How sound carries bits

A modem changes a sound to carry data. Signal Lab offers three custom-data modulations. Choose a signal in the encoder and match it on the receiver; transmission settings offer more options. All use the same packet and checksum; older Signal Lab versions understand BFSK only.

Frequency or phase?

BFSK uses a low tone for 0 and a high tone for 1. DBPSK keeps one carrier frequency and encodes each bit in the change in phase from one symbol to the next. DQPSK uses four possible phase changes, carrying two bits per symbol. In ordinary BPSK the two states have opposite carrier phases; DBPSK instead compares adjacent symbols, so it does not need a shared starting phase.

At the same bit rate, DQPSK uses half as many symbols per second as BFSK or DBPSK; it is not twice as fast at a fixed bit rate. Its four phase choices are closer together, so noise and echoes can make them harder to distinguish. Phase modes are experimental: start with a short message and a quiet room. Results depend on the devices and environment.

The buttons below demonstrate BFSK’s two pitches. Both phase modes can use an audible 1.8 kHz carrier or an experimental 20.5 kHz carrier. High-frequency signals may be less audible to many people, but are not guaranteed silent: modulation and equipment can produce audible components, and some people and animals hear higher frequencies. Some speakers, microphones and audio routes cannot reproduce them reliably. Start at a low volume and match the carrier at both ends. Read about differential phase detection ↗

Tap either bit to hear its tone for half a second.

Why do emoji work?

Text becomes UTF-8 bytes, which become bits. Emoji use several bytes, so the byte count can be larger than the character count. Files already consist of bytes; their type does not matter to the modem.

Speed is a trade-off.

At 300 bit/s, BFSK and DBPSK symbols last about 3.33 milliseconds; DQPSK symbols last about 6.67 milliseconds and carry two bits each. At 100 bit/s, symbols last three times as long. A slower transfer takes longer but can be easier to decode. File sizes matter: 8 KiB takes several minutes even before repeated copies.

What actually travels?

The preamble gives the receiver time to find the bit clock. The sync word marks a packet. The header declares the transfer ID, total copies, and copy number. Metadata describes the payload; the final checksum checks the whole frame.

Received is not always verified.

The custom protocol includes a transfer ID, a total copy count (1, 2, or 3), and a copy index in its opening header. That header has its own CRC-32 checksum, so the receiver can read the declared plan before the payload finishes arriving.

Each copy has to earn its check.

First, its full frame must pass CRC-32, an error-detection checksum. Next, its metadata and payload bytes must match the other copies of that transfer. Only distinct copy indices count; the receiver’s multiple timing candidates cannot accidentally count as extra copies.

1×: one checksum-valid copy. 2×: two checksum-valid, matching copies. 3×: three checksum-valid, matching copies. A missing copy leaves the transfer incomplete. A differing copy produces a conflict. Valid bytes may still be downloaded from an incomplete transfer, with that status shown.

What a checksum cannot tell you.

CRC-32 detects accidental corruption; it does not authenticate the sender or encrypt the message. Repetition adds independent consistency checks, but neither repetition nor a checksum proves who sent a message. This is a one-way experiment, without automatic acknowledgements or retries.

SAME is different.

SAME’s existing format sends three header bursts and has no CRC-32 field. Signal Lab counts physical bursts and compares the complete parsed header. It reports matching copies and missing bursts separately, without adding custom fields to SAME audio.

A small header with a very specific job.

The U.S. Emergency Alert System (EAS) uses Specific Area Message Encoding (SAME) to label alerts: who originated them, what kind of event they describe, where they apply, and how long they remain valid. The spoken announcement is a separate audio segment.

Read a header.

Location codes are not ZIP codes.

The six-digit SAME location code uses PSSCCC. SS and CCC are FIPS state/equivalent territory and county/equivalent area identifiers. FIPS expands to Federal Information Processing Standards. P adds a county-partition selector to that five-digit identifier.

A leading 0 covers all or unspecified portions. Non-zero values select locally defined partitions, conventionally numbered 1–9 from northwest to southeast. These are not Census county-subdivision codes; boundaries and which values are valid depend on local alerting plans. The encoder’s place search uses a bundled NOAA directory and explains every digit of your selected codes.

The sound of SAME.

The data rate is about 520.83 bit/s. A 0 uses 1,562.5 Hz; a 1 uses 2,083⅓ Hz. Each burst begins with sixteen 0xAB preamble bytes. Bytes travel least significant bit first. Identical headers are sent three times, with one-second gaps between bursts; the same repetition is used for end markers.

