Designer’s Guide Community

hiSkill_11

Community Member

Offline

Posts: 35

strange "noise" riding at frequency measurement
Aug 12^th, 2015, 2:10am

When I simulated a RC oscillator with digital controller, and use calculator freq funtion to plot out clock freqency, I found unrelated noise riding at frequency result? Is it realistic or just tool accuracy issue? I use spectre with maxmium step 10ps and other default setting. The target frequency is about 72Mhz.
Thanks, any comment?

Back to top

wave_001.png

IP Logged

sheldon

Community Fellow

Offline

Posts: 751

Re: strange "noise" riding at frequency measurement
Reply #1 - Aug 13^th, 2015, 10:26pm

hiSkill,

Do you mean the pulse or the small variations in frequency? For
the small variations, the accuracy is 10ps. Using my handy paper
napkin to help with the calculations. Adding ad subtracting 10ps
for the period at 72MHz, the expected variation is frequency is
about 100kHz[+/- 50kHz variation from the center frequency]
so the low level noise seems to be consistent with the numerical
accuracy control. Not sure what is causing the step.

Sheldon

Back to top

IP Logged

hiSkill_11

Community Member

Offline

Posts: 35

Re: strange "noise" riding at frequency measurement
Reply #2 - Aug 14^th, 2015, 2:38am

Hi Sheldon,

Thanks for your good calculation which guiding me well understood the phenomenon. The step is due to digital trimming algortith which drives a nonaccurate clock to a target one.

So the frequency jittering is due to accuracy setting, right? If i set maximum step 1ps, then the expected varaition is 10kHz, right?

Could you also please explain a little why the maximum step impact on the frequency solution of circuit?

Is there any ways to kill the "jitter" without affecting simulation speed?

Thanks very much.

sheldon wrote on Aug 13^th, 2015, 10:26pm:

hiSkill,

� �Do you mean the pulse or the small variations in frequency? For
the small variations, the accuracy is 10ps. Using my handy paper
napkin �to help with the calculations. Adding ad subtracting 10ps
for the period at 72MHz, the expected variation is frequency is
about 100kHz[+/- 50kHz variation from the center frequency]
so the low level noise seems to be consistent with the numerical
accuracy control. Not sure what is causing the step.

� Sheldon

Back to top

IP Logged

Ken Kundert

Global Moderator

Offline

Posts: 2384
Silicon Valley

Re: strange "noise" riding at frequency measurement
Reply #3 - Aug 14^th, 2015, 9:17am

You would be better served if you stopped adjusting maxstep (return it to its default value) and instead moved the frequency measurement to inside the simulator. Using the calculator to measure frequency �is innately low accuracy because it occurs outside the simulator and employs linear interpolation. A better approach is to use Verilog-A to measure the frequency. An example how to is given in the period measurement example on the Verilog-AMS page.

In general, adjusting maxstep is a bad way of controlling simulation error. Instead you should adjust reltol. In addition, it is better to perform timing measurements in verilog-a instead of the calculator. If you do these two things you will find that your simulations are both much faster and much more accurate.

-Ken

Back to top

IP Logged

hiSkill_11

Community Member

Offline

Posts: 35

Re: strange "noise" riding at frequency measurement
Reply #4 - Aug 19^th, 2015, 2:39am

Many thanks to you !

Ken Kundert wrote on Aug 14^th, 2015, 9:17am:

You would be better served if you stopped adjusting maxstep (return it to its default value) and instead moved the frequency measurement to inside the simulator. Using the calculator to measure frequency �is innately low accuracy because it occurs outside the simulator and employs linear interpolation. A better approach is to use Verilog-A to measure the frequency. An example how to is given in the period measurement example on the Verilog-AMS page.

In general, adjusting maxstep is a bad way of controlling simulation error. Instead you should adjust reltol. In addition, it is better to perform timing measurements in verilog-a instead of the calculator. If you do these two things you will find that your simulations are both much faster and much more accurate.

-Ken

Back to top

IP Logged