The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> pss+pnoise: cascaded system with 2 freq translations
https://designers-guide.org/forum/YaBB.pl?num=1409840321

Message started by mixed_signal on Sep 4^th, 2014, 7:18am

Title: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Sep 4^th, 2014, 7:18am

Hi,

I have done pss+pnoise analysis and found noise gain of drain noise of M1 to Vout1 and Vout2 from sideband = +1.

Does pss+pnoise do a second frequency translation using -LO while finding the gain from M1 to Vout2? What does sideband=+1 mean at the second output

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by Ken Kundert on Sep 6^th, 2014, 11:55pm

I cannot understand your question.

-Ken

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Sep 22^nd, 2014, 7:47pm

Hi Ken,
Sorry for the late reply.
The system has 2 mixers in cascade, 1st driven by +LO and 2nd driven by –LO. The input noise profile of a MOS device in the system is shown. I want to know how much noise at fRF (red) of M1 appears at the output at fIF where fIF=fRF-fLO.

I tried to find it out using pss+pnoise with noise separation by following these steps in ADE
Results>Direct plot> noise separation> source noise gain>sideband=+1>select device M1>plot

Is my procedure correct?
There are two frequency translations at X and Y. If I choose sideband=+1, will everything be taken care of?

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by Ken Kundert on Sep 22^nd, 2014, 8:25pm

Still having trouble understanding your question. Are you trying to determine how to compute the output noise of a chopper stabilized amplifier?

-Ken

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Sep 22^nd, 2014, 9:30pm

Hi Ken,

It is not a chopper stabilized amplifier but a circuit under research, some sort of mixer with differential LO.
If the scenerio is as depicted as in the figure, are the simulation steps correct to find the noise?

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by Ken Kundert on Sep 23^rd, 2014, 4:45am

Okay, well you have drawn a chopper stabilized amplifier. So, it seems that this circuit can be thought of as a chopper stabilized amplifier where the input signal is at the first upper sideband of the clock. This is the same as a down-conversion mixer and would be simulated in the same way.

-Ken

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Sep 23^rd, 2014, 8:10am

Hi Ken,
Thank you very much for the reply. One more question.

Please check the attached single gate nonlinearity based mixer. The LO frequency is 2.4 GHz.

I have simulated noise transfer function of thermal & flicker noise of M1 from RF to IF using pss+pnoise and noise separation. (sideband=+1). Why are they different? I believe both thermal and flicker are modeled as current source and hence transfer function should be same. The transistor model is BSIM 4.5 and noise model is holistic. Is it because contribution from noise source modulation is different for them?

I tried modeling the noise using a current source (red) and use PXF analysis. But the gain is around 150 in contrast with 1.2K from pnoise simulation. Why is pxf resulting in such low value? Is this method of modeling noise faulty?

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Sep 23^rd, 2014, 8:12am

Simulations are here

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by Ken Kundert on Sep 23^rd, 2014, 8:33pm

There are two important differences between the thermal (channel) noise and flicker noise produced by a MOSFET. Of course, flicker noise as a 1/f characteristic whereas the channel noise is white. The second is that the thermal noise is modeled as coming from a voltage source in series with the gate whereas the channel noise is modeled as coming from a current source across the channel. Thus, the flicker noise is being modulated by a possibly time varying gm.

-Ken

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Sep 23^rd, 2014, 8:57pm

Hi Ken,
Thanks again for the reply. I have still some doubts:
1. Even if channel/thermal noise is modeled as a current source, it is expressed as 4kTgm and hence modulated gm will also affect it in the same way as flicker.

2. The 1/f should not be an issue because I plotted noise transfer function and not the noise. The transfer function should be independent of input noise profile.

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by Ken Kundert on Sep 23^rd, 2014, 10:31pm

The gm term in the channel noise is considered to be part of the noise source and so is not included in the transfer function. The gm term in the flicker noise is consider part of the circuit not the noise source, and so is included in the transfer function.

-Ken

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Sep 24^th, 2014, 8:19pm

Hi Ken,

But flicker is modeled as current source in my kit that uses BSIM 4.5. I have attached the equations from BSIM 4.5 documentation. Also spectre plots transfer function in voltage/ampere units which indicates current noise model fro flicker.

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by Ken Kundert on Sep 25^th, 2014, 8:30pm

I believe SpectreRF treats the gm term in the channel noise as part of the noise source and so is not included in the transfer function. The gm term in the flicker noise is treated as part of the circuit not the noise source, and so is included in the transfer function. That suggests that the two transfer functions differ by a factor of gm. Is that what you are seeing?

