The dial-up sound room

Read this, then press the button

Two machines that have never met are about to agree on how to talk across a telephone line, a channel designed for the human voice and nothing else. The line passes frequencies from roughly three hundred to three thousand cycles a second and mangles everything outside that. So a modem does the only thing it can: it speaks in tones. Pitch, timing and phase carry the data. What the world remembers as a screech was two computers negotiating, out loud, in a language made of pitches.

This exhibit is a stylised version. The pitches and the order are modelled on how such a call proceeded, but the details, such as exact timing and the way the signals overlapped, are simplified so that each stage is easy to pick out. Real sessions varied from line to line and from machine to machine. Use the “slow motion” switch to stretch every stage to three times its length.

What each phase did

Dial tone. Lift the handset and the exchange sends two steady tones at once, a low one and a slightly higher one, as a way of saying the line is yours. A modem listened for it before doing anything, because dialling into silence would be rude and pointless.

Dialling. Each key sends a pair of tones at the same moment, one from a row of four pitches and one from a column of three or four. That makes twelve or sixteen unique pairs from only seven or eight pitches, and the exchange can recognise each pair even on a noisy line. You can hear the same twelve pairs by pressing the keypad. The number in the box is made up; no call is placed.

Ringing. The ring you hear is the exchange telling you the other end is being rung, not the sound of the other phone. At the far end, a modem sits waiting for the line to ring, then picks it up.

Answer tone. The answering modem sends a long, high, steady tone. Every so often its phase flips, which you cannot hear but which a machine can detect easily. The tone serves two jobs. It tells equipment on the line, such as echo suppressors that exist to protect speech, to switch off and stop interfering. And it tells the caller that the other end is a machine.

First words. Both sides now exchange short, slow messages by switching between two pitches, a method called frequency-shift keying. One pitch means zero and the other means one. These early messages are plain: a list of speeds and error-correction methods each modem understands. They use slow, robust signals because they cannot yet assume the line is any good.

Training. This is the part that sounds like static. Each side sends known test patterns, bursts of noise and stepped tones, while the other listens. The receiving modem measures how the line distorts them: echoes, a loss of high frequencies, hum. It then builds a filter that undoes the damage, which is why the sounds are so strange. They are designed to reveal flaws, not to be pleasant.

Agreeing on a speed. Based on what they have measured, the modems pick the fastest method the line can carry reliably. The signals now carry several bits at a time, using combinations of pitch, strength and phase. The sound gets denser and more complicated, because there is more information in every moment.

Connected. The squeal ends in a rush of hiss, then quiet. What you hear after that is the carrier: a steady tone that holds the channel open and is modulated by the data. Most modems let you mute it, and most people did.