The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl Design >> RF Design >> CMOS passive mixer https://designers-guide.org/forum/YaBB.pl?num=1152870909 Message started by Prabhu on Jul 14th, 2006, 2:55am |
Title: CMOS passive mixer Post by Prabhu on Jul 14th, 2006, 2:55am hi, I am designing a double balanced CMOS passive mixer. Input to the mixer is a LCR tank circuit ( i.e. LNA load ). Output of mixer is a 100f F cap . I am using a sinusoidal LO drive. My problem is this: how should i get the input impedance ( equivalent capacitance in parallel with equivalent resistance) offered by the mixer? I need the equivalent capacitance to choose the inductance value at the LNA output to resonate at the desired frequency. I am asking this because, the input impedance will be time varying with a period of one half of the LO period. I need help in this regard. regards, Prabhu |
Title: Re: CMOS passive mixer Post by TT on Jul 14th, 2006, 6:08pm maybe for the input impedance, you can approximate the triode mixer impedance by it's fundamental value (i.e. impedance at the RF freq). Maybe you would want to add a source follower to isolate mixer and LNA. |
Title: Re: CMOS passive mixer Post by ACWWong on Jul 17th, 2006, 8:34am I agree with TT. A small signal sweep (ac or sp) centred on the LO switching point will give you a good enough idea of the value of the capacitance to add, as non-linear large signal changes in the capacitance tend (in most cases!) to cancel over a cycle. You can confirm the small signal load capacitance you derive for loading your LNA (i.e mixer Cin) is correct by comparing the ac gain reponse when resonating with your inductor against a frequency swept transient large signal simulation. In the past I have often found they match quite well. |
Title: Re: CMOS passive mixer Post by Prabhu on Jul 23rd, 2006, 8:57am hi, I estimated the input impedance of the passive mixer and simulated the LNA + mixer combination. Output impedance of the LNA (at the mixer input) is 1600 ohms differential. I am using the passive mixer as a voltage switch. Now if the spectrum at the mixer input is seen, a tone at 2*flo - frf of comparable strength wrt carrier (frf) is present. Since the mixer is designed for a direct conversion receiver, the spur "2*flo - frf" will also get downconverted to the baseband. what could be the reason for this spurious tone at the mixer input? one thing I observed was that as the output impedance of the LNA is reduced, the spur strength reduces. I need help in this regard. |
Title: Re: CMOS passive mixer Post by TT on Jul 24th, 2006, 11:57am By having a direct connection MixerLNA, you probably have a too high impedance to really drive the mixer switches in voltage mode but rather current mode. As far as I can imagine your design, it seems to me that the LNA output impedance is probably to high. Before any conclusion: can you check by introducing a VCVS with gain = 1 if this effect diappear ? TT |
Title: Re: CMOS passive mixer Post by Prabhu on Jul 26th, 2006, 11:38am I dont think the spur is due to the high output impedance of the LNA I am using a tuned load ( including switch parasitic capacitances ) to get the desired rf amplitude at the mixer input. Actually the presence of a ( large ) source resistance at the mixer input should improve its linearity ( voltage drop across the switch is now less ). I tried using VCVS between LNA and mixer. No spur will appear at the mixer input (since VCVS has Rin = 0 ) But when I observed the spectrum of the mixer input current ( current through VCVS ) a tone at 2*flo - frf of comparable strength wrt carrier (frf) is present. I think the source for this 3rd order spur may be attributed to the non-linear capacitance of the switching mosfet. In the literature I came across many voltage switching passive mixers ( with the input loading conditions similar to the above case ). But I dont find any reference to the above spur. I am wondering whether I am doing something wrong. |
Title: Re: CMOS passive mixer Post by ACWWong on Jul 27th, 2006, 8:24am how exactly are you stimulating your circuit and in what simulation ? are you using large signal (transient) for both RF and LO. What are the RF and LO freq are you using ? Is everything differental and balanced ? anyway assuming there isn't a simple bobo somewhere i suggest you sweep the LO drive power to see the effect on the tone. You can repeat with the RF tone if required. reducing your LO power by 1dB should give you a larger (ideally 3dB) reduction in any 3rd order distortion introduced by that stage. |
