It's about the control-output measurement of a flyback converter with multiple outputs. The point is to show how theory and practice don't line up with each other, and you ultimately have to design based on empirical measurements.

In this case, the control-output transfer function has an unusual slope somewhere in between -20 dB/decade and -40 dB/decade, when it should just be -20 dB/decade.

I've seen this effect repeatedly over 30 years of making measurements. You don't see many papers on modeling multiple output converters because it is tough to write a research paper when the theory falls apart. However, this is reality, and we often have to design with what we've got, not what the simulation software says we should have.

The data shown in this paper is really just the tip of the iceberg of of how weird multiple converters can get. One of the more interesting cases I encountered was a situation where the loop design was determined by the esr of a small rectifier diode which caused an unintended 30 dB extra gain at the crossover frequency.

Absolutely. It is very rare that we actually ever use a single output flyback -- almost all are multiples. I guess that is where the rubber meets the road -- between trying to figure out loop stability across both DCM/CCM modes, and especially cross regulation with very dissimilar and dynamic loads on the multiple outputs!

Cross regulation can be a major headache with very different power rating rails in my experience. We did a design a while back which was 100W on main rails and then 5 off 1W rails for control supplies. Resistor zener clamps were the only way we could stop the low power rails rising very high when they were lightly loaded and the main rail was heavily loaded. It made me wonder whether I should have used a different architecture....

Dr. Ray straddles the fence of computer modeling and the hands-on experience. I entered the field long before the PC-AT and my first MPU design was the 4004, 2 years after graduating. So our experiences are quite different.
That aside, the flyback is my favorite topology. It is eternally fascinating. What Dr. Ray may be experiencing is the differences between the loading of each output. I in my more younger days, I made a fully synchronously rectified multiple output flyback converter. It was very educational. The flyback, diode rectified, is inherently 85% efficient. The fully loaded outputs may enter continuous mode while the lighter outputs are definitely operating in the discontinuous mode. So don't try it, because it is much less efficient than diodes.

This presents a very piecemeal influence in the control-to-output characteristic. With fixed resistive loads, it can be measured, but in the real world where loads are constantly changing, it is a challenge. There can be nominal load ranges, which may yield an area (frequency range) of potential problems. Having a -30 to -40 dB slope in the gain over a narrow range, is not necessarily bad, if your phase is under control. I am a phase monger.

The traditional signal injection point for network analyzers is in the upper branch of a two resistor divider. This make perfect sense for the single output power supply. For a multiple output supply, perhaps the injection point should be in the lower resistor at a much smaller signal level. In that way all outputs will be affected by their respective weighing currents.

interesting idea, but i'd have to think about that. I don't think it can be done that simply.

I remember measuring the weighted feedback system for the first time and being surprised at how complicated it was. The intuitive design approach didn't work.

Perturbing the constant-current lower sense resistor is a very untraditional method of loop stimulation , but I believe, if done at the right signal magnitude, it also can be done to view all of the magnitude/phase points of the multiple outputs important to loop stability.

I have had the privilege, in the olden days, of having time on the bench to actually play with my designs. Mutually coupled multiple output filter chokes, transformer winding techniques, and multiple output voltage sensing. What the heck, it may end-up being a white paper for you.
Nominal output loads are assumed.

A 50:1 or 100:1 current transformer comes to mind, Where the input impedance is near zero and the output impedance can be driven from a reasonable source impedance (say 50 ohms).
The constant-current (lower) sense resistor's current could be perturbed only slighty, while influencing the loop greatly, The current summing of the sensing node ( above the resistor) would combine the influences of the mutually combined sense currents of the sensing network thus producing a combined network feedback solution.
Play, have fun.