With QAM you're more thinking about the error vector magnitude (EVM), rather then the signal-to-noise ratio. From the EVM you get a bit error rate, which determines maximum sensible data speeds. How high QAM will go is also a strong function of what frequency band: at low frequency, I've seen 512-QAM and higher, while at >100GHz they're struggeling for 32-QAM
Modern ADSL and VDSL version typically use DMT, which is more or less the same thing as OFDM. Interesting part of that in context of xDSL is that each of the subcarriers (typically QAM-64 modulated) can have different amount of actual payload data assigned to them depending on the loop quality and interference (many xDSL CPEs have spectrogram somewhere in their web interface which directly shows how the payload is distributed over the subcarriers).
It's a good read, but it highlights to me the fact that you really need to understand the compact maths that describes all this (in this case the Fourier transform) at which point it's all rather simple. I notice that much of signal processing discussion at the hardware level involves lots of hand waving and attempts to avoid even complex numbers.
> For OFDM with many subcarriers this can get to be a very large value. It is quite likely that M_OFDM can exceed 10^50. With such high signal orders, we expect that OFDM will not be tolerant of waveform distortion, since in OFDM, changing the waveform directly changes the data.
> Comparing signal order between OFDM and QAM provides a measure of relative tolerance to waveform distortion, corresponding to linearity specifications on the receiver and transmitter circuits, particularly the PA linearity.
And this directly caused high-performance headphone amplifiers.
Don't forget that the QAMs can go pretty high, 128/256 points iif the SNR is sensible.
I studied radio TDMA and it's still magical to me.