The optional EAS attention signal combines 853 and 960 Hz. NOAA Weather Radio uses a 1,050 Hz attention tone. They share SAME metadata encoding but use different attention signals.

Signal Lab is an educational instrument, independent of emergency alert networks. Its SAME decoder is experimental and does not transcribe speech. In the U.S., 47 CFR § 11.45 restricts transmission of EAS codes and attention signals, including simulations and recordings. A disclaimer does not authorize their use. Review the FCC rule before playback or distribution.

Read the rhythm.

Morse keys a single tone on and off. In radio this is often called CW (continuous wave). Signal Lab uses a 700 Hz audio tone: a dot is a short mark; a dash lasts three dots. The tables below use International Morse characters from ITU-R M.1677-1.

Every silence counts.

Marks within a letter have a one-unit gap. Letters have three units between them; words have seven. In the written preview, spaces separate letters and / separates words. A literal slash in your text is itself encoded as −··−·.

Character tables.

Tap a character to hear it at 15 WPM. WPM means words per minute, conventionally based on the 50-unit word PARIS. At 15 WPM, a dot is 80 milliseconds. These tables cover this app’s supported characters, including É and punctuation.

Try it across the room.

1. Choose Encoder → Morse and start with HELLO at 15 WPM.
2. On the receiver, set Morse receive speed → 15 WPM, start the microphone and wait until it is ready.
3. Transmit. Letters appear as gaps arrive. A longer quiet gap ends the message; the microphone keeps listening.

The decoder targets a 700 Hz tone at the selected speed. It is experimental; irregular hand keying, other pitches, interference and mismatched speed can change letters. Unknown patterns appear as �. A decoded result has no checksum; review the text. For files, emoji or checked copies, use Custom data.

Keep learning.

Start close. Keep it short.

• Open the app on two devices. Choose Decoder → Start microphone on one; allow microphone access.
• On the other, choose Encoder, type a short message, and transmit at moderate volume.
• Keep devices nearby and the receiving tab in the foreground. Loud playback can distort, while quiet playback can disappear into background noise.
• If reception fails, use Resilient (100 bit/s). Try moving the devices closer and reducing room noise.
• For a controlled experiment, use Test in decoder to decode the generated waveform silently. Custom data and Morse can also be exported as WAV and imported in Decoder.

Playback volume on iPhone.

On supporting browsers, Signal Lab declares media playback so transmitting can use media volume even with Ring/Silent enabled. Before playback, iPhone may show ringtone volume; that indicator alone does not prove media playback is muted. Adjust your device’s volume while the signal plays; Signal Lab uses your system media volume, with no separate app volume control. During microphone use, the app requests a recording-capable session. Older or embedded browsers may still impose their own audio routing.

Microphone access.

Microphone capture requires a secure origin (HTTPS or localhost) and permission. Embedded browsers may impose extra limits. The receiver shows the actual input level in dBFS and whether it detects custom or SAME tone energy. The waveform automatically zooms quiet audio for visibility; this does not amplify the audio sent to the decoder. Diagnostics also show the audio-context state, capture engine, processing settings, input device, and sample rate.

Signal Lab first uses AudioWorklet, loaded from its bundled source. If that is blocked, it can use a compatibility capture engine. The microphone is never played back through the speaker, avoiding a monitoring feedback loop. Listening continues between messages and after verified or incomplete results. Each transfer has its own log entry. While listening, the app requests screen wake lock where available; iOS can still interrupt background or locked-screen capture. Stop closes the microphone’s dedicated audio context and releases the wake lock.

Sensitivity and startup.

Wait for Microphone ready before transmitting. The first custom copy begins with about 1.8 seconds of training tones so the input and tone-level estimates can settle. Repeated copies use a shorter 0.32–0.64 second lead. Normal sensitivity allows one damaged sync bit after a recognizable preamble; Higher lowers the detection floor and allows two. Lower requires exact sync. These settings change acquisition, not the CRC check, and cannot recover audio that was never captured.

Reading an incomplete result.

A result marked 1 / 3 means the header declared three copies and only one checksum-valid copy arrived. Each copy slot names its outcome: Not detected, Header only, CRC failed, or Verified. A copy can be audible yet fail data checks. The app will not silently call that a fully verified transfer. Import processing reports incomplete results at end of file; live capture also watches the expected transfer window.

Import supports recordings up to five minutes and 50 MB; uncompressed WAV is the most predictable format. No audio or payload is uploaded by this app. Receptions stay in this tab’s memory until cleared or reloaded.