|
ULTRASONIC
DOMESTIC GAS
METERS-A REVIEW
N. Bignell
National Measurement Laboratory
CSIRO Telecommunications and Industrial Physics
Lindfield, NSW, Australia
Page 2/4
3
DUCTS, FLOW AND ULTRASOUND
xxxxIn the construction
of a gas meter, the gas must flow through a duct
and the ultrasound must pass through the flowing
gas. When ultrasound propagates in a duct it generally
does so as a series of modes. The exact nature
of these modes depends very much on the geometry
of the duct, but they travel at speeds that decrease
as their complexity increases. The plane wave
mode is the simplest and always propagates. Other
modes have a cut-off frequency and they will not
propagate in a given duct, below this frequency.
As a mode approaches its cut-off frequency, its
speed approaches zero.
xxxxThe waveform
that is received after travelling through a duct
is comprised of the combined signals from many
modes and this has the effect of prolonging the
signal seen. The use of a transducer with a large
Q also prolongs the signal. This is of importance
in the pulse repetition method of timing.
xxxxThe presence
of flow in the duct changes the shape of the received
waveform. This can be illustrated for a cylindrical
duct such as that shown in Figure 1, by the waveforms
shown in Figure 2.
xxxxHere the relative
peak heights of the signal transmitted with the
flow are different from those
transmitted against the flow. This is very inconvenient
for timing based on a particular zero crossing
of the received waveform. If this crossing is
being detected using an electronic threshold,
illustrated by the heavy horizontal line in Figure
2, then it will select a different crossing depending
on the direction of transmission. This causes
a very serious timing error.
Figure 1. A cylindrical metering tube with transducers
and mode control devices (A).
xxxxA
solution to this problem has been to modify the
flow path by inserting mode control devices [3]
into the duct. These are shown in Figure 1. Other
meters use an element down the axis of the tube.
Sometimes the transducer is made considerably
larger than the duct to avoid the generation of
modes other than the plane wave [5].
Figure 2. Upstream (light trace) and downstream
waveforms, with gain adjusted to make them of
equal height. The origin of the time axis is arbitrary.
The heavy horizontal line represents a threshold.
4
TIMING TECHNIQUES
xxxxThe timing of
the signal in the two directions must be done
with an uncertainty of about 3 ns if the specification
is to be met for the uncertainty at low flow rates.
This is quite difficult to achieve when the restriction
of low power consumption is applied. A timing
clock of even 10 MHz will allow direct timing
to only 100 ns. An advantage is the very large
number of measurements made in the billing period.
If these measurements are truly random a high
single measurement uncertainty can be tolerated
while still achieving a low uncertainty in the
mean value. The meters developed so far do not
rely on this averaging to achieve their required
uncertainty. This is probably because they must
show their performance ability over a much shorter
period during the calibration and testing part
of their operation.
4.1 Pulse repetition
xxxxIn this technique
[3], a low frequency clock is used and the time
to be measured is increased by sending the signal
down the tube a number of times. A timer is started
as the first pulse is sent down the tube. When
it arrives it is detected and another pulse is
immediately sent down, in the same direction,
and so on for, say 100 pulses. When the 100th
pulse arrives the timer is stopped. Thus the time
that is measured is 100 times that for one pulse
and so a clock frequency 100 times less may be
used. This works well and allows resolutions of
a few nanoseconds with a clock period of 100 ns.
There are, however, some drawbacks detailed in
section 5.
4.2 Phase methods
xxxxAnother method
[8] of timing uses a special drive signal of 24
cycles of a sinusoid with a phase
reversal built into it two thirds of the way through.
The drive signal is generated from a 1.44 MHz
clock by counting down to 180 kHz so that it is
phase locked to it. Members of a group of eight
capacitors are switched in turn by the clock to
sample the received signal. During the 16 cycles
before the phase switch they form a good average
of the incoming waveform. A phase detector is
used to compare the incoming wave with this average
and hence the reversal is detected and the sampling
stopped. This measurement establishes the time
to one clock pulse but this is not nearly accurate
enough. The phase of the stored waveform on the
capacitors is then investigated. The voltage on
each of the eight capacitors is read by an analogue
to digital converter. If the phase reversal stopped
the data collection at exactly the start of the
phase of the received signal then the received
signal and the driving signal ( and the clock)
would be in phase and an integral number of clock
pulses would correspond to the transit time to
be measured. Usually there is a phase difference
that needs to be determined by the curve fittingxprocedure
used. It is claimed that this can be done to one
thousandth of a period of the signal thus achieving
an accuracy of several nanoseconds.
xxxxIn a similar
technique [2], the transducer is excited with
a tone burst of 8 cycles at 40 kHz. The
received waveform is sampled at 320 kHz to give
the data set y(ti). The phase is given by
 xxxxxxxxxxxx(2)
which
is more easily calculated than might appear since
the sine and cosine values for eight samples per
period are constrained to be either zero or ±1
or  .
It can only be determined between 0 and 2p. To
remove the phase ambiguity (or to do "phase
unwrapping") a separate, direct measurement
of the time of flight is done using a threshold
and comparator, exciting the transducer with a
single pulse.
4.3 Clock period interpolation
xxxxA portion of
the received waveform is digitised at a rate equal
to the clock rate and these data are stored. If
timing is done to a zero crossing it is easy to
find the integral number of clock pulses that
finish just before that crossing. Then an interpolation
is done to determine that fraction of a clock
period to the crossing.
xxxxIt is also possible
to interpolate by using a fast voltage ramp lasting
one clock period with a circuit that samples this
voltage at the instant of the event being timed.
The voltage sampled divided by the maximum voltage
for the ramp, is the fraction of the clock period
required.
|
