Gneiss is a tightly-coupled GNSS and Inertial Navigation System (INS) engine written in Rust. It fuses raw satellite observations (pseudorange, carrier phase, and Doppler) with high-rate inertial sensors (accelerometer and gyroscope) using an Extended Kalman Filter (EKF).

The engine is designed for robust operation in multi-path environments, utilizing empirical statistical methods for outlier rejection and variance scaling.

• Tightly-Coupled Integration: Direct fusion of raw GNSS observations and IMU data within the primary state vector.
• Adaptive Estimation: Implements Innovation-based Adaptive Estimation (IAE) and Median Absolute Deviation (MAD) RAIM to scale observation variances dynamically.
• Ambiguity Resolution: Uses the LAMBDA (Least-squares AMBiguity Decorrelation Adjustment) algorithm for carrier-phase integer ambiguity resolution.
• Post-Processing (PPK): Supports forward-backward Rauch-Tung-Striebel (RTS) smoothing to produce continuous trajectories from static files.
• Real-Time Streaming: #![no_std] compatible core engine, with a tokio-based live subcommand for streaming data via local UART and NTRIP casters.

The engine's architecture provides a mathematical scaffold for multiple GNSS processing modes. Currently, the project is heavily focused on optimizing RTK + INS workflows for urban canyon and challenging multipath environments.

Supported Modes:

• Single Point Positioning (SPP)
• Real-Time Kinematic (RTK)
• RTK + INS (Tightly-Coupled)
• Precise Point Positioning (PPP)
• PPP + INS

Features Included:

• Ionosphere-Free Linear Combinations for multi-frequency correction.
• Zenith Wet Delay (ZWD) tropospheric estimation.
• Geophysical corrections (Solid Earth Tides, Satellite Phase Wind-Up).
• CDDIS SP3 and precise clock (.clk) parsing via 10th-order Lagrange interpolation.
• Clock Jump State Preservation algorithms to prevent EKF divergence during TCXO adjustments.

The engine operates on a causal, recursive filtering architecture:

graph LR
    A[Raw Satellite Data] --> B(Gneiss Engine)
    C[Raw Inertial Data] --> B
    B --> D{Extended Kalman Filter}
    D -->|Float State| H[LAMBDA Ambiguity Resolution]
    H -->|Fixed Ambiguities| I{FFRT Validation}
    I -->|Pass| J[Apply Fix & Hold]
    I -->|Fail| D
    D --> E[Position Trajectory]
    D --> F[Calibrated Sensor Biases]
    D --> G[Attitude & Heading]

Gneiss provides a command-line interface for testing and processing datasets. To run a tightly-coupled fusion pipeline with backward smoothing on a static dataset:

cargo run --release -p gneiss-cli -- process \
    --rover datasets/my_data/rover.obs \
    --base datasets/my_data/base.obs \
    --output trajectory.pos \
    --enable-imu-fusion \
    --enable-backward-smoothing \
    --lever-arm "0.1,0.0,-0.2"

To run the engine in real-time streaming mode using a serial port and an NTRIP base station:

cargo run --release -p gneiss-cli -- live \
    --port /dev/ttyACM0 \
    --baud 460800 \
    --ntrip-url rtk2go.com \
    --ntrip-mount MOUNT \
    --ntrip-user USER \
    --ntrip-pass PASS

• --rover <PATH>: (Required) Path to the rover's raw GNSS observations (RINEX .obs).

• --base <PATH>: Path to the base station's raw GNSS observations (RINEX .obs). Required for RTK/PPK. If omitted, the engine defaults to SPP (Single Point Positioning).

• --output <PATH>: (Required) Path for the resulting .pos trajectory file.

• --config <PATH>: Path to a JSON configuration file to override default EKF process noise and tuning parameters.

• --enable-backward-smoothing: Enables the Rauch-Tung-Striebel (RTS) backward smoother.

• --lambda-ratio <FLOAT>: Minimum ratio for the LAMBDA Partial Ambiguity Resolution (PAR) test (default: 3.0).

• --lambda-subset <INT>: Minimum number of satellites required for PAR (default: 7).

• --lever-arm <X,Y,Z>: Translation vector (in meters) from the IMU center of navigation to the GNSS antenna phase center in the vehicle body frame.

• --calibrate-imu: Enables state-estimation of IMU mounting rotations (Roll, Pitch, Yaw) relative to the vehicle frame.

• --raim-outlier-m <FLOAT>: SPP Receiver Autonomous Integrity Monitoring (RAIM) threshold in meters.

• --chi-square-pr <FLOAT>: EKF Chi-Square threshold for pseudorange measurement rejection.

• --chi-square-cp <FLOAT>: EKF Chi-Square threshold for carrier phase measurement rejection.

For technical implementation details, see the following documents:

• Architecture Details
• Benchmark Methodology