[time-nuts] Re: Fixing PN degradation via ADEV measurement

[Bob kb8tq]

Hi

With an audio spectrum analyzer, and an RPD-1 mixer, and +10 dbm on 
each side of the mixer …. you should be able to get away with about 20 db
of “preamp gain”. Yes, that’s dependent on exactly what analyzer you have.

The typical sound card may well need closer to 50 db of gain. That’s better 
done with two op-amp stages than trying to get it all in one shot. Again there’s
an assumption involved about wanting 40 to 100 KHz sort of “top end” for the
measurement. 

Ground loops ( and power line noise ) are indeed a *very* big thing in all these
measurements. ADEV / phase noise / whatever. Being able to see the crud makes
getting rid of it much easier. That’s not to in any way to say it’s easy, only that it’s
easier. 

Fun !!!

Bob

> On Jun 21, 2022, at 1:26 AM, Erik Kaashoek <erik at kaashoek.com> wrote:
> 
> Bob, Using the approach you described I was able to verify the noise floor of my initial measurement setup to be at -80dBc at 10Hz offset from carrier and -100dBc above 1kHz with a strong peaks at 50Hz, 60Hz and harmonics. This all with an RBW in the FFT of around 4Hz.
> This level of noise is way to high to do phase noise measurement so I'm now going to work on removing ground loops and adding low noise amplification.
> Erik.
> 
> On 21-6-2022 2:04, Bob kb8tq wrote:
>> Hi
>> 
>> Ok, single mixer phase noise basics:
>> 
>> First thing is to womp the mixer up to the point it almost smokes. Putting +7 dbm into
>> both ports on a “7 dbm” mixer is very normal in this case. Watching for the fact that the
>> mixer likely is *not* a 50 ohm load is part of the process ( = pads might help out) as well
>> as understanding that it does not have a monster amount of isolation ( = isolation amps
>> may be needed ).
>> 
>> Next one generates a beat note by offsetting the two oscillators a bit. This gives you a nice
>> 360 degree sweep function ( 360 degrees per cycle :) ). From that you can work out the
>> system sensitivity in volts per degree ( or better yet per radian since that’s what you actually
>> want as the “magic number” …. it’s phase modulation so radian is king …).
>> 
>> Next you lock the two oscillators together via a DC feed out of the mixer to one or the
>> other of them. You adjust the “lock point” so that it is at zero volts out of the mixer. This
>> puts the two oscillators in quadrature. Yes, there is that messy 2X the input frequency RF
>> output and the inevitable leakage. Those are handled with a lowpass filter.
>> 
>> The output of the mixer is now “just noise”. There is no nasty carrier to deal with. There is
>> no messy fold over to wonder about. What you get is the DSB noise ( so both sides of
>> carrier) from the sum of the two oscillators.
>> 
>> Output of the mixer goes up if you terminate it in an “high” load. Something like 500 ohms
>> on a 50 ohm mixer or 5K ohms on an RPD-1 is often used. The isolation seems to be ok
>> either way and the added gain / better floor is “free”.
>> 
>> Simply put you add 3 db when you look at DSB if it’s uncorrelated, and another 3 db if the
>> oscillators are identical. Your “output” is 6 db higher than the single sideband / single oscillator
>> phase noise. You can argue that close in noise is likely correlated due to it being a modulation
>> on the carrier. The standard convention is to use 3 db.
>> 
>> Amplify the noise up and you can measure very low levels of phase noise. Low noise
>> audio op-amps are pretty easy to find spec sheets on. With anything these days finding
>> them on the shelf may be “interesting”. The OP-27 / OP-37 with low resistance in the
>> feedback path go way back for this application. There are a lot of other candidates.
>> 
>> The cutoff of the lock signal typically is adjustable to keep it below the lowest point
>> of interest for your noise testing. If that is impractical, there are ways to calibrate and
>> read “inside the loop”.
>> 
>> The HP 3048 phase noise analyzer was based on this approach. The original app note most
>> folks started from came from Fluke back in the early 1970’s. I have not (yet) found a good
>> copy of it on the internet.
>> 
>> Fun !!!
>> 
>> Bob
>> 
>> 
>> 
>>> On Jun 20, 2022, at 9:43 AM, Erik Kaashoek via time-nuts <time-nuts at lists.febo.com> wrote:
>>> 
>>> Bob, Magnus,
>>> Thanks, clear. A counter is for ADEV, not for phase noise.
>>> I made a test setup to learn how to use the mixer/PLL approach.
>>> First using 10MHz from both outputs of a DSS (Rigol DG990) to observe the DC shift with shifting the phase between the two signal.
>>> Then by modulating one output with FM or PM.
>>> There is a low pass filter after the mixer to get rid of the 10 MHz and its harmonics but the LPF is measured flat till about 10kHz.
>>> The output signal from the mixer was kept within 10% of the full voltage swing to stay in the (hopefully) linear range.
>>> Using PM creates a low frequency output from the mixer that is proportional to the phase shift (region 0-1 degree) and constant in amplitude with change of frequency. Also when using external modulation from an audio signal generator created the expected behavior with drive level and no frequency impact
>>> Using FM with 0.1 Hz frequency deviation the mixer output amplitude decreases very fast with increasing frequency (range 0.1 to 10 Hz)
>>> Also when using 1 Hz or more frequency deviation. The higher frequency deviation leads to higher output levels as expected.
>>> Can someone help me understand how this FM signal (0.1 to 1000 Hz modulation and 0.1 to 1 Hz frequency deviation) translates to the calibration example mentioned in the document on phase noise measurement as linked by Bob. (0.1 Hz deviation at 1 kHz rate leading to a sideband (at 1kHz?) level of -86 dBc)
>>> At a 1kHz rate I see (yet) no output from the mixer where at 1Hz there is a lot of output. Why is this output frequency dependency?
>>> Is this a problem with the signal generator?  Or the mixer?
>>> Then I tried to use the modulated signal from the SG PLL locked to a 10MHz VCO. Results where the same. FM output signal is frequency dependent, PM not.
>>> Erik.
>>> 
>>> _______________________________________________
>>> time-nuts mailing list -- time-nuts at lists.febo.com
>>> To unsubscribe send an email to time-nuts-leave at lists.febo.com
>