gnss-signal-simulator-rs

🌐 English · Русский

Multi-constellation GNSS IF signal simulator written in Rust

Генерация IQ для GPS, Galileo, BeiDou и GLONASS из RINEX 3.04. Основной путь использует CPU, Rayon и SIMD через wide; CUDA kernels пока не подключены к потоковому синтезу.

Актуальный статус: трекер, контракт генерации, матрица приёмников. Исторические z-score ниже подтверждают acquisition в конкретном тесте; они не доказывают корректность nav/CRC/PVT всех перечисленных сигналов. Выбор эфемерид выполняется отдельно для спутника/семейства сообщения, видимость обновляется раз в секунду. IQ сопровождается TXT и SHA-256 manifest.

---

Supported Signals

System	Signal	Frequency	Modulation	Code Length	Verified z-score
GPS	L1CA	1575.42 MHz	BPSK(1)	1023	58–88
GPS	L5	1176.45 MHz	BPSK(10)	10230	44–59
GPS	L2C	1227.60 MHz	BPSK(1)	10230	Verified
GPS	L1C	1575.42 MHz	TMBOC / BOC(1,1)	10230	Verified
Galileo	E1	1575.42 MHz	CBOC(6,1,1/11)	4092	63–103
Galileo	E5a	1176.45 MHz	BPSK(10)	10230	48–64
Galileo	E5b	1207.14 MHz	BPSK(10)	10230	449–514
Galileo	E6	1278.75 MHz	BPSK(5)	5115	270–299
BeiDou	B1C	1575.42 MHz	BOC(1,1)+QMBOC(6,1,4/33)	10230	132–247
BeiDou	B1I	1561.098 MHz	BPSK(2)	2046	65–96
BeiDou	B2a	1176.45 MHz	BPSK(10)	10230	48–72
GLONASS	G1	1602+k×0.5625 MHz	BPSK(0.5) FDMA	511	54–56

z-score = (correlation peak − mean) / std, threshold ≥ 30. Values above 50 indicate reliable detection.

Key Features

• RINEX 3.04 ephemeris parsing — GPS (7-line), BeiDou (7-line with BDT correction), Galileo (7-line), GLONASS (3-line with km→m)
• Keplerian orbit propagation — GPS ICD-200 algorithm; GLONASS RK4 in PZ-90
• Multithreading — Rayon-based parallel satellite processing
• IQ8 output — signed 8-bit I/Q interleaved samples, compatible with SDR receivers (GNURadio, SDR#)
• Python verification tool — 3-page PDF report with PSD, acquisition, correlation analysis
• Minimal dependencies — only rand, rayon, wide, serde_json, sha2 (+ optional cudarc)

Quick Start

Requirements

• Stable Rust (edition 2021) and cargo
• Рекомендуется x86_64; .cargo/config.toml использует target-cpu=native, такой бинарь непереносим на более старые CPU.
• Экспериментальная GPU-фича: CUDA 12.5; основной генератор остаётся CPU.

Build and Run

# Build optimized release
cargo build --release

# Generate GPS L1CA signal (simplest example)
cargo run --release -- presets/gps_l1ca.json

# Generate triple-system GPS + BeiDou + Galileo
cargo run --release -- presets/gps_bds_gal_l1.json

# Generate quad-system GPS + BeiDou + Galileo + GLONASS (46.5 MHz)
cargo run --release -- presets/quad_l1g1.json

# Verbose output
cargo run --release -- -v presets/gps_l1ca.json

Output: binary IQ file (e.g. generated_files/gps_l1ca.C8).

Build with GPU (optional)

cargo build --release --features gpu
cargo run --release --features gpu -- presets/gps_l1ca.json

Экспериментальная сборка требует CUDA 12.5/NVRTC. Она не включает GPU в основной потоковый синтез; runtime CUDA здесь не проверен.

Presets

Ready-to-use JSON configurations in presets/:

Preset	Systems	Sample Rate
gps_l1ca.json	GPS L1CA	5 MHz
gps_l5.json	GPS L5	21 MHz
bds_b1c.json	BeiDou B1C	5 MHz
gal_e1.json	Galileo E1	5 MHz
gal_e5a.json	Galileo E5a	21 MHz
gal_e5b.json	Galileo E5b	21 MHz
gal_e6.json	Galileo E6	11 MHz
glo_g1.json	GLONASS G1	10 MHz
gps_bds_gal_l1.json	GPS+BDS+GAL L1	5 MHz
gps_bds_gal_l1_300s.json	GPS+BDS+GAL L1 (over-air, 52 dBHz)	8 MHz
quad_l1g1.json	GPS+BDS+GAL+GLO L1/G1	46.5 MHz

The repository ships 46 presets. Most use the Montana location, 10s duration, elevation mask 5°, full RINEX; the moscow_* and usa_* presets are longer real-receiver scenarios (60–300s) at their respective locations.

Custom presets: copy any JSON, edit receiver position (LLA), sample rate, RINEX path, duration, and signal selection.

