1. Inertial Navigation Systems
1.1 Fundamentals of Inertial Navigation
Inertial Navigation Systems (INS) provide self-contained navigation by integrating measurements from
accelerometers and gyroscopes. The fundamental principle is dead reckoning: measuring acceleration and
rotation to compute position and orientation changes over time.
Core Principle: An INS measures specific force (acceleration minus gravity) and angular
velocity, then integrates these measurements to compute velocity, position, and attitude.
1.2 Navigation Equations
The mechanization equations in the navigation frame (NED - North-East-Down) are:
$$\dot{\mathbf{v}}^n = \mathbf{C}_b^n \mathbf{f}^b - (2\boldsymbol{\omega}_{ie}^n + \boldsymbol{\omega}_{en}^n) \times \mathbf{v}^n + \mathbf{g}^n$$
$$\dot{\mathbf{p}}^n = \mathbf{v}^n$$
$$\dot{\mathbf{C}}_b^n = \mathbf{C}_b^n [\boldsymbol{\omega}_{ib}^b \times]$$
Where:
- \(\mathbf{v}^n\): Velocity in navigation frame
- \(\mathbf{p}^n\): Position in navigation frame
- \(\mathbf{C}_b^n\): Direction cosine matrix (body to navigation)
- \(\mathbf{f}^b\): Specific force measured by accelerometers
- \(\boldsymbol{\omega}_{ib}^b\): Angular velocity measured by gyros
- \(\mathbf{g}^n\): Gravity vector
1.3 Error Propagation
INS errors grow unbounded without external aiding. Position error grows approximately linearly with distance
traveled due to velocity errors, which themselves accumulate from attitude and sensor bias errors.
$$\sigma_{pos}(t) \approx \sigma_{vel,0} \cdot t + \frac{1}{2}\sigma_{accel} \cdot t^2 + \sigma_{attitude} \cdot g \cdot t^2$$
For practical underwater navigation:
$$\text{Position Drift} \approx 0.1\% \text{ to } 2\% \text{ of distance traveled}$$
2. Acoustic Positioning Systems
2.1 Long Baseline (LBL)
LBL systems use a network of seafloor transponders to provide absolute position fixes. The vehicle ranges
to multiple transponders and computes position via trilateration.
$$r_i = ||\mathbf{p}_{vehicle} - \mathbf{p}_{transponder,i}|| + n_i$$
Where \(r_i\) is measured range and \(n_i \sim \mathcal{N}(0, \sigma_r^2)\) is measurement noise.
Position is found by minimizing:
$$\min_{\mathbf{p}} \sum_{i=1}^{N} (r_i - ||\mathbf{p} - \mathbf{p}_i||)^2$$
2.2 Geometric Dilution of Precision (GDOP)
GDOP quantifies how transponder geometry affects positioning accuracy. It multiplies measurement error to
give position error:
$$\sigma_{position} = GDOP \times \sigma_{range}$$
$$GDOP = \sqrt{trace[(H^T H)^{-1}]}$$
Where \(H\) is the observation matrix with rows: \(H_i = (\mathbf{p} - \mathbf{p}_i)^T / r_i\)
2.3 Ultra-Short Baseline (USBL)
USBL systems use a compact transceiver array on a surface vessel to track underwater vehicles. Range is
measured via travel time, and bearing via phase interferometry.
USBL Accuracy: Typically 1-2% of slant range. Bearing accuracy improves with longer
baselines: \(\sigma_\theta \approx \lambda / (2\pi d)\) where \(\lambda\) is wavelength and \(d\) is baseline.
3. Sensor Fusion & Kalman Filtering
3.1 Extended Kalman Filter (EKF)
The EKF fuses INS, DVL, LBL, and depth measurements to provide optimal state estimates. The state vector
typically includes position, velocity, attitude, and sensor biases.
