billcalley said:
While I have you two gentlemen online, another quick two
questions, if you have a minute: 1. Is the "charge pump gain"
specification on a PLL's data sheet the exact same thing as the
"charge pump current" specification on other PLL data sheets?; 2.
Does increasing the charge pump current setting on the PLL from 1mA to
10mA decrease the lock time by 10x, or just "significantly"?
If the loop is nice and stable to begin with, the transfer function
(gain vs frequency) crosses unity gain with a slope of pretty nearly 6
dB per octave, i.e. near there, the loop gain is proportional to 1/f.
(*) Small changes in gain therefore change the loop bandwidth in the
same proportion--a 5% increase in gain will increase the bandwidth by
pretty close to 5%. Factors of ten are another matter, though--it could
easily become unstable, as Andrew said. It all depends on the details
of your frequency compensation scheme.
Most of the time, PLLs use an op amp in the loop filter, so you can get
as much bandwidth as you can handle, irrespective of Ipd. Diode-bridge
phase detectors have better SNR if you drive them hard, but once you put
something as jittery as logic in there, the rest of the noise
contributions are much less important. (Not that you can't still screw
it up, but the PD output current isn't the important thing in general.)
It's actually very convenient to use op amps--since we're always
building loops that are much slower than the op amp, we can use the
ideal op amp laws and put poles and zeros wherever we like by using
appropriate RCs in the input and feedback elements.
The thing that always messed me up when I started building PLLs many
years ago was that loop gain looked like it had _units_. Kvco is in
Hz/V, Kpd is in V/radian, and the loop amplifier's gain is
dimensionless. Multiply them together, and you get units of Hz/radian,
i.e. 2*pi/seconds. The thing is that the VCO is actually an ideal
integrator--if you change its control voltage, the phase starts
increasing by 2*pi*Kvco per second, and keeps on going. That means that
there's a factor of 1/(2*pi*f) that you have to put in to take account
of the frequency rolloff due to the integration. You also have to use
the same units throughout, i.e. use volts per cycle and hertz per volt,
or volts per radian and radians per second per volt. That way the loop
gain comes out dimensionless, which of course it must.
Cheers,
Phil Hobbs
(*) This is because if the rolloff is much faster than this, the loop
phase shift becomes too large, and the loop stability suffers.