Spectrum Analyzer phase-noise and/or "shape function"

[awallin]

Hi all,
I've been struggling a bit to understand the output at 1 GHz from an ADF4350 PLL+VCO or at 100 MHz from an AD9912 DDS:
http://www.anderswallin.net/2014/03/adf4350-pllvco-and-ad9912-dds-power-spectra/

It seems I'm getting much wider peaks and possibly higher noise-floors than in the AD datasheets.

update: added offset frequency and link to specs
These were measured with a RIGOL DSA1030A with a quoted phase-noise of -88 dBc/Hz @ 10 kHz. (specs http://www.rigolna.com/products/spectrum-analyzers/dsa1000a/dsa1030a/)
If we imagine an ideal sine-wave input, which ideally has a delta-peak spectrum, what are the specifications on a spectrum analyzer that determine how the peak really looks like on the SA screen?
Any good way to test this?

The spectra for a 1 GHz clock and another DDS (AD9910) shown over here look much much nicer:
http://martein.home.xs4all.nl/pa3ake/hmode/dds_refclock.html
http://martein.home.xs4all.nl/pa3ake/hmode/dds_ad9910_amnoise.html
I wonder if I am doing something wrong or if these results are because a better SA was used (Wandel & Goltermann SNA-62)?

thanks for any replies!
AW

[cyr]

Note: I'm a bit of a noob in this area, so don't take what I say as absolute truth - but a few points:

1) -88 dBc/Hz at what offset? Is there not a table or graph rather than just a single number in the spec? This table/graph could give you some idea of what the limit of the analyzer is, and what shape you should expect to see with an ideal input signal. Another factor of course is the shape of the RBW filter.
2) The graph from the datasheet is normalized to 1Hz bandwidth, your screen shot is at 100Hz RBW so shows a higher noise level (20dB).
3) If you have a signal source available which you know is much cleaner than the analyzer can show, use that to get a baseline.

[KJDS]

From the datasheet, the phase noise is spec'd as, and these are typical values at 500MHz,

Carrier Offset 10 kHz: <-88 dBc/Hz,
nominal
Carrier Offset 100 kHz?<-100 dBc/
Hz, nominal
Carrier Offset 1 MHz?<-110 dBc/
Hz, nominal

The fact that it isn't spec'd below 10kHz implies to a cynical old fart like me that it's poor. Don't forget that the noise is spec'd in a 1Hz bw and you'll need to adjust the reading depending on what rbw you are using.

So measuring phase noise on a Rygol, it's a bit like women playing cricket. That is that it's not a surprise to see it done badly, the real surprise is that it is done at all.

I've used some simple home brew reciprocal mixing set-ups over the years, as well as some state of the art Agilent systems built into eight foot high cabinets. I've even used a decent high end Agilent spec ana, something like the 856x series has a phase noise measurement option.

For home use, I'd put together a reciprocal mixing system if I needed a reasonable measurement.

[rfbroadband]

the ADF4350 is one of lowest phase noise PLL on the market. You will need very special (and expensive) equipment to easily measure phase at these levels (especially better than -155dBc/Hz far away from the carrier). Equipment to this kind of phase noise measurements:
Agilent E5052A/B = 90,000 US$ list price
R&S FSUP = similar price range
or you can design your own phase noise measurement setup and develop a solid theoretical understanding on how the math works out by
a) mixing the target carrier down to DC with a source that is better than the ADF4350. A simplified viewpoint is: "Everything" that is not DC (in this setup) is the phase noise. You can then use an a high end FFT analyzer  and measure the phase noise provided you understand the math involved in it

b) you can use two  ADF4350, configure them at the same carrier freq, mix both of them down to DC and knowing you use two identical devices, you can make assumptions in your equations and use an FFT analyzer again to measure the phase noise.

It is complicated...Agilent published app-notes on that topic.

back to your graphs:
a) phase noise is measured in dBc/Hz at an offset freq. A standard spectrum analyzer measurement result needs to be converted to dBc/Hz not only based on the RBW setting of the SA but also based on the filter shape
b) phase noise is typically plotted on logarithmic freq scales, your spectrum analyzer plot is on a liner freq. scale, so don't compare the shapes of both graphs, it is not apples with apples.
c) you can't use a spectrun analyzer to measure those kind of phase noise levels, especially far away from the carrier. The internal LO of your spectrum analyzer will mask the measurement result. One of the reason your Rigol SA is so "cheap" is because the phase noise of the internal LO is really bad/high. Very high end SA have may good enough phase noise capabilities you could get away with close in (close to the carrier) measurements...
d) the only SA who could measure those levels is the R&S FSW (don't ask for the price). Even the most expensive Agilent PXA analyzer would not be able to accurately measure phase noise of the ADF4350.

Sorry, I wish I had better news.....

[G0HZU]

Because the ADF4350 VCO is quite noisy the PLL noise performance you will see at 1GHz across your 500kHz span will heavily depend on your choice of PD freq and loop bandwidth.

The orange /4 trace on the datasheet (i.e. your linked image) shows the typical ADF4350 performance when divided by 4 to 1.1GHz with 25MHz PDF and 40kHz loop bandwidth. This could be measured (out to 250kHz offset) on a top of the range 1980s lab analyser from HP or my old 1980s Takeda/Advantest analyser from Japan. If you pushed the loop BW out a bit further it would become a bit harder for the analyser though...

Obviously, the Rigol will be way too noisy to make these measurements but a decent 35 year old lab grade analyser could just about manage it. These can be bought for a few hundred pounds these days if you are brave enough to take the reliability risk on something this old and big and heavy!

However, the real challenge will be measuring the AD9912 DDS at 100MHz when clocked with this 1GHz signal.

You would need to use a decent SSA like the Agilent 5052 or some other method. Or you could just predict it using a few simple sums based on the analyser measurement of the ADF4350 clock up at 1GHz.

[G0HZU]

Hi all,
I've been struggling a bit to understand the output at 1 GHz from an ADF4350 PLL+VCO or at 100 MHz from an AD9912 DDS:
http://www.anderswallin.net/2014/03/adf4350-pllvco-and-ad9912-dds-power-spectra/

It seems I'm getting much wider peaks and possibly higher noise-floors than in the AD datasheets.

Also, have you simulated your ADF4350 using ADISimPLL? If you do this you will see that the choice of PDF and loop bandwidth are important.

I simulated a 1GHz output using integer N and 25MHz PDF and 40kHz loop bandwidth and ended up with a very similar phase noise plot as the datasheet.
I also dug out an official ADF4351 evaluation PCB from Analog Devices and programmed it as above and it showed slightly better phase noise than the simulation on a 500kHz span. But it was within a couple of dB in most places.

This was measured on my old Advantest TR4172 spectrum analyser from the 1980s. In its day it cost an obscene amount of money to buy new but you can pick these up very cheaply nowadays. I paid £400 for this one about 10yrs ago and bought a second one for £250 last year as a cheap spares backup unit. Your modern Rigol SA won't be able to make this measurement because it will use basic low cost circuits for its LO generation. So the Rigol LOs will be way too noisy. The old Advantest analyser was designed in the days when performance was the primary goal with little regard/compromise for the required size, power, weight and cost of the analyser.  So it is a HUGE unit. I think it weighs over 50kg.