Propagation/Timing Delay Question

[George W]

I'm developing a product based on ordinary low-cost handheld VHF FM
transceiver equipment with modifications.

The transceiver will receive a short carrier burst containing an audio
tone pulse (1 kHz or so) that serves as a timing reference. It must
then turn on its transmitter and transmit an audio tone pulse a precise
amount of time following the timing reference pulse. Transceiver
operation will be simplex mode.

The amount of time between the two pulses is not important. It can be
200 msec or more, as needed to let the transmitter and receiver settle
after turning on and off. The only requirement is that the time between
pulses be kept to a close tolerance of 50 nanoseconds if possible.

The transceiver filters and logic would contribute to the tolerance.
Can anyone give me an educated guess as to:

How uncertain would the timing of the control logic be? That is, the
time between when the incoming audio reference burst is detected and the
time when the following burst is sent? I assume this is a clocking
stability question. The circuit would utilize a common low-cost
clocking reference and components.

How much variation can be expected in propagation time through the RF
interstage filters? Would this vary much from unit to unit and over
temperature? If so, calibrating each transceiver during manufacture is
a possibility.

Thanks for any input.

George

 

[John Larkin]

I'm developing a product based on ordinary low-cost handheld VHF FM
transceiver equipment with modifications.

The transceiver will receive a short carrier burst containing an audio
tone pulse (1 kHz or so) that serves as a timing reference. It must
then turn on its transmitter and transmit an audio tone pulse a precise
amount of time following the timing reference pulse. Transceiver
operation will be simplex mode.

The amount of time between the two pulses is not important. It can be
200 msec or more, as needed to let the transmitter and receiver settle
after turning on and off. The only requirement is that the time between
pulses be kept to a close tolerance of 50 nanoseconds if possible.

The transceiver filters and logic would contribute to the tolerance.
Can anyone give me an educated guess as to:

How uncertain would the timing of the control logic be? That is, the
time between when the incoming audio reference burst is detected and the
time when the following burst is sent? I assume this is a clocking
stability question. The circuit would utilize a common low-cost
clocking reference and components.

How much variation can be expected in propagation time through the RF
interstage filters? Would this vary much from unit to unit and over
temperature? If so, calibrating each transceiver during manufacture is
a possibility.

Thanks for any input.

George

I'd prefer not to use the word "impossible", but it's just about
there. Timing the receive event to 50 ns is extreme, and generating a
200 ms delay accurate to 50 ns is a delay accurate to 0.25 PPM, hardly
"common low-cost clock" turf. Temperature and noise will be nasty
problems, too.

John

 

[ChrisGibboGibson]

I'd prefer not to use the word "impossible", but it's just about
there. Timing the receive event to 50 ns is extreme, and generating a
200 ms delay accurate to 50 ns is a delay accurate to 0.25 PPM, hardly
"common low-cost clock" turf. Temperature and noise will be nasty
problems, too.

John

Trying to detect the exact time that a 1KHz sinewave was switched on after
having passed it through a few 3KHz bandwidth (assuming communication
equipment) channels is, I believe, impossible to the accuracy required by OP.

Now let's watch someone tell us exactly how to do it

Gibbo

 

[Joerg]

Hi Chris,

Trying to detect the exact time that a 1KHz sinewave was switched on after
having passed it through a few 3KHz bandwidth (assuming communication
equipment) channels is, I believe, impossible to the accuracy required by OP.

Now let's watch someone tell us exactly how to do it

It can theoretically be done if you know all parameters of the chain.
For example prop delay in the IF filter, whether it is 3KHz bandwidth or
more. There will also be the ramp time of the detector that detects the
presence of the 1KHz tone. The behavior of that AM detector that sits
after the FM demodulator needs to be known precisely. If you don't know
all of this there would be the method of testing the total delay and
ramping pattern but that would be venturing onto thin ice. Since we are
talking 50nsec here about 1/50th of a degree of phase error in the 1KHz
signal would throw everything off kilter already. Not a whole lot.

