独立的 Iridium 卫星突发检测器和解调器 (C)
A standalone Iridium satellite burst detector and demodulator written in C. It provides an alternative to gr-iridium by eliminating the GNU Radio dependency, while producing the same iridium-toolkit compatible RAW output on stdout. For users who want a lighter-weight, dependency-free option or need embedded deployment, this offers similar functionality with a different architectural approach.
Supports HackRF, BladeRF, USRP (UHD), SDRplay (native API), and SoapySDR for live capture, or processes IQ recordings from file. Optional GPU-accelerated burst detection is available via OpenCL (NVIDIA, AMD, Intel) as a runtime plugin -- the main binary works on systems without GPU support. Runs on Raspberry Pi 5 and other ARM boards in CPU-only mode with FFTW wisdom pre-generation.
A built-in web map (--web, beta) provides a real-time Leaflet.js visualization of decoded ring alert positions and active satellites -- no external tools or Python required.
Built-in ACARS/SBD decoding (--acars) extracts aviation messages directly from IDA frames. When libacars-2 is installed, ARINC-622 application payloads (ADS-C, CPDLC, OHMA) are fully decoded -- no Python pipeline needed.
Native GSMTAP output (--gsmtap) sends decoded IDA (Iridium Data) frames directly to Wireshark via UDP, eliminating the need for the Python iridium-parser.py -m gsmtap pipeline.
--acars) with optional libacars-2 ARINC-622/ADS-C/CPDLC support--parsed) for direct reassembler.py piping (ACARS/SBD recovery)--gsmtap) for IDA frame analysis--position)--simd)--zmq) for multi-consumer iridium-toolkit compatibilityDragonOS Noble ships with HackRF, BladeRF, USRP (UHD), SoapySDR, and OpenCL drivers pre-installed. Just clone and build:
git clone https://github.com/alphafox02/iridium-sniffer.git
cd iridium-sniffer
mkdir build && cd build
cmake ..
make -j$(nproc)CMake auto-detects the available SDR libraries, GPU support, and libacars. All SDR backends, OpenCL GPU acceleration, and ACARS ARINC-622 decoding should be enabled automatically.
…CMake output shows what was detected:
-- HackRF: enabled
-- BladeRF: enabled
-- USRP (UHD): enabled
-- SoapySDR: enabled
-- libacars: enabled (ARINC-622/ADS-C/CPDLC decoding)
-- ZMQ: enabled (multi-consumer PUB/SUB output)
-- GPU acceleration: OpenCLThe downmix worker count auto-scales based on CPU cores: 4 workers on 8+ cores, 2 on 4 cores, 1 on 2 cores. On low-power CPUs like the Intel N100 (4 cores), this prevents 100% CPU saturation that occurs when 4 workers compete for 4 cores. Override with --workers=N if needed.
The Pi 5's VideoCore VII GPU passes basic Vulkan compute tests but cannot sustain the throughput needed for real-time FFT batch processing. Build CPU-only and use --no-gpu:
git clone https://github.com/alphafox02/iridium-sniffer.git
cd iridium-sniffer
sudo apt install build-essential cmake libfftw3-dev libsoapysdr-dev
mkdir build && cd build
cmake .. -DUSE_OPENCL=OFF
make -j$(nproc)FFTW wisdom (important for ARM): FFTW uses FFTW_MEASURE to benchmark FFT algorithms at plan creation time. On x86 this is fast and unnoticeable. On ARM it can block for 30-60+ seconds per plan, causing q_max to climb during live capture as samples queue up while plans are being built.
Pre-generate a wisdom file to avoid this. iridium-sniffer automatically loads wisdom from ~/.iridium-sniffer-fftw-wisdom at startup and saves updated wisdom on shutdown. After the first successful run (or the command below), subsequent starts are immediate.
The required wisdom entries depend on sample rate. The burst detection FFT size varies, while the downmix FFTs are always the same (cof4096 for CFO estimation, cof2048/cob2048 for correlation):
| Sample Rate | Burst FFT | Wisdom Command |
|---|---|---|
| 2-2.4 MHz (RTL-SDR) | 2048 | fftwf-wisdom -v -o ~/.iridium-sniffer-fftw-wisdom cof2048 cof4096 cob2048 |
| 6 MHz (Airspy Mini) | 8192 | fftwf-wisdom -v -o ~/.iridium-sniffer-fftw-wisdom cof8192 cof4096 cof2048 cob2048 |
| 10 MHz (default) | 8192 | fftwf-wisdom -v -o ~/.iridium-sniffer-fftw-wisdom cof8192 cof4096 cof2048 cob2048 |
| 12 MHz (extended) | 16384 | fftwf-wisdom -v -o ~/.iridium-sniffer-fftw-wisdom cof16384 cof4096 cof2048 cob2048 |
The naming convention: cof = complex forward, cob = complex backward, followed by the FFT size. If running multiple sample rates on the same system, include all burst FFT sizes in one command (e.g., cof2048 cof8192 cof4096 cob2048). The "system-wisdom import failed" warning from fftwf-wisdom is normal on a fresh system and can be ignored.
GNU Radio manages FFTW wisdom automatically (in ~/.gr_fftw_wisdom), which is why gr-iridium users never encounter this issue. Since iridium-sniffer replaces the GNU Radio dependency, it handles wisdom directly.
