Description

Real-world UWB setups often have NLoS links. These links introduce time-of-flight errors, leading to inaccurate TDoA estimations. Machine learning methodologies enable us to understand the environment, forecast, and rectify the TDoA estimation, ensuring an accurate resulting TDoA.

NOTE: Transfer learning for new environments supported with few samples.

Advantages

1. Improved accuracy
2. Resilient against NLOS

input: T1, T2, CIR1, CIR2, FP_index1, FP_index2, etc.(other channel information)
output: corrected_tdoa

Supported languages

1. Python / Tensorflow
2. C++ / Tensorflow Lite for Microcontrollers (TinyML)

Supported hardware

Machine learning model inference can be executed on the ultra-Wideband (UWB) node, an edge device, or within the cloud. The training process must be conducted on a robust edge device or within a cloud environment.