Your Voice's Journey

Say something into your phone and it crosses a city, or an ocean, in a blink. Record your own voice and follow it, stop by stop: from a shaking diaphragm, through copper and numbers, into a laser in a thread of glass.

Stop 1 of 6

You talk, and the air shakes

Your voice is air pushing back and forth very fast. Inside the phone, a tiny thin sheet called a diaphragm gets pushed by that air. As it moves, it changes an electric current. Now your voice is a wiggling electrical signal.

Louder means a bigger wiggle. Higher pitch means a faster wiggle. Record three seconds and see for yourself. Try a hum, then an “sss”.

Using a sample voice ("aah… sss") until you record.

Zoomed in on the shaded 30 milliseconds: this is the current in the wire, wiggling with your voice.

At the other end, the same trick runs backwards. The current pushes a speaker, the speaker pushes air, and your friend hears you. That works for one call on one wire. But a phone company can't run a wire from every house to every other house. So what next?

Stop 2 of 6

The phone throws most of your voice away

Your ears hear from about 20 Hz up to 20,000 Hz (20 kHz). But to understand words, you only need the middle bit: 300 to 3,400 Hz.

So the old phone network cut off everything else. Less to carry means more calls fit on a wire. Below is your clip split into its pitches. Tall means loud. Slide the cutoff, then press play and hear the difference.

100 Hz300 Hz1 kHz3.4 kHz10 kHz

Why does “f” sound like “s” on the phone?

The hiss in “s” and “f” lives mostly above 4 kHz, right where the phone cuts. With the hiss gone, the two sound alike. That's why people spell names as “S for Sierra”.
Modern calls: HD Voice

Calls over 4G and 5G (VoLTE) usually keep 50 Hz to 7 kHz, which is why they sound so much clearer. They are still far short of the full 20 kHz your ears can hear. Try the “HD Voice” preset.

Stop 3 of 6

Many calls, one wire

If every call used the same 0 to 4 kHz on one wire, they would all land on top of each other. You'd hear everyone at once.

The trick: give each call its own lane. Your call uses 0 to 4 kHz, the next one is moved up to 4 to 8 kHz, the next to 8 to 12 kHz, and so on. At the far end, a filter picks out one lane, like tuning a radio to one station, and moves it back down. Engineers call this frequency division multiplexing (FDM): sharing a wire by splitting up its frequencies.

0 kHz8 kHz16 kHz24 kHz32 kHz40 kHz48 kHz

Tune in to a call:

The filter keeps 0 to 4 kHz, shifts it down to 0 to 4 kHz, and sends only that to the speaker. You hear one clear voice.

How is a call moved up?

By mixing it with a steady high tone called a carrier. That is modulation, the same idea a radio station uses. See Modulation Fundamentals.

Twelve calls stacked like this were a standard block called a group, sent at 60 to 108 kHz. Groups were stacked into bigger blocks, and by the 1970s one copper coax tube carried over ten thousand calls, using frequencies up to about 60 MHz.

Stop 4 of 6

Copper hits a wall

More calls means more lanes, and more lanes means climbing to higher and higher frequencies. Copper does not like that.

A fast-wiggling current makes tiny swirling currents inside the wire (called eddy currents). They push the signal out of the middle of the wire and onto its skin. This is the skin effect. Less copper doing the work means more resistance, and the signal fades faster. Slide the frequency up and watch the wire go hollow.

Frequency

3.16 kHz

Skin depth

1.16 mm

Wire doing the work

1.0e+2%

Resistance vs. DC

1.0×

A 1 mm copper wire. Orange carries the current; grey is copper that is just along for the ride.

Phone companies fought this by boosting the signal every few kilometres. But above about 65 MHz the signal died in under 2 km. That was the ceiling for copper.

Why they needed more frequencies, not more power

In 1948 Claude Shannon worked out the most data a channel can ever carry. It depends on two things: how wide the frequency range is (the bandwidth) and how loud the signal is compared to the noise (the SNR, signal-to-noise ratio).

Try it. Which helps more?

