Figure 1 illustrates the typical setup used to provide so-called quiet power for a sensitive analog circuit. Good applications for this LC-filter structure include oscillators, PLLs, and fiber-optic receivers. This filter reduces the differential noise that the analog component X between terminals AVCC and DGND perceives.
What about the absolute noise on AVCC? Compared with a true center-of-the-Earth ground-reference point, does the filter reduce the absolute noise on AVCC? Careful consideration of this question may lead you to a better understanding of the purpose of power-supply filtering.
First, consider the matter of a 0V potential reference. First-year electrical-engineering texts normally teach this concept in conjunction with the study of Kirchoff's laws, which form the basis of all modern electrical engineering. According to Kirchoff, every circuit contains one 0V reference node. His equations then define all other voltages in terms of their potential differences from the reference node. Kirchoff's equations are so generally useful and so widely taught that engineers rarely stop to question their applicability.
Unfortunately, the problems of ground noise, electromagnetic radiation, and ESD susceptibility do not succumb to Kirchoff's analysis. These problems violate one of Kirchoff's first and most important assumptions, namely, that all the electromagnetic fields in a circuit must be well-contained within compact, discrete circuit elements. When electromagnetic fields ravage the territory between circuit elements, the zero-potential concept evaporates.
For example, when measuring the potential difference between two points on the ground plane of a high-speed digital processor card, you must connect wires (or probes) to these two points and then feed the wires over to the inputs of your measuring equipment. Already, you have a problem. As any EMI professional will tell you, the space surrounding any processor card is filled with intense, high-frequency electromagnetic fields. These fields interact with your wires, inducing noise. The induced noise shows up in your measurement, and there's no way to eliminate it. Even worse, when you move the wires, the noise changes. The measurement and the measurement technique influence each other. Just as in relativistic physics, you, the observer, become part of the circuit.
With electromagnetic noise present, you can talk sensibly about potential differences only between points that are collocated, that is, points so close that the total field strength between those points is negligible. Global 0V reference potentials do not exist within large, high-speed digital systems.
Lacking a good global 0V reference, then, does it make sense to talk about reducing the noise on AVCC? Yes, provided that you are interested only in reducing the differential noise between AVCC and DGND in the local vicinity of X, a job that the circuit in Figure 1 admirably performs.
To see how this circuit works, assume that at operational frequencies, the impedance of L1 is much greater than the impedance of X, and the impedance of C2 is much less than that of X. Component L1 thus operates as an open circuit, and node AVCC is shorted more or less directly through C2 to DGND. Given unavoidable high-frequency noise on DGND, C2 serves to inject that same noise directly onto AVCC, ensuring that AVCC and DGND perfectly track each other at high frequencies.
This filter does not eliminate noise on AVCC. It merely makes AVCC and DGND the same , reducing the differential noise between the two. Power-supply filters always work that way. They copy junk from one circuit onto another so that the two match. For circuits such as oscillators, PLLs, and fiber-optic receivers, which don't reference other external grounds, noise matching between AVCC and DGND is generally all you need.
Circuits such as A/D converters, which may reference two ground systems, can be much more complex. In those cases, you need to connect all the relevant grounds together while ensuring that no high-speed currents can flow through the attachment point. The absence of high-speed currents eliminates local magnetic fields, ensuring the applicability of Kirchoff's laws near the attachment point. All the grounded metal in the vicinity of your ADC can then truly rest at the same potential, and the circuit works.
Engineers often talk about "cleaning up" the AVCC supply. Power-supply filters don't do that. If you want to clean up your AVCC plane, use soap. If you want to minimize the differential noise between AVCC and DGND, use a filter.