One low cost way to do this in the real world is to detect the umpteenth
zero crossing of the 1KHz signal. Just pick one and stick to that. It
has to be the exact zero crossing, not a few millivolts up or down. That
could be achieved by a differential measurement. This is the way it used
to be done in Doppler ultrasound where you sometimes have to measure
minute movements of blood or vessel walls. This strategy would get rid
of the uncertainty, drift and what not of an additional AM detector.

Then, as John said, the whole circuitry must contain a good timing
reference. I would roll my own crystal oscillator in a case like this.
An off the shelf oscillator in a can isn't likely going to cut the
mustard. After all, a gourmet dinner doesn't come in a can either. And
watch out for that FM detector. The slightest DC offset (by frequency
drift etc.) and the precision is toast. The same goes for the modulator
in the transmitter.

Regards, Joerg

 

[Rich Grise]

Hi Chris,
....
Then, as John said, the whole circuitry must contain a good timing
reference. I would roll my own crystal oscillator in a case like this.
An off the shelf oscillator in a can isn't likely going to cut the
mustard. After all, a gourmet dinner doesn't come in a can either. And
watch out for that FM detector. The slightest DC offset (by frequency
drift etc.) and the precision is toast. The same goes for the modulator
in the transmitter.

Hasn't somebody, somewhere, already figured out how to make a
transponder? Is the OP trying to reinvent the wheel?

http://www.google.com/search?q=transponder&btnG=Search

Good Luck!
Rich

 

[Joerg]

Hi Rich,

Hasn't somebody, somewhere, already figured out how to make a
transponder? Is the OP trying to reinvent the wheel?

http://www.google.com/search?q=transponder&btnG=Search

Yes, but if you look at the specs the OP desires and narrow the search
down to those

http://www.google.com/search?as_q=t...s_occt=any&as_dt=i&as_sitesearch=&safe=images

then you are quickly relegated to very high frequencies and not a
regular radio. Doesn't mean it can't be done but it kind of hints that
it might indeed be a challenge.

Regards, Joerg

 

[George W]

Joerg wrote:

One low cost way to do this in the real world is to detect the umpteenth
zero crossing of the 1KHz signal. Just pick one and stick to that. It
has to be the exact zero crossing, not a few millivolts up or down. That
could be achieved by a differential measurement. This is the way it used
to be done in Doppler ultrasound where you sometimes have to measure
minute movements of blood or vessel walls. This strategy would get rid
of the uncertainty, drift and what not of an additional AM detector.

If I understand, I don't think you're saying to count zero crossings
until you get to the one you've picked, and use that crossing as a time
marker, are you? I think you're saying instead to take sufficient time
to be able to make an accurate differential measurement of zero
crossings, and the amount of time needed can be guesstimated as waiting
until X zero crosings have occurred. Then that last zero crossing time
can be used as the time marker that is needed. Right?

So, assume 1 kHz tone amplitude of 1 Vp-p. In 50 nsec after crossing
zero its amplitude will be 0.2 mV, and that's what has to be detected in
the presence of a noise floor and all the other stuff.

I had a similar idea in mind, and would appreciate your comments.
Suppose you let the tone run for a while (hundreds of msec would be OK)
so that a PLL can be locked to it. At the source, the tone is then cut
off abruptly. It would be cut off at the peak of its waveform so that
the transition from on to off would be as abrupt as possible.
(Unfortunately filtering ahead of the PLL detector would round off the
transition contributing to errors.)

Given that you've had a long time to close a PLL on the tone, its abrupt
transition straight down from the voltage maximum, and the fact that
this measurement is being made at 1 Volt rather than mucking around in
the receiver's noise floor, might this approach work?

Thanks for taking the time to read all this ... )

George

 

[ChrisGibboGibson]

George W wrote:

[large snip]

So, assume 1 kHz tone amplitude of 1 Vp-p. In 50 nsec after crossing
zero its amplitude will be 0.2 mV, and that's what has to be detected in
the presence of a noise floor and all the other stuff.