Title: Re: CMOS passive mixer Post by milkdragon on Jul 27th, 2006, 12:28pm Could you check if there is a 2*flo tune at the input/output of the mixer (without the present of the LNA)? is it big? It is quite strange that you will see 2*flo-frf with a balanced mixer. I agree with TT that your impedance (1600 Ohm differential) is too high, could you check also your LNA reverse isolation, S12)? there might be a case that your LO tone is leaked to the LNA input (circuit-wise, not talking about coupling via substrate) and gain back to LNA output with strong second harmonic (base on your information that reducing the output impedance helps) |
Title: Re: CMOS passive mixer Post by aaron_do on Jul 27th, 2006, 7:48pm Hi, you have a balanced passive mixer right? If you downconvert frf by high side injection for example you will get flo-frf. But i believe because of the symmetry of the passive mixer (S21=S12) your flo-frf will get upconverted (from output to input) and appear at the mixer input port (remember your input port impedance is high). The upconverted signal will have the signal + an image: flo-(flo-frf)=frf and flo+(flo-frf)=2flo-frf which is close to your original frf. Naturally the higher the output impedance from the LNA, the higher the image signal... The solution to this problem is to use a quadrature balanced passive mixer. That way the image when re-upconverted will be cancelled. Remember the receiver mixer is basically opposite to a transmitter mixer which inputs quadrature IF and outputs differential RF. Hope that helps! Aaron |
Title: Re: CMOS passive mixer Post by aaron_do on Jul 27th, 2006, 7:51pm By the way... If you don't use quadrature the simulator will actually show much more than 0 dB gain for the passive mixer. It seems like there's a lot of forward and backward mixing. From power conservation it may actually be possible but i wouldn't trust it. Anyway I think you need quadrature downconversion for most modulation schemes... Aaron |
Title: Re: CMOS passive mixer Post by Prabhu on Jul 28th, 2006, 11:53pm I am using BSIM3 model for 0.18um technology. I am using Spectre simulator to do QPSS analysis with LO being the large tone and RF being the medium tone. fLO = 2.44 GHz, fRF = 2.45 GHz. Circuit Setup: Drain and Source of the 4 transistors are biased at 0.7 V through large biasing resistors (100k). Gates are biased at 1.2 V . LO drive is square / sine wave of ±0.5 Vp around 1.2 V. Mixer input ( ac coupled ) is differentialy driven with a sine wave source ( 500 uVdifferential ) and 1600 ohms differential source resistance. Mixer output has a load capacitance of 50 fF on either side. Observations: At the mixer input I get a comparable ( one third wrt RF ) tone at 2.43 GHz. Transient analysis and DFT of the mixer input also gives similar result. Circuit is fully differential and balanced. I sweeped the LO amplitude and found that there is not much change in the amplitude of 2.43 GHz ( 2LO - RF ) signal at the mixer input. So this means the problem is not due to any third order non-linearity but due to the switching action of the passive mixer itself. As Aaron says, Is it be possible for the downconverted signal at the mixer output to mix backwards and add to the actual mixer input signal? How can that happen? 1) There is no active source at the mixer output. 2) Also there will be no reflections since we are using lumped components in simulation. |
Title: Re: CMOS passive mixer Post by ACWWong on Jul 29th, 2006, 5:18am Prabhu wrote on Jul 28th, 2006, 11:53pm:
I don't understand the two points you make, i can't see why active sources at the mixer output or lumped components make any difference to what you are observing. I think Aaron's original explanation seems pretty good. So (without going to quadrature mix) you should be able to decrease the relative amplitude of the unwanted tone by decreasing mixer Rs, or fiddling with the fet sizing, bias points (including back gate!) to improve the situation. |
Title: Re: CMOS passive mixer Post by aaron_do on Jul 29th, 2006, 7:38am Hi Prabhu, I guess there's more than one way of looking at it...if you simply observe a transient analysis then the current will only be flowing in one direction since as you say there is no source at the output...however if you split it into frequency components and use the principal of superposition it is possible to observe currents flowing in both directions on the same wire in a nonlinear/time varying element. I don't think your point on lumped elements and reflections is valid since even if you have assumed wave equations it doesn't physically mean current is flowing in both directions. Wave equations always apply and treating each element as a lumped element is a simplifying tool...just like a transmission line can be treated as a network of Ls, Rs and Cs. cheers, Aaron |
The Designer's Guide Community Forum » Powered by YaBB 2.2.2! YaBB © 2000-2008. All Rights Reserved. |