[time-nuts] Updated Divider Jitter Results - 74HC390
John Ackermann N8UR
jra at febo.com
Sat Apr 4 20:44:39 UTC 2009
Hi Brian --
It's good to collect this data; thanks. It's interesting that your std
dev in the first test seems to increase significantly with the number of
samples; I haven't seen that kind of scaling here (1K sample and 10k
sample turned in very similar std dev). From what Poul-Henning said
earlier, your first run may suffer the same distortion as my data at the
bottom of this thread.
I just finished rerunning the TADD-2 test using a Wavecrest DTS-2075
(the first real use I've had for that box!) and with 1 PPS input on the
start channel, 10 MHz from the same source on the stop channel, and 10K
samples, I got 22.0 ps of jitter, and a 92 ps min/max range. (As far as
I can determine, the Wavecrest doesn't allow you to use an external
reference, and its internal reference runs at 100 MHz so it probably
wouldn't be useful in this measurement.)
That's consistent with what I measured earlier with the 5370B when I
didn't have the reference and the inputs in coherence. It appears that
the test below, where I used the same reference for *everything*
triggered the problem that Poul-Henning warned about, so those results
should be disregarded.
While I haven't done any testing to validate this, I think the complaint
about the 74HC390 dividers isn't so much their jitter in normal use, but
the tempco problems the cascaded stages can cause. If you can do it, it
would be interesting to measure the phase change over temperature --
I've done a preliminary experiment on that for the TADD-2, but plan to
rerun it with much better measurement technique.
I'm also hoping to do a jitter and tempco test of the Wenzel input
conditioning circuit by itself. I really like that circuit for its wide
input amplitude range.
Brian Kirby said the following on 04/04/2009 04:18 PM:
> I will report some results on a asynchronous divider, which I basically
> copied from Dr. Thomas Clark's designs, which everybody likes to report
> as a bad design.
> The 10 MHz input signal is coupled thru a resistor and capacitor. On
> the other side of the capacitor is the resistive divider that is tied to
> Vcc and ground - it biases the signal to 2.5 volts, which is feed to the
> input of the 74HC132. The output of the 74HC132 feeds several 74HC390s
> until it becomes a buffered 1 pulse per second signal. I also have
> buffered 5 MHz and 1 MHz outputs. The other 3/4 of the 74HC132 are used
> to externally synchronize the 74HC390s.
> I used the Thunderbolt as the source of 10 MHz and it was feed to the
> divider, and the stop input on the HP5370B. The 5370B was run on
> internal clock. The 1 PPS from the divider feed the start input on the
> 100 seconds TI 79.865 nS MIN 79.80 nS MAX 79.98 nS STD 36.4 pS.
> 1000 seconds TI 79.831 nS MIN 79.71 nS MAX 80.00 nS STD 49.9 pS
> 10K seconds TI 80.1552 nS MIN 79.79 nS MAX 80.88 nS STD 271 pS
> 100K planned
> Also a second test, using the Thunderbolt as a source of 10 MHz and it
> was feed to the divider, the stop input on the 5370B and the external
> clock of the 5370B. The 1 PPS from the divider feed the start input on
> the 5370B.
> 100 seconds TI 75.002 nS MIN 74.96 nS MAX 75.04 nS STD 22.5 pS
> 1000 seconds TI 74.931 nS MIN 74.80 nS MAX 75.04 nS STD 56.8 pS
> 10K seconds TI 77.5135 nS MIN 77.40 nS MAX 77.62 nS STD 35.9 pS
> 100K measurement in progress.
> I believe having STD in parts of 10-14th is fairly respectable for
> amateur designs..
> Brian KD4FM
> John Ackermann N8UR wrote:
>> I just finished a jitter test of the first TADD-2 built on the
>> production circuit board.
>> The configuration was somewhat optimized from what I used for the
>> earlier tests.
>> A single 10 MHz source was daisy-chained to the TADD-2 input, to the
>> 5370B external reference input, and to the 5370B STOP channel. The 1
>> PPS output from the TADD-2 was connected to the 5370B START channel.
>> Thus any reference jitter shouldn't be common-mode, and using the
>> reference clock on the STOP channel avoids the need for a second
>> divider, and ensures that the time interval is small (always less than
>> 100 ns; in this case, about 90 ns).
>> For a 10,000 sample run, the standard deviation was 12.1 picoseconds,
>> and the peak-to-peak variation was 70 picoseconds. Based on experiments
>> I ran a few years ago, I think this is pretty much the noise floor of
>> the 5370B and the divider could be better than this.
>> time-nuts mailing list -- time-nuts at febo.com
>> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>> and follow the instructions there.
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
More information about the time-nuts