I am sorry to re-post it here from Measurement.

1. I had tried to characterize the jitter for driven logic gates (for insance, a simple inverter) due to intrinsic device noise. According to Ken Kundert's paper - An Introduction to Cyclostationary Noise, there are three approaches to do that (on page 38). The first approach needs to determine the instantaneous noise power at the time of threshold crossing time. It is written that "SpectreRF can do this", but I wonder what SpectreRF analysis can do this.

2. On the other hand, I had used 'pnoise' for this purpose, but 'pnoise' gives the time-average of the noise at the output of the circuit in the form of a spectral density versus frequency, according to the reference guide. I was wondering how to convert the resulting noise spectrum to actual time-domain jitter?

Thanks.

Hi Frank, thanks a lot for your help.  I will look into the thread to get started. But meanwhile, what is this 'pnoise jitter' analysis? I have IC5141 USR5 (released June 2007), I click on 'pnoise' but do not know how to have 'pnoise jitter' analysis? I really appreciate it.

thanks again, Frank.
I had a simple inverter circuit driven by a clock to generate an inverse clock, and I just want to see what is amount of jitter for the inverse clock due to device noise.

I used 'pss' analysis first. I do not know how to set either 'beat frequency' or 'beat period'. It seems by default it is the clock frequency or clock period.

After 'pss', I run 'pnoise' and choose 'noise type' as jitter as you suggested. I select the output as the inverse clock voltage output and input source as the 'vpulse' from analogLib. 'Sidebands' is set to 100 and 'reference sideband' is 0 since the inverse clock output has the same frequency as the input clock. One thing I am not sure is the 'sweeptype', what is the difference between absolute and relative?

However, in the compute power spectral density plot for the voltage output, the reported 'jittereventtime' (about 4ns) does not seem to be the correct transition time for inverse clock? I thought it should be 5ns for rising transitions based on the transient output. Maybe I either misunderstand something or I set something wrong for my simulation. Also, how do I then convert the PSD to time-domain jitter? Which ADE functions I should use? Thanks very much.

Frank, thanks for your continued help.

I have used the method your introduced in that thread for jitter computation. So, set noise type in 'pnoise' to 'timedomain', then complete the simulation. From Direct Plot, I select 'tdnoise', 'integ output noise', 'total noise' and 'magnitude', and was able to get the RMS noise power at the jitter event time. Dividing that by slew rate, I obtained the jitter amount, which seems very reasonable. However, one thing confuses me is the units of measure for 'magnitude' shown in Direct Plot. I thought it should be 'V', not 'V/sqrt(Hz)', because the slew rate is by default 'V/s' so that the division would give 's' for jitter amount. Why is it shown as 'V/sqrt(Hz)'?

Then, to confirm my jitter amout, I set noise type in 'pnoise' as 'jitter'. And I suppose that I should get the same jitter amount as before. Afer simulation, in Direct Plot, I select 'pnoise jitter', 'Jee' and it gives the jitter spectral density after clicking Plot. However, it did not give me the 'Jee' amount, as claimed in page 30 of the application note you gave me in last reply.  Also I checked 'Jc' or 'Jcc', clicking Plot did not give any value either. What may be the problem?
Thanks.

Frank, again thanks are due to you.

I got the ‘Jee, Jc, Jcc’ as mentioned in my previous reply. Somehow, they show in my AWD plot, but not in WaveScan.

Assuming the output of a driven circuit is a periodic clock-like signal. But due to device noise, there will be variations for this periodic signal. ‘Jee’ is edge-to-edge jitter, so it characterizes the jitter amount for a single edge. But the ‘Jee’ could be different for a rising or a falling edge due to different noise injection at those moments or even different slew rates, right?

‘Jc’ is the period jitter and a period is defined as the time duration of two consecutive rising/falling edges. But for the time duration defined by the rising edge following by the closest falling edge, (for example, this time duration is almost X% of the period if the ideal periodic clock-like output signal is designed to have X% duty cycle), how to calculate its jitter in ‘pnoise’? I think I can not simply find the ‘Jee’ for both edges and then take RMS of them, because they are correlated. In fact, I feel that this question is very similar to asking how ‘Jc’ is computed in ‘pnoise’ when there is correlation between two consecutive jittered rising/falling edges?
Please correct me when I am wrong. thanks.

Thanks, Frank. I guess that the calculator expression you gave me last time is for the so called 'K-cycle jitter' (K=3) (defined on page 12 of the application note)?

I was also thinking that if the output signal is NOT periodic, then can we apply ‘pss’ and ‘pnoise’? For example, I have a driven circuit which is a 2-input NAND gate and the two inputs are non-periodic, so that the output is not periodic either, can we still apply ‘pss’ and ‘pnoise’ for this circuit? My guess is possibly yes, because I feel the essence of ‘pss’ is not exactly a periodic time-varying operating point, but a finite number of operating points (such as the low-high transition, high-low transition, high state and low state)? Or this case is where ‘qpss’ and ‘qpnoise’ would kick in?

Further, if the output signal is non-periodic for a general driven circuit, then I guess only ‘Jee’ would make sense? What do you think? Thanks again.

thanks for your reply, Frank. I think I know how to calculate time jitter due to intrinsic device noise. One more question I have is that how to simulate the time jitter caused by power supply noise and substrate noise? I looked into the user guide and I am under the impression that I need to model these noise as small signal noise and run 'pxf' analysis in SprectreRF?
Also, how to measure power supply reject ratio for a general deriven logic circuit?

Are there any other noise source causing jitter?

thanks for your patience for helping me through this.

thank you, Frank.
I looked into the thread and can understand the idea. I also found an application note at sourcelink which taught how to measure PSRR, but that 'modulated pxf' is used instead of 'sampled pxf' analysis. So, still 'pss' first and then 'pxf'? what is the sweep type in 'pxf'? I seem to get a plot of the gain transfer function of the output with respect to the each supply voltage source in my driven logic circuit (in Results Browser), but how to transform it to the time-domain jitter? Still use 'pnoise' to do that? Your brief description is highly appreciated.