HB 2-tone puzzling results - Some basic understanding missing?

The initial transient results from hb/qpss are to be expected, because only the "large" tone (or first tone for hb) are enabled during the initial transient. It would be rather pointless having both enabled, because the purpose of the initial transient is to give a transient assist, and you'd have to simulate over (at least) the common frequency of the tones which would be very slow. So instead the aim is to assist the solving of the first tone, with the second expected to have lower linearity effects.

Can you please share the input.scs so I can check with the component values you have to make sure I see what you see? Both for your hb and your qpss setups would help.

Andrew

Hi Andrew,

I think I know why I'm seeing the effect I'm seeing. It's maybe worth a talk.

First, thanks for the comment on the HB/QPSS second tone being zero "as to be expected". I am not fully agreeing with the argument that you would otherwise be simulating for a long time because, well, you do decide how long the transient should go by settting the tstab parameter to your liking, posslbly exaclty because you want to see "slow"  effects (although not in full precision).

I must admit however, although it should be clear from the question, that I am not used to run QPSS or HB with a tran before, so I was expecting from the simulator the same behaviour it displays for a tstab transient in PSS.

OK, now I know

My current setup does not allow export of files outside the simulation environment unfortunately, I will share a couple of snapshots and I hope you can work with those.

But prior to that, let me tell you that I did manage to get 'meaningful' results with PAC, which was not the case in the beginning, and I think HB is no different.

The circuit I am simulating is a 2 phase mixer (positive and negative, 180 deg apart). This type of mixer is sensitive to phase relationship between LO and RF when they are isofrequential. I think this might have to be at the core of the issue.

When fLO = fRF (isofrequential case), a "cosine based" LO (i.e. symmetric around zero) and an RF sine will give rise to a zero swing at the vx node.

TRAN and PSS do give the same, correct result.

In the pic above, the red vx node is virtually zero because of the phase relationship between the LO and the RF tone.

However, when fRF != fLO, then the 'phase' concept drops altogether and this does not play a role anymore.

TRAN understands and reflects this, while PSS does not.

If I rotate the phase of fRF, though, magic happens and now PSS understands the circuit too:

Still I don't get why TRAN and PSS do not fully agree on the behaviour of such a simple circuit, as one can see that TRAN exhibits a beat at fRF-fLO, as it should be, while PSS does not, neither in the preliminary transient, nor in the final result.

Also, note that TRAN results are constant vs this phase rotation, PSS are not.

This, although the beat frequency is set at 1M - maybe because I should then increase the harmonic number? Still, if the beat frequency is (autocalculated) to be 1M, why am I seeing a 10GHz repeating waveform?

OK, while preparing to send you the netlists, I noticed something that of course changes the whole picture.

While the above is my input interface, you see two tones and a beat frequency of 1M and all that, this is the corresponding netlist:

So it looks like it's not simulating the way it is written through the GUI. What am I doing wrong here?

I think this is the first relevant thing to address.

However, since we discussed HB so far, here's a result for it vs TRAN, it's same old thing (i.e. not working):

and its corresponding netlist:

phase phi = 90 in the last two simulations.

Regards,

Michele

Michele,

AncisMichele said:

First, thanks for the comment on the HB/QPSS second tone being zero "as to be expected". I am not fully agreeing with the argument that you would otherwise be simulating for a long time because, well, you do decide how long the transient should go by settting the tstab parameter to your liking, posslbly exaclty because you want to see "slow"  effects (although not in full precision).

The tstab is not intended to be for your benefit to observe transient like behaviour. That's what transient analysis is for. The whole point of tstab is to assist the analysis to have a better starting point, and given that the main reason for using a multi-tone harmonic balance or a quasi-periodic steady state analysis is that the frequencies involved are not commensurate and so the beat frequency (for a reasonable size circuit) would make it unfeasibly slow to simulate with PSS (which only simulates over a single common frequency) and having to run that common frequency in the initial transient would rather defeat the major benefit of a multi-tone analysis.

Anyway, screen shots of parts of the netlist (knowing parameter values would have helped a bit, but screenshots are pretty much useless as I'm not going to re-type in the netlist). I also cannot see any reason why the choose analysis settings wouldn't be reflected in the netlist unless something has gone horribly wrong with the setup and they're being ignored (it seems unlikely to me though as I've not seen this before).

AncisMichele said:

This, although the beat frequency is set at 1M - maybe because I should then increase the harmonic number? Still, if the beat frequency is (autocalculated) to be 1M, why am I seeing a 10GHz repeating waveform?

Increasing the number of harmonics probably won't help much (that would affect the frequency domain results because you'd only be observing 15 harmonics of the common frequency which is not terribly interesting with PSS). The time domain results with PSS should be similar to the transient though.

AncisMichele said:

So it looks like it's not simulating the way it is written through the GUI. What am I doing wrong here?

No idea - as I said above. This seems like a symptom I've not seen before.

Trying to guess what's wrong without having something to explore around is very hard, especially as you're now looking at something different from the original setup you described. I suggest you contact customer support. 

Andrew

Thanks Andrew

I do think something is really going wrong, as I said the GUI and the netlist out of the "view netlist" RMB menu do not agree.

With that netlist actually simulated, I have no doubt things go completely wrong.

On the other hand, a RMB on the test and "netlist--> recreate" does bring up a correct one:

Why that is, is a mistery to me.

However, with the correct netlist, meaning with the intended simulation parameters for PSS, everything works and simulates as expected.

Thanks for pointing me to the netlists Andrew, I somehow gave for granted that the GUI was connected to the netlist generator.

Apparently that was not the case.

Regards,

Michele

Thanks Andrew

I do think something is really going wrong, as I said the GUI and the netlist out of the "view netlist" RMB menu do not agree.

With that netlist actually simulated, I have no doubt things go completely wrong.

On the other hand, a RMB on the test and "netlist--> recreate" does bring up a correct one:

Why that is, is a mistery to me.

However, with the correct netlist, meaning with the intended simulation parameters for PSS, everything works and simulates as expected.

Thanks for pointing me to the netlists Andrew, I somehow gave for granted that the GUI was connected to the netlist generator.

Apparently that was not the case.

Regards,

Michele