[Help] Gilbert Cell Mixer in RFIC

[cmosbjt]

Gilbert Cell Mixer is the most popular double balanced mixer used in RFIC, but it requires all the input and output to be differential, that means balun is needed if the RF input signal is single ended. My question are:

1. how to implement balun in IC?
2. Is In the real design of wireless LAN, GPS, Bluetooth tranceiver, is Gilbert Cell Mixer, or double balanced mixer popular? Is there other kind of mixer in use?

Thanks.

 

[Mazz]

There two way to implement baluns:
1. passive. Like inductors there are some coupled spirals, commonly on same metal layer, hard to design (electromagnetic simulator)
2. active. Of course in this case linearity can be a problem.

In BiCMOS the Gilbert cell is really popular and the most used. In CMOS both are used. The circuits have really different performances (the first one has positive gain, the second one has loss, to say only one).

I hope it can help.
Mazz

 

[vfone]

Below you can see how to implement a balun in RFIC:
http://www.ece.uic.edu/pdfs/transformer.pdf
Double balanced is common, but to improve linearity, one of the most used mixer in WLAN and Bluetooth is ring passive mixer (Folded Switching Mixer), using four FETs.

 

[dsjomo]

Connect a decoupling cap. between one of the input terminal and ground, add a tail current source beneath the differential pair.

 

[cmosbjt]

Connect a decoupling cap. between one of the input terminal and ground, add a tail current source beneath the differential pair.

Good idea, I've seen this method before (in P298, "High-Frequency Analog Integrated Circuit Design", Ravender Goyal). Thanks.

 

[dsjomo]

That is a practical way.
Plz see the following paper written by Osama Shana'a in 2001.
This might be the mixer you want.

 

[Puppet1]

look in figure 11 of that paper (gilbert cell).

what is the purpose of C1,C2 and why is one of the RF inputs shorted to ground ?

 

[dsjomo]

C1 : Bias voltage decoupling
C2 : RF virtual ground

Single -> Differentail

 

[Nonsense]

Good idea, I've seen this method before (in P298, "High-Frequency Analog Integrated Circuit Design", Ravender Goyal). Thanks.

Would you be so kind to share this book please? I cannot find it in this forum.

Thanks.

 

[cmosbjt]

Hoho, I only have a hard copy of this book. I don't know how to share it.

 

[vfone]

Hoho, I only have a hard copy of this book. I don't know how to share it.

Scan and upload it...

 

[cmosbjt]

Ok, i will try later, but need time.

 

[DonJ]

Hi

Double-balanced Gilbert mixers are used A LOT in commercial products. Please have a look at the publications on Bluetooth, GPS, WLAN, Zigbee etc. front-ends.
Usually, the LNA is differential: Differential input of mixer works fine.
The output of the mixer should also be differential to allow differential signal of the IF and baseband stages.

All the best

 

[cmosbjt]

But I have checked with some GPS receiver designs using Low IF architecture, they prefer single balence mixer, not Gillbert cell. Why?

 

[Mazz]

The explanation for first paper is in the paper itself. They have choosen a single end LNA (with a certain level of risk in term of noise coupling from other part of the IC), so the more natural choice for the mixer is the single bal. one.
This reduces current consumption and in today's GPS market this is a must. So to have a good NF with only 4 mA for all the Front End a differential structure is not applicable (we don't know a lot of details of course, but it seems reasonable).

Look how noisy is the output spectrum with BB on respect to BB off!

I hope it can help.
Mazz

 

[cmosbjt]

I don't think single-ended LNA output is a very sound reason why they choose single balence mixer. There are several samples of using single-ended LNA together with Gillbert cell mixer. They just connect one RF input of the Gillbert cell to ground through a cap, by this means, they convert single end signal to differential one. Please refer to "High-Frequency Analog Integrated Circuit Design", Ravender Goyal, Page 298. And also the paper posted by dsjomo in this topic (Frequency-Scalable SiGe Bipolar RF Front-End Design).

But I do think this single end to differential things is not balance enough.

What's your opinion?

 

[DonJ]

I agree in general with Mazz about the noise issue. You get much better noise immunity by using differential structure. For RF design, there are m a n y other reasons for using differential circuits.
One obvious penalty is power consumption. Power reduction could probably be the motivation in the GPS design you mentioned (?).

 

[cmosbjt]

I agree in general with Mazz about the noise issue. You get much better noise immunity by using differential structure. For RF design, there are m a n y other reasons for using differential circuits.
One obvious penalty is power consumption. Power reduction could probably be the motivation in the GPS design you mentioned (?).

Yes, Noise is an big issue. But differential structure can cancel the noise from "outside", that means noise from the power supply, baseband or sth like that, since the noises from each mosfet, even the match pair, are not correlated, they can not be canceled by using a differential structure, and on the contrary will introduce some more noise comparing with single-ended LNA. In my attached paper "A complete single-chip GPS receiver with 1.6-V 24-mW radio in 0.18um CMOS", the Sony guys apply the deep n-well to isolate the noise from baseband. In case when only an RF front-end IC is designed, this problem is relaxed a lot.

Another thing is that comparing the single balance mixer with double balance Gillbert cell, I don't think there is much difference in term of power consumption.

I still do not get a convincing explanation why they use single balance mixer in Low IF GPS RF front-end. Any idea about that?

Thanks.

 

[mrm]

Take a Look at Grebene's book (also on this site).

 

[Mazz]

At a certain point of the receiver path the structure can be made differential, but this will bring to more current consumption for interested stages. The balancing can be a problem, also.
Once they have the LNA "corrupted" by substrate noise, maybe is not really convenient make others block differential.
As I've said we know very little the design.
Furthermore, and this is a general consideration, take all informations contained in papers with a GRAET care. A lot of people making papers, lie, lie, lie...

I hope it can help.
Mazz