifies this equipment. Since attenuation and recovery times characterize AGC action RCA provided the values of the pertinent components. The 56 K resistor and the 0.22 MF capacitor set the attenuation time constant to [(5.6 X 10⁴ Ohm) (2.2 X 10^-7 farad)] or 12.3 millisecond. For the recovery characteristic of the AGC action, the 4.7 MEG resistor and the 0.22 MF capacitor set the time constant to [(4.7 X 10⁶ Ohm) (2.2 X 10^-7 farad)] or 1.03 second. These results show excellent agreement with reported attenuation and recovery times.

An analysis of the most prominent and ignored signal from the Dictabelt, rationalizes the denial by BB&N of their own evidence.

Characteristics of the loud interference changed often and with unexpected rapidity. Fortunately between 99.6 and 101.0 second these changes relaxed and enabled a Fourier analysis to provide useful information. A magnitude spectrograph shows a strong and nearly narrow-band sinusoid at 900 Hz.

Spectral Characteristics of the Relaxed Interference

However, the sound of this signal indicates a loud noise. Calculating the spectrum over shorter and contiguous intervals show that the sinusoid persisted with slightly varying frequency throughout the duration of the longer spectrum. These results show that simultaneous amplitude and frequency modulation of audio was the primary component of the loud signal during this interval. This composition describes a frequency modulated heterodyne in the presence of noise and prompted a systematic study of the spectral changes of the loud interference.

On longer time scales, the spectra of the interference show ordered changes. The dominant frequencies fall into two bands. Lower frequencies appear near 375 Hz and the higher frequencies cluster around 1000 Hz. The relative powers in these bands change with time. These characteristics suggest a ratio of relative powers as a measure of these spectral changes. Dividing the relative power in the low frequency band by the sum of relative powers in both bands gives a measure that is independent of the unknown scale factor relating power to spectral magnitude. This ratio provides an absolute measure of power in the low frequency band as a proper fraction of power in both bands. Changes in this ratio measure changes in the spectra of the interference. However, the frequency selectivity of the DPD radio system and widths of the selected bands influence the magnitude of this ratio. Diminishing the ratios by the mean value of the ratios produce measures, called skewness, that are independent of the frequency response of the system and widths of the selected bands. This study used limits of 300 Hz to 450 Hz for the low frequency band and limits of 800 Hz to 1200 Hz for the high frequency band.

A bar graph of skewness shows that the spectrum of the relaxed interference fits the pattern of neighboring spectra.

Spectral Skewness of the Synthetic Interference

More important the skewness of the relaxed interference coincides with the global minimum of the spectra family. This later consideration is overwhelmingly strong evidence that the loud interference is a simultaneous amplitude and frequency modulation of audio. On this basis the character of the loud interference becomes the second cardinal piece of the puzzle.

The spectral study produced two surprises. By construction the mean value of skewness tends toward zero. However, the bar graph shows two distinct means separated by an abrupt discontinuity. Clearly this discontinuity shows a rapid change in either the radio or recording system.

This surprising change occurs between 225 and 228 second, which precedes the first observation of the itinerant AGC activity at 233 second. The increase of skewness at the discontinuity shows a considerable shift toward a larger fraction of lower frequencies. Both considerations are consistent with the sudden introduction of intermodulation distortion by excessive AGC activity. The magnitude of this distortion excludes AGC circuits of tape recorders as the cause of the itinerant behavior.

The skewness of the loud interference began with a small dip followed by an abnormally large peak that lead to an extended and deep decline. Following an interruption of the loud interference, the skewness reached its minimum value and produced the spectra of the relaxed interference. The skewness started to increase, ascended to a lesser peak and declined. These changes typify the warmup drifts of vacuum tube oscillators during their first minute of operation. Identification of spectral changes that mimic the thermal shock of warming oscillators is the second surprise.

During the fifties the standard design for jamming sources of audio used two main oscillators, which operated over different portions of the spectrum. Two secondary oscillators modulated the main oscillators and produced simultaneous amplitude and wide band frequency modulation of audio. This combination yielded a filter resistant and non-stationary noise. By the early sixties transistors replaced vacuum tubes in jamming sources and the earlier designs were available on the surplus market.

