Noise Figure of LNA with an input tone (using hb/hbnoise)

There's no need to apply a large signal input for the RF signal when computing the noise, unless you are trying to investigate the effect of a large-signal blocker on the noise response (e.g. if the blocker is driving the circuit into saturation, or you want to look at how the noise intermodulates with the large signal applied). In this case because your RF input is very small, there's no need to have the signal.

What is happening is that the flicker noise at DC (which would be infinite) mixes with the RF input frequency resulting in a very large signal at that frequency. You'd see the same at any multiple of any of the large signal frequencies in the circuit.

Regards.

Andrew.

Hi Andrew,

Thanks for the response.

Exactly, my intention was to check the Noise Figure of LNA in the presence of a large-signal blocker. 

I agree with you about the mixing of flicker noise and RF input frequency.

But, if you see the document in SpectreRF Workshop, LNA Design Using SpectreRF (MMSIM 13.1, September 2013), Lab2, pages 29-35 where the same simulation is illustrated, the noise figure result does not show this effect. Any ideas why is not showing up there? Is it related to the number of steps in sweeping ?

Regards

Yes, you'll notice the the step in the pnoise sweep misses the 2.4GHz point - there's one at 2.125GHz and one at 2.5GHz. Far enough away from the peak at 2.4GHz to see the peak. That said, using a 0.2GHz step also doesn't show the peak because spectre shows:

Warning from spectre at freq = 2.4 GHz during HBNOISE analysis `hbnoise'.
WARNING (SPCRTRF-15037): Infinite flicker noise is ignored.

I don't get it even if I take a very fine sweep near the frequency point (I'm using a much newer spectre though, so maybe that's why it's really omitting it; or the models I have in the workshop may not really model flicker noise). 

I'm sure it's the reason though...

Andrew.

Thanks Andrew,

So I assume, I can ommit this effect by selecting in hbnoise all the sidebands except the one that folds the flicker noise inside my band of interest.

So, the setup of hbnoise is like that:

Regards

Yes, but then  you're omitting all noise contributions from baseband which get unconverted. Not sure that's necessarily wise. You could just step over the problematic point...

I am a bit short of time today otherwise I'd have tried doing experiments with different versions or trying to get to the bottom of why it doesn't show up with the workshop LNA if I choose a sweep which gets close to the RF input.

Andrew.

Sure. Okay, Iif you find time to check that it will be really helpful. Thanks!

Hi,

Any progress on the above issue?

Regards

No. To be honest I wasn't planning to as I had really explained the root cause - figuring out why it was not showing with the LNA in the workshop was rather secondary. Almost certainly it's something to do with a small numerical rounding error meaning your point is not exactly at the beat frequency but extremely close. With the workshop LNA I get this if I do a zoomed in sweep:

In my case the exact point has a dip (because of the omitted infinite flicker noise), but you'll see the noise figure is very peaky either side.

Omitting the -1 sideband removes this peakiness. The flicker noise in the models I'm using drops quite quickly and ceases to be a big contributor (unsurprisingly) at these frequencies, but a small error doesn't make much sense.

Either way, hitting the exact input frequency doesn't make a great deal of sense anyway.

If this had been urgent (not sure why, because I'd explained the phenomenon - perhaps you didn't believe me?) you should have gone to customer support - I do this in my (fairly limited) spare time, and I'd already stated that I was rather busy...

Regards,

Andrew.

No. To be honest I wasn't planning to as I had really explained the root cause - figuring out why it was not showing with the LNA in the workshop was rather secondary. Almost certainly it's something to do with a small numerical rounding error meaning your point is not exactly at the beat frequency but extremely close. With the workshop LNA I get this if I do a zoomed in sweep:

In my case the exact point has a dip (because of the omitted infinite flicker noise), but you'll see the noise figure is very peaky either side.

Omitting the -1 sideband removes this peakiness. The flicker noise in the models I'm using drops quite quickly and ceases to be a big contributor (unsurprisingly) at these frequencies, but a small error doesn't make much sense.

Either way, hitting the exact input frequency doesn't make a great deal of sense anyway.

If this had been urgent (not sure why, because I'd explained the phenomenon - perhaps you didn't believe me?) you should have gone to customer support - I do this in my (fairly limited) spare time, and I'd already stated that I was rather busy...

Regards,

Andrew.

Hi Andrew, sorry for the misunderstanding. I thought that you will have a look on that and post your findings.

Anyway, thanks for finding time to answer.

Regards