At 11:30 AM 4/2/2007, Alan Bloom wrote:
The other issue is that the phase shifter needs a full 360-degree range. It would be easy to do at the reference frequency with an NCO.
It is easy to get full 360 degree phase control. You just buy this chip that does it.
It is also possible to put the phase shifter before the PLL, in which case you do not need 360 degree phase control, but 360/n where n is the PLL multiplication factor.
Good phase noise shouldn't be too hard. With a high reference frequency you can use a wide loop bandwidth and basically get the same phase noise as the reference +20log(Frf/Fref).
Yep. There are a couple of issues here. I would do the data modulation after the PLL, although it can be done either way. If the data modulation is done before the PLL, you would want the PLL bandwidth to be many times wider than the modulation, because you want the PLL output phase to change in a small fraction of a modulation symbol time. (The numbers might work out ok.) If you put the modulation after the PLL then the modulation puts no constraint on the PLL loop bandwidth. Why would you care? Well because if we have a 1 Msymbol/sec modulation, and you want the phase to change in less than 1/10th of a modulation symbol, then you might choose a PLL bandwidth 20x the modulation rate, which would get you to 20MHz, but if the reference frequency of the PLL is only 100 MHz, and you want the bandwidth of the PLL to be a small fraction of the reference frequency (sampling issue) then you're already stuck with limited choices. Might work ok, but its not like there's a lot of room, but you'd need more careful analysis. I prefer an approach with a lot of room, and easy analysis, so I was preferring the modulation after the PLL approach.
As for phase noise, as long as the PLL loop bandwidth is wider than the modulation, then the phase noise THAT MATTERS (ie the phase noise within the modulation bandwidth) will be controlled by the reference, not the VCO in the PLL.
I forget what our modulation symbol rate was gonna be. Suppose it is 1 Msymbol/second. Then the loop bandwidth needs to be wider than about 1 MHz to ensure that the phase noise is dominated by the reference, and that seems trivially easy to accomplish. Hey, you can throw a factor of 2 or 3 in there for good measure. A 3 MHz loop bandwidth works out fine with a 100 MHz reference frequency, as that's 33:1.
So phase noise would seem to be not a problem. You can use cheap VCOs. The loops are wide enough to make 'em track the reference.