Very often an engineer resolves a stubborn EMI problem by just 'playing' with the transformer. The transformer comes into the picture in the following ways
With its windings carrying high-frequency current, it becomes an effective H-field antenna. These fields can impinge upon nearby traces and cables, and enlist their help in getting transported out of the enclosure via conduction or radiation.
Since parts of the windings have a swinging voltage across them, they can also become effective E-field antennas.
The parasitic capacitance between the Primary and Secondary carries noise across the isolation boundary. Since the Secondary side ground is usually connected to the chassis, this noise returns via the Earth plane, in the form of CM noise. The situation is very similar to the tradeoffs required in heatsink mounting issues. In this case, we want to couple the Primary and Secondary very close to each other in order to reduce leakage inductance, but this also increases their mutual capacitance, and thus the CM noise.
Here are some standard techniques that help prevent the above
In a safety-approved transformer, there are three layers of safety-approved polyester ('Mylar?'') tape between the Primary and Secondary windings, for example the popular #1298 from the 3M Company. In addition to these layers, a copper 'Faraday shield' may be inserted to 'collect' the noise currents arriving at the isolation boundary, and diverting them (usually to the Primary ground). Note that this shield should be a very thin piece of copper to avoid eddy current losses and also to keep leakage inductance down. It is typically 2-4 mils thick, consisting of one turn wound around the center limb. A wire is soldered close to its approximate geometric center and goes to the Primary ground. Note that the ends of the copper shield should not be galvanically connected, as that would constitute a shorted turn as far as the transformer is concerned. Some designs also use an additional Faraday shield on the Secondary side (after the 3 layers of insulation), and that is connected to the Secondary ground. However, most commercial ITE power supply designs don't need these shields, provided adequate thought has gone into the winding and construction, as we will discuss.
There is usually also a circumferential copper shield (or 'band') around the entire transformer. It serves primarily as a radiation shield. It is often left floating in low cost designs, though it may be connected to the Secondary ground if desired. If so connected, safety issues may need to be considered in regards to the requirement of reinforced insulation between Primary and Secondary, and also the required Primary to Secondary 'creepages' (distance along the insulating surface) and 'clearances' (shortest distance through air) as applicable. When the transformer uses an air gap on its outer limbs, the fringing flux emanating from the gap causes severe eddy current losses in the band. So this band is also usually only 2-4 mil thick. Note that the ends of this band can, and should be, soldered together, because this is an outer shield, and can never constitute a shorted turn for the transformer. But like the Faraday shield, this too can be omitted by good winding techniques.
To reiterate, even from the point of view of EMI, a flyback transformer should be preferably center-gapped. i.e. no gap on its outer limbs. The fringing fields from exposed air gaps become strong sources of radiated EMI besides causing significant eddy current losses in the copper band.
There is usually an auxiliary winding present on the Primary side which provides a low voltage rail for the controller and related circuitry. One end of it is connected to Primary ground. Therefore, it can actually double over as a crude Faraday shield if we a) wind it spread out over the available bobbin width, and b) we help it collect and divert more noise by AC coupling its opposite end (i.e. the diode end) to Primary ground, through a small 22pF-100pF ceramic capacitor as shown in Figure 1.
Figure 1 shows also shows a low noise construction principle as applied to a flyback transformer. In the discussion below, we note that transformers with split windings are not being discussed, though the principles described here can be easily extended and applied to them too. We should also continually compare the left-hand side diagram of Figure 1 with its corresponding schematic on the right-hand side.
Since the Drain of the Fet is swinging, it is a good idea to keep this end of the Primary winding buried as deep as possible, i.e. it should be the first layer to be wound. The outer layers then tend to shield the field emanating from this. The Drain end of this winding should definitely not be adjacent to the 'safety barrier' (the three layers of tape). Noise current injected is proportional to the net dV/dt across the two 'plates' of the parasitic capacitor. Since we really cannot reduce the capacitance much, without adversely impacting the leakage inductance, we should at least try to reduce the net dV/dt across this capacitor.
Comparing the diagram on the left with its schematic on the right, we see that the start and finish ends of any winding have been indicated. In particular, the start ends have been shown with dots in the schematic. In a typical production sequence, the coil winding machine always spins the bobbin in the same direction for every winding placed, therefore a) all the start ends (dotted ends) are magnetically equivalent (so if one dotted end goes high, the other dots also go high at the same moment, as compared to their opposite ends). We can also see that from the point of view of actual physical proximity, every dotted end of a winding is close to the non-dotted end of the next winding.
Which means that for the flyback transformer of Figure 1, the diode end of the Secondary winding will necessarily fall adjacent to the safety barrier. Therefore, we have a very small dV/dt on the Primary side, though we have some dV/dt on the Secondary side, and therefore a small net dV/dt across the barrier. But this dV/dt is much smaller than if the Drain end of the Primary winding was adjacent to the safety barrier. The latter situation can be created by winding the transformer the 'wrong' way, i.e. reversing all the start and finish ends shown in Figure 1). That would have brought the Drain end of the Primary winding right next to the safety barrier, with the Secondary ground end (which is connected to the chassis) directly across it. With this arrangement, we would have a healthy dose of CM noise injected directly into the chassis/Earth. Not the best way to achieve compliance for sure.
The transformer in Figure 1 has the advantage that the quiet end (ground) of the Secondary is now the outermost layer. That is by itself a good shield. So we can safely drop the ubiquitous circumferential shield (copper band).
When we go through the same reasoning for a forward converter transformer, we will find that with the described winding sequence, we will automatically have the quiet ends of both Primary and Secondary overlooking each other across the safety barrier. This is good from the viewpoint of conducted EMI since very little noise will be injected through the parasitic capacitance. But the outermost layer is not 'quiet' anymore, and we could have a radiation problem. In this case, the circumferential shield becomes necessary.
A way out of this forward converter outer surface radiation problem is to ask our production to reverse the direction of the Secondary winding (only). So for example, if till that point the machine was spinning clockwise, for the Secondary we specify an anticlockwise direction. With this, the reasoning given above for the flyback will apply to the forward converter transformer too. We would then have a quiet exterior (without a shield). That is usually more helpful than trying to reduce the amount of noise injection through the barrier, because that is something we can control quite well by various tricks --- like having the auxiliary winding double over as a Faraday shield etc. We do note however, that a forward converter transformer has no (or very small) air gap, so it is generally considered 'quieter' to start with.