State vector: \(\mathbf{x} = [\mathbf{p}, \mathbf{v}, \boldsymbol{\psi}, \mathbf{b}_g, \mathbf{b}_a]^T\)
Prediction:
$$\hat{\mathbf{x}}^- = f(\hat{\mathbf{x}}^+, \mathbf{u})$$
$$\mathbf{P}^- = \mathbf{F}\mathbf{P}^+\mathbf{F}^T + \mathbf{Q}$$
Update:
$$\mathbf{K} = \mathbf{P}^-\mathbf{H}^T(\mathbf{H}\mathbf{P}^-\mathbf{H}^T + \mathbf{R})^{-1}$$
$$\hat{\mathbf{x}}^+ = \hat{\mathbf{x}}^- + \mathbf{K}(\mathbf{z} - h(\hat{\mathbf{x}}^-))$$
$$\mathbf{P}^+ = (\mathbf{I} - \mathbf{K}\mathbf{H})\mathbf{P}^-$$
3.2 Process and Measurement Noise
Proper tuning of \(\mathbf{Q}\) (process noise) and \(\mathbf{R}\) (measurement noise) is critical for
filter performance:
- Q Matrix: Represents model uncertainty and unmodeled dynamics. Larger Q means filter
trusts measurements more.
- R Matrix: Represents sensor measurement noise. Should match actual sensor statistics.
- Tuning Rule: Start conservative (large Q and R), then reduce iteratively while
monitoring innovation sequence and consistency.
4. Error Propagation Models
4.1 Dead Reckoning Error Growth
Without external aiding, position uncertainty grows according to sensor characteristics:
For a mission of duration \(T\) and distance \(D\):
$$\sigma_{DR}(T) = \sqrt{(\sigma_0)^2 + (k_{gyro} \cdot D)^2 + (k_{accel} \cdot T^2)^2}$$
Where:
- \(\sigma_0\): Initial position uncertainty
- \(k_{gyro}\): Gyro-induced drift rate (% of distance)
- \(k_{accel}\): Accelerometer-induced drift
4.2 DVL-Aided Navigation
Doppler Velocity Log (DVL) provides velocity updates that bound error growth. The bounded error is:
$$\sigma_{DVL}(t) = \sqrt{\sigma_{DVL}^2 \cdot \Delta t_{DVL} + (\sigma_{INS} \cdot \Delta t_{DVL})^2}$$
Typical performance: Position error < 0.2% of distance with 5 Hz DVL updates
4.3 LBL-Aided Navigation
LBL fixes reset absolute position uncertainty. Between fixes, uncertainty grows according to INS drift:
$$\sigma(t) = \begin{cases}
\sigma_{LBL} & \text{at fix} \\
\sqrt{\sigma_{LBL}^2 + (k_{drift} \cdot d(t))^2} & \text{between fixes}
\end{cases}$$
5. Coordinate Systems & Transformations
5.1 Common Reference Frames
- ECEF (Earth-Centered Earth-Fixed): Origin at Earth's center, rotates with Earth
- NED (North-East-Down): Local level frame, navigation computations
- Body Frame: Fixed to vehicle, sensor measurements
- Geographic (LLA): Latitude, Longitude, Altitude
5.2 Rotation Representations
Attitude can be represented using:
Direction Cosine Matrix (DCM):
$$\mathbf{C}_b^n = \begin{bmatrix}
c_\psi c_\theta & c_\psi s_\theta s_\phi - s_\psi c_\phi & c_\psi s_\theta c_\phi + s_\psi s_\phi \\
s_\psi c_\theta & s_\psi s_\theta s_\phi + c_\psi c_\phi & s_\psi s_\theta c_\phi - c_\psi s_\phi \\
-s_\theta & c_\theta s_\phi & c_\theta c_\phi
\end{bmatrix}$$
Quaternions: \(\mathbf{q} = [q_0, q_1, q_2, q_3]^T\), \(||\mathbf{q}|| = 1\)
5.3 Gravity Modeling
Accurate gravity modeling is essential for INS. The WGS84 ellipsoidal gravity model:
$$g(\phi, h) = g_e \cdot \frac{1 + k \sin^2\phi}{\sqrt{1 - e^2\sin^2\phi}} - 2g_e \cdot \frac{h}{a}$$
Where \(g_e = 9.7803267714\) m/s², \(k = 0.00193185138639\), \(e^2 = 0.00669437999013\)
Further Reading
- Groves, P. D. (2013). Principles of GNSS, Inertial, and Multisensor Integrated Navigation Systems
- Titterton, D. H., & Weston, J. L. (2004). Strapdown Inertial Navigation Technology
- Kinsey, J. C., & Whitcomb, L. L. (2007). Underwater Vehicle Navigation: Recent Advances
- Farrell, J. A. (2008). Aided Navigation: GPS with High Rate Sensors