I understand there are both coupled and un-coupled versions of differential mode chokes as shown below.
Which one is better? I see most of the designs having un-coupled version. What is the reason?
Coupled differential mode inductors:
Un-Coupled differential mode inductors:
Coupled common mode choke real model includes stray inductances, and these inductances work as differential mode chokes and thus they help to tame the differential mode noise.
If these stray inductances are not enough the designer may want to put discrete inductors for better differential mode noise filtering (2nd image is a good example).
As a direct comparison, coupled ones generally have higher inductance in a smaller volume.
Saturation may be a concern. In a common-mode choke the power supply outbound and return currents cancel out. In a differential mode choke they add together. If the ferrite starts to saturate its inductance will be greatly reduced.
I can think of several reasons:
The HF modes are probably a subtle point that bears more detail. As is the case with most anything in the topic of EMC...
In the same way that the CMC has self-resonant modes, it also has modes coupling between windings. A dual-winding DMC will have this as well.
Consider this curve fit of a T604050-R6161-X504:
This was solved by superimposing the datasheet plot on the analysis output, and tweaking values until they match. I gave up towards the HF end as you can see; the data probably aren't too meaningful up there anyway.
Note that this is the common mode equivalent, so acts in parallel with the magnetizing inductance of the CMC (or rather, is the magnetizing impedance). Most of the inductance will be shared between the two windings, but some leakage and capacitive elements will couple across them.
Note that the legend says "1 phase"; sometimes the windings are wired in parallel for these measurements, but evidently they left the other open-circuit. This will cause it to act as a resonant trap, its terminal capacitance (approx. C1) resonating with leakage inductance (approx. L2 + L4), hence the notch and peak at (and a bit above) 10MHz.
There will be higher-order effects as you go up from 30MHz, of course; effects that depend ever more critically on what's nearby, and which radiate into space, not remaining confined to the measurement terminals. While we can measure components up there, the measurements become less and less meaningful as we go up. So this is a fine stopping point; and indeed, most datasheets end around 30MHz (conducted emissions limit), or even 10MHz.
Anyway, in the same way that the CM windings couple, giving a blip in the transfer function here -- so too, the DM windings will couple to each other. What would be CM effects, become DM effects, and vice versa, as we cross-connect the windings for the respective purposes. A highly reactive notch at 10MHz might have a substantial impact on the CM response of the filter otherwise. We can't simply stack these impedance plots together as if they were resistors; we must account for their reactance, and avoid creating notches in the overall filter response that lead to emissions (or susceptibility) problems.
In contrast, if we have single inductors, even if they have a response like the above -- that is, with ugly peaks and notches at some points -- we have a better chance of working with them, because we can for example dampen peaks by placing an R+C in parallel with it, and dampen valleys by putting an L||R in series with it. In general, where an inductive component becomes dominant capacitive (such as the above, >600kHz), we can add another inductor in series with it, and the added inductor resonates with the effective capacitance of the first; and we can ensure the series combination remains well-behaved by damping the resonance. And we can do all this without having to concern ourselves with more than two pins at a time, multiple modes between multiple pins.
