Designer’s Guide Community

sandman

Community Member

Offline

Posts: 73

White noise in a current commutating passive mixer
Oct 17^th, 2012, 3:23pm

Hi all,

I recently came across a paper which suggested that the the FET switches in a current-commutating passive-mixer contribute noise ONLY during an overlap of the non-ideal complementary LO signals (i.e., LO signals which have finite transition time and momentarily turn 'on' complementary switch paths) that drive the switches.

To my understanding the switches contribute white noise to the RF port (output) whenver they turn on (i.e., whenever the channel exists), irrespective of the flow of RF current, and irrespective of whether the other switch is On or Off. The assumption in the paper was that the passive-mixer switches act as cascode stages to the transconductance stage and therefore do not contribute any noise.

This, to me seemed wrong on two accounts. This degeneration of passive mixer switches due to a transconductance stage that does not share any DC current with the cascode (Mixer) stage, did seem right to me. Is there any degeneration here?

Secondly, degeneration does not reduce noise, it only reduces the gain for input referred noise.

If my understanding is incorrect could anyone explain why one switch of a current-commutating passive mixer does not contribute channel-resistance (white) noise to the RF output when the other is the Off state ?

The paper I refer to is "Noise in Current-Commutating Passive FET Mixers" - Saeed Chehrazi, Ahmad Mirzaei and Asad A. Abidi, IEEE Transactions, February 2010.

I am experimenting with a mixer in a up-conversion from non-zero IF. While I haven't had a chance to experiment with this in a simulator, I was hoping to get some comments here before I got started! Any inputs are welcome.

Thanks and cheers !

Back to top

IP Logged

RFICDUDE

Community Fellow

Offline

Posts: 323

Re: White noise in a current commutating passive mixer
Reply #1 - Oct 17^th, 2012, 5:50pm

I briefly looked at the paper.

The only reference to degeneration is regarding a common gate current buffer that loads the current mode passive mixer at base-band. There is some discussion about the flicker noise of this baseband buffer, but not much on the degeneration aspect you are concerned about.

My experience is that the thermal noise of the "on" switch does contribute, but no more than any other thermal noise contributor.

Back to top

IP Logged

sandman

Community Member

Offline

Posts: 73

Re: White noise in a current commutating passive mixer
Reply #2 - Oct 18^th, 2012, 12:26am

Thanks RFICDUDE.

There's a section in the paper (IV.B Switch Noise), which has the following text. "The switches contribute noise
to the mixer output during the overlap time when they are both ON. If one switch is OFF, it obviously contributes no noise, and neither does the other switch that is ON because its current is fixed by the RF input transconductance stage."

Thermal noise doesn't depend presence or absence of current flowing through the device. I take the section above to mean cascoding. Cascoding reduces output current noise of the cascode by degeneration (by ro of the previous/lower stage).

Any thoughts? I am still not sure why one switch of a current-commutating passive mixer does not contribute channel-resistance (white) noise to the RF output when the other is the Off state ? Is the current-driven passive mixer in cascode with the transconductance stage?

RFICDUDE wrote on Oct 17^th, 2012, 5:50pm:

My experience is that the thermal noise of the "on" switch does contribute, but no more than any other thermal noise contributor.

Back to top

IP Logged

RFICDUDE

Community Fellow

Offline

Posts: 323

Re: White noise in a current commutating passive mixer
Reply #3 - Oct 20^th, 2012, 6:46am

Let's not think of the switch as a "cascode" device. It is just a low value resistor when it is in the ON state.

The switch resistor is connected to a high impedance node (gm output) and a low impedance node on the other side. For any resistor, thermal noise current cannot flow out of the terminals if one end is open. However, there is a thermal noise voltage that is, of course, equal to the noise voltage of the resistor. This might look like the noise current is cancelled, but really its just physics (if one end of the current source is open then there can't be current flow).

In reality the gm side is not open because it has some resistance (1/gds). But if the output resistance is high compared to the switch resistance, it will limit the noise current of the switch (i.e. most of the thermal noise voltage of the switch is dropped across the output resistance of the gm device).

