IMU-Assisted RSSI Vector Prediction under Android Wi-Fi Scan Throttling

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


MPhil Thesis Defence


Title: "IMU-Assisted RSSI Vector Prediction under Android Wi-Fi Scan Throttling"

By

Mr. Adriel Abraham INTAN


Abstract:

Android has deployed Wi-Fi scan throttling since 2019 (which limits foreground
Wi-Fi scans to at most four scans in two minutes), adversely impacting the
user experience of many real-time applications such as localization and network
handover. Although many prediction models for Wi-Fi RSSI (Received Signal
Strength Indicator) vectors have been proposed, their performance degrades
significantly under scan throttling due to the long interval between scans. We
propose WIPI, a novel, effective, and light-weight Wi-Fi RSSI vector predictor
assisted by IMU (Inertial Measurement Unit) and trained with Wi-Fi and IMU
data crowdsourced using an Android app. To support the training, we present a
customized scanning strategy for crowdsourcing and a pooler to adaptively
clean and aggregate the crowdsourced data. We then design a transformer-based
prediction model with a weighted loss function to effectively predict RSSI
vectors in short intervals (e.g., once every few seconds). We evaluate WIPI on
complex indoor environments totaling more than 7,000 m^2 and show that it
predicts RSSI vectors with substantially lower MAE (by 52%) and higher F1-score
in RSSI vector completeness (by 0.25) than state-of-the-art approaches.


Date:                   Thursday, 23 April 2026

Time:                   2:00pm - 4:00pm

Venue:                  Room 2132C
                        Lifts 22

Chairman:               Prof. Raymond WONG

Committee Members:      Prof. Gary CHAN (Supervisor)
                        Dr. Chaojian LI