I had a similar idea in mind, and would appreciate your comments.
Suppose you let the tone run for a while (hundreds of msec would be OK)
so that a PLL can be locked to it. At the source, the tone is then cut
off abruptly. It would be cut off at the peak of its waveform so that
the transition from on to off would be as abrupt as possible.

It sounds like you're expecting this signal to now suddenly fall from 1 volt to
zero volts in zero time ?

How fast do you think it can do this through a 3KHz (or 10KHz or whatever)
bandwidth channel ?

Gibbo

 

[George W]

Rich Grise wrote:

Hasn't somebody, somewhere, already figured out how to make a
transponder? Is the OP trying to reinvent the wheel?

Hi Rich. I'd love to find something that exists but I believe the
requirements in this case call for a home grown solution.

If this were a conventional transponder (RX straight thru to TX in real
time) the radio would have to operate in duplex mode. Unfortunately I'm
limited to operating in simplex. I agree with Joerg that it's probably
doable but a challenge!

George

 

[Joerg]

Hi George,

If I understand, I don't think you're saying to count zero crossings
until you get to the one you've picked, and use that crossing as a
time marker, are you? I think you're saying instead to take
sufficient time to be able to make an accurate differential
measurement of zero crossings, and the amount of time needed can be
guesstimated as waiting until X zero crosings have occurred. Then
that last zero crossing time can be used as the time marker that is
needed. Right?

Actually in your case you'd have to also detect the onset of the 1KHz
packet and this detection needs to be accurate enough never to slip a
cycle. This is because, if I understand correctly, the timing of the
event is unknown. After that, you can pick any zero crossing as long as
it is always the same. Integrating the measurement routine over several
crossings would improve the SNR and thus your accuracy. So the longer
the tone lasts the more accuracy you can squeeze out.

So, assume 1 kHz tone amplitude of 1 Vp-p. In 50 nsec after crossing
zero its amplitude will be 0.2 mV, and that's what has to be detected
in the presence of a noise floor and all the other stuff.

There is one of the challenges. It can be done but it is best to use a
differential method to avoid offset errors and you need a good strong RF
signal coming in. Detecting several crossings and integrating the
results will improve the SNR and jitter.

I had a similar idea in mind, and would appreciate your comments.
Suppose you let the tone run for a while (hundreds of msec would be
OK) so that a PLL can be locked to it. At the source, the tone is
then cut off abruptly. It would be cut off at the peak of its
waveform so that the transition from on to off would be as abrupt as
possible. (Unfortunately filtering ahead of the PLL detector would
round off the transition contributing to errors.)

Given that you've had a long time to close a PLL on the tone, its
abrupt transition straight down from the voltage maximum, and the fact
that this measurement is being made at 1 Volt rather than mucking
around in the receiver's noise floor, might this approach work?

The abrupt shut-off would be of no concern if you stop using the PLL
just before that event. It should be no problem because the length of
the transmission is under your design control so the receiver would
"know" when to quit logging the PLL.

A PLL is a good idea but more work to design. However, considering all
the chips you can get nowadays it might have become the easier option by
now. The tone needs to be long enough for the PLL to level off at the
desired accuracy. Here you have the usual trade-off between slower loop
response or more noise. I am afraid a few 100msec won't cut it at all
here. It is only a few hundred cycles with a 1KHz tone. Not a whole lot.
The fancy approach would be a multi-loop PLL but now we might enter the
terrain of "over-sophistication".

Yet another approach is to send out a unique binary sequence and run a
matched digital filter against it. There are several codes that are used
for this purpose and probably someone makes chips with such a code and
filter built in. That would give you absolute (time of onset) and
relative (50nsec spec) at the same time. I am partial to the zero
crossing scheme. But then again, maybe that is because I am an analog guy.

Regards, Joerg

 

[John Fields]

If this were a conventional transponder (RX straight thru to TX in real
time) the radio would have to operate in duplex mode. Unfortunately I'm
limited to operating in simplex. I agree with Joerg that it's probably
doable but a challenge!

 

[George W]

The abrupt shut-off would be of no concern if you stop using the PLL
just before that event. It should be no problem because the length of
the transmission is under your design control so the receiver would
"know" when to quit logging the PLL.

