How can I improve the gain of CMOS Gilbert cell mixer?

[Nadya]

Dear all,

I am a first time analogue designer and need help with my Gilbert cell mixer for Bluetooth (CMOS 0.18um technology). The problem is that my circuit has a negative gain though, I believe, 10 dB is achievable. I have tried the following:
1. impedance matching with a shunt inductor of 1.5nH on both RF and LO inputs. This improved linearity but not the gain.
2. To bias the bottom two switch trasistors so that they operate in the saturation region (Vgs< Vds, right?) but that makes the gain even worse. I tried to bias them by means of introducing an offset source at the gate of the transistors.
3. Make the transistors wider. This does help a bit but even with 300 um gates my gain is still negative.

I would very much appreciate any advice anybody can give me.

 

[fcfusion]

In CMOS amplifiers reducing current generally improves gain.

Perhaps the 2 bottom transistor's should be in Sub-Threshold. This also improves gain.

Increase the upper transistor's L (length). This increases channel modulation effect, thus increasing incremental resistance Ro which will increase your gain. If you can't change W/L ratio increase W and L by the same amount. The ratio will be the same but the channel modulation will increase anyway.

How are you connecting your upper transistor's? As "diode connection" or as "current sources"?

In "diode connection", for each differential pair, one of the transistors has his Gate connected to Drain. They're easier to bias but have reduced gain.

In "current source" connection, the upper transistor's gates are connected only to a voltage reference, for bias. They have increased Ro, thus increased gain too.

Hope it helped

 

[Nadya]

Thanks for the reply.
I have tried out the things you have suggested. Increasing the L of the transistors makes the gain worse as expected since gm is inversely proportional to sqrt(1/L) and the gain is theoretically (2/pi)*Rload/(Rs+ (1/gm)). So I want that gm as high as possible.
Biasing the bottom transistors (those connected to the RF input) either has no effect or makes the gain worse as well. Btw, I bias by hanging a DC offset source of roughly 0.2-0.6V on the gate. Is it correct? According to the model simulation the transistor hardly conducts till about 0.5 V VGS so I am guessing 0.4 V is subthreshold.
Biasing the top trasistors (the ones connected to the oscillator input) at 0.6 V increases the gain. The differential pair is connected as what you call the "current source" way.
So, right now I can make a gain of 7 dB if I use W=70um, L=0.18u and hang a 500 ohm impedance at the output. I believe W=70um is too wide (you get all sorts of secondary effects from a transistor that wide, right?) and it should really be 50 ohm at the output. My voltage swing is from 2.5 to -2.5V which I guess is also too large. So, please,please, if anyone has any suggestions, help is very much needed.

 

[fcfusion]

I have tried out the things you have suggested. Increasing the L of the transistors makes the gain worse as expected since gm is inversely proportional to sqrt(1/L) and the gain is theoretically (2/pi)*Rload/(Rs+ (1/gm)). So I want that gm as high as possible.

I was sugesting increasing L for the load transistors.

According to the model simulation the transistor hardly conducts till about 0.5 V VGS so I am guessing 0.4 V is subthreshold.

Yes, 0.4 V is subthreshold. Too bad it didn't improve your gain.

I believe W=70um is too wide (you get all sorts of secondary effects from a transistor that wide, right?)

Increasing transistors size decreases your bandwidth but also decreases your noise, so it's not necessarilly bad.

it should really be 50 ohm at the output

Are you buffering your Gilbert mixer? If not, the parallel Ro//Rload will decrease your gain because Rload is to small.

Besides that, have you measured your bandwith? What is the frequency you're using? Your bandwidth might be too low. If so, increasing your bandwidth is your priority.

 

[Nadya]

I was sugesting increasing L for the load transistors

You mean like using transistors for Rload? I am actually using resistors of 500 ohm. Is it better to use transistors for that?

Are you buffering your Gilbert mixer?

I am not sure what you mean by buffering. can you explain, please?

Besides that, have you measured your bandwith? What is the frequency you're using? Your bandwidth might be too low. If so, increasing your bandwidth is your priority.

Well, the RF signal is 2.4 GHz and the LO signal is 1.8 GHz. This is what I use for testing purposes.In the real bluetooth application it would probably be something like 2.4 GHZ RF and 2.2 GHz LO. I guess 200MHz is about right for the IF signal in the bluetooth receiver application. So I don't think bandwidth is the bottleneck factor here.

 

[fcfusion]

You mean like using transistors for Rload? I am actually using resistors of 500 ohm. Is it better to use transistors for that?

No, no. Perhaps I didn't explained my self clearly. Your load must be a resistor, no doubt about it. I was refering to the top transistor's. Like in a simple amplifier, your gain is defined by Gain=-gm*Ro. The Ro can either be a resistor or a transistor. That's what I call a load transistor. What are you using as Ro,resistors or transistors?

Post an image of your circuit so I can understand better. Besides that, I wanna know if it's a balanced mixer or not.

I am not sure what you mean by buffering. can you explain, please?

If your using load transistors (be carefull, no confusion) your output impedance will probably be very high (several kOhms, which is good). If you connect your circuit to a load (like 500 Ω resistor, antenna, or whatever) your gain will drop dramatically. That problem is usually solved connecting the output of your mixer to a buffer, like a common drain amplifier, which as voltage gain = 1 but very low output impedance (around 50 Ω, which is good)

Well, the RF signal is 2.4 GHz and the LO signal is 1.8 GHz. This is what I use for testing purposes.In the real bluetooth application it would probably be something like 2.4 GHZ RF and 2.2 GHz LO. I guess 200MHz is about right for the IF signal in the bluetooth receiver application. So I don't think bandwidth is the bottleneck factor here.

You're working with very high frequencies. This means you cannot use big transistors because they will reduce your bandwidth. You should plot 2 AC simulations, 1 applying only the LO signal and the other applying only the RF signal, and check the output bandwidth. If the cut-off frequency is bellow 2.4 GHz then you have a problem. If so you'll have to reduce all the transistor's sizes to reazonable values.

Let me tell a little bit of personal experience. Once I tried to build a balanced Gilbert mixer for 5 GHz using 0.38um technology, when I was still a noob. It failed miserably. After a while I discovered why: for the 0.38 technology, even the simplest inverter circuit with the smaller transistors had a maximum cut-off frequency of about 4 GHz (insufficient). The whole mixer circuit had an even worst frequency performance.

So, moral of the story, check the bandwidth first. It's useless to try improving your gain if your bandwidth is not enough.

 

[Nadya]

Yes, the mixer is fully balanced. I am posting a doc file with the ADS schematic. Maybe you can see some glaring errors there.

I shall try the bandwidth simulation as you have suggested. So when doing the simulation I will have to completely disconnect one of the inputs?

 

[ouali]

hi, I think 70um transistor width in 0.18um technologie is ok, you can use 200um if you want, but what is the input LO power level? the noise figure? the power consumption of your circuit. I think with your topologie you can reach more than 14dB CG an less than 8dB noise figure