A Digital Radio Mondiale (DRM30) receiver and transmitter written in Rust.

DecDRM decodes DRM robustness modes A–D in every channel bandwidth (4.5–20 kHz) from recordings or a live sound card (for example a virtual audio cable fed by a web SDR such as a KiwiSDR), and generates DRM signals from a station description. The signal processing is a spec-first reimplementation of ETSI ES 201 980 that borrows proven algorithms from Dream; audio uses Fraunhofer FDK-AAC (AAC, HE-AAC v1/v2, xHE-AAC decoding; AAC/HE-AAC encoding), libxaac (xHE-AAC encoding) and libopus.

Receiving a Deutsche Welle recording (mode B, 10 kHz, 64-QAM): HE-AAC v2 stereo audio, text messages, the broadcast clock and Journaline news pages.

Transmitting live into a virtual audio cable: an xHE-AAC stereo service with text messages, a slideshow and a programme guide in the more strongly protected part A (outlined), plus a Journaline news service; live input and output meters and the output spectrum on the right.

Status: the receiver decodes audio, text messages and data services from every DRM test recording available here (modes A/B/C, 9–20 kHz, real and I/Q inputs, inverted spectra, clock offsets up to 1250 ppm, AAC/HE-AAC/xHE-AAC/Opus). The transmitter produces complete multiplexes that the receiver decodes. See docs/DESIGN.md for the plan and milestone status.

• Sensitivity close to the theoretical limit. MSC bit error rate 10⁻⁴ after decoding (64-QAM, R = 0.6) at 14.8 dB SNR in AWGN (annex A's ideal receiver: 14.9 dB) and within 1.6 dB of ETSI ES 201 980 annex A's ideal-receiver figures on the DRM fading channels 2–5 — with real synchronisation and channel estimation (cargo run --release -p decdrm-core --example bercurve -- 1 A 14.6 14.8 15).
• Automatic frequency, robustness-mode, spectrum-occupancy and spectrum-inversion detection; no need to set the IF or flip the spectrum by hand.
• Sample-rate-offset estimation that is robust on fading channels (cross-correlation of impulse-response snapshots, lag-compensated tracking); pilot-slope acquisition for clock errors beyond 1000 ppm.
• Wiener channel estimation in time and frequency with Doppler and delay-spread adaptation, impulse-response based timing tracking, iterative multilevel decoding.
• Services: AAC, HE-AAC v1/v2, xHE-AAC, Opus (Dream's extension), text messages, MOT slideshow, Broadcast Website, Journaline, EPG, TPEG/raw capture (--data-dir), broadcast time, alternative frequencies.
• Transmitter: up to four services with AAC/HE-AAC/xHE-AAC/Opus audio, text, slideshow, website, Journaline, EPG, TPEG and raw data, alternative frequencies, every MSC mode incl. hierarchical 64-QAM and unequal protection, programme audio from a file, a sound card, a test tone or an internet radio stream (Icecast/SHOUTCAST over HTTP/HTTPS, MP3/AAC/Ogg/FLAC, its titles as text messages), WAV/FLAC or sound-card output (clock-drift compensated with a sound-card or stream input), plus a channel simulator (the DRM channel models 1–6, noise, frequency and clock offsets: [simulate] or decdrm tx … --channel-model 3 --snr 18) for testing receivers.
• Desktop GUI (egui): spectrum, waterfall, constellations, audio spectrum, channel, impulse response, SNR per carrier, reception history, and two displays Dream lacks: a fading map (channel gain per carrier over time) and a delay–Doppler map (each propagation path at its delay and Doppler shift), status LEDs, Dream-style service bars (codec, SBR/PS, bit rate, protection, data applications), text, slideshow, Journaline browser, broadcast website, EPG, broadcast clock, alternative frequencies, recording of the audio (WAV/FLAC), and a transmitter tab whose Journaline pages can be edited while on the air.
• KiwiSDR client: DecDRM tunes a KiwiSDR on the internet and decodes its I/Q directly (decdrm rx --kiwi HOST --freq KHZ; in the GUI a KiwiSDR source with a list of the public Kiwis whose owners allow apps, and a double-click in the Schedule tab). It respects the Kiwis' limits: busy, password or app-limited Kiwis are not retried, and sessions the Kiwi ends are not reconnected.
• Diversity reception, which Dream lacks: one station through two KiwiSDRs far apart, combined cell by cell before decoding (maximum-ratio combining weighted by each one's SNR; frames paired by content, so the Kiwis' different network delays and clocks do not matter). On simulated fading channels it decodes almost every frame where each KiwiSDR alone loses most of them (decdrm rx --kiwi A --kiwi2 B --freq KHZ; in the GUI a 2nd KiwiSDR field).
• MDI and RSCI (ETSI TS 102 820, 102 349 over DCP, TS 102 821), with Dream parity: the receiver decodes MDI or RSCI from UDP (multicast too) or recordings (Dream's .rsX, pcap, pcapng), with PFT fragments rebuilt by Reed–Solomon. It shows an RSCI receiver's status, spectrum and impulse response, retunes it by RCI, and accepts RCI itself. The transmitter works as a modulator for MDI from a content server ([mdi] in the station file), filling lost frames so the signal never stops.
• Station schedule, like Dream's Stations dialog: the DRM broadcasts on the air now from EiBi's or Dream's schedule (downloaded on request), in the GUI's Schedule tab and with decdrm schedule; a frequency in a recording's file name (KiwiSDR …_6140.00_iq.wav) picks out the station.
• Light on the CPU: the receiver decodes 10 kHz signals at 100–150× real time (20 kHz at ~45×) on one core; the GUI needs a few percent of a core while decoding live.
• DAC (the Descript Audio Codec, a neural codec) as a DecDRM-only audio codec: 1.5–24 kbit/s, CRC-protected layers and concealment — at 15 dB SNR it lost 1.4 % of audio frames where HE-AAC lost 37 %. On broadcast audio and speech, DAC at 3 kbit/s came about as close to the original as EnCodec (the neural codec of DecDRM 0.4.6 and earlier) at 12. It needs a reasonably modern CPU: decoding takes a quarter of real time on a 16-thread desktop, 0.83 on one core. Standard receivers (and Dream) ignore it.

