5.0 KiB
modemdab
A native DAB/DAB+ receiver (ETSI EN 300 401, transmission mode I), used by the DAB Demodulator plugin.
What it does
From complex baseband at 2.048 MS/s it produces:
- The ensemble: services, labels, PTy, language, user applications, and date and time.
- Audio: of the selected service, as 16-bit stereo PCM. DAB+ (HE-AAC v2) is decoded by FAAD2 and DAB (MPEG Layer II) by FFmpeg's libavcodec; without a codec, the service is still found, but no audio is produced.
- Data: the dynamic label with DL Plus tags, MOT SlideShow images and Journaline, from X-PAD or a packet mode component.
- Service and Programme Information (SPI): station names, descriptions and logos, and programme schedules, from the ensemble's SPI data service (MOT directory mode), whatever service is selected.
- Announcements: traffic, news, alarm and other announcements (FIG 0/18
and 0/19). The selected service's audio switches to them, for the types
chosen (
setAnnouncementTypes()); alarms always. - Service following: where a service can also be received: other ensembles (FIG 0/24), and linked DAB, FM and DRM services (FIG 0/6), with their frequencies (FIG 0/21).
- TII: the identifiers of the transmitters received.
- Statistics: including the channel bit error rate (estimated by re-encoding the Viterbi output), the carrier to noise ratio from the null symbol (overall and per carrier), and the MER of the demodulated symbols.
| Files | Responsibility |
|---|---|
dabtables |
Mode I constants, the phase reference, frequency interleaving, puncturing, UEP/EEP profiles, PRBS, CRCs, character sets, PTy and language names |
dabofdm |
DC offset removal, channel filter, null symbol and phase reference acquisition, carrier and clock tracking, FFT window placement from the impulse response (searching again within 0.3 s when input samples are lost), differential and three-symbol demodulation, and soft bits weighted by each carrier's noise (from the null symbol) |
dabconvolutional |
Mother code, depuncturing and the 64-state soft Viterbi decoder (shared with DRM, in modemdr), with an optional thresholded soft output (SOVA) |
dabensemble |
FIC decoding and the FIG database, with current and next configurations; announcements, service linking, frequency information and other ensembles' services |
dabmsc |
Sub-channel time deinterleaving and decoding; packet mode |
dabaudioframing |
DAB+ superframes (Fire code, RS(120,110), AUs) and Layer II frames (header CRC) |
dabaudiodecoder |
FAAD2 and FFmpeg backends with concealment |
dabpad |
F-PAD/X-PAD, the dynamic label, DL Plus and MSC data groups |
dabtii |
TII detection |
dabcoherent |
Decoder aided (near coherent) demodulation, from the whole multiplex decoded and re-encoded |
dabreceiver |
The receiver: service selection, detection, callbacks, status and an optional decoding thread |
dabfft |
FFTs: FFTW when built in, kissfft otherwise |
dabprofile |
Time per stage, with MODEMDAB_PROFILE set |
MOT (slideshow, Broadcast Website and SPI files), SPI, Journaline, the CRCs, inflate, the
character sets and the Viterbi decoder are shared with modemdrm, in modemdr.
Performance
One DAB+ service with its data takes about 2.3% of a core (44x real time) on
the development machine. The Viterbi decoder has SSE2 (x86) and NEON (AArch64)
forward passes (the scalar pass, for other architectures, has the same butterfly form;
MODEMDR_SCALAR or MODEMDAB_SCALAR selects it, to compare).
setThreaded(true) moves decoding to a second thread. Callbacks are still
made in the caller's thread, from process().
Detection
setDetection() chooses how the carriers are demodulated:
| Detection | AWGN threshold | CPU | Delay |
|---|---|---|---|
ThreeSymbol (multiple symbol differential, the default) |
5.1 dB | 3.2% | - |
Coherent (three-symbol, then decoder aided) |
4.4 dB | 8-9% | about 0.4 s more |
Differential (for comparison) |
5.3 dB | 2.3% | - |
The thresholds are for a DAB+ service at EEP 3-A in a full multiplex. CPU is for the 11B recording, a multiplex of 14 sub-channels (coherent detection takes more in weak signals: 13-14% at 6 dB). In fading, coherent detection is about 1 dB better near the threshold at 60 km/h, and about equal at 120 km/h.
Coherent decodes every sub-channel, re-encodes it, and demodulates the
frame again with references averaged over many symbols:
- Unreliable bits: bits that the Viterbi decoder's soft output marks
unreliable are not used (
setReliabilityThreshold()). - Fallback: where too few neighbours are known for the references to beat differential detection, the three-symbol soft bits are kept. In weak signals few decoded bits are reliable, so this keeps it from being worse than three-symbol.
- More passes:
setIterations()adds passes. Two gain about 0.2 dB in AWGN but lose about 0.3 dB in a static echo, at twice the cost.
The detection can be changed while running; the audio restarts. The DAB
Demodulator's Coherent button selects between ThreeSymbol and Coherent.