Hi aaron
Thanks for the link, I downloaded the thesis yesterday evening and I readed it before closing the day. At this moment I didn't go through replicating the analysis of mixer input impedance. I spended some time playing with the matching network and duplicated their results-+/-0.5dB here or there,I used the QL they indicated and assumed ideal capacitors- for figure 25 page 70 and with the NF they indicate for passive network I can say for sure that they are bad matched(s11 around -5dB) at 2.4G, then I did a little experiment I consulted the ITRS for the predicted Quality factor of inductors and I see that it's around 50 for 2020-and it's marked as red so at this moment they don't have a clue how to manufacture it-(it's not referred to the same inductance or band that we are speaking here) but just to made the experiment more interesting I assumed a QL=100 and repeated the simulations, as expected NF was at all efects null,voltage gain boosted but mismatching is worsened(in fact what I think that limits the voltage gain here is QC-ideal capacitors- that depends on Rs w and the values of the capacitors themselves, so maybe I could optimize a little bit this thing). My first impression is that the overall SNR is degraded compared with a typical matched amplifier due to the loss of power delivered by the antenna, but I must run more numbers to be sure.
About other parts of the thesis:I immediately discarded the low-IF architecture since they indicated only 20dB of IRR and I considered it too low and focused in the Direct Conversion architecture(that gives them a better NF after all and the power consumption is still good).
About the antennas is true that their impedance could be different of 50 ohms(I don't even remember an antenna right now that gives a real part of 50 ohms), but it's very common if you are an antenna manufacturer to match them to 50 or 75(for TV) this way one company usually designs the chip,another the antenna and another the final product(a friend of mine works for a company that designs bluetooth headseats and they do this way). If you are in a company that manufactures the whole module you have more freedom on the choice of impedances(but better check with the PCB designer because e ohms of characteristic impedance could be very funny).
I will try to scratch my head a little bit more this evening,
PS:finally I recovered my google powers I think that was a collateral effect of no drinking coffee for several hours...
In fact is a very interesting thesis to read and discuss, but one thing is clear:they traded off all(BW efficiency,hardware implementation-it reminds me a little bit to an old AM radio-) to make a good number in energy/per bit.

Hi didac.

your conclusion seems correct. Energy/Bit is the focus...

BTW when i say that the antenna impedance is different from 50 ohms, I mean that the antenna is designed to be 50 ohms, but in reality, the S11 of the antenna is better than around -10 dB.  I don't know this for sure because I don't know much about antenna design, but I remember looking at the data sheet for an antenna used for ZigBee, and the S11 was not great. This suggests that LC network used in this paper could have huge variations in the center frequency, bandwidth and gain. A normal active LNA doesn't have this problem because the matching network is usually much lower Q, and most of the gain comes from the active devices (i suspect in most cases this is true).

Now i'm going to read up on the Bode-Fan criteria

cheers,
Aaron

I think Lee’s comment is directly entirely to the comment in the interview article. I don’t think you can infer anything else from what he said in the paper.

Lee is making the point that here is a federally funded project on transceivers from a major research institution and we have comments obviously dismissing the well established fundamentals of MOSFET gate noise and impedance at high frequencies. Coming from a Ph.D. on transceiver design, it is an open mouth insert foot type of comment.

Cook says the gate is essentially a capacitor and does not require power matching, but what is the backplate of the cap? The channel! Cook’s comment would be correct if the backplate of the gate is simply connected to ground, but instead it is coupled (hopelessly) to the channel. Because of the channel coupling directly to the input, the output gain doesn’t diminish this noise source. In addition, the input impedance is not just a capacitance. There is a substantial real part at high frequencies and it is related to the channel conductance.

At low frequencies the passive gain argument holds, but higher frequencies it breaks down. Cook doesn’t bother to analyze the problem to tell us where this approximation breaks.

Having said all this, Cook does make a point and raises some food for thought, but more rigorous analysis would really enlighten us as to what is really being traded off from his approach.

Hi Aaron,

Just an aside. I believe that there is a simple way of extending the capabilities of a BSIM3v3 model to include NQS effects by just adding a resistance in
series with the gate. I don't recall all the details, but you can find it here:

"A-Simple Subcircuit Extension of the BSIM3v3 Model for CMOS RF Design"
Suet Fong Tin,....
IEEE Journal of solit-state circuits, Vol. 35, No. 4, April 2000

At the end, your fab still has to do this modelling for you of course.

Vivek

Hi,
First of all this a pretty interesting discussion. I haven't worked with this issue for a while but I think that if we can forget about the doomsday scenario of induced gate noise I see a big drawback from the point of view of matching, I think that Cook in his thesis is trading off directly SNR due to mismatching so I don't see any advantatge right now-as I said I must dig deeper in this issue-, but assuming that we can find passive gain structures with good matching I'm pretty sure they will be very narrowband so then we will face a architectural problem to cover the whole band.
PS:vivek I think that the BSIM3 extension that you reference it's already in use in RF transistors models, they add the gate resistance and the substrate network to model more accurately the high frequency behaviour, this is the reason that RF transistors are not very friendly in the sizing procedure-fixed width,length and variable number of fingers-.From what I know the gate resistance models correctly gate noise due to poly material but doesn't do anything with induced gate noise.