# OpenCL GPU (default when available)
cmake .. -DUSE_OPENCL=ON
# Vulkan GPU
cmake .. -DUSE_VULKAN=ON -DUSE_OPENCL=OFF
# CPU only
cmake .. -DUSE_OPENCL=OFF
# Debug build with AddressSanitizer
cmake .. -DCMAKE_BUILD_TYPE=Debug…Tested against gr-iridium on a 60-second IQ recording (cf32, 10 MHz, 1622 MHz center, USRP B210):
Default threshold (16 dB) -- maximum frame recovery:
| Metric | iridium-sniffer | gr-iridium |
|---|---|---|
| Detected bursts | 5468 | ~3666 |
| Demodulated RAW frames | 3701 | 2713 |
| Ok rate | 68% | 74% |
IDA frames (internal --parsed) |
743 | -- |
| IDA frames (external iridium-parser.py) | 373 | 690 |
The default 16 dB threshold detects more bursts than gr-iridium, including weaker signals at the noise floor. Many of these marginal bursts fail demodulation, which lowers the ok percentage -- but the absolute frame count is 36% higher (3701 vs 2713). This is the recommended setting for maximum data recovery.
Matched threshold (18 dB) -- apples-to-apples comparison:
| Metric | iridium-sniffer | gr-iridium |
|---|---|---|
| Detected bursts | 3668 | ~3666 |
| Demodulated RAW frames | 2737 | 2713 |
| Ok rate | 75% | 74% |
IDA frames (internal --parsed) |
605 | -- |
| IDA frames (external iridium-parser.py) | 361 | 690 |
At 18 dB (gr-iridium's default), burst detection counts are nearly identical. The ok rate now matches gr-iridium at 75% vs 74%. The external parser IDA gap (361 vs 690) reflects that gr-iridium's GNU Radio-based demodulator produces cleaner bits -- more frames survive standard BCH correction. Use --threshold=18 if ok rate percentage is more important than total frame count.
A note on live ok% rates: In live SDR capture, you may see ok_avg of 35-50% with iridium-sniffer compared to 70-80% shown in gr-iridium guides. This is expected and not a problem. iridium-sniffer uses a lower default detection threshold (16 dB vs gr-iridium's 18 dB), which catches more weak bursts at the noise floor. These marginal detections lower the ok percentage but increase the total number of successfully decoded frames. The ok% statistic measures what fraction of detected bursts decode -- not how many frames you are actually recovering. What matters is decoded frames per second, and iridium-sniffer typically recovers more usable data than gr-iridium despite the lower ok% figure.
x64 Processing speed (60s cf32 file, i7-11800H):
| Configuration | Wall time | CPU time | Realtime factor |
|---|---|---|---|
| AVX2 + GPU | 15.1s | 23.6s | 4.0x |
| AVX2 only | 12.0s | 21.5s | 5.0x |
| SSE4.2 only | - | - | ~3.5x est. |
| Scalar + GPU | 16.1s | 42.6s | 3.7x |
| Scalar only (baseline) | 13.0s | 40.6s | 4.6x |
ARM NEON performance (Apple M3 Max, 12 GB 10-min ci8 recording):
| Configuration | CPU Time (user) | Wall Time | Realtime Factor |
|---|---|---|---|
| NEON (default on AArch64) | 647.8s | 2:31.98 | 3.9x |
| Scalar | 995.2s | 3:58.02 | 2.5x |
Four SIMD tiers are available: AVX2+FMA (256-bit, x86), SSE4.2 (128-bit, x86), NEON (128-bit, AArch64), and scalar. On x86 the tier is selected at runtime via CPUID; on AArch64 NEON is the default. Use --simd=MODE to force a specific path (auto/avx2/sse42/neon/scalar) for testing or debugging. AVX2 provides ~1.9x CPU time reduction over scalar; SSE4.2 provides ~1.5x (helps on older Intel Celeron/Atom); NEON provides ~1.54x on Apple Silicon. GPU acceleration adds startup overhead for files this size but becomes beneficial for longer recordings and continuous live capture.
All SIMD configurations produce identical demodulated output (frame count, bit content). GPU vs CPU may differ by a few frames due to floating-point rounding in the burst detection FFT.
The IDA decoder uses BCH(31,20) t=2 hard-decision error correction, identical to iridium-toolkit's bch_repair(). Standard BCH corrects up to 2 bit errors per 31-bit block. An experimental Chase soft-decision extension is available via --chase=N (N=1..7), which uses per-bit amplitude scores from the demodulator to attempt recovery of blocks with more than 2 errors on IDA frames only (IRA/IBC Chase is completely disabled). Chase recovery is off by default.
The --web flag starts an embedded HTTP server that decodes IRA (ring alert) and IBC (broadcast) frames in real time and displays them on a map. This provides similar functionality to Iridium Live without any external dependencies.
Note: The web map feature is currently in beta. Position plotting and satellite tracking are functional but undergoing validation.
# Default port 8888
./iridium-sniffer -i soapy-0 --web
# Custom port
./iridium-sniffer -i soapy-0 --web=9090Then open http://localhost:8888 in a browser.
The map shows:
--position is active).All layers are individually toggleable via the layer control panel.
Data updates once per second via Server-Sent Eve
暂无开放 Issues,或尚未同步最近议题。