Over-the-Air Replay (HackRF / PortaPack → real receiver)

When replaying a generated file over the air into a real receiver (phone, u-blox), the defaults tuned for the software verifier are too weak. For a phone to acquire and sustain tracking (decode ephemeris → get a fix):

• initPower ≥ 52 dBHz in the preset. The AGC normalizes the file to a fixed output RMS regardless of initPower, so this sets the in-file CN0 (signal-to-noise margin), not the loudness. The default 45 dBHz acquires in the verifier but a phone cannot hold tracking.
• Sample rate ≥ 8 MHz. HackRF One does 2–20 Msps but <8 MHz is out of DAC spec; PortaPack replay is SD-throughput-limited, so ~8–10 Msps is the practical sweet spot.
• E1/B1C use pure BOC(1,1) below 12.276 MHz automatically. The true CBOC/QMBOC BOC(6,1) subcarrier (6.138 MHz) aliases into the data channel at low sample rates and blocks nav-message decode on real receivers; the generator falls back to clean BOC(1,1) (which a phone's narrow L1 front-end uses anyway) and switches to true CBOC only at Fs ≥ 12.276 MHz.
• BeiDou on phones is B1I (1561.098 MHz), not B1C (1575.42 MHz). B1I is outside an L1-centered file, so use a dedicated B1I preset for BeiDou on a phone.

presets/gps_bds_gal_l1_300s.json is preconfigured for this (8 MHz, 52 dBHz, 300 s).

Signal Verification

The included Python tool verify_signal_enhanced.py generates a 3-page PDF diagnostic report.

Usage

# Install dependencies
pip install numpy matplotlib

# Verify with auto-configuration from preset
python verify_signal_enhanced.py generated_files/gps_bds_gal_l1.C8 \
    --preset presets/gps_bds_gal_l1.json

# Fast mode (search only RINEX-visible satellites)
python verify_signal_enhanced.py generated_files/gps_l1ca.C8 \
    --preset presets/gps_l1ca.json --fast

# Custom parameters
python verify_signal_enhanced.py generated_files/output.C8 \
    --sample-rate 5.0 --threshold 30 --output report.pdf

PDF Report Contents

• Page 1 — Signal Overview: Welch PSD, I/Q histogram with Gaussian fit, constellation diagram, RMS stability
• Page 2 — Acquisition Results: z-score bar chart (color-coded by system), polar skyplot (expected vs detected), CN0 estimation, Doppler accuracy
• Page 3 — Correlation Analysis: Zoomed correlation peaks per system, 2D Doppler×Code heatmaps

Verification Results

Triple-system (GPS + BeiDou + Galileo, 5 MHz, 10s, Chicago):

System	RINEX Visible	Detected	z-score Range
GPS L1CA	11	11	58–88
BeiDou B1C	8	6	574–955
Galileo E1	10	6	364–425
Total	29	23	100% of generated

Galileo multi-band (Montana, 10s each):

Signal	Sample Rate	Detected	z-score Range
E5a	21 MHz	7/7	434–502
E5b	21 MHz	7/7	449–514
E6	11 MHz	7/7	270–299

Single-system results after the June 2026 passes (true CBOC/QMBOC waveforms + NH-robust verifier):

Signal	Sample Rate	RINEX Visible	Detected	z-score Range
Galileo E1 (CBOC)	5 MHz	10	10/10	63–103
BeiDou B1C (QMBOC, pilot acquisition)	5 MHz	12	12/12	132–247
BeiDou B1I	5 MHz	12	12/12	65–96
GPS L5	21 MHz	10	10/10	44–59
Galileo E5a	21 MHz	10	10/10	48–64
BeiDou B2a	21 MHz	12	12/12	48–72

PRN codes cross-verified bit-exact against independent references: B1I vs gnss-sdr (63/63 PRN); B1C data/pilot/secondary, E5a I/Q, B2a data/pilot and all NH/secondary codes vs PocketSDR. GPS L5 XA/XB is ICD-literal (free-running XB; the references' mod-10230 roll diverges in the last adv chips per PRN, <0.12 dB effect).

The verifier uses double-window linear correlation (immune to NH/secondary chip flips at code epochs) with code-Doppler drift compensation — the earlier 0/10 GPS L5 result was a verifier artifact, confirmed by a clean-room dump showing ideal per-period correlation and bit-exact NH wiring.

GLONASS G1 (10 MHz, Montana, 10s):

