I recently saw a demonstration of a harmonic balance simulator that the salesman claimed simulated phase noise more accurately than SpectreRF. He claimed his simulator accurately captured the effects power supply ripple and SpectreRF did not.  I did not buy the sales pitch for the following reasons and was wondering if my concerns were justified.

1. The power supply driving the VCO was a linear regulator supplied by an ideal source.  I failed to see where ripple in a linear regulator would physically come from.  If the ripple came from the VCO, SpectreRF should accurately capture it. The salesman claimed the LINEAR regulator had ripple independent of the VCO that SpectreRF missed. Did I miss something here? (Assuming the linear regulator was stable.)

2. Since harmonic balance is based on a truncated Fourier series, and a truncated Fourier series could have ripple, might harmonic balance introduce a fictitious ripple on the linear regulator output that would in turn falsely corrupt the phase noise simulation?

On different issue, the salesman claimed SpectreRF's noise floor was higher because the shooting Newton method uses a variable time step. If one strobes the output at the proper rate, doesn't that make SpectreRF's noise floor comparable to noise floor of a harmonic balance simulator?

Thanks for the response Ken.   I was hoping you would chime in on this one.  

The example was an autonomous circuit and I believe the HB tool requires a special element that gives the oscillator a "kick start". Couldn't the kick start cause the ripple problem or does the discontinuity have to be periodic?

-Eugene

Thanks Ken.  

Now I am just curious. Getting back to the linear regulator ripple issue, suppose the regulator had an internal instability that produced a small amount of ripple at its output. Assume this poorly designed regulator drove a VCO. In this case, the linear regulator contains an unwanted oscillator. Would either harmonic balance or shooting Newton methods accurately simulate the problem?

For the regulator/VCO circuit, the user might not know to place a second oscprobe (if it is even possible.) But for some circuits, the user would know to place a second oscprobe. Some frequency plans call for two VCOs operating at different frequencies. Would either simulator have a chance at simulating multiple VCOs? If both could do it, would one do it better?  Envelope methods would probably work for transients but what about phase noise?

Thanks Ken. This discussion was very informative.

-Eugene

Hi Vborich,

Thanks for the response but I was not referring to 60 Hz ripple. The ps ripple I was referring to was due to an unstable linear regulator running off a battery. In this case there are two unknown frequencies, the power supply oscillation frequency and the VCO frequency. I doubt any RF simulator to date is ready for this challenge.

-Jess

Art,

It is true that one would eventually want to eliminate the unwanted oscillation and proceed with a stable power supply. However, with ICs, we want to be as sure as we can that we've correctly identified and solved a problem before we tell our management that the next tape out will be OK. Before one can prove that a simulated solution works, one should simulate all observable symptoms of the problem and phase noise might be one of them. Furthermore, many circuits, like superhet transceivers, actually have two VCOs and one might want to simulate the phase noise at the final output. Even with direct conversion, if one wants to simulate the entire link, the transmit and receive VCOs will be independent and often at different frequencies. My point is that there are times when we need to simulate total phase noise from two or more independent sources of oscillation.

Thanks for the update on SpectreRF. I did not know that Cadence released an autonomous version of envelope following. That is indeed a welcomed step in the right direction.

-Eugene

Hi Eugene,

In principle, you can perform the signal analysis of multiple oscillators using multi-tone envelope simulation. ADS/RFDE features multi-tone envelope analysis. (As a disclaimer, I work for Agilent).

You set the tone frequencies as close as possible to the VCO and the parasitic oscillation frequency and it'll likely pick up the correct multi-tone oscillatory regime. If it doesn't, tickler sources may be needed to excite it.

It is also possible to run multi-tone envelope simulation with one unknown frequency, in which case a probe is needed and it virtually guarantees convergence to the correct frequency of the oscillator to which it is attached.

Hi Eugene,

You can run an HB noise analysis at the end of an envelope simulation. The operating point is taken to be the last envelope time sample. (You can also run a Monte-Carlo envelope noise analysis, where noise sources are represented as noisy AM/PM modulation of the HB carriers. Envelope noise is a Monte Carlo type analysis and so it carries with it many of the same disadvantages.)

I've never attempted a phase noise simulation with two unknown frequencies. For reasons that I can't get into in a public forum, I'd expect it to work under limited circumstances. One instance involves an arbitrary oscillator and another that is very high-Q.

There are no fundamental obstacles to multiple oscillator HB, but there are practical considerations that make it unappealing. Oscillator analysis is inherently less efficient than standard HB and it really shows in large systems where the whole thing becomes one big, usually multi-tone, oscillator. The simulation would get slower with multiple oscillators and the slowdown wouldn't scale in linear proportion to the number of oscillators. In fact, even with a single oscillator, most users choose to simulate it separately and use a behavioral representation to simulate the link.

I am nearly certain, after seeing a white paper describing what you saw in the sales presentation, that you didn't see a two-oscillator noise simulation. Rather, I am quite sure that the "ripple" was represented as an equivalent noise source, peaking at the fundamental frequency of the ripple. This can be done routinely in most HB simulators I am aware of.

Vuk