How much have you seen your AC mains affect PS for OCXO, Rb, and Cs measurements

[notfaded1]

I've been getting into measuring oscillators with precision counters and have noticed some issues when getting down in µs and ps.  I'm seeing some similar behavior despite which oscillator I'm measuring and I'm starting to think power from the mains is affecting my readings. I'm finding out that even when on linear power supplies you can still get noise from the AC mains granted less but it's still there.  I see many are using cheap switched power supplies to power theirs DUT's and my initial experience has been that this doesn't work very well although I have some too.  You're never going to better than your weakest link correct?  And if that weakest link is the AC mains how do you rectify this?  I was thinking possibly using a UPS and putting it in transfer mode and running the tests running off batteries?  Does anyone have any suggestions for how to do something like this or an alternative for really clean power for this kind of testing?  Better Fluke and HP linear power supplies is one thing I've found that helps.  What do you do when your AC in you area isn't great?  I know we do this with LTZ1000 volt measurements doesn't the same apply to frequency and time interval measurements?  Garbage in garbage out right?

What brought this on for me was seeing some patterns in test data regardless of the meter I used or the DUT.  The patterns start to look too consistent to be from a specific meter or DUT.  When I substitute or remove variables in the test and some of the patterns remain and the only consistent thing has been the AC mains power.  I suppose the only that way to be sure is to change the variable and to change this variable means getting off the AC mains for the test.

[Theboel]

Hi,

Try To find Your measurement setup noise floor do it several times, it good base for Your real measurement. 
If You use SR620 try to read the manual about calibration it will help alot.
I use online UPS and isolation transformer and avoid any interference source like LED lamp etc and use double shield cable for connecting my reference and DUT.
in case You try to compare GPSDO try to changed the delay at least 0,5 us

 

[notfaded1]

Hi,

Try To find Your measurement setup noise floor do it several times, it good base for Your real measurement. 
If You use SR620 try to read the manual about calibration it will help alot.
I use online UPS and isolation transformer and avoid any interference source like LED lamp etc and use double shield cable for connecting my reference and DUT.
in case You try to compare GPSDO try to changed the delay at least 0,5 us

I have an isolation transformer and haven't tried that yet.  I do now have SR620.  Do you put everything on the UPS and isolation transformer?  What kind of UPS do you use?  I'm familiar with some UPS stuff and for meter and DUT to be on UPS for extended period will require a really decent UPS.  For a few minutes is one thing but if you want to do 1 hour plus tests we're talking about a bigger (and usually really heavy) UPS.

[Theboel]

Hi,

i use eaton 9130 1KVa UPS I do not disconnect the ups from main line I believe because it an online UPS it clenaer than my main line.
to made sure I also use isolation transformer medical grade I salvage from old medical equipment.
Btw what is Your reference ? nS/pS for me is extremely hard I have 3 SR620, CNT-90, even after careful calibration the result is not same I really think to use DMTD to improve my measurement.
may I see Your noise floor I will post my noise floor when I got home.

[tkamiya]

I have not seen mains affect stuff I do.  That's what regulated power supplies are for...!

As to UPS, I have 3KVA, 2KVA, 1.5KVA and 2 1KVA rack mount models by APC.  Bought them all used and replaced batteries. 

I see "weird" fluctuations when I get right down to the last digit of my HP53132A but that's expected.  Timing standard (GPS or Rubidium) does not have great short term stability, and OCXO doesn't do well by itself beyond 100 second or so.  What you are seeing may be expected.

When I measured Cs, I took 24 hour measurement and averaged it.

[DaJMasta]

While it's important to have clean power when regulating precision timing, mains hum or switching noise doesn't really have the effect you're imagining - and the design of precision timesources and counters reflects that.  You don't even need a linear supply to do a good job, you can just regulate a switcher's output fine enough, and if you took a switching supply that makes noise at a few kHz, you wouldn't see timing noise that reflected the new noise frequency.  Also worth mentioning that when you're driving an OCXO, for example, the thermal mass/thermal time constant of the oven will be so long that it will be a low pass filter that blocks out anything over a few fractions of an Hertz in frequency - so even if the DC you were driving the OCXO's oven with had significant mains noise, that change in power applied to the heater would be smoothed out over so many seconds, the slower oven controller wouldn't even notice it.


What is your measurement method?  If you're counting edges, this could be the normal cyclic variation of the way two frequencies align - think the moire effect or very-near-DC mixing products from two very similar frequencies.

