ABSTRACT: When recording data from more than one source, using more
than one data type, and with
more than one sample rate, it is non-trivial to compare the information and
accurately determine if
features are indeed, simultaneous. An assumption that all the files start
at the same time can be made,
and this may or may not be valid, especially in a case where more than one
computer is used for
collecting the data. A simple method of recording distinctive markers in
the data is described which
ensures than data can be aligned at the beginning and end of the recording.
This method has an
accuracy as good as the lowest sample rate. The method is entirely electronic,
doesn't rely on moving
parts, and is repeatable and reliable. Schematic diagrams are included to
allow implementation.
INTRODUCTION
In a physiology laboratory, there are often several different types of
data being recorded simultaneously. It is
vital that the recorded data be time aligned, so that features can be compared.
Absolute data alignment is only
ensured if the data is gathered from one device on one computer and
saved in one file. If multiple data
sources are used, with different sample rates, and different data representations,
then feature alignment is
problematical and prone to error. This may invalidate all the data and lead
to incorrect interpretation. Making an
assumption about the beginning and end times is dangerous. Marking all the
data being gathered is as simple
as pressing a button. With the correct synchronising circuitry, a uniform
unambiguous and distinct marker can
be placed in all data. The beginning of the recording session can be
marked so that data can be aligned. The
end of the session can also be marked, to ensure that there are no missing
samples due to data under or over
runs, or sample clock slipping. The actual principle is quite straightforward,
and the devices used readily
available. The system is a combination of several simple devices, arranged
in various data streams, and gives
reliable, accurate, and distinct results. The integration into the entire
physiology laboratory system is unique.
The system has been tested and used for long data gathering sessions, recording
speech, E.P.G., E.M.A.,
E.M.G., and airflow data simultaneously. The recorded data was analog
and digital signals, with different
concurrent sample rates of 20khz, and 200hz.
CONTROL
The marking system is manually controlled, so that synchronising markers
can be added at any time, at the
discretion of the operator. There is a simple button that is pressed when
a synchronising marker is desired.
The button is connected to circuitry, which ensures that there is only one
synchronising pulse, of a fixed
duration. The press button switch has a debounce filter to prevent false
triggering and multiple pulses. This is
connected to a simple 555 timer IC (Integrated Circuit) which creates a fixed
width pulse, with clean and steep
attack and decay edges. This is the synchronising pulse. A transistor is
used in an open collector arrangement,
so that it can drive any and all synchronising slave devices in the laboratory.
The first device it drives is a LED
(Light Emitting Diode) which merely indicates that the circuitry is functioning.
A connector on the back of the
control panel, feeds the synchronising pulse to all the synchronising slaves
in all the data gathering devices in
the laboratory.
ANALOG SWITCHES
The devices used to insert the synchronising data, are analog switches.
These behave like a normal switch, but
switch much faster, have no contact bounce, and make reliable connections.
These are superior to mechanical
relays. The 4053 device was chosen for this function, as it has been available
for a long time, is well proven in
switching characteristics, has a useful dynamic range, and operates from
5 volt logic. This device has 3 change
over switches in the one package. Each analog switch has a low ON resistance,
in the order of 300 ohms. The
high impedance input of the A/D (Analog to Digital) converter is in the order
of megohms, and the output from
the devices being measured are amplifier IC devices which have a low output
impedance. The analog switch
thus does not load, or modify the analog waveform in any way. When an analog
switch is placed in an analog
line, the analog data passes straight through, (when there is no synchronising
pulse) so data can be measured
in the normal way. When the analog switch is activated, it switches, so that
the A/D converter is connected to a
distinctive signal, for the duration of the switch pulse.
ACOUSTIC DATA
It is easy to mark audio, just by tapping the microphone with your finger.
This places a loud noise in the audio
data, but the pulse is variable in duration and loudness, and bears no relation
to any of the other data being
gathered. A better way to achieve the same result, is to use an audio oscillator,
which is controlled from the
synchronising pulse. The oscillator is in the form of a simple 555 timer
IC which is connected as a free running
multivibrator. The circuit has a control to set the audio frequency. The
output from this IC drives a small
speaker which is mounted on the headrest of the dentists chair, where the
subject sits. When a microphone is
placed for recording the subjects speech, it is also perfectly placed to
pick up the tone from the speaker. The
tone appears in the audio data as a neat sinewave, of a distinct frequency
and duration, and with a uniform
amplitude. It is absolutely unique even when speech is occurring, and any
FFT analysis or spectrogram adds
further information, if there is any doubt. This marker is independent of
audio sampling frequency
ELECTROPALATOGRAPHY (E.P.G.)
