Question 1

How do I use this project with my own mmWave data?

Accepted Answer

You need to generate beamformed fingerprints using a ray-tracing simulator like Wireless InSite, preprocess the data with the provided C tools, and modify the YAML configuration files to run training and testing scripts, which involves significant setup and simulation expertise.

Question 2

What's better for tracking: LSTM or TCN in this system?

Accepted Answer

Both LSTM and TCN are supported for tracking; LSTM is better for capturing long-term dependencies in sequential data, while TCN uses convolutional layers for efficiency. The choice depends on your specific accuracy and computational needs, as shown in the example configurations.

Question 3

Can this work for indoor positioning instead of outdoor?

Accepted Answer

No, the system is designed and tested for outdoor urban scenarios using specific propagation models and 3D maps. Indoor environments have different multipath characteristics, so it would require major modifications and likely not perform as well without retraining.

Question 4

What hardware do I need to run this locally?

Accepted Answer

You need an Nvidia GPU with Compute Capability 3.5 or higher, at least 16GB of RAM, Python 3.6+, TensorFlow 2.4, and CUDA 11.0. For embedded evaluation, an Nvidia Jetson TX2 is required, as per the documentation.

Question 5

How does this compare to GPS for accuracy and power?

Accepted Answer

It achieves similar or better accuracy (1.78-3.30 meters) in urban areas but is 47x to 85x more energy-efficient than A-GPS. However, it relies on 5G mmWave coverage, which is less widespread than GPS signals, limiting deployment scenarios.

Question 6

Is the dataset available for other cities besides New York?

Accepted Answer

Currently, the dataset is specific to a New York City map; the author mentions emailing for other 3D files, but generating new data requires access to ray-tracing simulators and high-precision 3D maps, which may not be readily available.

mmWave-localization-learning

What is mmWave-localization-learning?

Overview

Use Cases

Best For

Related Projects

Found a gem we're missing?

Not Ideal For

Pros & Cons

Pros

Cons

Open Source Alternative To

Frequently Asked Questions