And in the olden days of genuine analog lines, it was done well. But we switched to digital trunk lines long ago, and those compress the hell out of the audio. And then with cellphones, we moved to even more severe digital compression (not to mention having to handle lost data).
I doubt if I have actually ever heard a call made on a true analog phone line from end-to-end. Digital trunk lines were the standard by the time I was born
[...]
Anyway, the real issue is standards: two cellphones making a call do not create a high-speed point-to-point data link, nor are they just shooting packets at each other (like FaceTime). They’re running a call via the voice network, which has strict standards that are essentially inviolable. So without upgrading the entire voice network infrastructure, you can’t just easily upgrade the voice quality.
Nonetheless, a few carriers have been doing this, allowing calls made within their networks, when using a supported handset on both ends, to have better sound quality. But support between carriers has been slow to roll out. See https://en.wikipedia.org/wiki/Wideband_audio
Old analog lines were, in my personal experience, ranging from horrible to bad.
When the land line was finally switched to a digital exchange (the last mile was still analog, and still is) quality went up quite a bit.
The only compression done on POTS on digital trunks was A-law or u-law (depending on your country), which causes very little loss in quality.
One of the initial phases in a mobile phone call is codec negotiation, where the phones and the network agree on which transcoder to use for compressing voice.
In some lucky case AMR-WB can be selected, but there also cases where the selected codec is worse than POTS quality.
For GSM and 3G, both circuit calls, it is possible that compressed voice is directly exchanged between the user terminals, with obvious advantages, instead of being:
1. compressed in one mobile
2. expanded in the network - where depends on 2G/3G and network architecture
3. transported as A/u-law samples
4. re-compressed before being sent to the other mobile
5. re-expanded in the mobile
Search for "transcoder free operation", but all the planets must be correctly aligned for that to work.
The GSM codecs greatly reduce the data rate by encoding the signal, just as mp3 etc does. The difference is the GSM codecs incorporate a model of the human voice tract, so it would be surprising if they transmitted arbitrary sounds well.
Exactly, so much that DTMF tones needs to be transported out of band. As above, they might or might not be transformed back to samples in the network.