I was wondering if a clock circuit could be made that employed transmission line design techniques rather than a standard crystal.
The impedance of a transmission line is dependant on the load impedance, the length of the line and the wavelength of the incoming signal. So if the load impedance and length of the line are fixed then the line would have a peak impedance for a particular frquency (and multiples of that frequency).
My idea is to use this to create a resonant circuit that would essentially have zero drift and zero jitter as the resonance is based only on the transmission line characteristics that stay constant.
The frequency of oscillation would, unfortunately, be high due to the fact that the length of the line has to comparable to the wavelength of the signal on it for transmission line techniques to be employed. A high frequency Schmidt-trigger and clock divider would have to be used somehow to make a usable frequency, although I don't see this as a huge problem?
Any input is welcome, although constructive critisism is prefered.
The impedance of a transmission line is dependant on the load impedance, the length of the line and the wavelength of the incoming signal. So if the load impedance and length of the line are fixed then the line would have a peak impedance for a particular frquency (and multiples of that frequency).
My idea is to use this to create a resonant circuit that would essentially have zero drift and zero jitter as the resonance is based only on the transmission line characteristics that stay constant.
The frequency of oscillation would, unfortunately, be high due to the fact that the length of the line has to comparable to the wavelength of the signal on it for transmission line techniques to be employed. A high frequency Schmidt-trigger and clock divider would have to be used somehow to make a usable frequency, although I don't see this as a huge problem?
Any input is welcome, although constructive critisism is prefered.
annex666 said:
I've used pieces of subminature coax as delay lines in ECL circuits in the past so the idea might work. Have a look at ring oscillators for some inspiration.
Unfortunately it won't be zero drift and zero jitter. The characteristics (length, impedance etc.) will vary with temperature and other environmental factors.
James
Thanks for the reply - I just wanted to bounce the idea off someone else before thinking about the actual design. I think I was a bit optimistic with the quote "essentially zero drift and jitter", although I realise environmental factors will effect things like track length, etc, there may even be ways to compensate for this in the design.
I'll try and come up with some sort of design.
Also while you're online, do you think there will be any problems using a Schmidt-trigger and clock divider at this sort of frequency?
i.e 10-20MHz (or there abouts)
I'll try and come up with some sort of design.
Also while you're online, do you think there will be any problems using a Schmidt-trigger and clock divider at this sort of frequency?
i.e 10-20MHz (or there abouts)
annex666 said:
It's been a quite a while since I did any of this so take all the figures below with a big pinch of salt.
You'll need to use a long line. The backplane tracks I used to work with on FR-4 had a propagation delay of 2.2ns/ft. With coax it will differ. Assuming a line length of 1metre, a propagation delay in the range of 6-9 ns will be obtained. This is a frequency in the 110-160MHz region. This will not be usable with standard gates e.g. HC, AC, VHC etc. So up the line length or use a faster logic family. If you are building a ring oscillator then the propagation times of the gates themselves may be more significant than that of the line.
James
Quick lashup
Hi,
was pottering about today feeling miserable with a head cold and decided to give this a whirl. The prototype could easily qualify for the worst looking prototype thread.
One 74F04 with four of the inputs grounded. One gate used as the oscillator element and one for buffering the output signal, taking its input from the output of the first gate to the transmission line. Output and input via 50ohm BNC sockets. 50ohm source termination from the output of the first gate to the BNC socket. Transmission line made up from 2m BNC terminated RG-58 coax jointed with BNC barrels.
With 2x2m length (4m in total) a frequency of 20.6MHz was obtained. With 3x2m length (6m) a frequency of 13.7MHz. With 2m a frequency of 36.3 MHz. Waveshape at various points in the circuit wasn't bad considering the construction methods used.
I haven't taken it any further than that, but at least it shows it's workable. Have fun!
James
Hi,
was pottering about today feeling miserable with a head cold and decided to give this a whirl. The prototype could easily qualify for the worst looking prototype thread.
One 74F04 with four of the inputs grounded. One gate used as the oscillator element and one for buffering the output signal, taking its input from the output of the first gate to the transmission line. Output and input via 50ohm BNC sockets. 50ohm source termination from the output of the first gate to the BNC socket. Transmission line made up from 2m BNC terminated RG-58 coax jointed with BNC barrels.
With 2x2m length (4m in total) a frequency of 20.6MHz was obtained. With 3x2m length (6m) a frequency of 13.7MHz. With 2m a frequency of 36.3 MHz. Waveshape at various points in the circuit wasn't bad considering the construction methods used.
I haven't taken it any further than that, but at least it shows it's workable. Have fun!
James
Further thoughts
Hi Jesus,
yes, it might not have been FR-4, it was about a decade ago. More likely the 2.2ns/ft number is due to my dodgy memory. I have Howard Johnson's books and have been at his courses. I would recommend his stuff for anyone who is interested in high speed digital design.
