Insights on Mini-Batch Alignment for WiFi-CSI Data Domain Factor Independent Feature Extraction

Unobtrusive sensing has the ambition to embed sensing into our daily lives. A way to achieve it is through repurposing technology that we are already used to having in our environments. Wireless Fidelity (WiFi) sensing which makes use of Channel State Information (CSI) measurement data seems to be a perfect fit for this, since WiFi networks are already omnipresent. A big challenge is that CSI data is very sensitive to ‘domain factors’ such as position and orientation, making its interpretation in different domains very difficult. We present a domain factor independent feature extraction pipeline called ‘mini-batch alignment’. Its goal is to develop models with domain factor independent latent representations. Unfortunately, based on extensive evaluations on a gesture recognition benchmark dataset, the proposed mini-batch alignment pipeline did not lead to better inference performance. We discuss the pitfalls that may have led to this result, as well as future research directions.