The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> Circuit Simulators >> PSS/PAC analysis with SpectreRF: double sampling
https://designers-guide.org/forum/YaBB.pl?num=1087304747

Message started by reuben on Jun 15^th, 2004, 6:05am

Title: PSS/PAC analysis with SpectreRF: double sampling
Post by reuben on Jun 15^th, 2004, 6:05am

Hi all,

I am using SpectreRf to run a PSS/PAC analysis on a switched-current circuit (a bit like SC). I have been using the simulator on these circuits for a couple of years now with great sucess. I have recently tried to simulate some circuits which have double sampling (sampling on both high and low phases of the clock) which means the clock is 1/2 of the sample frequency. The PSS form insists I use the clock frequency to do the operating point analysis about, but of course this does not work correctly since it is only half the sample frequency. I can put a divide by two after the clock so that the clock frequency equals the sample frequency, but PSS analysis does not seem to converge.

Any ideas?

Cheers,

Reuben

Title: Re: PSS/PAC analysis with SpectreRF: double sampli
Post by Frank Wiedmann on Jun 15^th, 2004, 7:33am

The PSS analysis cannot converge if the simulation period is not an integer multiple of all signal periods in the circuit (integers including 1). PSS analysis tries to match all signals at the beginning and at the end of the simulation period, which is impossible if this condition is not fulfilled. It is possible to fool the PSS form, but the PSS analysis will punish you with nonconvergence in this case.

In your circuit with double sampling, you definitely have two different sampling instants during the clock period and there is a priori no reason why the sampled values at both instants should be identical. You will probably have to live with this fact and find a way to compute your results from the values at both instants.

Title: Re: PSS/PAC analysis with SpectreRF: double sampli
Post by reuben on Jun 15^th, 2004, 8:42am

Frank, thanks for the reply. This is why I put a divide by two after the clock source, hence there is now one sample per clock period. But, it does not converge...

Title: Re: PSS/PAC analysis with SpectreRF: double sampli
Post by Andrew Beckett on Jun 15^th, 2004, 10:27am

Hi Reuben,

If you put a divide by two in the circuit, you'll definitely need to adjust the PSS frequency to compensate - because you then have a period which is twice the original clock period.

In the case of sampling on both phases of the clock, then this should still be periodic in the original clock frequency, and so I don't see that you should have to do anything special.

Unless you are unintentially introducing some division in there which you've not realised? Does it converge if you half the PSS frequency (just enter it by hand on the form) - missing out the divider which should be unnecessary?

Regards,

Andrew.

Title: Re: PSS/PAC analysis with SpectreRF: double sampli
Post by reuben on Jun 16^th, 2004, 4:43am

Andrew - this is why I was confused. Say I set the clock to 50MHz, then the circuit is actually sampling at 100MHz (because of the double sampling) but the PSS form will insist I use 50MHz for the PSS frequency. If I run this simulation (with 50MHz) then it converges and the PAC runs, but the results are way off expected because the thing is actually sampling twice every PSS period.

So, I put a divide by two after the clock, and I change the clock to 100MHz. Now the clock is at 100MHz but after the divide by two it is 50MHz (as before) and so the sampling is again 100MHz (double sampling). Now the PSS form allows me to put 100MHz in as the PSS frequency, but the PSS analysis does not converge. I really cannot see why to be honest - I have one sample for every clock period this way, which is exactly the same as a normal single sampled circuit.

I have tried changing the tstab parameter but with no joy.

Cheers,

Reuben

Title: Re: PSS/PAC analysis with SpectreRF: double sampli
Post by Ken Kundert on Jun 16^th, 2004, 9:22am

Reuben,
How is it that you are conveying to Spectre that you are sampling at all? You say that when you specify the fundamental frequency as 50 MHz that Spectre converges but gives unexpected results, but you neglect to say what results you are expecting, any why the results that Spectre gives looks wrong to you.

From what I have learned so far, it seems that the problem you are having is not due to having the wrong fundamental frequency (50 MHz is the correct frequency), but rather with either the way you are determining the desired results, or your interpretation of those results. Perhaps you can say more about the results.

-Ken

Title: Re: PSS/PAC analysis with SpectreRF: double sampli
Post by Andrew Beckett on Jun 16^th, 2004, 2:47pm

Not really adding much to what Ken said, but remember that the PSS frequency is the the greatest common divisor of all the frequencies in the circuit. We need to have an integer number of cycles of all frequencies simulated. So with an input frequency of 50MHz, then you need to have a PSS fundamental of 50MHz. You'll get 1 cycle of the 50MHz, but 2 cycles of 100Mhz (if things are occuring at 100MHz).

With your second example, specifying the input clock at 100MHz, and PSS fund at 100MHz, you have 1 cycle of the input clock, and 1 cycle of the part of your circuit that operates at the doubled frequency. However, you're only capturing half of the period at the output of the divider, which is why it won't converge.

So, what you were doing first is most likely to be the right thing.

As Ken said, please explain what it is about the results that weren't correct - what were you looking at? Perhaps you were looking at the wrong sidebands?

Andrew.