-Ken

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Sep 26^th, 2014, 6:16pm

Hi Ken,

Thanks again for your reply.
The gm is only 1.15mS. In contrast the ratio is 1.13 which is 1000 times more. I have thoroughly checked both the units (V/A).

I am more concerned with the failure of PXF analysis in determining the noise gain (150 in contrast with 1000). I have tried doing both pnoise and pxf before to find noise transfer function from resistor thermal noise to different circuit nodes and they always agree.

Is modeling noise by external current source not correct?

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by Ken Kundert on Sep 27^th, 2014, 4:14am

Hmmm. Okay, I am starting to see some of the subtlety. The simplest model for flicker noise is a stationary noise voltage in the gate, so I think of it that way, but the simulator does not actually put a voltage source in the gate because that would require extra nodes, which would make the simulation more expensive. However, the simulator can still compute the transfer function from the gate if it wanted to by dividing through by gm, but it seems like SpectreRF is not doing that. If it were, then the units should be V/V rather than V/A.

Having said that, the units given in your screen capture are very confusing to me. What is the meaning of :sqrt(Hz). A transfer function is not a noise and so should the units should not include :sqrt(Hz). I also do not know how to interpret the colon.

Now, as to your question as to why the transfer function for channel noise and flicker noise are different, I cannot say. You might ask Cadence about that.

Also, if the gm term is not being included in the flicker noise transfer function, then it does seem as if the PXF and PNoise transfer functions should be exactly the same. I cannot explain why they would be different. Perhaps there is something in the documentation that describes it. After all, there must be some reason why they provide PNoise separations as opposed to just advising you to use PXF.

-Ken

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Sep 27^th, 2014, 8:04am

Hi Ken,
I shall go through the manual in detail and try to figure out the issue.

Can you suggest me papers which analytically discuss how noise source modulation leads to cyclostationarity?
I want to quantitatively distinguish cyclostationary noise contributions from
1. Noise source modulation
2. Noise transfer function modulation

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by Ken Kundert on Sep 29^th, 2014, 5:19am

I do not have a good reference that covers cyclostationary noise analytically (actually there are the Gardner books on cyclostationary noise, but those are rather heavy). I have written something that presents cyclostationary noise from a more conceptual level (see Noise in mixers, oscillators, samplers, and logic: an introduction to cyclostationary noise). But fundamentally you cannot tell from the output whether a noise contribution to an output is cyclostationary because the underlying source is cyclostationary or the path from the source to the output periodically modulates the noise signal.

Have you approached Cadence and asked them for help?

-Ken

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Sep 29^th, 2014, 7:16pm

Hi Ken,

Thank you very much for the link! Highly appreciate it. Yes I have approached them for help and shall update once I hear from them.

I have facing difficulty in understanding the attached paragraph from page 5 of your paper. What is meant by "analyzer has small effective input bandwidth"? Do you mean the scanning bandwidth of the analyzer which is swept across the spectrum?

Why would noise modulation not show up?

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by Ken Kundert on Sep 29^th, 2014, 11:17pm

A spectrum analyzer effectively converts the signal of interest down to baseband and then passes it through a lowpass filter. Generally, we think of spot noise as the total power of a signal within a 1Hz bandwidth centered about the frequency of interest. In this case, the bandwidth of the lowpass filter will be 1Hz. Response time with a 1Hz filter is really slow, so the spectrum analyzer will allow you to specify the bandwidth of this filter. But lets assume that the BW of the spectrum analyzer is set to 1Hz.

Then if the modulation frequency is less than 1Hz, you will see the noise change on the screen at the modulation frequency. If instead, the modulation frequency is greater than 1Hz, as it usually is, then you will not see the noise change on the screen because it is changing too fast. Instead, you will see the time-average noise.

-Ken

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by mixed_signal on Nov 14^th, 2014, 12:04am

Hi Ken,
I am in touch with Cadence folks and shall update when I get satisfactory explanation.

1. By the way, I can have access to a modern FFT based spectrum analyzer with fast detectors and I can get a real time snapshot of the spectrum unlike traditional sweeping ones. Do you think I shall be able to visualize cyclostationarity in noise?

2. If cyclostationary noise is input to 2 sharp bandpass filters (1Hz BW) centered at f1 and f1+fLO and instantaneous output is plotted in time domain then what shall I get? I know magnitude and phases of the two components will be correlated. Will they be sines with amplitudes proportionally varying with time? How about their phase difference? Phase correlation is difficult to visualize since the phasors have different speeds.

Title: Re: pss+pnoise: cascaded system with 2 freq translations
Post by Ken Kundert on Nov 14^th, 2014, 6:00am

For a spectrum analyzer to see the cyclostationary nature of noise it must be synchronized to the modulation signal. Without that synchronization, you will always see some form of averaging.

-Ken