This discussion has been locked.

You can no longer post new replies to this discussion. If you have a question you can start a new discussion

PSS Shooting - High Q crystal oscillator - Simulator by mistake detects a frequency divider

Hi *,

I am simulating a 32kHz high Q crystal oscillator with a pulse shaping circuit. I set up a PSS analysis using the Shooting Newton engine. I set a beat frequency of 32k and used the crystal output and ground as reference nodes. After the initial transient the amplitude growth was already pretty much settled such that the shooting iterations could continue the job.

My problem is: In 5...10% of my PVT runs the simulator detects a frequency divider in the initial transient simulation. The output log says:

Frequency divided by 3 at node <xxx>

The Estimated oscillating frequency from Tstab Tran is = 11.0193 kHz .

However, the mentioned node is only part of the control logic and is always constant (but it has some ripples and glitches which are all less than 30uV). These glitches spoil my fundamental frequency (11kHz instead of 32kHz). Sometimes the simulator detects a frequency division by 2 or 3 and the mentioned node <xxx> is different depending on PVT - but the node is always a genuine high or low signal inside my control logic.

How can I tell the simulator that there is no frequency divider and it should only observe the given node pair in the PSS analysis setup to estimate the fundamental frequency? I have tried the following workarounds but none of them worked reliably:

- extended/reduced the initial transient simulation time

- decreased accuracy

- preset override with Euler integration method for the initial transient to damp glitches

- tried different initial conditions

- specified various oscillator nodes in the analysis setup form

By the way, I am using Spectre X (version 21.1.0.389.ISR8) with CX accuracy.

Thanks for your support and best regards

Stephan

Andrew Beckett 10 months ago

I moved this into an appropriate forum (the Feedback, Questions and Suggestions forum is for issues with the forum itself, not for technical questions).

The simulator (quite rightly) does not only use the specified node to detect the fundamental frequency - that would be highly dangerous and the result (if there are indeed non-commensurate frequencies at other nodes in the circuit) would be erroneous (it would be like doing an FFT on a non-periodic signal without windowing).

Why are there glitches in this control logic when it's supposed to be constant? Something sounds awry here.

I suggest you contact customer support (submit a support case after logging in) - that way an Application Engineer in our team can look at this together with you.

Regards,

Andrew
Cancel
Vote Up 0 Vote Down

Cancel
Stephan Bayerer 10 months ago in reply to Andrew Beckett

Hi Andrew,

thanks for your answer and sorry for posting my question to the wrong forum!

The pulse shaping circuit "transforms" the sinusoidal voltage on the XTAL pin to a square wave. Whenever this square wave shows a transition, it couples to the rest of the circuit whose nodes then show some glitches.
Cancel
Vote Up 0 Vote Down

Cancel
Stephan Bayerer 9 months ago in reply to Andrew Beckett

Hi all,

I am replying here again because I found the solution to my problem in the Cadence Help (see selected text in the attached screenshot).

Actually, increasing the beat frequency by a factor of 3.5 did not work for me because the detected fundamental frequency after the tstab interval was too high (37kHz instead of 32kHz) and thus convergence could not be achieved. Instead I used a factor of 2.5 which worked fine in my case. Anyway, I would find it more helpful if there was an option in PSS like the "freqdivide" for the QPSS analysis.

Best regards

Stephan
Cancel
Vote Up 0 Vote Down

Cancel
Tawna 9 months ago in reply to Stephan Bayerer

If you are simulating a high Q oscillator, use HB instead of Shooting PSS. You will need to increase the number of harmonics and set oversample to 2 or 4 if the oscillator is followed by circuitry that produces a square wave.

How to simulate a high Q (LC or crystal) oscillator followed by strongly nonlinear circuitry (e.g. hard switching buffers)

Also check out (for general best practices):

Getting the Most out of Spectre® X-RF 23.1 Maximizing Performance

best regards,

Tawna
Cancel
Vote Up 0 Vote Down

Cancel
Tawna 9 months ago in reply to Stephan Bayerer

If you are simulating a high Q oscillator, use HB instead of Shooting PSS. You will need to increase the number of harmonics and set oversample to 2 or 4 if the oscillator is followed by circuitry that produces a square wave.

How to simulate a high Q (LC or crystal) oscillator followed by strongly nonlinear circuitry (e.g. hard switching buffers)

Also check out (for general best practices):

Getting the Most out of Spectre® X-RF 23.1 Maximizing Performance

best regards,

Tawna
Cancel
Vote Up 0 Vote Down

Cancel

Stats

PSS Shooting - High Q crystal oscillator - Simulator by mistake detects a frequency divider

Community Guidelines