The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> The physical meaning of spike (Pnoise analysis)
https://designers-guide.org/forum/YaBB.pl?num=1185212271

Message started by JC on Jul 23^rd, 2007, 10:37am

Title: The physical meaning of spike (Pnoise analysis)
Post by JC on Jul 23^rd, 2007, 10:37am

When I simulate my oscillator by pnoise analysis (noiseType=sources) and try to investigate the output noise current by a probe as shown in the following Fig. I find there is a spike at the oscillating frequency in the output noise current. What's the reason for that spike? Is there any physical meaning or just due to some effects of the simulator.

As to the oscillator shown, bias current is from an ideal current source; the behavior of transistors are describled by Verilog-A without capacitive and noise effect; and I use a VCCS to model the thermal noise current with PSD 4kTR, which is the only source of noise.

In my understanding the output noise in my case is cyclostationary, and it is also a white noise with a flat PSD.

-JC

Title: Re: The physical meaning of spike (Pnoise analysis
Post by sheldon on Jul 25^th, 2007, 9:25pm

JC,

First question, the units of noise you are simulating are less than 1e-24 A^2/Hz,
are the simulator tolerances set appropriately for this accuracy? Next question,
what frequency does the oscillator oscillate at? Does it oscillate at the frequency
of the spike? Since there are still sources of noise in the simulation, Rload, could
the spike be due to insufficient frequency resolution? If you just plot the spike
does it look more like the expected phase noise plot of the oscillator?

Best Regards,

Sheldon

Title: Re: The physical meaning of spike (Pnoise analysis
Post by JC on Jul 26^th, 2007, 10:36am

sheldon,

Thanks for your reply.
1. My tolerance settings are: errpreset = "conservative" ; iabstol = "1e-13"; vabstol = "1e-7"; reltol = "1e-5". These settings are recommended by some people, but comparing with the result value it seems they are not enough.
2.The center of spike resides at the oscillating frequency. And the shape shows a little similarity with phase noise plot.
Since I use the ideal components, which means they don't generate noise e.g. for resistor with option "isnoisy=no", and the FETs use behavior model wrote by verilog-a without noise and capacitive effects, the only noise source is implemented by a vccs controled by a noisy resistor as shown in the Fig.
I wander it maybe caused by the positive feedback of OSC and the V->I conversion of FETs, which converts the noisy voltage near the oscillation frequency to noisy current with a gain large than 1. So the amplitude of that noise current PSD is large than that of the only noisy source. Is it correct?

-JC

Title: Re: The physical meaning of spike (Pnoise analysis
Post by Ken Kundert on Jul 26^th, 2007, 3:13pm

You should always start with using the default tolerances. The tolerances you are using are most likely way too tight. I have almost never found it necessary to use such tight tolerances.

The spike you are seeing is the phase noise of the oscillator. You'll get such a peak at each harmonic of the oscillator. The fact that you do not see the peaks at other frequencies is probably because the sampling is very course.

-Ken

Title: Re: The physical meaning of spike (Pnoise analysis
Post by JC on Jul 30^th, 2007, 6:01am

Ken Kundert wrote on Jul 26^th, 2007, 3:13pm:

..........
The spike you are seeing is the phase noise of the oscillator. You'll get such a peak at each harmonic of the oscillator. The fact that you do not see the peaks at other frequencies is probably because the sampling is very course.
...........
-Ken

Thanks for reply.
Yes, I think the spike should be the phase noise. And I still have some doubts about the procedure of the conversion from noise current source with flat PSD at tail to output noise current with that shap.

-JC

Title: Re: The physical meaning of spike (Pnoise analysis
Post by Ken Kundert on Jul 30^th, 2007, 11:01am

The noise is proportional to 1/Δf where Δf = f - f_o. It looks unusual because you are looking at in terms of absolute frequencies. If you plotted it versus relative frequency (Δf) in a log-log fashion, it would look more familiar.

-Ken