We have an EPG3+ in our laboratory that is integrated into our data gathering
system. This device measures
the subject's tongue contact with the palate. The E.P.G. data is gathered
in a digital form of eight 8 bit bytes, at
a slow sample rate of 200 hz. This can be displayed as a map of the
palate. The way the device works, is
determined by body conduction. The subject holds an electrode that is energised
by a low level oscillator. An
artificial palate, which is individually made for each subject, is placed
in the subject's mouth. The artificial palate
has 62 small metal contacts arranged in an eight by eight matrix, so that
the whole area of the roof of the
mouth is covered. When the tongue touches the roof of the mouth during speech,
the low oscillator energy is
transferred to the contacts, and then by small wires, to detectors in the
E.P.G. control unit. The control unit
changes this to a digital form, to be recorded by the computer. It
can then be displayed as a series of maps,
and also as a time varying waveform indicating total contacts in the alveolar,
palatal and velar areas. The maps
are displayed as a square of eight rows by eight columns. The front row near
the teeth, has the two outer
contacts missing from the palate, as the physical area in the mouth is quite
small. This is useful in orientation of
the palate map. The map shows free contacts as white, and contacts with the
tongue as black.
Marking the data presents a problem. There is no slowly varying analog data
that can be easily interrupted
before they are measured by an A/D converter. The digital data could be changed
by placing logic gates in the
device output, and pulsing them to logic 1 for synchronisation. I chose an
alternative method. A small P.C.B.
(Printed Circuit Board) is placed between the palate and the control box.
This involves no modifications to the
E.P.G., and can be easily disconnected for testing and maintenance. The contact
sensor lines, from the palate
to the control box detectors, pass through analog switches, which are connected
straight through for normal
data collection. During the synchronisation pulse, the switches change, and
connect the detectors directly to
the oscillator, and this mimics a tongue contact. The oscillator signal has
its level controlled so that it
represents a tongue touch. All the contacts are connected, and this results
in a completely black palate map.
This is unusual in two ways. Firstly, it is very difficult for the tongue
to touch the entire roof of the mouth at any
one time. Secondly, the outer two contacts for the front row (which
don't exist), are represented as being
connected. In this way, an entirely black palate map is unique, and easily
distinguishable as a synchronising
event.
Fig 1. ELECTROPALATOGRAPH SYNCHRONISER AND CONTROL SCHEMATIC DIAGRAM
ELECTROMAGNETICARTICULOGRAPHY (E.M.A.)
Our laboratory has a 10 channel Carstens AG100 Articulograph. This device
allows sensors attached to the
tongue, to have their movements recorded in real time by a computer. In its
original form, the Carstens
Articulograph AG100 has no easy method available to add a synchronising pulse
to the data. It is housed in a
desktop cabinet that has a backplane with plug in cards. The analog data
from each receiving card is
connected via the backplane to the processor card. The processor card contains
a multiplexer, A/D, I.E.E.E.
interface and a microprocessor. The card is sealed with a layer of compound,
to prevent contamination, and
this precludes modification. There is a separate backplane and processor card
for each 5 channels, our 10
channels version, having two backplanes. There are several methods by which
the data might be marked. A
large transmitter coil could be mounted on the helmet, and a pulse applied.
The disadvantages are the weight,
the exciting frequencies required, and the possible field exposure to the
subject. Another method could be a
box added in series with each transmitter, to momentarily interrupt or boost
the transmitter coil output. The
disadvantages are that it would require analog power switches as the transmitters
are driven by power
amplifiers. Mechanical relays are undesirable due to changeover times and
possible mechanical unreliablilty. A
further method could be a box added in series with each of the 10 sensor leads,
which interrupts the signal.
This may be a reasonable approach, providing the analog switches used have
a suitable frequency response.
The disadvantages are that it requires some messy cabling, and may interfere
with the sensor calibration.
