You can use the poly-phase splitter to generate quadratutre LO, and I have worked on chips where we did that... but it seems to gone out-of-fashion to my mind because these days (with fast processes) its is now more common to use Master-Salve divide-by 2 to get the quadrature.
With the divide-by-2 you can get as good quadrature (phase & especailly amplitude) but for a wider input range of signals. With the polyphase, you need to align your casacaded stages (3 maximum to avoid excessive loss) to minimise amplitude ripple/mismatch... which if you have a large LO frequency range can be quite tough. Also there is no power penalty as the current you would have needed to drive the passive polyphase is now probably better imployed in a div2. div2 usually has a lower area penalty.

The polyphase is much better suited in reverse (combiner mode) in low-IF image reject archtectures, where the frequency range is now just the IF bandwidth.

I have never used polyphase splitter in the RF  before downconversion....

cheers
aw

whether 2*flo VCO & div2 is better than quadrature VCO (or flo VCO & phase splitter) depends on the frequency, and the nature of the proposed tank.

for example, if the BOM means off chip L is not acceptable, then the VCO power will be dependant on spiral L losses (Q~3 to 10 in CMOS/RFCMOS). so in this case a VCO running at 2GHz will probably be more power efficient and divide-by-2 might be better than a quadrture 1 GHz VCO (as L value needed at 2GHz is lower than 1GHz and Q is usually better for lower L values at higher frequencies). Even off chip L's only really give good Q's and usuable values @ > 500MHz
The situation at mid-GHz points (where Lreq is now only low nH or less) gives bondwires as attractive option, moving the tank Q bottle neck to varactor (Q>20), so i think a 4GHz VCO using bondwires to be better than a 2 Ghz using spiral L from the noise viewpoint, without power penalty.

the fact that 2*flo does mean the direct buffering is more power hungry, but i have found that current is better imployed in the VCO and div2 (ie no buffer required) given that the VCO has large swing and div2 is basically a digital circuit. the only reason a buffer would be required is for isolation, but when vco is at flo*2, this may not be a big issue on RX, and for loIF/ZeroIF fVCO≠flo is a must.

remember also the div2 gives a 6dB noise improvement to the VCO phase noise.

hope my comments give you some idea of some of the trade-offs, but basically i think you should play around with your design kit to deduce the best architecture for your receiver..

aaron_do wrote on Aug 3^rd, 2006, 6:49am:


the VCO is usually quite powerful, as such its tone can usually be picked up all over a reciever chip (especially on bondwires and also via substarte, power supplies etc.). SO if you are trying to receive small signals on a snesitive LNA input or mixer input (at the same frequency as the VCO) you will find the signals you want will be swamped by the VCO leakage... this is the classic example of LO self mixing. being differential with good layout etc. helps but not enough in most cases....
So generally for ZIF and nearZIF (Very-low IF) one puts the VCO frequncy such that self-mixing products are out of the IF band.  i.e fVCO≠flo. The fVCO=2*flo and then divide-by-2 only suffers from self mixing across the mixer itself and the amount of flo power/devices switching at flo is much less so that in these cases its tolerable.