I wonder if there is someone on the list who’d be willing to let me call
them on the phone to talk about the feasibility of using DSP techniques
(i.e., GnuRadio) to very precisely measure the zero crossing point of a
slowly changing waveform with some noise on it.
More precisely, the goal is to measure the zero crossings of a pair of 1
Hertz sine waves, having an amplitude of perhaps 1 volt peak to peak,
with a precision of ideally 100 nanoseconds. We want to determine the
true zero crossing point, somehow eliminating the effect of any noise on
the signal.
The zero crossings are then used to calculate the time interval (or
phase) between the two waves. I need to capture each pair of zero
crossings, and can’t average over multiple intervals.
The challenge of doing this with conventional techniques like a time
interval counter is that trigger jitter becomes a huge problem with a 2
volt/second slew rate. I’m hoping that DSP techniques might provide a
way around this problem.
It might be easier to explain further details over the phone, so if
someone is interested in guiding me through this problem, please let me
know and I can get your number off-list.
Thanks!
John
John Ackermann N8UR wrote:
The zero crossings are then used to calculate the time interval (or
phase) between the two waves. I need to capture each pair of zero
crossings, and can’t average over multiple intervals.
If it is the phase difference that you are ultimately looking for, there
are alternatives to measuring the zero crossing difference in the time
domain.
How precisely are the two sine waves matched in frequency? How hard
would it be to come up with reference waveform representing “zero”
phase? How often do you need to measure the difference?
-Johnathan
Johnathan C. wrote:
How precisely are the two sine waves matched in frequency? How hard
would it be to come up with reference waveform representing “zero”
phase? How often do you need to measure the difference?
Good questions, and adding a bit more background may help.
What I am trying to do is measure the short term frequency stability
(Allan Deviation) of a device under test compared to a reference. The
two signals are at nominally the same frequency, though in actuality
they will be a small fraction of a Hz off from each other.
The signals start out at HF – typically 10MHz – but are each mixed in
the analog domain with a common local oscillator that’s offset by about
1Hz (ie, LO is 10.000 001Hz) to yield a pair of 1 Hz difference
frequencies. Using a common local oscillator signal makes any noise on
the LO common mode to the other two signals, and thereby reduces its
impact.
The result is that those 1Hz signals carry the same noise components as
the 10MHz signals did. This improves the measurement resolution ) by
the ratio of original to output frequency – as much as seven decades
(using a divider would divide the noise, too, so that wouldn’t gain
anything).
The goal is to measure the stability in terms of the time difference
between the two 1Hz outputs; more correctly, we measure the delta of the
time difference second-to-second to determine the stability. It’s
possible to measure stability to parts in 10e-13 for one second using
this method.
This “dual mixer time difference” technique is the standard one used for
frequency stability measurement over short (0.1 to perhaps 1000 second)
intervals. The problem is that the trigger jitter in a normal time
interval counter masks most of the resolution that’s gained by the
process; finding the zero crossing to 100ns accuracy against a 1Hz, 1
volt p-p signal means the trigger jitter must be down in the nanovolts,
and that’s hard to accomplish.
Most systems currently use a series of limiting amplifiers to increase
the slew rate and reduce the trigger jitter, but those amplifiers
contribute their own noise and other complications like thermal
instability. My hope is that DSP might yield a better way to accurately
determine the zerio crossing of a very slow slew rate signal.
Thanks!
John
John Ackermann N8UR wrote:
The signals start out at HF – typically 10MHz – but are each mixed in
the analog domain with a common local oscillator that’s offset by about
1Hz (ie, LO is 10.000 001Hz) to yield a pair of 1 Hz difference
The LO is at 10.000 001 MHz, not Hz. Sorry.