The simplest and neatest way to add a synchronisation module, without modifying,
voiding warranty, or
damaging the Articulograph cards in any way, is to add an extender card to
each processor card. An extender
card is normally used to move a card out of the card frame, to make it accessible,
so that it can be examined
for faults. It is in the form of a blank card, with a connector on each end,
with pin for pin wiring. An extender
card can be made, which contains the synchronising circuitry. The card need
not be the full depth of the card
cage, but only enough to hold the synchronising circuitry. The processor card
will be projecting proud of the
front panel by this distance. In this way the Articulograph can be easily
restored to normal, merely by removing
the synchronising extender. The synchronising pulse cable can be plugged into
the card with a flying lead, or
attached to the back panel with a connector. 
Fig 2. ELECTROMAGNETICARTICULOGRAPHY SYNCHRONISER SCHEMATIC
DIAGRAM
Analog switches are placed in each analog line. Every other power, digital
and earth line is connected directly
through. When a synchronising pulse occurs, each switch is activated, and
each analog line is disconnected
from the sensor, and connected to +5 volts, so the A/D converter input suddenly
changes. This is far greater
than the normal voltage from the sensor, and appears as a distinctive pulse.
The height of the pulse is unique,
and the steep rising and falling edges make it appear like a digital waveform,
which is obviously different from
the relatively slow changing analog waveform. The analog switches are powered
from the backplane.
In the S.H.L.R.C. physiology laboratory, the data gathering system does not
use the Articulograph I.E.E.E.
interface. The analog signals from the backplane are connected to the normal
S.H.L.R.C. A/D converters. The
Articulograph processor cards are removed, and a S.H.L.R.C. designed and
built card is inserted that connects
the Articulograph analog signals to the system A/D via a cable. The interface
card contains synchronising
circuitry operating in the same way. A circuit has already been published
(ROBINSON 2000).
AIRFLOW AND ELECTROMYOGRAPHY (E.M.G.)
The laboratory has three air sensors which measure the nasal and oral airflow
and the intra-oral pressure
during speech. The analog outputs of these are connected to an A/D converter.
There are also four E.M.G.
amplifiers which have muscle sensing electrodes which can be attached to
the skin on the face to detect
muscle activity during speech. The analog outputs of these are also
connected to an A/D converter. The A/D
converter has sixteen inputs available for any physiology data collection,
but the seven mentioned are always
connected and recorded. The adding of synchronising markers is done in the
same way as previously
mentioned. Analog switches are used in each analog line. The switches are
on a P.C.B. that is placed between
the sensor output and the A/D input.
Fig 3. AIRFLOW AND ELECTROMYOGRAPH SYNCHRONISER SCHEMATIC DIAGRAM
ACCURACY
The marker accuracy is as good as the sample rates used. In the case of
the 20khz sample rate used for
speech, the beginning of the acoustic tone marker, can be determined with
50 microsecond accuracy. In the
case of the 200hz sample rate used for E.M.A. and E.M.G. signals, the beginning
of the markers can be
determined with 5 millisecond accuracy. In the case of the 200hz sample rate
used for the E.P.G. digital data,
the marker can be seen for at least two palate maps and may begin and end
midway through a palate. This is
because the complete palate map is started at the sample clock initiation,
but takes a finite time to access all
palate contacts. This does not affect the accuracy because the complete palate
contact map access time is
much less than the sample rate.
CONCLUSION
The physiology synchroniser is an essential device which verifies the integrity
of simultaneous multiple
disparate data gathering, and removes any doubt regarding data concurrency.
The design is simple, easy to
understand, and easy implement. The operation is easy, non restrictive,
and reliable. The markers are distinct,
unambiguous and contribute to the sanity of the data analysis. Any data not
gathered using a reliable
synchronising technique may be suspect and open to argument and misinterpretation.
ACKNOWLEDGMENTS
Thank you to all those that helped in this project. Chris Callaghan from
S.H.L.R.C. did the layout and
construction of all the printed circuit boards. He also drew all the final
drawings from my rough sketches, and
designed the pulse generator and audio oscillator section of the E.P.G.
synchroniser board.
REFERENCES
Robinson, R.E.E. (2000). Articulograph Interface, Proceedings of the Eight
Australian Conference on Speech
Science and Technology, Canberra, December 2000, 484 - 489.
Fig 4. EXAMPLE OF SYNCHRONISING MARKERS (ACOUSTIC,
EMA, EPG)