Joerg, I think you're saying that a PLL with small noise BW would be
useless to detect the sharp OFF transition in the waveform, so you gotta
get it out of there at the right time and use something else that can
detect the sharp OFF transition quickly.

I'm starting to think that the way to go is to use a DSP to first lock
on to the tone like a PLL, then to look at the tone in, say, 50 nsec
time slices and detect when the waveform drops below the value that the
DSP predicted would occur based on the waveform's history. That would
mean it's headed south fast, and that's the OFF event needed for timing
reference.

A PLL is a good idea but more work to design. However, considering all
the chips you can get nowadays it might have become the easier option by
now. The tone needs to be long enough for the PLL to level off at the
desired accuracy. Here you have the usual trade-off between slower loop
response or more noise. I am afraid a few 100msec won't cut it at all
here. It is only a few hundred cycles with a 1KHz tone. Not a whole lot.
The fancy approach would be a multi-loop PLL but now we might enter the
terrain of "over-sophistication".

Sounds like the PLL may not work anyway if its BW makes it too slow to
detect the sharp OFF transition.

Yet another approach is to send out a unique binary sequence and run a
matched digital filter against it. There are several codes that are used
for this purpose and probably someone makes chips with such a code and
filter built in. That would give you absolute (time of onset) and
relative (50nsec spec) at the same time. I am partial to the zero
crossing scheme. But then again, maybe that is because I am an analog guy.

Wow - that's beyond my level!

Thanks again.

George

 

[Joerg]

Hi George,

Joerg, I think you're saying that a PLL with small noise BW would be
useless to detect the sharp OFF transition in the waveform, so you
gotta get it out of there at the right time and use something else
that can detect the sharp OFF transition quickly.

The PLL would not detect the sharp OFF transition, it would not be very
suitable for that. The PLL would detect the exact phase position of the
sine wave. Something else would have to detect the ON or OFF transistion
but that is rather easy because for that you only need a few hundred
usec precision. The PLL would do the fine stuff. But again, I'd first
look into zero crossing techniques. I have only done these at much
higher frequencies above 2MHz but it didn't require a lot of hardware,
just good low noise design.

I'm starting to think that the way to go is to use a DSP to first lock
on to the tone like a PLL, then to look at the tone in, say, 50 nsec
time slices and detect when the waveform drops below the value that
the DSP predicted would occur based on the waveform's history. That
would mean it's headed south fast, and that's the OFF event needed for
timing reference.

I don't think determining a 50nsec accuracy with the ON or OFF events is
going to work. You really need to determine the exact position of the
phase of your 1KHz signal.

a matched digital filter against it. There are several codes that are
used for this purpose and probably someone makes chips with such a
code and filter built in. That would give you absolute (time of onset)
and relative (50nsec spec) at the same time. I am partial to the zero
crossing scheme. But then again, maybe that is because I am an analog
guy.

Wow - that's beyond my level!

If you use a DSP it isn't really rocket science. But the trick will be,
as with the 1KHz method, to treat the signal with silk gloves.

Another idea: Why don't you do all that at the RF level? Fire up the
transmitter, gently modulate it with something unique, wait until
stabilized, then close an RF gate with good isolation and synchronized
to the RF signal.

Now the receiver would measure the exact arrival of the RF and also
verify the unique modulation, to avoid triggering on a foreign signal.
Measuring zero crossings increases in precision as the frequency
increases. Just look at the frequency: At, say, 200MHz there are already
10 cycles in a 50nsec period. A simple threshold comparator circuit
could already achieve a timing accuracy far beyond of what you need. Of
course you'd have to know the group delay of all the filters in the path
up to this comparator. The wider the bandwidth of your total path the
lesser the effect of group delays.

Regards, Joerg

 

[George W]

Hi George,

Another idea: Why don't you do all that at the RF level? Fire up the
transmitter, gently modulate it with something unique, wait until
stabilized, then close an RF gate with good isolation and synchronized
to the RF signal.

