Distorted/deformed output signal downconversion mixer

Dear all,

I am designing a down conversion mixer for a wireless transceiver course. I started with just a simple single balanced mixer to get to know the basics. The circuits (mixer circuitry and test bench) are shown below. A mixer multiplies 2 sines. The result should look like this. However I noticed that my output is somewhat different. This is the output of my circuit when feeding it with a (sufficiently large) 900MHz LO and -20 dBm 918MHz RF signal:

You can see that the 18 MHz down converted signal has some strange deformations around the peaks of the signal (marked with a red circle). I increased the power of the RF source, to look if this changes the case, and it only made it worse. When I changed the power of the RF source to 0 dBm, the differential output signal of the mixer became:

As you can see, the deformation/distortion increased significantly. I was thinking about what the cause could be and come up with the following explanation (and that is my actual question): Would it be a fair assumption to attribute this distortion to the (mainly second order, namely the square law) non-linearity of the V to I converter (tail current source MN2)? Because when the input power is increased, the RF MOSFET (MN2) will be driven more out of it's "linear region" (of course this is not strictly non-linear but for small changes in gate voltage it can be assumed as linear) at the peaks of the input signal, resulting in more distortion/deformation at the peaks of the output signal?

Thank you very much in advance!

Test bench:

Analog Circuitry:

There are two very different types of mixers, switching mixers and multiplying mixers.

Your circuit does not seem to fall readily into either type, so do not be too surprised if the output looks distorted.

Dear Warpspeed (Tony),

Thanks for your reply. I believe it is a switching type of mixer. I based the design on the RF-Micro Electronics book of Behzad Razavi. It is one of its active single balanced mixer. The LO signal is in my case (for simulation purposes) a sine, but with such a large amplitude that the switching pair directs the current to either one branch or the other.

The distortion is directly related to the gm of the tail pair. If it's non-linear, mixer output will be non-linear. It should be biased with a large Vgs-Vt to lower the distortion.

Dear deba_fire,

What do you exactly main with "tail pair"? There is only 1 tail current source (the V-I converter MN2).

Thank you for your reply. I have been thinking about this. e.g. by increasing the length of MN2. This will decrease Gm0/Gmds (which was fairly large for the minimum length devices). However this increase the quality of the output signal.

Furthermore I have been simulating and thinking a lot yesterday evening and I noticed that the weird 'v'-shaped peaks occur at the moment of minimum current (instead of maximum current, which I thought). At that moment the Gm of the tail current source (MN2) is small. I would expect that the most distortion is experienced when the Gm is at it largest point instead at it lowest point. deba_fire, could you explain why this is not the case to me?

It also seems not to be related to the current source getting out of the strong inversion/saturated regime. I'll give it a try to use a larger Vgs-Vt, but I am afraid I am getting problems with the headroom claimed by the tail current device (MN2) becomes too large (so either cannot guarantee the saturation condition of MN2 or the load resistances should be decreased which is devastating for the gain).

I meant the tail transistor. Sorry about the typo. For higher linearity, input gm must be as linear as possible.

Thanks deba_fire. It looks like you are indeed right. I eventually managed to plot the Gm of MN2 and the Gm does indeed change a lot (from ~8 mS to almost 18 mS for the case I am currently looking at). So there is also some sinusoidal behaviour in Gm. So the input signal is amplified (converted to current) with a changing factor (with some sinusoidal properties). So the Gm has some y+sin(x+phi), in which y is larger than 2 (because my Gm is in under all circumstances positive, a negative Gm actually doesn't make sense?). So you could for example get some sin(x)+(2.1+sin(x)) kind of behavious. Plotting this in Wolframalpha results in shapes that look like the behaviour I am experiencing. So I think the cause of the distortion/deformations is found.

Thank you very much for the help deba_fire!