PLL reference model for PSS+PNOISE

Dear HoWei,

I'm not sure if this might satisfy your need for a somewhat programmable PLL model to validate your circuit simulations, but Ken Kundert provides a set of programmable PLL modules that can be combined to simulate a frequency synthesis PLL using veriloga. He explains each module and their use oh his paper at URL:

https://kenkundert.com/docs/aacd97.pdf

As for your 6 dB (exactly) discrepancy, that does seem highly suspicious of just a difference in a factor of 2. I would really check to make sure the phase noise is defined correctly. Please recall it is defined with a factor of 2.

L(f) = 1/2 *S(f)

and I am wondering if your custom MatLab or ADI simulation results include this.

Shawn

Hi Shawn,

can you confirm that Kens VerilogA models are working with PSS+PNOISE ?

I am always a bit puzzled among the different definitions of phase-noise spectral densities.

Can you tell which PHN-definition I get when doing PSS+PNOISE simulations and plot "phase-noise" ?

Is it L(f), S(f), Sphi(f), or Sy(f) ?

What I understand is, that Sphi(f) contains the double-sided power spectral density of phase fluctuations, whereas L(f) only uses the one-sided spectral power density and this is defined by IEEE as phase-noise.

I also think the spectrum-analyzer measures L(f) - right ?

In my Matlab-script, I derive all the transfer-functions from the individual blocks (VCO, FBDIV, PFD, REF, LPF) to the output node.

Then I calculate the contribution to the output phase-noise of each individual block at the output node and sum-up the output power - the result is the total output phase-noise spectral denstiy.

When I am using L(f) for the individual blocks, then the output-noise will be L(f).

When I am using Sphi(f) for the individual block, thn the output will be Sphi(f).

There are 3 unknowns here:

- the VCOs phase-noise has been simulated with PSS+PNOISE - is it Sphi(f) or L(f) ?

- the LPF noise is caused by the resistors, which is flat thermal noise only - is this a Sphi(f) of L(f) spectrum ?

- the PLL intrisic phase-noise (PFD, FBDIV, CP) is usually determined by measurement and provided in PLL datasheets and has a 1/f and a flat component given in dBc/Hz - are those values given Sphi(f) or L(f) ?

Another question arising from this is, from which phase-noise spectrum the integrated rms-jitter is derived - from Sphi(f) or L(f)  ?

Thanks to your comment, I found that I need to dig deeper into this topic......

Dear HoWei,

First, you are asking some very good questions in my opinion given the magnitude of the discrepancy you are observing!! Let me try to do my best to provide my thoughts on your questions and comments...I hope they are somewhat useful to you!!

HoWei said:

can you confirm that Kens VerilogA models are working with PSS+PNOISE ?

Yes. In fact, there is an article on the On-line support site that recommends the use of Ken Kundert's macromodels in the simulation of the phase noise of a behavioral PLL. The article is at URL:

https://support.cadence.com/apex/ArticleAttachmentPortal?id=a1O0V00000679z2UAA&pageName=ArticleContent

HoWei said:

Can you tell which PHN-definition I get when doing PSS+PNOISE simulations and plot "phase-noise" ?

Is it L(f), S(f), Sphi(f), or Sy(f) ?

What I understand is, that Sphi(f) contains the double-sided power spectral density of phase fluctuations, whereas L(f) only uses the one-sided spectral power density and this is defined by IEEE as phase-noise.

If you plot phase noise from the Direct Plot form for an autonomous circuit and select both the PM and then the dBc/Hz axis option, these choices  will default to the SSB option on the Direct Plot form and this represents the L(f) function defined by IEEE.

HoWei said:

I also think the spectrum-analyzer measures L(f) - right ?

Actually, a spectrum analyzer, as a standalone piece of test equipment does not measure and then plot phase noise. Phase noise is measured by a phase noise test set. In a spectrum analyzer, you are measuring the amplitude of a signal as a function of frequency and this is not phase noise. There are several vendors that provide phase noise measurement routines for use with a spectrum analyzer - or you can attempt to do the calculations manually - but these are only approximate estimates of phase noise. Typically, the dynamic range and noise floor of a spectrum analyzer is simply not sufficient to guarantee a good estimate of the actual phase noise. The most accurate instrument, which is a also considered the "standard" by many is manufactured by Keysight as the 

https://www.keysight.com/us/en/product/N5511A/phase-noise-test-system.html

They also provide a much lower cost instrument, albeit not as accurate, which I purchased and made a LOT of use of called the E5052B

https://www.keysight.com/us/en/product/E5052B/signal-source-analyzer-ssa.html

HoWei said:

There are 3 unknowns here:

- the VCOs phase-noise has been simulated with PSS+PNOISE - is it Sphi(f) or L(f) ?

