joaquins said:
Wait a second, let's go for basics...
L+R=M
L-R=S
M+S=L+R+L-R=2L
M-S=L+R-L+R=2R
So, if it would be made of 1:1 transformers level would be doubled at the end of the chain, if you use some pull down in one stage and the same pull up in the other also would be doubled.
So you need to lose 3dB in each stage, both the same, unless transformers loads are contemplated or don't pay atention to 6dB boost in the chain both shouldn't be mirrored and be pull down.
Maybe I'm not making sense and losing some fact, just an observation.
JS
That's an almost centennial debate. Some advocate losing 3dB because signals combine quadratically (do they?), others advocate losing 6dB because the L & R signals are strongly correlated, so they combine almost arithmetically. No one has given a perfect universal answer. Using 1:1+1 runs the risk of overloading the M channel; what's more, the L & R input impedance becomes 300 ohms when both signals are correlated (the impedance varies from half-nominal to near-infinity in function of the correlation factor). Using 1:0.7+0.7 (-3dB) maintains proper impedance and level for all sorts of signals. IMO 1:0.5+0.5 is overkill. The only advantage is it is easier in terms of manufacturability (quad-filar winding is easy and accurate).
Now regarding the M-S to L-R side, there are disadvantages to the 1+1:1 approach. The L & R signals recovered from a nominal M are at -6dB, which may force the equipment inserted in the M chain to work a little hot. Using 0.5+0.5:1 makes the reflected impedance too low, which the inserted chain may not like. Again, the impedance of the M & S inputs vary considerably against correlation. In typical cases, where the correlation factor is about 0.8, the impedance of the M input is about nominal and the S input is about 5 times larger.
As you can see, the (apparently) more trivial subjects hide some real complexities. No wonder all major manufacturers have gone active. Any transformer is much more expensive than a handful of IC's, which also offer much more flexibility.