Tell me please : what does it mean 20dB balance at 30Mhz and what is a "series of transmission line transformers ? What does that look like ?
20dB balance, meaning, the common mode (error, imbalance) is 20dB below the differential mode (intended signal).
Common mode is the average of two lines, (V1 + V2) / 2. Differential mode is the difference, V1 - V2.
Error arises due to how the transformer is wound. Ideally, it would be geometrically symmetrical, but the primary is itself unbalanced, so this isn't possible. The next best construction would be a flat (cylindrical, not toroidal) winding, primary in a single layer, then a metal shield (a cylinder, tied to ground, slitted so as not to form a shorted turn), then the secondary (in one single layer, CT in the middle). Transformers of this construction have been used for bridge instruments.
For a construction like I suggested (three wires twisted, wound together), you have the transmission line mode coupling the primary directly into the same ends of the secondaries. But one secondary is turned completely around. So what you will actually get is, above some crossover frequency, the negative secondary side will droop down, phase shift, and flip around to ~0° phase. Then the transformer output will be ~100% common mode! Down below this resonant frequency, in the passband, differential mode will be dominant, but not perfect, and because we know a resonance is two poles, it will have a -40dB/dec asymptotic response. That is, if it's ~0dB balance (e.g., one side of secondary zero, other side nominal) at Fmax, then at Fmax/10 it will be closer to 40dB balance.
Whereas this kind of transformer design kind of grudgingly accepts transmission lines despite their problems, there is a better way.
The ideal construction method harnesses transmission lines in the best way possible. An ideal transmission line has no common mode current: it has two ports (input and output, say; but mind they're both bidirectional, so implying an input and output is kind of disingenuous), only the differential voltage of which matters. The relative voltage between the two ports can be whatever, it doesn't matter. That's an ideal port.
A real transmission line doesn't have ideal ports, but we can make one look arbitrarily good by wrapping it in ferrite and other core materials.
So, suppose we take the input signal, delay it by some time T (a transmission line is an ideal delay element), and hook up the far port backwards. Now its signal comes out inverted, with respect to ground. An inverting transformer! If we look at the voltage from input to inverted-output, we see it's pretty good (as a centertapped transformer), but one pin is delayed and the other is not. So, let's add another transmission line, that doesn't do anything (it's wired positive-to-positive on both ends), just adds delay. (This line doesn't need any ferrite cores on it; the other one does.) Great, now we get a perfect (delay matched) positive-and-negative (centertapped winding) version of the input!
Note that, because the transmission lines are straight through, there is only delay added -- this type of transformer does not depend on frequency. It has infinite bandwidth, no Fmax!
In practice, of course, Fmax is limited by the properties of the transmission line. It might be too leaky, or the transmission line itself breaks down into higher (waveguide) modes. (It's rather hard to perfectly invert the geometry, along an inverting transmission line winding, from one end to the other. Usually, twisted pair is used, or two pieces of coax with a crossover joint in the middle. These will inevitably radiate some imbalance, particularly at very high frequencies.)
That's the first step, a Guanella TLT. But it's not so much a transformer as an autoformer. We don't have isolation.
To obtain isolation, follow it with a 1:1 isolation transformer. An isolation transformer cannot have unlimited bandwidth, sadly, but we can do better with a single (two winding, 4 terminal; or 6 terminal if CT:CT) transformer, than we can with doing everything at once*.
This gives good balance, isolation and CMRR. Note that it requires two cores, so each transformer needs to be designed for double the inductance, to get the same Fmin (LF cutoff frequency).
At LF to HF, it's probably not a big deal where the isolation transformer is placed, on the primary or secondary side. Primary might be better, because of convenience and bandwidth (easier to get a 4 terminal transformer to behave nicely?). A VHF+ IF output might want it the other way, though (less stuff floating above ground!).
*At this point, we're really not much, if any, better off, compared to the shielded transformer case. That may get better with other tweaks, though.
What tweaks would those be?
We can further improve balance and CMRR by adding chokes. Again, at some expense to loss, and Fmin or balance. An isolation transformer has isolation capacitance; this can be improved by introducing a common mode choke. A CMC is just a length of ordinary transmission line -- a pair (or triple, or whatever) of wires wound up on ferrite as needed. Downside, it's more (electrical) distance from the CT winding, which means balance can be worse (not from the primary side -- but because unequal loading will result in unequal voltages).
Putting the CMC on the single ended (primary) side would seem to be the best option. Again, if it's a priority to have fewer components floating at RF/LO/IF frequencies, it would be better placed on the secondary, and then you have to deal with balance.
Finally, balance can be improved (at expense of Fmin) by adding a CT choke. Just another TL, this time with a twist back to itself so it's a 1:1 inverting autoformer. With no other windings on the core, these can be quite compact, giving good balance. Again, downside is more stuff loading down the signal -- with fully three cores loading the signal now, you need triple the inductance for each, to get the same Fmin.
You might not use all these options, but some combination is often very beneficial when shooting for a particular impedance ratio, isolation requirement and so on. Here are two examples:
https://www.seventransistorlabs.com/Images/Wideband_Amp_4W_50MHz_Sch.pnghttps://www.seventransistorlabs.com/Images/Wideband_Amp_4W_50MHz.jpgL1 is on the phenolic former, just a plain transmission line to match delay. L4 is an inverting transformer, wound on the blue and gray cores. Thus, L1 and L4 form a 1:1+1 Guanella TLT.
L3 is the central pot core, wound with red-white-green wires (enough strands wired in parallel to give the specified characteristic impedance). Note that opposite ends of L3 are bypassed to ground, and supplied with +20V: this is a 1:1 inverting Guanella TLT, which enforces balance on the two output phases.
L2 is the sideways pot core, wound with, uh, whatever other colors, six strands total (connected in parallel and series as shown, to give the characteristic impedance shown). This is a 1:2 transformer, connected in such a way that one phase's voltage gets doubled and the other's goes to ground (the input terminals are +V/2 and -V/2 at 12.5 ohms, the output is +V and 0 at 50 ohms).
This was rather overkill for what bandwidth the amplifier ended up with (about 50MHz); I made these transformers for it, more as an exercise.
Tim