OK, if the receiver has a DSP up front at the VHF input and the transmit
carrier is gated on cleanly at the source after it stabilizes then we should
be able to tell within a few RF cycles when the carrier arrives if S/N is
reasonable.

Now the receiver would measure the exact arrival of the RF and also
verify the unique modulation, to avoid triggering on a foreign signal.
Measuring zero crossings increases in precision as the frequency
increases. Just look at the frequency: At, say, 200MHz there are already
10 cycles in a 50nsec period. A simple threshold comparator circuit
could already achieve a timing accuracy far beyond of what you need.

Now I remember. About 10 years ago my company developed a satellite modem
using a DSP to do PLL threshold-extended FM demodulation. (We would need
just a simple discriminator in the present application.) I don't remember
whether the DSP was at the 70 MHz input or at a lower frequency after a
mixer. It might have been at 10.7 MHz I think.

It's been a few years since I looked at prices, but DSPs capable of 200 MHz
operation were pretty pricey then and they may still be. May reduce battery
life too. Attractive, though, because they can do a lot in addition to
serving as the FM demodulator. If too expensive at VHF one could be
mplemented after a mixer stage but then you have to pay for the mixer,
filter etc in addition to the DSP and it starts to lose it advantages for
counting RF cycles accurately as the frequency goes lower. Hmmmm ....

George

 

[Mac]

OK, if the receiver has a DSP up front at the VHF input and the transmit
carrier is gated on cleanly at the source after it stabilizes then we should
be able to tell within a few RF cycles when the carrier arrives if S/N is
reasonable.


Now I remember. About 10 years ago my company developed a satellite modem
using a DSP to do PLL threshold-extended FM demodulation. (We would need
just a simple discriminator in the present application.) I don't remember
whether the DSP was at the 70 MHz input or at a lower frequency after a
mixer. It might have been at 10.7 MHz I think.

It's been a few years since I looked at prices, but DSPs capable of 200 MHz
operation were pretty pricey then and they may still be. May reduce battery
life too. Attractive, though, because they can do a lot in addition to
serving as the FM demodulator. If too expensive at VHF one could be
mplemented after a mixer stage but then you have to pay for the mixer,
filter etc in addition to the DSP and it starts to lose it advantages for
counting RF cycles accurately as the frequency goes lower. Hmmmm ....

George

Well, if you want to directly sample VHF of 200 MHz, you need a 500 MHz
ADC, and a steep filter to block all the higher frequencies.

Or you could under sample, but then you'll still need a pretty fast
ADC with a bandpass filter in front.

I don't think you want to go down this path.

--Mac

 

[George W]

Well, if you want to directly sample VHF of 200 MHz, you need a 500 MHz
ADC, and a steep filter to block all the higher frequencies.

Or you could under sample, but then you'll still need a pretty fast
ADC with a bandpass filter in front.

I don't think you want to go down this path.

Sounds like you'd also need a preselector / BPF up front of the ADC to
allow just the wanted RF channel to get through? The BPF would have to
be down, say, 60 dB or so at 15 kHz away (the channel spacing) from the
wanted channel. Seems hard to do at 200 MHz.

George

 

[George W]

Well, if you want to directly sample VHF of 200 MHz, you need a 500 MHz
ADC, and a steep filter to block all the higher frequencies.

Or you could under sample, but then you'll still need a pretty fast
ADC with a bandpass filter in front.

I don't think you want to go down this path.

--Mac

Oops, you did mentionthe bandpass filter. Sorry ...

 

[Joerg]

Hi George,

OK, if the receiver has a DSP up front at the VHF input and the transmit
carrier is gated on cleanly at the source after it stabilizes then we should
be able to tell within a few RF cycles when the carrier arrives if S/N is
reasonable.

You don't have to do this with a DSP. It can also be done with a
detector. The trick is to provide a narrow enough filter to avoid false
triggers and there are many ways to do that. Directly at the RF level it
may be cumbersome because large resonators are bulky and expensive. I'd
probably down-convert to something reasonable, maybe in the 50MHz range
where you can achieve enough precision but can obtain or build filters
at reasonable expense. This has been a long time ago but TV sets contain
IF filters of about 5MHz bandwidth and because they are made by the
gazillion should be cheap. They are also carefully designed for a nice
flat group delay.