As mentioned above, if you make the proper Direct Plot GUI selections, the output in dBc/Hz will be L(f). This is also shown in the Cadence training video (about 7 minutes) on performing phase noise simulations on oscillators at the On-line support URL:

https://support.cadence.com/apex/ArticleAttachmentPortal?id=a1O3w00000AGxpNEAT&pageName=ArticleContent (about 7 minutes)

The video may be helpful to supplement or add to my comments HoWei!

HoWei said:

- the LPF noise is caused by the resistors, which is flat thermal noise only - is this a Sphi(f) of L(f) spectrum ?

Actually, the resistor noise on its own is not phase noise at all. The resistor noise results in phase noise to the extent that the noise causes a voltage variation in the control voltage of a VCO. Hence, the voltage noise which is applied to the VCO control voltage over a 1 Hz bandwidth must be computed from the resistor noise and then filtered by the high-pass loop transfer function of the loop to determine its contribution to the PLL output phase noise.  If you need a rather basic review of this computation, you might consider reviewing the article Mr. Lance Lascari wrote for Applied Microwave & Wireless magazine in 2000 at URL:

https://www.linkedin.com/redir/redirect?url=http%3A%2F%2Fwww%2Erfdude%2Ecom%2Flance_resume%2FLL_apr2000pg30%2Epdf&urlhash=Hgfj&trk=public_profile_publication-button

HoWei said:

- the PLL intrisic phase-noise (PFD, FBDIV, CP) is usually determined by measurement and provided in PLL datasheets and has a 1/f and a flat component given in dBc/Hz - are those values given Sphi(f) or L(f) ?

The answer to this is found in my last comment concerning the resistor noise contribution to PLL output phase noise. Their noise must be translated to voltages that are applied to the VCO control voltage and then the impact of the PLL error transfer function must be applied to determine their contribution to PLL output phase noise. Keep in mind that since many VCO have very, very large values for the Kvco, even small amounts of voltage noise can be quite significant contributors to PLL output phase noise.

HoWei said:

Another question arising from this is, from which phase-noise spectrum the integrated rms-jitter is derived - from Sphi(f) or L(f)  ?

The integrated rms jitter is computed as a direct integral of the L(f), SSB, phase noise over the desired bandwidth with whatever transfer function may be required by the standard you are trying to meet.

I hope I managed to both understand and provide thoughts on all of your questions and the comments provide at least some help in resolving your observed discrepancy HoWei!

Shawn

Hi Shawn,

your comments just remind me that I was using the HP3048A phase-noise test-system about 18 years ago to measure some GHz VCOs I designed. I also remember that we were using an very old analog oscilloscope to observe the beat-note frequency of the mixer output whilst we were manually tuning the reference oscillator close to the DUT frequency. Once the HP3048A PLL was locked, we were using this very high accurate  FFT analyzer instrument to measure the phasenoise spectra in baseband (offset from DC).

I agree that a simple spectrum analyzer with phase-noise module cannot provide this kind of measurement.

Now, I am motivated enough to read thru the old HP manuals and application notes again (e.g. https://www.hpmemoryproject.org/an/pdf/HP_PN_seminar.pdf) !

Thanks for that :-)

Seems you running a business on designing clock-sources and/or synthesizers *curious* ?

Dear HoWei,

Wow! Great memories I am sure - and your comments sparked a few memories of mine too! We did have a full phase noise system at one time some years ago that I used to characterize and measure quartz resonator based oscillators and VCXO. I was to surprised to read your background is in using the HP3048A too - great!! 

I actually am not running a business in clock-sources nor in synthesizers but have designed them for longer than I care to admit!! I However, the original design for the Vectron series of the S-type XO/VCXO was from the quartz resonator  group in which I was a member for years...I still keep in touch with those Vectron employees who were in our group...

Thank you HoWei!!!

Shawn