Designer’s Guide Community

AK

New Member

Offline

Posts: 2

Noise Fmin in the Transient Noise Analysis settings
May 16^th, 2013, 2:44pm

In Spectre, what is the significance of Noise Fmin in the Transient Noise Analysis settings?

I have a test circuit of a Resistor and I am analysing its White Noise. I expected the noise to be limited to frequencies between Noise Fmin and Noise Fmax.
When I do the Transient Noise Analysis with different Noise Fmin (say 10K and 800M) and a Noise Fmax of 1GHz, I do not see any difference in the Spectral Components (DFT) of resulting noise and PSD plot. For both the cases of Noise Fmin, DFT components exist from dc upto Noise Fmax. So what may be the error.

In addition, which fiter (type and order) is modeled in Transient Noise Analysis to limit the White noise to Noise Fmax?

Back to top

IP Logged

RFICDUDE

Community Fellow

Offline

Posts: 323

Re: Noise Fmin in the Transient Noise Analysis settings
Reply #1 - May 16^th, 2013, 6:17pm

Interesting question. I was not aware of these options for transient noise simulation, but after reading the Spectre help information I see that it is important to know what these parameters do.

noisefmax and noisefmin control the spectral density of all the transient noise sources in the circuit.

Below noisefmin the spectral density of all noise sources is forced to be flat. By default only thermal noise (flat spectral density), so effectively noisefmax=noisefmin. This is ok if you are not concerned with flicker noise (shaped noise source) sources.

If you are concerned about the impact of frequency shaped noise sources then noisefmin needs to be set to the lowest frequency of interest for considering the shaped noise source. Also, the simulation stop time has to be long enough to resolve this minimum frequency for the noise sources (i.e. tstop>=1/noisefmin).

I have not attempted to use transient noise to look at flicker noise sources, but if I need to in the future I'll now know what to set in the simulator.

noisefmax will impact the timestep to accurately solve for the noise at noisefmax. The spectral density of the result may not be accurate above noise fmax. However, setting noisefmax too high may greatly increase simulation time since (I think) it is like putting a sinusoidal source with frequency noisefmax into the simulation.

Back to top

IP Logged

sheldon Community Fellow Offline Posts: 751	Re: Noise Fmin in the Transient Noise Analysis settings Reply #2 - May 17^th, 2013, 6:45am Noisefmin=0 is white noise only Noisefmin=XXX defines the minimum frequency of the noise, this is how you include flicker noise in a simulation If your model does not include flicker noise, then of course even if you specify noisefmin, there will be no flicker noise
Back to top	IP Logged

AK

New Member

Offline

Posts: 2

Re: Noise Fmin in the Transient Noise Analysis settings
Reply #3 - May 17^th, 2013, 12:46pm

Many thanks for the reply.

************************************************************
@sheldon: I am not sure if Noisefmin=0 implies white noise only. I tried to simulate and Noisefmin=0 resulted in error.

************************************************************

The problem I still have is that, suppose I have a 1k resistance from analoglib and simulate it for thermal noise (4kTR) only (flicker noise not modeled) by transient noise analysis for noisefmax= 1G, I should be able to see flat PSD=16.56*10^-18 V²/Hz @T=300K upto 1G. But I see the above mentioned PSD for only fraction of 1G bandwidth with roll off starting at about 150 MHz.

For Noisefmax=1GHz, I think the simulator filters off noise above 1GHz. So is it sharp roll off at 1GHz or roll off starts well before 1Ghz (depending on Filter type and order)??

Back to top

IP Logged

ywguo

Community Fellow

Offline

Posts: 943
Shanghai, PRC

Re: Noise Fmin in the Transient Noise Analysis settings
Reply #4 - Jun 22^nd, 2013, 2:36am

Hi AK,

It is not surprising that PSD rolls off at about 150MHz because the noise sources is injected in every nodes/branches as ideal Sample and Hold. Strictly, it is first-order sample and hold. So its PSD must be shaped by sinc function. Do you have Cadence application note Application Notes on Direct Time-Domain Noise Analysis using Virtuoso Spectre?

If you do NOT have that application note, please google the following Ph.D disseration.
Low-Phase-Noise, Low-Timing-Jitter Design Techniques for Delay Cell Based VCOs and Frequency Synthesizers
Todd C. Weigandt [1998]

He used similar method to investigate the VCO jitter. I recall that the principle is explained in the appendix.

Best Regards,
Yawei

Back to top

IP Logged