Generated	Detected	z-score Range
7 SVs	5	54–56

Architecture

src/
├── main.rs              # Entry point — reads JSON preset, runs generation
├── lib.rs               # Public module exports
├── ifdatagen.rs         # Core IF data generation (AGC, satellite signals, noise)
├── sat_if_signal.rs     # Per-satellite IF sample generation (PRN codes, BOC)
├── coordinate.rs        # Keplerian propagator, ECEF/LLA/ENU conversions
├── json_interpreter.rs  # RINEX 3.04 parser, JSON preset parser
├── gnsstime.rs          # GPS/BDT/GST/GLONASS time system conversions
├── types.rs             # Core types: GpsEphemeris, BeiDouEphemeris, KinematicInfo
├── constants.rs         # WGS84, PZ-90, CGCS2000 constants
├── prngenerate.rs       # PRN code generation (Gold, Weil/Legendre, memory codes)
├── memory_code_e1.rs    # Galileo E1 memory codes (4092 chips)
├── memory_code_e6.rs    # Galileo E6 memory codes (5115 chips)
├── fastmath.rs          # Optimized sin/cos/atan2 approximations
├── complex_number.rs    # Lightweight complex arithmetic
├── satellite_param.rs   # Satellite parameters and signal configuration
├── satellite_signal.rs  # Signal-level satellite processing
├── almanac.rs           # Almanac parsing and type detection
├── trajectory.rs        # Receiver trajectory generation
├── *navbit.rs           # Navigation message generators (GPS LNAV, Galileo I/NAV,
│                        #   F/NAV, BeiDou BCNav1/2/3, GLONASS G-NAV, etc.)
├── nav_decode.rs        # Reverse decoders + ICD parity/CRC verifiers for nav messages
├── gps_pilot/mod.rs     # Clean minimal phase-continuous GPS L1CA generator
└── bin/
    ├── spectrum_analyzer.rs  # IF spectrum analysis (PSD, peaks, CSV export)
    ├── nav_diagnostics.rs    # Nav message encode→decode→compare vs RINEX (ICD verify)
    ├── decode_iq.rs          # GPS L1CA IQ8 decoder — acquisition, tracking, LNAV decode
    ├── gnss_pilot.rs         # Clean GPS L1CA + Galileo E1 + BeiDou B1C L1-band generator
    ├── gps_pilot.rs          # Clean minimal GPS L1CA generator
    ├── nav_test.rs           # Navigation bit unit tests
    ├── bench.rs              # CPU/AVX-512/GPU benchmarks
    └── extreme_bench.rs      # Stress benchmarks

RINEX Data

• Rinex_Data/rinex_v3_20251560000.rnx — 2025-06-05, GPS/BDS/GAL
• Rinex_Data/BRDC00IGS_R_20251560000_01D_MN.rnx — 2025-06-05, full merged GPS/BDS/GAL/GLO

Output Format

IQ8: signed 8-bit interleaved I/Q samples (.C8 extension)

[I₀, Q₀, I₁, Q₁, I₂, Q₂, ...]   — each sample is int8 (-128..+127)

• AGC target RMS: 0.25 (quantized: ~31.75 in int8 units)
• Compatible with GNURadio (file_source → char_to_float), SDR#, and custom SDR receivers
• File size: 2 × sample_rate × duration bytes (e.g. 5 MHz × 10s = 100 MB)

Spectrum Analyzer

Built-in IF spectrum analysis tool:

cargo run --release --bin spectrum_analyzer -- generated_files/gps_l1ca.C8 iq8 5000000 0 --csv

Outputs PSD to terminal and optionally to spectrum.csv.

Development

cargo check          # Fast syntax/type check
cargo test           # Run unit tests
cargo fmt --all      # Format code
cargo clippy         # Lint

Detailed verification status is tracked in docs/verification_report.md and docs/gnss_icd_conformance_report.md.

Troubleshooting

Problem	Solution
"Configuration file not found"	Use presets from presets/ or edit generated config.json
"RINEX file not found"	Check relative paths in preset, ensure files exist in Rinex_Data/
CUDA not found	Install CUDA 12.5, verify NVRTC in PATH/LD_LIBRARY_PATH
Slow generation	Use --release, use --features gpu
0 visible satellites	Check RINEX date matches preset time, verify receiver coordinates

License

See source files for license information.

---

Описание (Русский)

Мультисистемный симулятор GNSS-сигналов на Rust

Генерация реалистичных IQ-отсчётов промежуточной частоты для GPS, Galileo, BeiDou и GLONASS из навигационных данных RINEX 3.04.

Поддерживаемые сигналы

• GPS: L1CA, L5, L2C, L1C
• Galileo: E1 (CBOC), E5a, E5b, E6
• BeiDou: B1C (BOC+QMBOC), B1I, B2a
• GLONASS: G1 (FDMA)

Быстрый старт

# Сборка
cargo build --release

# Генерация GPS L1CA
cargo run --release -- presets/gps_l1ca.json

# Тройная система GPS + BeiDou + Galileo
cargo run --release -- presets/gps_bds_gal_l1.json

# Верификация сигнала (Python)
python verify_signal_enhanced.py generated_files/gps_l1ca.C8 \
    --preset presets/gps_l1ca.json

Основные возможности

• 4 навигационные системы, 10 сигналов, одновременная генерация
• Парсинг эфемерид RINEX 3.04 (GPS/BDS/GAL/GLO)
• Кеплеровская пропагация орбит + ГЛОНАСС RK4 в ПЗ-90
• Физически корректная модель АРУ (AGC)
• Ускорение AVX-512 + многопоточность (Rayon)
• Экспериментальные CUDA kernels (не основной путь генерации)
• Выход IQ8 (8-бит I/Q), совместимый с SDR-приёмниками
• Инструмент верификации с 3-страничным PDF-отчётом
• ~55 000 строк Rust, минимум зависимостей