A cascode device, being a three terminal device, has a feedback mechanism that acts to cancel the noise in the channel regardless of the load on the drain. Of course, as you mention, the resistance at the source of the CG device has a big impact on the feedback that acts to cancel the channel noise.

Hope this is helpful.

Back to top

ResistorNoise.png

IP Logged

nrk1

Community Member

Offline

Posts: 81

Re: White noise in a current commutating passive mixer
Reply #4 - Oct 20^th, 2012, 8:47pm

sandman wrote on Oct 17^th, 2012, 3:23pm:

Hi all,

I recently came across a paper which suggested that the the FET switches in a current-commutating passive-mixer contribute noise ONLY during an overlap of the non-ideal complementary LO signals (i.e., LO signals which have finite transition time and momentarily turn 'on' complementary switch paths) that drive the switches.

To my understanding the switches contribute white noise to the RF port (output) whenver they turn on (i.e., whenever the channel exists), irrespective of the flow of RF current, and irrespective of whether the other switch is On or Off. The assumption in the paper was that the passive-mixer switches act as cascode stages to the transconductance stage and therefore do not contribute any noise.

I think the paper is only saying that the noise from a resistor (switch) placed in series with a current source(gm) doesn't affect the current coming out on the other (load-usu. virtual short) side. This is correct. If a passive switch mixer is driven by a voltage source, the switch noise will contribute to the output whenever it is on.

Back to top

IP Logged

sandman

Community Member

Offline

Posts: 73

Re: White noise in a current commutating passive mixer
Reply #5 - Oct 22^nd, 2012, 2:18pm

nrk1 wrote on Oct 20^th, 2012, 8:47pm:

I think the paper is only saying that the noise from a resistor (switch) placed in series with a current source(gm) doesn't affect the current coming out on the other (load-usu. virtual short) side. This is correct. If a passive switch mixer is driven by a voltage source, the switch noise will contribute to the output whenever it is on.

Hi nrk1,

Thanks for your thoughts. Although I'm not sure I fully agree with interpreting the statement in the paper. Just to make things clear, since we're talking of a passive current-driven mixer, the output from the mixer is a current. Measuring voltage across a low-load makes little sense.

Thermal noise across the switch is a current source which contributes as 4kT2gm/3. This is the intrinsic noise of the switch which will reach the output if the preferred path is towards the output (low impedance). To add, cascoding reduces the gain, but the cascode stage (switch) noise still goes to the output. It only affects the SNR lesser.

If the switch is driven by a voltage source at Source/Drain, then the voltage source presents a low impedance but the series resistance of the voltage source is high. If it is low, it makes little design sense as the mixer's noise will then dominate. This is simple analysis.

Thanks RFICDUDE, I think you are referring to a current divider ? I've made the same analysis before, but, unless I misunderstood all the comments, I'm still stumped by ".. neither does the other switch that is ON because its current is fixed by the RF input transconductance stage". The switch is driven by a current source (current-driven mixer). The switch noise current, when ON, flows to the low impedance Mixer output load. This should not matter whether the other switch is ON or OFF.

If I'm wrong, I'd appreciate your patience in explaining in some more detail !

Back to top

IP Logged

RFICDUDE

Community Fellow

Offline

Posts: 323

Re: White noise in a current commutating passive mixer
Reply #6 - Oct 22^nd, 2012, 5:43pm

I can only guess that the author meant that the noise current is dominated by the transconductance stage. Both the gm stage and the switch contribute, but as I said the ON switch noise current into the transresistance load is small.

Yes, essentially there is a current divider such that the noise current, due to Rsw, that flows into the transresistance load is the same as the current flowing through 1/gds. There is, essentially, a current divider between Ron of the switch and 1/gds from the transconductance stage.

However, nearly all of the drain current noise of the gm stage flows through the switch and into the transresistance amplifier.

Back to top

IP Logged

nrk1

Community Member

Offline

Posts: 81

Re: White noise in a current commutating passive mixer
Reply #7 - Oct 23^rd, 2012, 8:14pm

sandman wrote on Oct 22^nd, 2012, 2:18pm:

... �the output from the mixer is a current. Measuring voltage across a low-load makes little sense.

