How to improve mixer's conversion gain

I am designing a Gilbert Mixer, and I use 3v to supply the power. I find my conversion gain is very low, about -20dB.

I use resistors for bias and use on-chip transformer for differential inputs of LO and RF.

Who can tell me how to improve the conversion gain? Thanks a lot.

At –20dB conversion gain you have a problem there. Could be a phase unbalance.
This document may help you how to improve conversion gain at low powers.
http://www.ifh.ee.ethz.ch/Microwave/...SanDiego95.pdf

A circuit schematic will help a lot to understand the problem.
Mazz

Dear vfone,

I could not open the link you posted in the post.

thanks

wccheng

To Vfone,

I cannot open the link you posted.

Added after 23 minutes:

To Mazz,

The attachment is my schematic for Gilbert Mixer.

The DC working point is there.

Thanks

Karen

For some reasons the link that I posted is not working.
I am attaching that document.

To Vfone,

Thank you. I will check to find a way to improve.

Best Regard

At a first glance it seems that the input transconductor bjt is in saturation. So its gain is very small. The base of the quad should be at least 300mV higher.
Furthermore the LO signal is not applied in simmetrical way. One path has LO_RB2 in series, the other path no.
I hope it can help.
Mazz

Oooops... the font he used is really small, I can hardly read any value.

I agree with Mazz, and I'd like to address some comments.

Don't bias the transistor with resistive divider and feed the divided voltage directly to base or gate node, especially when you are using BJT or HBT. BJT and HBT have exponential relation between Vbe and Ic, it's very hard to tune the resistance. And while the IC suffers process variation, you'll get a totally different bias current.

You can reserve a pin for the bias voltage for external tuning , but don't do that with on-chip two-resistor voltage divider.

To Dsjomo and Mazz,

Thank you for answers. I have fixed my problem yesterday.

I am a new RF designer. I can only find books for how to do bias in LNA and amplifier. They only use resistor divider. Could you recommend some basic books about mixer bias?

Thank you every much.

Now I got the LO and RF input reflection coefficient(or input impedance) using large signal simulation @ ADS. However, when I use Smith Chart to conjugate match this impedance to 50 Ohm. I find it's totally wrong. It is does not improve the reflection coefficient, but it makes it worse. I don't know why. Could you tell me why?

Thanks a lot.

Are you sure you are reading correct values from Large signal simulation?
To start you can use simple S parameters for both Mixer in and LO in and do conjugate math there.
Mixer bias is not really different respect other circuits. Active bias can be used.
I hope it can help.
Mazz