Designer’s Guide Community

BobDyl

New Member

Offline

Posts: 7

CONVERSION PROBLEM--> SC Time-Domain Simulation
May 05^th, 2006, 8:38am

Hi Everyone,

I am designing a fully-differential switched-capacitor (SC) voltage amplifier for an input frequency of 100 KHz and a sampling clock of 1 MHz. The core opamp is a fully-differential folded-cascode OTA with a UGBW of 30 MHz and DC gain of 80 dB. So, everything seems to be set for the Spectre simulation. However, I have a serious CONVERSION PROBLEM! Here are my time-domain settings:
Accuracy Defaults: Liberal
Integration Method: gear2
Iteratio: 100 or 500 or 1000
maxiters: 100 or 500 or 1000
But, I still can not get a clean sampled data at the outputs of my circuit. A simulation plot is attached. Please, let me know how I can solve this problem through correct setting of the time-domain options.

Thank you,
BobDyl

Back to top

diff_out.jpg

IP Logged

Ken Kundert

Global Moderator

Offline

Posts: 2386
Silicon Valley

Re: CONVERSION PROBLEM--> SC Time-Domain Simulatio
Reply #1 - May 5^th, 2006, 11:34pm

What is a conversion problem? Are you saying that the results are not accurate enough? If so, might I suggest starting with the default tolerances and then tightening them if you need more accuracy. You have loosened tolerances by setting errpreset to liberal, the you largely turn off the time step control by setting LTEratio to 1000.

I suggest you stop fiddling with all those settings, and start with the defaults. They generally work pretty well. If you need more accuracy, either try errpreset=conservative or tighten reltol a bit.

Once you get things looking good, strobe at the sample points to get the best accuracy.

-Ken

Back to top

IP Logged

sheldon

Community Fellow

Offline

Posts: 751

Re: CONVERSION PROBLEM--> SC Time-Domain Simulatio
Reply #2 - May 6^th, 2006, 1:32am

BobDyl,

It might be better to go back to the beginning, i.e., use the moderate
error preset. Does that clean up the waveforms? The settings you are
using make it difficult to comment about whether there is a circuit issue
or a simulator issue. Next question, is the circuit around the op-amp
configured properly? From the waveforms you are showing, it looks
like the circuit maybe oscillating. It appears to go from rail-to-rail with
every period. It would also be helpful, if you provided the target
closed-loop gain or the input sine magnitude so we had some idea of
what the expected results are. Also what is the slew rate of the amplifier?
Is the slew rate sufficient for the application?

Best Regards,

Scheldon

Back to top

IP Logged

BobDyl

New Member

Offline

Posts: 7

Re: CONVERSION PROBLEM--> SC Time-Domain Simulatio
Reply #3 - May 7^th, 2006, 3:43pm

Thank you for the replies.
Based on your suggestions, I used the simulation defaults and set a simple circuit to verify if the circuit was oscillating. This time, I used the attached differential amplifier (gain of -1) to check for the slew rate. I saw that oscillations were obvious when I used the defaults and it got much better when I used liberal (attached file)! It was in the case that the OTA had a PM of 85 deg and a UGBW of 30 MHz and the output-common-mode feedback was working properly. I still believe there is something wrong with my simulation settings and the circuit is not oscillating. The same problem shows up when I simulate the SC amplifier, which is more complex though. Please, let me know if you have had similar problem before.

Thank you again.
BobDyl

Back to top

test_bench.jpg

IP Logged

BobDyl New Member Offline Posts: 7	Re: CONVERSION PROBLEM--> SC Time-Domain Simulatio Reply #4 - May 7^th, 2006, 3:47pm BTW, I have used Pulse supplies instead of Sin supplies. Thanks. BobDyl
Back to top	IP Logged

Ken Kundert

Global Moderator

Offline

Posts: 2386
Silicon Valley

Re: CONVERSION PROBLEM--> SC Time-Domain Simulatio
Reply #5 - May 7^th, 2006, 6:51pm

If you are concerned about the simulation settings you should be using errpreset=conservative. That is what conservative is there for, it is the conservative settings.

I can see trapezoidal ringing in your first waveforms, so you should avoid trapezoidal. Using conservative does this. I would also avoid setting maxiters. Setting maxiters might help if you were having convergence problems, but this seems to be an issue with accuracy. Switched capacitor circuits are charge-based circuits, where charge conservation is critical. Errpreset=liberal is rarely an appropriate choice on such circuits.

-Ken

Back to top

IP Logged

sheldon

Community Fellow

Offline

Posts: 751

Re: CONVERSION PROBLEM--> SC Time-Domain Simulatio
Reply #6 - May 8^th, 2006, 1:06am

BobDyl,

Some additional comments/questions,

1) What is the gain and phase margin of the common-mode feedback loop?
The differential characteristics look reasonable, but there is definitely
an oscillation.

2) It is standard procedure to run both an ac analysis and a transient analysis
to verify circuit stability. For example, circuits with positive feedback,
inputs reversed, will often converge to a meta-stable state for dc op analysis
and have good gain and phase margin. Running a transient simulation is useful
in order to confirm the results of ac analysis.

3) While it is appropriate to use simulator settings to supress trapeziodal ringing,
changing the simulator settings is not an appropriate method for reducing circuit
oscillation. Circuit oscillations should be corrected by improve the circuits
gain and phase margin.

Best Regards,

Sheldon

Back to top

IP Logged

BobDyl

New Member

Offline

Posts: 7

Re: CONVERSION PROBLEM--> SC Time-Domain Simulatio
Reply #7 - May 8^th, 2006, 9:48am

I should mention that I am learning alot through our discussions and I really appreciate that. I checked the CMFB's DC gain and PM. Gain was about 75dB but PM was not good! I redesigned the OTA and had to limit its UGBW to 7 MHz to meet a CMFB gain of 75dB, PM of 45 deg and UGBW of 2 MHz. So, now the CMFB has an acceptabe specs. But still the trapezoidal response is showing up. The strange thing is that the circuit works perfectly when I have sinusoidal input signals (output swing plot is attached) and I was able to get a max differential output swing of better than 3 V from a 3 V supply. At this moment, I think I should try to increas the PM of CMFB to see how it affects the simulation results. Your suggestions are highly appreciated.

Thank you.
BobDyl

Back to top

output_swing.jpg

IP Logged

Geoffrey_Coram Senior Fellow Offline Posts: 1999 Massachusetts, USA	Re: CONVERSION PROBLEM--> SC Time-Domain Simulatio Reply #8 - May 16^th, 2006, 7:32am BobDyl - Are you completely ignoring Ken's advice to use errpreset=conservative??
Back to top	If at first you do succeed, STOP, raise your standards, and stop wasting your time. IP Logged

BobDyl New Member Offline Posts: 7	Re: CONVERSION PROBLEM--> SC Time-Domain Simulatio Reply #9 - May 16^th, 2006, 8:49am Hi Geoffrey_Coram, After improving the CMFB and using errpreset=conservative, the simulation was converging with no problem. Sorry that I didn't get back to the Designers-Guide for update! Thanks. BobDyl
Back to top	IP Logged