Of course the filter will shallow the slope of the onset. The lower the
bandwidth the more difficult it will be to obtain 50nsec precision. You
would need to accomodate for variances in RF signal strength which can
be done by registering the RF level once the detector has ramped up and
then adjust the measured slope position to compensate if the RF level
changed because of obstacles or whatever may be in the path.

I don't know the application but if it is short range and you can use an
ISM band with plenty of bandwidth there should not be too many issues
with interference.

It's been a few years since I looked at prices, but DSPs capable of 200 MHz
operation were pretty pricey then and they may still be. May reduce battery
life too. Attractive, though, because they can do a lot in addition to
serving as the FM demodulator. If too expensive at VHF one could be
mplemented after a mixer stage but then you have to pay for the mixer,
filter etc in addition to the DSP and it starts to lose it advantages for
counting RF cycles accurately as the frequency goes lower. Hmmmm ....

Prices have come down big time, mostly because of the advent of DSL and
cable modems. Check out TI. Yes, these DSP can do all the math you want
to perform afterwards on the same chip that does the detection. Also,
you may be able to pull routines from a library if the manufacturer
provides them. But in this application a DSP may still be overkill.

Regards, Joerg

 

[Tim Shoppa]

I'm developing a product based on ordinary low-cost handheld VHF FM
transceiver equipment with modifications.

The transceiver will receive a short carrier burst containing an audio
tone pulse (1 kHz or so) that serves as a timing reference. It must
then turn on its transmitter and transmit an audio tone pulse a precise
amount of time following the timing reference pulse. Transceiver
operation will be simplex mode.

The amount of time between the two pulses is not important. It can be
200 msec or more, as needed to let the transmitter and receiver settle
after turning on and off. The only requirement is that the time between
pulses be kept to a close tolerance of 50 nanoseconds if possible.

50 nsec is going to be impossible with a 1kHz audio tone burst. Well,
theoretically it is possible with an insanely high signal-to-noise
ratio or a long enough integrating time but you won't ever get there.

Going to more bandwidth will make 50ns possible. As an example, GPS
receivers do timing to within a few tens of nanoseconds (and in fact
they measure the timing for multiple satellites simultaneously thanks
to clever correlator coding). Look up how a GPS correlator works.
You will have to send and receive signals much more complicated than
a simple tone burst but the technology to do all this is well
developed and consumerized.

Doing all this with a common VHF handie-talkie probably won't work,
group delay through the filters is completely uncontrolled. Unlike
a GPS, where you need to know the time difference between received
signals and thus are largely insensitive to slow-changing delays in your
receiver, you are very sensitive to any delays. The
"modern" way of doing this is all at the IF frequency with DSP's.

Tim.

 

[George W]

50 nsec is going to be impossible with a 1kHz audio tone burst. Well,
theoretically it is possible with an insanely high signal-to-noise
ratio or a long enough integrating time but you won't ever get there.

Going to more bandwidth will make 50ns possible. As an example, GPS
receivers do timing to within a few tens of nanoseconds (and in fact
they measure the timing for multiple satellites simultaneously thanks
to clever correlator coding). Look up how a GPS correlator works.
You will have to send and receive signals much more complicated than
a simple tone burst but the technology to do all this is well
developed and consumerized.

Doing all this with a common VHF handie-talkie probably won't work,
group delay through the filters is completely uncontrolled. Unlike
a GPS, where you need to know the time difference between received
signals and thus are largely insensitive to slow-changing delays in your
receiver, you are very sensitive to any delays. The
"modern" way of doing this is all at the IF frequency with DSP's.

You're probably right. On the positive side, a DSP will let me do a lot
of things like optimize the design (including some of the filters)
without hardware revisions, but the downside is cost and power
consumption/battery life. I was hoping to build this transceiver for
about $25 in quantity. I bet the DSP alone will cost over ten bucks.
Oh well ...

George