I didn't say you measure the voltage across the low-Z load. You measure the current through the low-Z load(which is what effectively the transimpedance amplifier is presenting to the mixer).

sandman wrote on Oct 22^nd, 2012, 2:18pm:

Thermal noise across the switch is a current source which contributes as 4kT2gm/3. This is the intrinsic noise of the switch which will reach the output if the preferred path is towards the output (low impedance). To add, cascoding reduces the gain, but the cascode stage (switch) noise still goes to the output. It only affects the SNR lesser.

I wouldn't call this a cascode as the switch(in triode region) has a very low gm. Imagine an ideal transconductor(infinite Rout) with a resistance in series with the output. The output current will still be identical to gm*vin regardless of the resistor's noise. The resistor's noise will only change the voltage waveform across the transconductor's output. As RFICDUDE pointed out, if the transconductor has a finite Rout, there will be small noise current due to the switch. BTW, a MOS in triode has a noise spectral density of 4kT*gds(This gds is measured in deep triode, and with square law, equals the gm in saturation, i.e. mu*Cox*(VGS-VT)) . There is no factor of 2/3.

sandman wrote on Oct 22^nd, 2012, 2:18pm:

".. neither does the other switch that is ON because its current is fixed by the RF input transconductance stage". The switch is driven by a current source (current-driven mixer). The switch noise current, when ON, flows to the low impedance Mixer output load.

This is not correct. The switch noise current will not flow into the low-Z load if the impedance on the driving side(transconductor) is very high.

Back to top

IP Logged

sandman

Community Member

Offline

Posts: 73

Re: White noise in a current commutating passive mixer
Reply #8 - Oct 25^th, 2012, 6:52am

nrk1 wrote on Oct 23^rd, 2012, 8:14pm:

BTW, a MOS in triode has a noise spectral density of 4kT*gds(This gds is measured in deep triode, and with square law, equals the gm in saturation, i.e. mu*Cox*(VGS-VT)) . There is no factor of 2/3.

Thanks nrk1, I correct myself, I suppose it is rather 4kT(gm+gds) in triode, with the gm being rather small.

nrk1 wrote on Oct 23^rd, 2012, 8:14pm:

sandman wrote on Oct 22^nd, 2012, 2:18pm:

".. neither does the other switch that is ON because its current is fixed by the RF input transconductance stage". The switch is driven by a current source (current-driven mixer). The switch noise current, when ON, flows to the low impedance Mixer output load.

This is not correct. The switch noise current will not flow into the low-Z load if the impedance on the driving side(transconductor) is very high.

Perhaps what you say is not completely true. There is a current divide at the output of the transconductor, much like what RFICDUDE also pointed out. If the intention is to use it in current mode, the load resistance in parallel with ro of the gm stage is typically small as no voltage gain is necessary here. I am referring to the load that connects it to the supply.

This prevents large voltage fluctuations at the input of the Mixer. (Furthermore, irrespective of upconversion or downconversion, the 'mixer' is AC coupled to the gm stage, in that there is no DC current from gm to output_load.)

This is not as small as the Ron (1/gds) of the Mixer-switch, but is definitely smaller than the ro of the gm. The current gain (if you will) of the switch noise current to the output load is

Iout,load Rswitch
---------- = -------------------------------------------
Iswitch Rswitch + ro//Rload_gm + Rout,load

My apologies for the poor illustration, but, you can see, this depends on the ratio between Rload_gm (the load of the gm stage) and the Rswitch (Ron of the switch in triode), although, this is actually Rswitch(t), assuming Rout,load is rather small!

As for the 'cascoding', I agree, that was only an abstraction, I guess it makes things a bit more complicated! I was only trying to figure out the assumption being the following statement in the paper for which it would be true ".. neither does the other switch that is ON because its current is fixed by the RF input transconductance stage"... which was actually my question.

Thanks for all your comments. Much appreciated!

Back to top

IP Logged