One purpose of the PDN is self-impedance, yes. The other purpose is to measure the influence of varying demand from one side of the circuit to another. That is, the transfer impedance. (Or other equivalent representations thereof.)
In general, we can represent a supply net (not a SCH/PCB "net" as such, but the connected DC path, including inductors and low-value resistors) as a coupling matrix. (Which by port theory, can be any of multiple equivalent kinds, but consider the impedance matrix for example.) The self-impedance is simply the \$z_{i,i}\$ element of that matrix -- give or take assumptions on what the other ports are doing; but just as well, a well-damped PDN will have low sensitivity to the impedances connected to other ports, so we can quantify properties of both isolation and damping in this way.
If we convert to a transmission matrix, we can read off the gain factors between ports; or other matrices, the relative sensitivity to \$v\$ or \$i\$, from one port, to another port's \$i\$ or \$v\$ (e.g. H or G matrices).
In general, port theory provides a powerful and flexible way to analyze multivariate impedance networks.
As for using ferrite beads, or other series impedances, if the specification requires some minimum impedance at a given port, then clearly we must form a low-pass filter with \$f_c\$ such that |Z(ferrite bead)| ≤ |Z(required)|. We connect capacitor(s) in shunt afterwards, to maintain the impedance at high frequencies.
For typical values of say single ohms, this will be fairly low frequencies, sub MHz perhaps. Which already tells us two things: 1, we can expect reasonable attenuation for frequencies above here, and 2. what capacitance, and damping ESR or RC, we need to complete the circuit.
Say we have a ferrite bead of approx. 1µH at <1MHz, and suppose we need |Z| < 1Ω. We need \$f_c\$ < 160kHz, so C > 1µF, and R = \$\sqrt{\frac{L}{C}}\$, or 1Ω at these values. (Some R may come from the FB, both as DCR and core loss; I'm assuming lossless "FB" for now.) Just putting these together, we likely miss the peak impedance, but increasing C by 2-4× and reducing R by 30-50%, gives us ample wiggle room. And with \$f_c\$ at 160kHz, we expect good attenuation (>10dB say) at say SMPS ripple or harmonic frequencies (low MHz), and even more attenuation at say MCU core or peripheral clock frequencies (10s, 100s MHz). Thus we would obtain a nice quiet supply, when in the presence of other noise sources like these.
HF attenuation can be further improved by using a C || (R+C) element instead of R+C, where the parallel (no-R) C is less than 1/3 the value of the "bulk" (lossy) C. Typically the bulk cap will be something naturally lossy like electrolytic or (dry MnO2) tantalum (or various polymer types: both Al and Ta polymers are available in a range of ESRs, mostly but not exclusively low ESRs; shop around), or ceramic with an explicit external resistor; and the HF bypass will be ceramic, perhaps device local bypass (i.e. placed at the device power pins).
