Measuring phase noise of non-autonomous circuits (DLL) with SpectreRF

Hi,

I need to measure phase noise of the non-autonomous circuit - VCDL
(voltage controlled delay line) with spectreRF. How should I do that?
I have some results but I'm not sure if they are trustworthy - I
measured the phase noise using two methods but the results are
completely different.

Let's say that I'm using following configuration for spectre
simulation:

pssout ( ) pss fund=0.45G harms=10 errpreset="moderate" tstab=200n
maxstep=50ps
pnoiseout (out gnd) pnoise sweeptype="relative" relharmnum=1 start=100
stop=100M dec=5 maxsideband=21 +noisetype="sources"

Is this ok? My main concern is the pnoiseout simulation and its
noisetype parameter. Is this ok to use noisetype="sources"? Or maybe I
should use "timedomain"?

When the simulation is finished I get results in V/sqrt(Hz) units. Is
this enough to convert such result to dB20? Or maybe there is a need
for different conversions/computations? Should I take into account the
shape of the signal? (divide by slew rate or something like that). If
yes, there is another problem, the signal shape changes in the circuit
-> The circuit input is rail to rail clock signal. Inside the VCDL the
signal's amplitude is limited and signal is less square like. Then
signal is buffered so the output of VCDL is again rail to rail square
signal.

Thanks in advance for any advice.

Regards,
Bart

 

I've used SpectreRF a lot for my simulations for a RF VCO design. I
was actually under the impression that SpectreRF is the most accurate
simulator for this type of analysis (phase noise). After a
considerable period of time, I finally concluded that other simulators
like EldoRF and ADS are more accurate when it comes to phase noise
analysis.. The main differentiator here is the harmonic balance method
used to analyze the circuit... PSS as used by Spectre has turned out
to be not as accurate as the harmonic balance..

My advice would be to switch to a harmonic balance simulator to avoid
any sort of confusion that would be related at the end to non-accuracy
in SpectreRF.

Thanks.

 

It largely depends on the kind of circuit. Very linear oscillators are likely to
benefit from harmonic balance simulators, whereas more non-linear oscillators
(and things like ring oscillators) are more likely to benefit from shooting -
and harmonic balance will be less accurate.

Since SpectreRF has both shooting and harmonic balance, you can use the most
appropriate algorithm for the circuit.

SpectreRF has many years of showing excellent results for phase noise. Having
harmonic balance as well allows you to analyse circuits with very high dynamic
range (i.e. very linear circuits) more efficiently.

So I really wouldn't make sweeping statements about SpectreRF being inaccurate.

You could equally well say the same thing about inappropriate use of harmonic
balance.

As for Bart's question, you may well want to look at the "jitter" and
"modulated" modes of pnoise analysis. Not sure if you're running from the
command line, or using ADE - but doing it via ADE gives you a whole bunch of
post-processing capabilities which could be very useful in this case. For
switching circuits, you don't really care about the noise during the periods
where it is not switching - since that's only AM noise. Using the method you've
suggested, you're finding the time-averaged noise, and then you can convert it
to dBc by dividing by the signal amplitude and plotting in dB. However, that
will then contain both the AM and PM components, which is probably not what you
want. The "modulated" mode will split the noise into AM and PM components, and
the jitter mode is a smart tdnoise mode where it automatically performs the
noise at the time of the threshold crossing (try doing that in a harmonic
balance-only simulator!).

I'd also suggest taking a look at Ken Kundert's paper on analysing phase noise
and jitter in PLLs on http://www.designers-guide.org - in the Analysis (I think)
section.

Regards,

Andrew.

 

Thanks Andrew for the valuable info, it's always beneficial to learn
about the right analysis for the appropriate type of circuit.

I never knew that SpectreRF supports a HB kernel. Was that recently
introduced ? I actually used it like 3 years ago, and the tight spot I
will always remember was getting a high phase noise (bad performance)
with an ideal LC tank filter when simulating that on SpectreRF and
that was a turning point for me believing that SpectreRF should not be
considered as accurate...

Do you have any useful comments concerning this issue ?

One thing more, about the "shooting" method, is that the one
introduced by Hajimiri ?

Thanks,

 

LC tanks are a particularly good application for HB in general, because they are
rather linear and have very high dynamic range. HB has been in spectre for the
last 2.5 years - it was introduced during MMSIM601/MMSIM602 timeframe. Switching
circuits and non-linear circuits can be very innaccurate in HB though...

The shooting method is nothing to do with Hajimiri as such - I found some papers
which he is talking about this (and he references Ken Kundert et al's book on
simulator methods - I'll reference in a moment):

K. S. Kundert, J. K. White and A. Sangiovanni-Vincentelli,
Steady-State Methods for Simulating Analog and Microwave
Circuits, Kluwer, Academic Publishers, 1990

Note that the pnoise analysis in SpectreRF is closer to the methods of Demir,
rather than Hajimiri (who has more of an empirical noise model).

There are quite a few discussions about this on http://www.designers-guide.org
in the Forum there. You might want to take a look.

Regards,

Andrew.