How do autotransformers affect sound?

Is the only reason many other [tube] amp designers don't do this because the autotramsformer is another component in the signal path? What is the trade off?

The big tradeoff is cost. And weight. Which is the same as cost. Oh and did I mention price? Sure there are technical disadvantages, but those can be fairly effectively solved by more weight, and/or cost.

The modern Mac solid-state amplifier may not be everybody's cup of tea, but they are standing on extremely solid technical ground with their use of an output autoformer. There's no reason why they should have flabby bass (i.e. they have good LF response and pretty high damping factor) . . . I just think that most SS amps have thin-sounding bass, and this what many speakers are well suited to.

The cost/benefit equation is probably different for McIntosh as well, because they manufacture the autoformers in-house. Rumor has it that the winding machines they're using today were specially built by Frank McIntosh and Gordon Gow when the company was founded . . . and they've always had autoformers in their solid-state amps. This means that all the costs required to build the tooling, and recruit, train, and manage skilled people to do the work, is already fully depreciated.

For another company that would have to invest in this or outsourse it, it's hugely, vastly cheaper to build a direct-coupled amp to the lowest expected impedance . . . and there's also sound logic in the idea that there's better places to put manufacturing resources than into an output autoformer.

Eldartford, the impedance relationship between the output stage and the autoformer on McIntosh solid-state amps has changed somewhat over the decades . . . and it directly parallels the changes in what good power semiconductors have been available.

Early SS amps like the MC 2105, 2300, etc. presented a low-impedance load (like 2-3 ohms) to the output stage, which was a quasi-complementary (all NPN) design. This is likely to be because transistors at the time were limited in their voltage capability, and good complementary pairs (NPN and PNP) weren't available. Virtually all other high-powered SS amps at the time used bootstraped pairs of output transistors (in series) to divide the voltage between them, allowing them to use high enough power-supply rails to get the output power but keeping the output transistors within their limits. At this time, Mac used no feedback around the transformer.

This remained relatively unchanged through the 1970s, even as complementary (NPN/PNP) EF output stages were adopted. In the early 1980s with amps like the MC2250, they added a little bit of feedback around the autoformer, but the output stage was still loaded at a low impedance. The 2250 was big step and the schematic is great to study . . . classic implementations of a diff-amp current mirror, active current-source for the tail and for the VAS, etc.

This arrangement stayed the same until the early 1990s, with amps like the MC7300, at which point the design was changed to where the autoformer presents a higher-impedance load to the amp, like 6-7 ohms. This makes sense as the output devices they used (MJ15003/4 and MJ15024/5) were now capable of handling both the voltage and current required for loading at a higher impedance.

The MC1000 in the late 1990s shared this approach, but was basically two amplifiers bridged around a single autoformer, which allowed the output power to grow without needing higher voltage capability in the output stage. The MC1000 was a commercial success, and now this bridged arrangement is common to all their amps that use autoformers.

As far as their linearity goes . . . I've looked at the distortion residuals of a ton of McIntosh amps, and with the earlier amps, it's dominated by output-stage (crossover) distortion. In modern SS Macs, the actual THD is completely buried in the noise floor, unless you really torture the thing.

I'm not saying that McIntosh solid-state amps are perfect, but they are damn good . . . and it takes very little time with any of them on the test bench to figure out that the linearity of the autoformer itself is simply not an issue.

Arthur, just a couple of minor clarifications . . . bifilar winding refers to the process of placing multiple transformer windings side-by-side as they are wound around the bobbin. Since an autoformer has only one winding, it technically can't be "bifilar wound".

You are correct in the point that McIntosh has increased the bandwidth of the autoformers to the point where HF audio response simply isn't an issue, but there's actually a nice side effect in the fact that the autoformer tends to roll off before the frequencies where solid-state amplifiers usually have parasitic oscillations and stability problems. If these factors are balanced correctly, then the autoformer can actually improve the stability of the amp, without the need for a silly Zobel network or the like.

Another nice benefit is the fact that the autoformer (when combined with proper current limiting) gives excellent DC-offset protection for the speaker, without needing a relay . . . even though there are indeed a few Mac autoformer-based amps with speaker relays for some reason.

Wow, really interesting thread.

Atmasphere, I get the impression that what you're describing is simply the tendency of certiain topologies to present non-linear loads to their preceeding stages. If this is the case, then how is this substantially different than a tube output stage running in class A2 or AB2 whereby grid current becomes significant for part of the cycle?

Shadorne, yours is a great question . . . to me, it still may boil down to cost. Winding high-linearity audio magnetics is expensive no matter what; could it be that for the typical values used in a loudspeaker crossover, it's simply cheaper to get the same or better linearity using an air core?

Eldartford, the optimum load for the complete amplifier output stage depends on the load it's designed to drive, or vice-versa. A single pair of bipolar transistors in an emitter-follower output stage tends to be really comfy with loads of 8 ohms or above, but it's a simple matter to lower this by paralleling output devices, as is the case with virtually all the Mac autoformer-based amps. The autoformer simply gives the designer more flexibility to balance both the total current and dissapation requirements of the output stage.

Arthur, thanks for helping clear up some of the terminology . . . I have only superficial familiarity with transformer winding techniques. But I always thought that the process of splitting a single winding as you describe was called "interleaving", and that two separate windings wound together was called "bifilar winding". By this, the Acrosound/Dynaco transformers were interleaved but not bifilar, but the McIntosh transformers were both interleaved and bifilar?

And finally, I think that an autoformer would be an interesting addition to a Class D amplifier, as their output filters tend to give best transient response at a particular output load impedance. But most of these are designed to be small and lightweight . . . and an autoformer is of course a poor choice if those are major design criteria.