[time-nuts] HP 53131A question
magnus at rubidium.dyndns.org
Sat Nov 22 17:32:42 UTC 2008
Tom Van Baak skrev:
>> Hi all,
>> I'm planning to use a HP 53131A for phase measurements.
>> In a limit case, just to understand, suppose that I'm
>> reading, say, 100 degrees, and the next reading gives again
>> 100 degrees, but meanwhile the signal under test jumped 360
>> degrees. Is there any way to detect such a case?
>> Thanks in advance,
>> Antonio I8IOV
> That's a good question for the purpose of understanding.
> In general, no. So you have to pick your sampling rate to be
> somewhat faster than the expected maximum cycle drift rate
> (Nyquist has something to say related to this).
> As a further example, the same thing would happen to the earth,
> if while you were sleeping it quickly sped up 3x normal speed
> and slowed back down to normal. You'd wake up, see a sunrise,
> and think it was Saturday morning. Instead it would actually be
> Sunday morning. You were sleeping; how would you know?
> Now if there were a special application where detecting phase
> jumps like this was critical you could drive two free-running
> counters, one from the input signal, one from a reference clock,
> and compare them as often or as infrequent as you wish. A
> "continuous count" or "zero dead time" counter works this way.
> But a 53131A is isn't of this design.
Hmm... come to think of it, I don't recall that I have seen any
documentation of the counter core of the HP53131A/53132A.
The "zero dead time" aspect of a counter has been somewhat debated. What
it means is that the counter makes a pause between two measurements,
read out the values, resets its counters and arm the trigger circuit for
the next trigger event. The event counter counts the number of gated
signal, i.e. for a clock the number of cycles. The time counter counts
the number of reference clock cycles of the reference clock, so this is
an event counter for the reference clock. The reference clock is often
higher than the 10 MHz local reference, but the HP5335A is an example of
a counter using the 10 MHz directly. Modern counters also use
interpolators to increase time-resolution, so the interpolated start and
stop value is also stored.
The problem is that during the pause, you are not monitoring the signal,
so you can't see if it does any jumps or anything. A zero dead time
counter solves this by keeping the event counter and time counter
running continously. Rather, the start and stop events sample the state
of these counters and store it. By taking the difference between the
start and stop event counter values, you get the number of events
between the start and stop, and similarly for the time counter. If you
now record a number of these into a memory, what is the stop of one
measurement can be the start of the next, so the same trigger event
becomes both stop and start. There simply is no dead time. What is
stored in the memory is the absolute phase of the signal. What many does
not realize is that the reference input at the back is infact a input
channel, and that some measurements is better performed directly with
that. When using two channels it becomes a transfer oscillator with its
benefits and drawbacks.
A great way to learn about many of the aspects is to read the HP5372A
programmers manual, since the binary format is really the pure binary
pattern stored in the 8192 sample long memory. It details how you need
to process it to get the same result as the display shows when you have
selected that measurement on the front. It also allows for other forms
of analysis where the pure TI data is important.
> Magnus' suggestion about dividers is perfect. It is what many of
> us do with our clocks -- instead of measuring phase angle between
> two 5 MHz RF signals, we divide 5 MHz down to 1 Hz and leisurely
> compare the time interval of 1PPS signals, as often as once a
> second or even as infrequently as once an hour or day.
This method simulates much of the continous aspect, as the division
circuit does not have the dead time of the counter. However, the benefit
of dividing less is that you can get a quicker update time. Higher
frequencies of jitter/wander (phase deviations) can be exposed and
treated separately. The division converts the signal phase it into a
form which a +/- TI capable counter with dead time can handle. To cover
dead time, the minimum period of the output signal should be longer than
the effective deadtime for a complete guarantee, but for many cases
higher frequencies can be used anyway. 1 Hz is not necessarilly needed.
More information about the time-nuts