Transient noise noise_fmax vs. sampling rate

Dear Jere,

Jere Jarvinen said:

If I set the noise_fmax parameter to a higher frequency than my Fs/2, the noise levels increase in the spectrum and SNR decreases.

I am not sure what your fs value is - are you assuming it corresponds to the inverse of the strobeperiod?

Jere Jarvinen said:

So the question is, should the noise_fmax parameter always be set to Fs/2, when sampling the output? I use stobeonly-option to output only values defined by strobeperiod.

In any case,if your signal source from your 50 ohm port is sinusoidal in nature, then your entire circuit is basically linear and your output is bandlimited via your use of the RC circuit load element. There is a Rapid Adoption Kit covering this specific case with respect to transient noise. In essence, for a linear and bandlimited system, the value of noisefmax can be set as the system bandwidth. However, this may not provide sufficient simulation accuracy since the maximum timestep for a transient noise analysis is set by the value of noisefmax as 1/(2*noisefmax). Hence, as discussed in the RAK from the On-line support portal at URL:

https://support.cadence.com/apex/articleattachmentportal?id=a1O0V000009MqHrUAK&pageName=ArticleContent&attachId=0690V000007JLMXQA4&sq=null&caseSessionKey=null

you may need to increase simulator accuracy by setting a lower value of reltol than the default.

The concept of what one should choose for the value of noisefmax in a transient noise simulation, from my experience, is highly circuit and performance dependent. The subject is covered on pages 17 - 20 of the transient noise background document at URL:

https://support.cadence.com/apex/ArticleAttachmentPortal?id=a1O0V000007Mp6nUAC&pageName=ArticleContent

An exact "answer" is really not provided - but rather a number of considerations and a possible set of experiments to try to validate your choice. In the former RAK URL, a comparison is provided between the power spectral density of a transient noise simulation result and a more conventional AC noise simulation result that indicates their reasonable correlation for a linear, bandlimited circuit - such as yours - when using a noisefmax corresponding to the circuit's bandwidth and adjusting the reltol.

I hope this provides some thoughts and insights Jere!

Shawn

Hi Shawn,

Thank you for your response. Yes you are correct that I have Fs=1/strobeperiod, which I want to control depending on how wide spectrum I want to have. My test case is just a simple one to test this, but in reality we have a complex system that has also a sampling clock.

I have already looked up that RAK you posted link to, also this one:

support.cadence.com/.../ArticleAttachmentPortal

However, I was not able to find out direct connection between sample rate and noise Fmax.

In my testbench, I compare the transient noise result also to an AC-noise result. I can see that the accuracy matches with the table on page 19 in the RAK you pointed out.

If I look at the examples in both of these RAKs, it turns out that sample rate (1/strobeperiod) is always 2xnoise Fmax. For example on page 18 of the advanced RAK sampling settings have been done this way.

However, in any point has this been clearly mentioned. After thinking about this, it seems clear, since transient noise generates noise points up to a frequency given for the noise Fmax parameter. I assume those should be treated the same way as any signal without any harmonics to consider? Therefore, sampling rate should be 2x noise Fmax value to avoid aliasing. Lower noise Fmax value does not make sense either, since it only reduces accuracy and does not help with the simulation time when strobe is forcing shorter transient steps.

In my continuous-time test, If I keep noise Fmax = 2xFs, I can increase Fs so that I get high enough accuracy. Of course when there is sampling clock in the system, it determines the sampling frequency and should be high enough to fullfil all the requirements. Moreover, like in the RAK you pointed out, on page 46 is an example how signal harmonics fold back to the signal band if noise Fmax is not high enough. I have seen the same phenomena and if I increase Fs enough, the results stabilize at some point.

So do you agree, that noise Fmax should be set to a same (or lower) value than Fs/2?

Thanks,

Jere

Hi Shawn,

Thank you for your response. Yes you are correct that I have Fs=1/strobeperiod, which I want to control depending on how wide spectrum I want to have. My test case is just a simple one to test this, but in reality we have a complex system that has also a sampling clock.

I have already looked up that RAK you posted link to, also this one:

support.cadence.com/.../ArticleAttachmentPortal

However, I was not able to find out direct connection between sample rate and noise Fmax.

In my testbench, I compare the transient noise result also to an AC-noise result. I can see that the accuracy matches with the table on page 19 in the RAK you pointed out.

If I look at the examples in both of these RAKs, it turns out that sample rate (1/strobeperiod) is always 2xnoise Fmax. For example on page 18 of the advanced RAK sampling settings have been done this way.

However, in any point has this been clearly mentioned. After thinking about this, it seems clear, since transient noise generates noise points up to a frequency given for the noise Fmax parameter. I assume those should be treated the same way as any signal without any harmonics to consider? Therefore, sampling rate should be 2x noise Fmax value to avoid aliasing. Lower noise Fmax value does not make sense either, since it only reduces accuracy and does not help with the simulation time when strobe is forcing shorter transient steps.

In my continuous-time test, If I keep noise Fmax = 2xFs, I can increase Fs so that I get high enough accuracy. Of course when there is sampling clock in the system, it determines the sampling frequency and should be high enough to fullfil all the requirements. Moreover, like in the RAK you pointed out, on page 46 is an example how signal harmonics fold back to the signal band if noise Fmax is not high enough. I have seen the same phenomena and if I increase Fs enough, the results stabilize at some point.

So do you agree, that noise Fmax should be set to a same (or lower) value than Fs/2?

Thanks,

Jere