Most data possible

30.9 kbit/s

Double the bandwidth

61.8 kbit/s

Double the power

34 kbit/s

Doubling the bandwidth doubles the data. Doubling the power adds only a little. So engineers wanted a much wider band, and copper could not give it to them.

The formula

C = B · log₂(1 + S/N). C is bits per second, B is bandwidth in Hz, S/N is the power ratio (30 dB = 1,000×). Light has frequencies in the hundreds of terahertz, millions of times higher than copper's ceiling. Engineers called it “a fabulous El Dorado”.

Play with your own wire on the skin depth calculator.

Stop 5 of 6

Turn your voice into numbers

Here is the big idea that saved the phone network. Don't send the wiggle. Send numbers that describe it.

Measure the wave over and over (that's sampling). Round each measurement to the nearest step on a ladder (that's quantizing). Write each step as 1s and 0s. A phone call does this 8,000 times a second with 8 bits per sample. Make it worse and listen to what breaks.

Grey: the real wave. Red dots: the samples. Purple: what the far end rebuilds from the numbers.

Data rate

64 kbit/s

Phone standard

8,000 × 8 = 64 kbit/s

The real phone trick: μ-law

Phones don't space the 256 steps evenly. Steps are small near zero and big for loud sounds (called μ-law or A-law). Quiet voices stay clean, and 8 bits sound about as good as 12 even steps.

Calls take turns, like a zipper

Numbers have a superpower: calls can now share a wire by taking turns in time. Eight bits from call 1, then eight from call 2, then call 3, then back to call 1. This is time division multiplexing (TDM). The far end knows the order, so it can sort every bit back to the right call.

On glass fiber, each bit is a flash of laser light: on for 1, off for 0.

011100101000100010000101

Line speed needed

192 kbit/s

Frames per second

8,000

The first fiber under the Atlantic (1988)

Its lasers flashed 280 million times a second. With compression squeezing each call below 64 kbit/s, it carried about 40,000 calls at once: ten times the copper cable before it.

Stop 6 of 6

Into a thread of glass

Light wiggles at hundreds of terahertz: room for a huge number of calls. The laser was invented in 1960 and gave a strong, narrow beam. Now the light needed a pipe.

Glass can trap light. When light inside a dense material hits the edge at a shallow enough angle, it can't get out. It bounces back in, every time. This is total internal reflection. Tilt the beam and find where it starts to leak.

Inside / outside

1.43 / 1

Steepest beam that stays in

45.6°

The numbers are each material's refractive index: how much it slows light down. The bigger the gap between inside and outside, the steeper a beam can be and still stay trapped.

Why fiber has two kinds of glass

In air, a finger or a speck of dust on the glass lets light leak out. So a fiber puts its core inside a second glass, the cladding, with a slightly lower index. The light stays in the core even when the fiber is coated, bent and buried. The gap is tiny, so only beams within a few degrees of the line stay in. A laser beam is narrow enough.

The critical angle comes from Snell's law: sin θc = n₂ / n₁.

The leaky sand truck

The real problem was not light escaping out the sides. Glass itself soaked up the light. Engineers measure this loss in decibels per kilometre.

Picture a leaky truck of sand. Lose 20 dB over a kilometre and 100 kg becomes 1 kg. In the 1960s the best glass lost 1,000 dB/km. Pick a glass and drive.

Light left

1 part in 10^100

Distance until 1% is left

20 m

The fix: make the glass ridiculously pure

Charles Kao showed in 1966 that the loss came from tiny bits of metal, like iron, in the glass, not the glass itself. In 1970 a team at Corning made ultra-pure fiber at 17 dB/km. Today's fiber loses about 0.2 dB/km, so light still has 1% left after 100 km. Kao won the Nobel Prize in 2009.

The whole trip

Your voice shakes a diaphragm. The phone keeps only the pitches that carry words. The signal becomes 64,000 bits a second, takes turns with thousands of other calls, and rides a laser through glass purer than anything that existed in 1965. Then it all runs backwards in your friend's ear.

Inspired by Veritasium's The Absurd Engineering Behind the Internet.