Annex666,
You didn't say what the clock source is for. Just a few back of the envelope calculations for a CD clock source:
Looking at www.amphenolrf.com for RG-58 with a solid polyethylene dielectric gives 1.54ns/ft; this translates to ~5.05ns/m.
Assuming a 44.1 kHz CD sampling frequency, a common clock frequency is 256x this at 11.2896MHz. Also assume that a frequency of twice this at 22.5792 MHz will be used. This reduces the required line length and the output is then divided by two. Further assuming that the gate propagation delay is zero (ideal case). This is not true in practice but simplifies the calculation. 22.5792MHz corresponds to ~44.29ns. Using the simple inverter described in the previous post requires that this time be divided by two again to give ~22.15ns.
A line length of 4.38m is therefore required. For a CD clock source, the class II accuracy specification is +/- 1000 parts per million (ppm). In length terms at 4.38m this is +/- 4.4 mm. This may be achievable with careful trimming.
The class I accuracy spec. is +/- 50ppm. This is what most add-on clocks are achieving (or should be). For the transmission line based oscillator this is now +/- 0.2 mm.
Trimming a 4m cable to 0.2 mm tolerances is not feasible. In addition the following factors must be accounted for:
(1) Thermal expansion - in 4m of cable 0.2mm of expansion might be possible.
(2) Propagation delay of ICs. This will have to be carefully measured. At these close tolerances the delay between individual samples of the same part will probably be significant. It may mean that each driver would have to be measured and the transmission line length trimmed to suit. In addition the propagation delay of any individual driver will also vary with temperature and voltage (other factors?)
(3) Mechanical. The path of the 4m cable will need to be tightly controlled to ensure that it remains constant over time. Winding on a former will be mandatory. The use of RG-58 was an example only - more suitable minature coax, perhaps semi-rigid would be a vast improvement, though the semi-rigid type may suffer additionally from poorer thermal expansion characteristics. The entire structure should then be secured by potting with epoxy or similar.
In summary, if you're trying to make a class I accuracy CD replacement clock this won't work (or will require far too much work). If initial frequency accuracy and drift over time is not important then have fun. However, there are easier ways to achieve the same goal.
Regards,
James
Jesús Puerto said:
Hi Jesus,
yes, it might not have been FR-4, it was about a decade ago. More likely the 2.2ns/ft number is due to my dodgy memory. I have Howard Johnson's books and have been at his courses. I would recommend his stuff for anyone who is interested in high speed digital design.
Annex666,
You didn't say what the clock source is for. Just a few back of the envelope calculations for a CD clock source:
Looking at www.amphenolrf.com for RG-58 with a solid polyethylene dielectric gives 1.54ns/ft; this translates to ~5.05ns/m.
Assuming a 44.1 kHz CD sampling frequency, a common clock frequency is 256x this at 11.2896MHz. Also assume that a frequency of twice this at 22.5792 MHz will be used. This reduces the required line length and the output is then divided by two. Further assuming that the gate propagation delay is zero (ideal case). This is not true in practice but simplifies the calculation. 22.5792MHz corresponds to ~44.29ns. Using the simple inverter described in the previous post requires that this time be divided by two again to give ~22.15ns.
A line length of 4.38m is therefore required. For a CD clock source, the class II accuracy specification is +/- 1000 parts per million (ppm). In length terms at 4.38m this is +/- 4.4 mm. This may be achievable with careful trimming.
The class I accuracy spec. is +/- 50ppm. This is what most add-on clocks are achieving (or should be). For the transmission line based oscillator this is now +/- 0.2 mm.
Trimming a 4m cable to 0.2 mm tolerances is not feasible. In addition the following factors must be accounted for:
(1) Thermal expansion - in 4m of cable 0.2mm of expansion might be possible.
(2) Propagation delay of ICs. This will have to be carefully measured. At these close tolerances the delay between individual samples of the same part will probably be significant. It may mean that each driver would have to be measured and the transmission line length trimmed to suit. In addition the propagation delay of any individual driver will also vary with temperature and voltage (other factors?)
(3) Mechanical. The path of the 4m cable will need to be tightly controlled to ensure that it remains constant over time. Winding on a former will be mandatory. The use of RG-58 was an example only - more suitable minature coax, perhaps semi-rigid would be a vast improvement, though the semi-rigid type may suffer additionally from poorer thermal expansion characteristics. The entire structure should then be secured by potting with epoxy or similar.
In summary, if you're trying to make a class I accuracy CD replacement clock this won't work (or will require far too much work). If initial frequency accuracy and drift over time is not important then have fun. However, there are easier ways to achieve the same goal.
Regards,
James
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.