Three private boxes sit on top of public radio and charge rent: Nielsen & Arbitron's PPM watermark, and Gracenote's disc identity — which started life as CDDB, a database built by the public and then fenced off and sold. None of it is magic. Here is exactly how each one works, in the open — including a calculator that reproduces Gracenote's disc ID bit for bit, no database of theirs required.
This is how a pager-sized meter "hears" a station: not by recognizing the music, but by reading an inaudible code hidden inside the program audio itself — slipped in before broadcast, tucked under the ear's masking threshold so you never hear it, but a microphone across the room can.
| Carrier band | 1000–3000 Hz (inside the voice/music range, masked by it) |
| Code bands | 10 parallel frequency bands |
| Tones per band | 18 — 16 data tones (4 bits/symbol) + 1 SYNC + 1 STOP |
| Symbol size | 4 bits (8 bits/sec) |
| Message | station/distributor identity + timestamp, repeated 12–13×/min, updated once per minute |
| Why it survives | all 10 bands carry the same message; each bit uses a different tone per band, so the code survives EQ, noise, and a phone mic across the room |
When you put a CD in, the "who is this?" lookup starts with a disc ID.
Gracenote treats it as their crown jewel. It isn't — it's a plain arithmetic function of the disc's track list
(XXYYYYZZ: a checksum of track start-times, the total seconds, the track count). CDDB was community
data; the enclosure was of people's unpaid work, not of any real secret. Type a disc's table of contents and watch
the exact same ID fall out:
The open path back: the community re-forked the last free CDDB as gnudb, and MusicBrainz rebuilt the whole thing under an open license — the same identity, owned by no one. Machine-readable: /rdns/reverse/cddb
The third box is the audience estimate — and we already took it apart. The metric definitions (AQH, Cume, TSL, Share) are Arbitron's own published arithmetic; the only thing paywalled is the thin panel behind them. See the Census Book for the panel exposed in Nielsen's own numbers, and a census computed from real ears instead.