Hi Tom!

On 05/09/2017 03:02 AM, Tom Knox wrote:

When you provide an external reference, the long term stability becomes
relatively unimportant. Let's view the two cases:

1. Bypass internal oscillator: If you bypass the internal oscillator and
   always supply an external oscillator, don't waste money on the internal
   reference, it has no use.

2. Lock internal oscillator: If the oscillator is PLL steered, the short
   term noise, i.e. phase-noise, is of relevance, but the long-term only
   cares in the aspect that you can maintain lock, and you can probably
   trim the oscillator to every 5 years to ensure lock, but other than that
   you don't need to waste mony on it. If the phase noise is an issue, you
   could possibly see that in the datahseet/performance spec, but for all I
   have seen, only long term performance is given, so I'd say that you
   should not waste your money there either.

So, unless it will operate as a stand-alone, don't waste money on
internal reference.

Also, there is a reason that you can buy different internal oscillators,
this is one of them. For labs with their frequency distribution there is
no need to waste money on stand-alone performance.

Look at the HP5335A for instance. Standard performance is an XO, and you
need to trim that regularly to have any kind of performance. The high
stability option is a 10811. That is quite a jump in performance.

Cheers,
Magnus

In message bad861e5-00a6-5ffc-e031-be6372a1306f@rubidium.dyndns.org, Magnus Danielson writes:

Please don't forget that running the counter of the same signal as
the circuits being measured runs the risk of masking noise processes.

I'm sure that modern counters like 53230 are better at this than
the 5370, but you should always sanity-check your measurements
using the counters internal, undisturbed timebase.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

Hi,

On 05/09/2017 09:05 AM, Poul-Henning Kamp wrote:

Indeed. On the other hand some measurements is hard not to do that way.
One should think about how isolation is created if needed, and this
includes AC and DC, common mode and differential mode. Isolation
amplifier should be part of the arsenal.

Indeed. When you think you have a risc of disturbance, please do
measurements to verify the setup.

Cheers,
Magnus

Hi

If you grab the schematic of pretty much any of these counters, you find a fairly common approach.
Somewhere inside is a VHF oscillator. The internal TCXO, OCXO, or Rb acts as a phase lock source
for that VHF oscillator. Typical PLL bandwidths are pretty low (10’s of Hz). When you put in an external
reference, it acts as the reference to the same PLL.

This does a couple of things. You take care of spurs on the external reference. That is the same
thing you do locking up a VHF oscillator to your GPSD for microwave work. You get the long term
accuracy (1 second and out) of the external reference. Your phase noise (and thus broadband mask
jitter) is improved.

Does every counter on the planet work this way? - of course not. Somewhere somebody did it a
different way. As long as they did do it this way, the internal reference does not matter once you
switch to an external standard. Even if they did it in some other way, if they did it right, the same
statement would apply.

Bob

In message 60c74d75-41e5-7112-d8b6-721664b895af@rubidium.dyndns.org, Magnus Danielson writes:

EMI is part of it, but there raw synchronism is a significant
source of systematic errors because the "noise" is no longer
gaussian.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

Hi Poul-Henning,

On 05/09/2017 02:39 PM, Poul-Henning Kamp wrote:

Wither it is proper side bands or other systematic noise, it is bad
indeed, that would be part of the RF differential mode disturbances.
This assumes the source is "clean", and if it is not it should not be
used to start with. What remains is really issues from the setup, and
that's when signal integrity/EMI care and isolation amplifiers comes in.

Cheers,
Magnus

Hi,

In HP5370A/B you mux sources. In SR620 there is a lockup PLL, as far as
I remember.

Cheers,
Magnus

On 05/09/2017 01:11 PM, Bob kb8tq wrote:

In message 54ee7bdd-0a49-cc49-540d-2812e70dd4f3@rubidium.dyndns.org, Magnus Danielson writes:

It's more subtle than that, it can be things like thresholds in
the counters trigger being sensitive to the phase of the internal
clock because of (very slight) cross-talk.

I reported data on this a couple of years ago, as I recall my setup
were something like this:

House-10MHz --> HP3336C clk-in and HP5370B clk-in.

HP3336C output -> HP5370B start then 3m coax then HP5370B stop.

Measure TI(start->stop), for different settings of output phase angle on the HP3336C.

Theoretically that plot should be a flat line.

In practice it is not even close.

I think I convinced myself that the majority of the problem was trigger-noise the HP5370B,
but my notes are not accessible at this time.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

Hi Thomas,

About the 53230A -- like many of us I figured there was no point in buying the expensive OCXO versions since I have plenty of good 10 MHz references around here. So years ago Keysight loaned me two of them for a month: one with, and one without OCXO. Like most counters, the 53230A has an ext-ref input. But it is horrible. And the ref output is even worse. I think we've discussed this on the list before, including ADEV and PN plots. I decided not to buy one, ever, until they fixed the problem.

So it could be a nice counter. Someone at Agilent / Keysight needs to re-design the int/ext/mux/pll clock handling of the 53230A; someone who understands phase noise and short-term stability; someone who honors the intent of ext-ref-in and ref-out; someone who keeps stupid microprocessor / LCD / comms / power supply noise out of the clock path. Meanwhile I keep my old 53132A's and older SR620's and stay away from the 53230A.

If you have your own measurements, please share. Maybe I just got a bad sample, I don't know. But the ext-ref performance smelled like poor engineering to me. My hope is that someone on the list will hack a 53230A, fix the problem, and share the solution with us, or even with Keysight.

/tvb