Is artificial intelligence the key to unlocking the multipath problem in GNSS?

Graphics and correlation functions for no multipath, line-of-sight multipath, and non-line-of-sight multipath.

GNSS navigation can be significantly degraded by buildings and other surroundings that reflect and obstruct GNSS signals. The arrival of reflected signals alongside the line of sight (LOS) signal is known as multipath.

The error introduced by multipath is difficult to predict or mitigate, and when unaccounted for, can introduce significant errors to positioning, navigation, and timing (PNT) solutions. Many multipath mitigation techniques exist that can reduce the effect of, but not eliminate, multipath error. These techniques alter the correlator and discriminator used in the delay lock loop of a GNSS receiver and include the Narrow Correlator, Pulse Aperture Correlator, Vision Correlator, and Multipath Estimation and Correction technique.

A GNSS receiver will perceive a correlation function that is a composite of direct and reflected signals. These reflected signals can be parameterised based on their amplitude, carrier phase, and propagation delay relative to the direct signal. Each of these parameters will affect the degree to which the correlation function is distorted.

A convolutional neural network (CNN) was developed to explore its effectiveness for not only detecting measurements degraded by multipath, but also to explore how the detection results can be leveraged to improve the positioning performance in high multipath environments. The network was designed to detect measurements that have multipath range error, based on the shape of the correlation function.

The approaches were tested in real urban environments where signal reception conditions were extremely poor, reducing the mean error by up to 90%. This work shows the promise of AI for the detection and mitigation of metre-level pseudorange multipath error and for improving the ability of GNSS receivers to operate in environments with high volumes of non-line of sight and multipath signals.

What you will learn

Written by Christian Phillips, geomatics designer, Hexagon’s Autonomous Solutions Business Area, this article explores how AI can be leveraged to detect and reduce the impact of multipath on GNSS positioning. Techniques for the detection and exclusion of multipath measurements, and for the weighting of measurements for a weighted least squares (WLS) positioning solution using AI are presented.

Example architecture of a convolutional neural network.

Figure showing an example architecture of a convolutional neural network.

Read the full article by downloading Velocity 2025



Questions? Contact our sales team


Contact Sales
Find Dealer