Why High‑Precision GPS Still Struggles in Urban Canyons and Tunnels

In September 2013 the national navigation white paper highlighted China’s Beidou system, which provides regional positioning with 10 m accuracy and, when combined with indoor signals, can achieve sub‑meter outdoor and sub‑3 m indoor precision.

1. Plain‑language overview of differential GPS

Typical GPS errors (satellite ephemeris, troposphere, ionosphere, multipath) can be mitigated by using a reference station that measures the same satellite signals simultaneously; the difference between observed and true values is broadcast to nearby receivers, allowing them to correct their positions in real time.

Assume positioning error is mainly caused by local environmental factors.

Therefore, a small area shares similar error characteristics.

If a baseline station knows the true satellite ranges, the instantaneous error can be computed.

Broadcasting this error to all receivers in the area lets them cancel the error and achieve high precision.

Real‑world differential techniques include simple “black‑box” corrections, pseudorange differencing, and carrier‑phase differencing.

2. Conditions required for high‑precision positioning

Consistency of local signal bias.

Nationwide deployment of sufficient baseline stations.

Real‑time communication of error data from stations to mobile receivers.

Receiver firmware capable of applying fine‑grained adjustments.

In practice, the first and fourth conditions are the most difficult to satisfy.

Urban high‑rise areas suffer severe multipath effects, causing errors of dozens to hundreds of meters, which breaks the assumption of local bias consistency.

Mobile device size also limits the integration of high‑precision hardware, making true decimeter‑level positioning on smartphones currently unrealistic.

3. Scenario‑specific solutions

Bridge (over‑/under‑pass) problem

Using satellite‑derived altitude to distinguish bridge levels fails because height errors are larger than planar errors, and multipath under bridges produces random altitude offsets.

Barometric sensors can detect the ~10 m pressure difference between bridge levels, but atmospheric variations and sensor drift make absolute pressure unreliable.

Instead, detecting signal blockage patterns (e.g., reduced satellite count, abnormal geometry) can help filter out bridge‑under scenarios, reducing the problem to planar positioning.

Tunnel problem

GPS signals are unavailable inside tunnels; possible mitigations include deploying Wi‑Fi/iBeacon beacons or using vehicle‑dead‑reckoning (VDR) that fuses inertial sensor data with the last known GPS fix.

Smartphone VDR is limited by sensor noise, but combining VDR estimates with map‑matching (projecting the trajectory onto known tunnel geometry) can improve accuracy.

Main/sub‑road identification

High‑precision differential GPS could resolve parallel lanes, but without it, accurate road maps (≤1 m error) or aerial imagery can help.

Gyroscope and accelerometer data can detect lane‑change maneuvers; machine‑learning classifiers (SVM, decision trees, neural networks) can infer whether a vehicle is on a main or side road based on speed, steering, and deceleration patterns.

4. Vision of ultimate navigation

The author envisions a future where navigation relies on computer vision combined with high‑precision positioning, despite challenges such as continuous camera usage, power consumption, and limited mobile compute resources.