Hello Attila,

It seems to me that an AC simulation could never work since the
very generation of phase noise by the mechanisms that matter is
a modulation process at heart, automatically forcing one into the
realm of transient simulations.

But I am surprised about the simulation times that you speak of.
Would you be willing to post some information detailing your
methodology and an example "simple" circuit?

Dana

On Sun, Oct 22, 2017 at 7:53 AM, Attila Kinali attila@kinali.ch wrote:

Dear Attila,

you haven't specified what sort of circuits would you like to simulate,
but maybe the answer is Harmonic Balance. Have a look at
http://qucs.sourceforge.net/ and
http://qucs.sourceforge.net/tech/node36.html

HSPICE from Synopsis and ADS from Keysight (which I use) also have the
HB engine.

Best regards,
Rafael Gajanec

On 22-Oct-17 4:23 PM, Dana Whitlow wrote:

Hoi Attila

Since close in phase noise can result from up conversion of supply noise etc via circuit non linearities, using an AC analysis won't work.

Only transient simulation or perhaps analytical modelling of the various non linearities will provide accurate estimates of upconverted PN. If you use transient simulation techniques increasing the level of the various noise sources above the actual levels encountered in real circuits and then correcting the resultant PN back to the level that would be encountered in the actual circuit (using the results of analytical modelling) may be a useful way to reduce simulation time or at least overcome some of the challenges associated with accurately determining low level PN from a simulation.

There are some in the LTSpice Yahoo group attempting this but they seem way out of touch with the amount of simulation data required. I've provided them with the appropriate formulae to extract PN from the the amplitude spectra. At the moment they appear bogged down with some somewhat trivial peripheral issues.

Bruce

If one for example wishes to estimate PN down to an offset of 1Hz then an equivalent filter noise bandwidth of 0.1Hz or perhaps less is desirable (the PN spectrum  at low offsets is far from flat). To achive accurate noise estimates a simulation time of at least 100 x the reciprocal of the equivalent noise bandwidth is required. The resultant simulation for 1000 sec or more takes considerably longer than 1000 sec to  run.

Bruce

One has to provide noise models that work with the Spice transient simulation for all devices including resistors. Random number generators can be used but they need to be independent and must not repeat during the entire simulation.

Bruce

Am 22.10.2017 um 22:58 schrieb Bruce Griffiths:

In a previous life, when I was an EE&CS student, we had to write all the
relevant algorithms ourselves, like building the conductance matrix,
finding the operating point, linearizing nonlinear devices around the
OP, doing the integration over time, companion models etc, b4 we were
given the Spice 2G4 sources...

(Attila, that was a few 100 meters from where you seem to work right
now. There was a beautiful TR440!)

Given that we often enough see convergence problems in integration over
time to the point that the simulator gives up altogether, especially
when there are high Q resonances or nonlinearities around, and that
these errors look like phase noise, I would never ever trust a FFT
result at, say, the -140 dBc level. And there it just starts to be
interesting.

As much as I like to use LTspice, it's easy availability blocks any fast
progress in the public spices like adding HB, s-params by diverting
people to experiment with add-ons instead of solving the fundamental
issues. X/Ngspice and QUCS are nice but understaffed for sure.

regards, Gerhard.

(who was designing a chopper amplifier in the 140 pV/rt Hz league this
rainy weekend and did not even try to simulate its noise. The
interesting part of it would never make it through the pot core
transformer.)

Hi

One would guess that they put them in parallel to get more drive. If that’s correct,
details of the loading are going to get into the simulation pretty quickly.

In a lot of cases, these amplifiers were designed against a specific need. If you have
a signal source that is in the -180 dbc / Hz range, they are unlikely do perform well.
In many cases a floor in the -140 dbc / Hz range was considered “good enough”.
If you are simply driving common test gear, it probably is good enough. If the
application was video rather than a standard the specs could have been very different.

In the case of an amp with a LMH6702, you are not going to get super  close in
phase noise. The device is very noisy under 1 MHz. It also starts to increase distortion
by 10 MHz so you will see up conversion. It probably did quite well against the intended
design spec.

=====

If you need a system that will distribute one frequency today and a totally different
frequency tomorrow, broadband makes sense. For the more common task of
something like “only 10 MHz”, it does not make much sense at all. Gain other
frequencies is just going to spread around noise from this or that source
of crud. Driving filters with op amps can be problematic. It often is easier to go another
way.

Bob

I never had much luck with current feedback amplifiers such as the
LMH6702.  Their input current noise (at the time) was too high for
my needs and their output peaks at higher frequencies if the
feedback resistors aren't optimal for the part.

I had the best results with voltage feedback op amps like the
MAX4104/MAX3404 when I needed gain on the input stage and the
LMH6609 when I needed a buffer.

My applications were broadband. If I remember correctly,
aggressive bandwidth limiting can cause phase shift problems due to
temperature changes unless one is careful in the design of the filter.

I've successfully put as many as four op amps in parallel in an
input stage to reduce phase noise.

Bob M (another bob)

On 10/24/2017 6:24 AM, Bob kb8tq wrote: