A question on Cartesian feedback loop.Thanks

In the article "lineariztion technique using digital cartesian loop transmitter" , the author said "as with all feedback systems, its performance is limited by the delay around the loop and hence the linearity improvement which can be obtained depends upon the bandwidth over which the feedback must operate." I don't know the "bandwidth" is RF bandwidth or baseband bandwidth. Other articles say BB bandwidth has effect on stability by magnitude and phase margin. But I think the "bandwidth" in the article refers to RF operation bandwidth.
Can someone give me a hand to understand the words.
I'm a new comer to do cartesian feedback loop research, hope freinds mail me discussing cartesian loop.
my e-mail: foxgangq@yahoo.com.cn

Various amplifier linearizers work by generating a spurious product that is equal in magnitude to, but 180 degree phase shifted from, the products that are internally generated in the amplifier device. In other words, you are trying to vector cancel some generated noise with another vector of equal magnitude.

Unfortunately, if there is too much time delay in a system, the unwanted spurious signal is generated, then some time later you generate a cancelling signal--so now when you try to vectorially cancel the two "events" you find that you do not get a good cancellation because the two do not match up in time. ALSO, if you have a broadband modulation, it is easy to get the vectors to cancel exactly at one modulation frequency, but at other modulation frequencies the cancellation is not exact (usually since the phases are not 180 degrees out anymore)