[jpb]

While it's important to have clean power when regulating precision timing, mains hum or switching noise doesn't really have the effect you're imagining - and the design of precision timesources and counters reflects that.  You don't even need a linear supply to do a good job, you can just regulate a switcher's output fine enough, and if you took a switching supply that makes noise at a few kHz, you wouldn't see timing noise that reflected the new noise frequency.  Also worth mentioning that when you're driving an OCXO, for example, the thermal mass/thermal time constant of the oven will be so long that it will be a low pass filter that blocks out anything over a few fractions of an Hertz in frequency - so even if the DC you were driving the OCXO's oven with had significant mains noise, that change in power applied to the heater would be smoothed out over so many seconds, the slower oven controller wouldn't even notice it.


What is your measurement method?  If you're counting edges, this could be the normal cyclic variation of the way two frequencies align - think the moire effect or very-near-DC mixing products from two very similar frequencies.

What you say makes sense, but I have found that the quality of power supply does seem to affect the measurements I've made in the past. I use mainly  good quality linear supplies but I also have a Hameg supply which is a linearly regulated switch mode I think which is low noise and ripple but I found on one measurement a difference between it and my TTi straight linear supply.

I think there are two factors that you haven't covered. One is that the supply noise, though not affecting an OCXO through the power supply itself might well have an effect on the Voltage controlling the frequency where there isn't the same long time constants involved.
Secondly, if the noise translates to AM this may translate to inaccuracies in the phase measured by the counter. Certainly the internal noise of the counter affects its accuracy.

I find it interesting that the LPFRS Rubidium standard that I have contains three dc-dc converters  but:

The synchronization of the three converters is achieved by the use of a common ramp generator given by an internal 125kHz signal derived by direct division of the 20MHz main VCXO.

So the designers here felt it was important to avoid differently synched frequencies being present in the internal power supplies.

[notfaded1]

One things is for sure... when you get into ns and ps a LOT becomes more visible.  Rb do move around in frequency... it's not a panacea.  I have a bunch of good counters SRS SR620, HP 53132A, Pendulum CNT-90.  I'm going to try testing phase noise with the updated FSA3011 Frequency Stability Analyzer next.  I've been inspired to search for a better clock.

[thermistor-guy]

I've been getting into measuring oscillators with precision counters and have noticed some issues when getting down in µs and ps.  I'm seeing some similar behavior despite which oscillator I'm measuring and I'm starting to think power from the mains is affecting my readings....

What brought this on for me was seeing some patterns in test data regardless of the meter I used or the DUT.  The patterns start to look too consistent to be from a specific meter or DUT.  When I substitute or remove variables in the test and some of the patterns remain and the only consistent thing has been the AC mains power....

In my home lab, I set up my tests and run them remotely, with the lab lights off and the door closed. Any test equipment not required is switched off at the mains wall socket. Any WiFi or bluetooth gear is similarly unpowered.  There is a wired LAN connection to my monitoring point, via an rpi and small ethernet switch in the room. All other computing devices in the room are switched off at the mains wall socket. With doors and windows closed, window shades drawn, and no-one in the room, the room temperature gets stable and stays that way.

Fluorescent lights, some led ights, and light dimmers can generate a lot of RFI, as can switch-mode power supplies in nearby equipment. They can also inject noise back into the mains (conducted emissions), and that can effect the equipment running the test.

So turn off the lights (or have simple battery-powered lights), disconnect all unneeded equipment, and check if that makes a difference. Your mains supply may be an issue, but these other factors may be contributing.

[notfaded1]

What's your best method for noise floor calculation?  Do you recommend using PPS or 10MHz signal for the noise floor measurement?  What do you use as the source?  How do you wire up your counters?  Is it:

source -> BNC T on port A -> cable to port B?  This would be a separate ext in reference on the ext ref in port.

There are other methods using ext reference and port A too correct as in:

source -> BNC-T on ext reference in -> cable to port A

How do you configure your two input ports and what do you trigger on?  What impedance do you set on your input ports 50 Ohm or High Z?

Thanks,

Bill

[ArthurDent]

I've said before I haven't seen where the possible difference between a linear and a SMPS made any difference that would affect anything I was doing. There probably is a difference but for what I'm doing it is small enough so I just don't notice the effect.

Yesterday I received a nice Spectracom 8183 GPS clock I got on eBay (for well less than $100) with 10Mhz, 1 PPS, and IRIG outputs along with a nice big and bright red LED time display on the front. It used an external wall wart that plugged in the back and I don't like these setups because I'm always losing the power adapter. The first thing I did after checking it out was to ditch the wall wart that powered it and build in a 24 VDC SMPS with a detachable IEC cord. There were 2 screws holding the 2 halves of the PS shell together and by carefully drilling 2 holes in the bottom of the clock case I could use those screws to mount the PS. I cut a rectangular hole in the back of the clock case so I could plug an IEC cord into the PS and wired the PS output to the circuit board. The SMPS is internally fused and I plan on running the clock 24/7 so I didn't use a switch and fuse on the back panel.