git clone --recursive https://github.com/CasualArclamp/DecDRM.git
cd DecDRM
cargo build --release

Requirements: Rust 1.88 or newer (1.95 for the GUI), a C/C++ compiler (MSVC Build Tools on Windows, gcc on Linux) and CMake for the vendored codecs, and on Linux the ALSA development package (libasound2-dev).

Portable Windows executables: decdrm-gui.exe and decdrm.exe that run on any 64-bit Windows 10/11 with nothing installed (C runtime linked statically, DAC with its weights built in, about 320 MB each) are attached to the releases, or built into exe\ with powershell -ExecutionPolicy Bypass -File scripts\build-portable.ps1 (after decdrm models download dac).

The user guide covers receiving from a KiwiSDR directly or from other web SDRs through a virtual audio cable, the displays, data services, logs, the station file and troubleshooting.

# decode a recording (real IF / audio input at any sample rate)
decdrm rx recording.flac

# I/Q recording; save the audio, the data objects and a reception log
decdrm rx iq_recording.wav --format iq --out audio.wav --data-dir data --log rx.csv

# live from a sound-card input, e.g. a virtual audio cable, with playback
decdrm devices
decdrm rx --device "CABLE-A Output" --play

# straight from a KiwiSDR on the internet (DecDRM tunes it and takes its I/Q)
decdrm rx --kiwi kiwisdr.example.org --freq 6140 --play

# a station whose encoder switches its SBR band (the treble) on and off frame by frame
decdrm rx --kiwi kiwisdr.example.org --freq 13835 --play --smooth-sbr

# which DRM stations are on the air now (--update first downloads EiBi's schedule)
decdrm schedule --update

# transmit: the station file describes services, codecs, data and the output
decdrm tx crates/decdrm-station/examples/station.toml --check     # show the multiplex
decdrm tx crates/decdrm-station/examples/station.toml --duration 30 --output drm.wav

# desktop GUI
cargo run --release -p decdrm-gui -- recording.flac

Korean Central Broadcasting (6140 kHz) sends EVS speech inside a data service. DecDRM recognises it and shows it as EVS audio, but does not decode it (see the user guide).

DAC (optional, pulls in the candle ML library):

cargo build --release -p decdrm-cli --features dac
decdrm models download dac          # ~299 MB weights, SHA-256 checked
# station.toml: [service.audio] codec = "dac"   (optional: bandwidth_kbps = 6)

--log writes one metrics row per second of signal (SNR, MER, Doppler, delay, clock offset, FAC/SDC/MSC/audio counters, service) as CSV, or JSON Lines (.jsonl) with text messages, data objects and log lines as events.

GPL-2.0-or-later (it derives from Dream, which is GPL); the licence text is in LICENSE. The vendored libxaac (xHE-AAC encoder) is Apache-2.0, which is compatible with GPLv3 but not GPLv2, so binaries that include it are distributed under GPL-3.0; its LICENSE and NOTICE must ship with them. The vendored FDK-AAC library has its own licence (see third_party/fdk-aac/NOTICE), which is generally regarded as incompatible with the GPL: fine for personal use, but check before distributing binaries. AAC, xHE-AAC (MPEG-D USAC) and DRM technologies are covered by patents licensed through Via LA.