Designer’s Guide Community

‹ Previous topic | Next topic ›

Pages: 1

Pnoise warning: "Relative residual of linear ..." (Read 4484 times)

Thomas

New Member

Offline

Posts: 3

Pnoise warning: "Relative residual of linear ..."
Feb 15^th, 2007, 8:30am

Hi,

I'm simulating a chopper amplifier. It works quite well, but when I include a sensor as a Verilog-A model a warning appears during the Pnoise simulation.

It says �Relative residual of linear system is significantly above tolerance. Results may be inaccurate.�

The simulation finishes and the results seem reasonable, but I'm not sure how much I can trust them. Does somebody have any idea how to solve this problem? I already increased maxacfreq to a high value and varied reltol, vabstol and iabstol. I have set the solver to std instead to turbo. The only thing what avoids this warning is to reduce the tolerance from 1e-9 (default) to 1e-7. But that doesn't really solve the problem, it just doesn't forces spectre to spit out some warnings.

Other question. How many sidebands should I use in the Pnoise. I thought that I increase the number of sidebands until there are no significantly changes. So I simulated up to 1000 sidebands, but it doesn't really look like its going to converge. Does somebody have some experience values which make sense to use?

Thank you so much.

Cheers
Thomas

IP Logged

Tawna Wilsey

Junior Member

Offline

Posts: 11
Snohomish, WA

Re: Pnoise warning: "Relative residual of linear .
Reply #1 - Apr 11^th, 2007, 3:29pm

The warning message about residual means that the simulator is not able to fully
converge the circuit at that point and that the data may not be not reliable. You
can get this message as well if you have chosen to perform a periodic small signal
analysis that is very close to the fundamental.

This warning indicates that the periodic small signal analysis being run needs a
more accurate PSS solution. One way to get a more accurate solution is to make the
PSS take more time points. You might need to set the PSS parameter "maxacfreq"
to a higher value. This will force spectre to take more timepoints and therefore
improve the accuracy of the Periodic small signal solution.

Why does this happen? SpectreRF injects the small signal tone in the time domain
then demodulates out the PSS harmonics to leave a signal that is just harmonics of
the Periodic small signal analysis tones. SpectreRF then takes a fourier analysis
of this solution to give the resultant small signal harmonics (sidebands). If the
small signal tone is close to a PSS harmonic the algorithm might not be able to
resolve the small signal tone accurately enough. To get more accuracy, you can
set maxacfreq (PSS option) to a value greater than 40*fundamental (default value).

Another tip is to set the solver method in your Periodic small signal analysis
option to "std" (standard). The default is "turbo" (recycled Krylov Sub-space
technique). The standard method is useful for frequency points close to PSS
harmonics (10-100Hz) in any of the small signal analyses.

Tawna Wilsey

IP Logged

Geoffrey_Coram

Senior Fellow

Offline

Posts: 1999
Massachusetts, USA

Re: Pnoise warning: "Relative residual of linear .
Reply #2 - Apr 18^th, 2007, 4:17am

Very nice explanation, Tawna.

However, it seemed to me that Thomas was saying he *doesn't* get the residual warning when the Verilog-A sensor is omitted. Did I read that right, Thomas? In which case, you might need to take a close look at the Verilog-A code and see if you're doing something funny there -- eg, if the sensor output is the wrong order of magnitude for the tolerances, like you have an output voltage that is 1e-9 but you have a voltage tolerance of 1e-6.

If at first you do succeed, STOP, raise your standards, and stop wasting your time.

IP Logged

Pages: 1

‹ Previous topic | Next topic ›