Without doubt, the signals on the Dictabelt are the cardinal pieces of this puzzle.

References

1. Failure of adaptive filtering

1. Further discussion of adaptive filtering

1. Failure of autocorrelation analysis

2. Failure of spectral subtraction

3. Report of succcessful filtering

4. RCA-BA-25A AGC Program Amplifier

Adaptive Filter

An electronic filter separates one frequency from another. However it cannot determine whether any given frequency belongs to the desired or the interfering signal. Hence the characters of the signals determine the effectiveness of filtering.

When the signals have most frequencies in common, filtering is useless. Effectiveness of filtering improves as the number of common frequencies decreases.

An adaptive filter uses autocorrelation to identify the simple spectrum of a periodic signal and adjust its performance to track changes in the period. Hence the techniques employed by BB&N were highly effective at recognizing and filtering out the sounds of engines.

AGC

Automatic gain control, AGC, counteracts overloading of circuits by excessively strong signals. AGC may act upon either audio or radio signals. The impression of audio on amplitude modulated radio signals limits the rapidity of AGC attenuation. Ten milliseconds represents a reasonable compromise between loss of audio due to sluggish response and inaction upon the explosive syllables of speech. Ideally the recovery time should be comparable with the attenuation time. However, this requirement significantly complicates design. Typically they make the recovery time two or three times longer that the attenuation time by adding a five-cent resistor.

Amplitude modulation, AM, impresses audio on a radio signal by varying its amplitude. Successful recovery of this audio requires AGC to prevent the amplitude of the radio signal from exceeding the capacity of the receiver. Hence, AM receivers use AGC at the RF and IF stages to prevent overloading.

Radio signal strength at the receiver affects the volume of recovered audio. AGC levels differences between strong and powerful stations. However, it does not boost the audio from weak stations. Hence AGC of audio is useful with a AM receiver.

Autocorrelation

Periodic signals repeat the same pattern at regular intervals. This characteristic enables recognition of periodicity by sliding an oscillograph over a copy of itself. A coincidence of signal shapes shows periodicity and the time interval between one coincidence and the next equals the period. Normally people perform this test mathematically and call it an autocorrelation.

Frequency modulation, FM, impresses audio on a radio signal by varying its frequency. This design permits full recovery of the audio when strong signals overload stages. For this reason, they use very high gain amplifiers without AGC in the IF stages. In fact many FM receivers include a limiter stage between the IF amplifiers and the discriminator, which recovers the audio from the frequency modulated signal. The purpose of this limiter is to overload and suppress amplitude noise.

The frequency deviation of the transmitter determines the magnitude of the recovered audio. Thus for equal deviations the strongest transmitter produces audio whose volume is comparable with the weakest transmitter. This characteristic of a FM system renders AGC control of received audio unnecessary and even harmful.

Intermodulation Distortion

Distortion is the production of frequencies not found in the original signal. Intermodulation introduces new frequencies that are the sum and difference of all frequencies contained in the original signal. By contrast harmonic distortion adds frequencies that are whole number multiplies of the frequencies in the uncontaminated signal.

From these considerations it follows that harmonic distortion cannot and intermodulation distortion may increase the low frequency content of a signal. When the sums of signal frequencies exceed the high frequency cutoff of the system then intermodulation distortion increases the low frequency content of the signal.

Clearly with 1 khz high frequency cutoff of the DPD radio system, intermodulation distortion would decisively increase the low frequency content of the signal.

Magnitude Spectrogram

A magnitude spectrogram plots magnitude on the y axis against frequency on the horizontal axis. Usually this plot presents the short term spectrum of the signal. Under these conditions, it produces a snapshot of the spectrum. This enables a person to flip a stack of short term spectrographs and view the changes in the spectra as a motion picture.

Periodic

Periodic signals, no matter how complicated, have simple frequency spectra. They consist of narrow-band sinusoids whose frequencies are whole number multiples of a fundamental frequency. The amplitude and phase of the narrow-band sinusoids contain all the complications of the periodic signal.