Hi,

I am designing a differential voltage mode cmos passive mixer test chip.  The mixer RF/LO frequency is from 400M to 4G Hz.  IF is 30M to 250MHz (low side injection).  For the test chip, mixer I/Os are coming and going off chip.  The mixer consistes of 4 swtiching NMOS with the gate (LO) biased at 1.25 (with VDD=2.5) and the source at 650m-700mV.  The souces of these FET are also ac coupled to the incoming RF signal.
My questions are :
1.  Since it's kind of broad band, can I add a 50 ohms resistor at the each input instead of a matching network.  And how would I do that? (after the ac couping cap with the two 50ohms across the two inputs?)
2.  Can I do something similar to the output?  The mixer differential outputs are ac coupled to an external differential filter.  
3. Do I need to have both drains and sources biasd at 650mV or just either is ok?

I am used to seeing the passive mixer as part of the integraded LNA -mixer-filter-VGaA combo.  That's why I a little incertain about mixer I/O since now they need to go off chip.  

I am new to the community and have been benefitting from reading the posts here.  Any help from anyone is greatly appreciated.

Kelly

Hi rfmagic,

My mixer does not have a gm stage, the inputs are the sources of the switching FETs.  The input impedance is the channel resistance looking back to the sources.  I was just trying to see whether I can size the devices without worry about the input impedance.  My thinking is size te switching devices to give me the bast IIP3, and take care of the input by using a real resistor.  

The gate is the LO inputs, I plan to do what you said to resonant out the input cap in order to have more voltage swing.

Kelly

Hi RFICDUDE,

The reason the RF input impendance depends on the baseband/IF is because when you look into the RF ports you see the Ron(the switching pairs)+IF pads + input impedance of the IF filter, right?  Since the IF filter is off chip (for this passive mixer test chip), I assume I need to do an output match to the IF filter input, no? (IF filter is a saparate part with differential imput impedance of eiher 50, 100 or 200 ohms depending on which part the system guy chooses, but it's negotiable).

Since my mixer really just the 4 FETs switch devices, I thought I could at least just put a resistor across the IF ports (since it's low frequency).  But how does that change the input impedance look into the RF prots?  Would you just see the Ron + Rif?  Am I understanding your point correctly?

Thanks so much for replying.
Kelly

You are just about there.
I'm trying to say that if the Ron of switches are small then the impedance of the IF filter will be upconverted to the RF frequency and vice versa such that, except for a possible scaling factor, you can treat the problem as matching at RF to the IF filter impedance or attempt to select the IF filter impedance such that the RF match is achieved. In either case, you should not necessarily to resort to resistive loading to help achieve the match.

Now if the mixer was active then you would need to use resistors to load the IF because the RF input and IF output would be electrically isolated through the transistors. A passive mixer connects the RF to the IF through the periodic switching action of the mixer, so KCL/KVL has to be balanced on both sides. It is just more complicated because now you have to balance components that are converted in frequency due to the periodic switching.

Hope this helps.

RFICDUDE,

To add to your correct assessment of the impedance at the IF.  In the past I also noticed that a baseband impedance can be superimposed back to the RF input.  Based on this observation  I experimented with different types of baseband impedances for example a series LC in parallel with a real load and I was able to observe the filter being being superimposed back to the RF input which was interesting because the effective Q was relatively high as a fractional bandwidth to LO frequency although filtering greater then 6 to 10 dB was all I could achieve in simulation (I never actually built the filter part since the inductors would be enormous).  Concluding that the filtering is only a result of mismatch outside of the superimposed baseband  bandwidth I realized that impedance at the RF Source could negate some of these desired effects.  

Another point is that if an impedance at baseband can be seen at the RF then a signal at baseband can be seen at the the RF and thus noise at baseband can be seen at the RF.  Passive mixers suffer from being bidirectional and the noise at baseband will get mixed up to RF as a double sideband noise per I or Q and then get remixed down on the opposite quadrature and this is vice versa as well.  In the end I tried to decipherer this mechanism from sim results and concluded that this enigma was beyond simple to me and I just wanted to be sure that it was accounted for in my sim results which seemed to match measured within a margin of error.

http://rfcooltools.com

Hi vp1953,

I believe noise conversion  has a lot to do with the baseband receiver circuitry and the how the RF is split to I and Q.  Let me try and formulate a reasonable example.... if at base band you where to use two opamps with feedback to provide a certain final gain then there will be a noise associated with each at its input uncorrelated with the other.  This is due to the noise at the opamp output being feedback to the input with the sign reversed and depends on the feed back factor and the amount of noise at its output.  
If the RF is actively split (good reverse isolation) then there will be very little issue.  But if it is passively split then the noise of one opamp for example the I will be seen at the input of the opamp at Q through the mechanism  described.  Depending on the level of noise above the thermal noise floor the effect can be neglect-able to observable.   As you said the conversion up and back down has loss and this will put a bound on it.  But remember a opamps own noise is cancelled through its own feedback, but noise from the other will be amplified upon conversion.

http://rfcooltools.com