Short name: Fed4AI
Long name: Experimental validation of AI models for the prediction of wireless links performance
Company: ALLBESMART LDA | allbesmart.pt
SUMMARY REMARKS & TESTBEDS
Accurate prediction of wireless performance links can be very useful to optimize radio resource allocations. However, the vast variety of possible wireless configurations and propagation scenarios make it notoriously hard to design explicit\theoretical models to forecast the performance of wireless links.
The main objective of Stage 1 of the Fed4AI experiment was the experimental validation of Machine Learning (ML) algorithms for predicting the throughput of WiFi radio links. We build implicit “black box” models using real-world measurements and we tested them systematically, by asking them to predict the throughput for WiFi links in configurations that have never been observed during the initial measurement phase. In this Stage 2 of the Fed4AI experiment, we will extend the work done in Stage 1 to the vehicular communications domain including: LTE-A for the long-range/cellular component and ITS-G5/IEEE 802.11p for the short-range communications component.
The experiment requires the combination of two Fed4FIRE+ testbeds: PerformLTE (UMA) and the Smart Highway which is part of the CityLab testbed (imec). The work in Fed4AI-Stage 2 is structured in three main objectives:
- Objective 1: Lab test of Machine Learning models to forecast the capacity of LTE radio links without active/intrusive transmission – Testbed: PerformLTE (UMA)
- Objective 2: Field test of Machine Learning models to forecast LTE connectivity gaps for the long-range/cellular component of vehicular communications – Testbed: CityLab/Smart Highway (imec)
- Objective 3: Field test of Machine Learning models to forecast the latency of V2X radio links for ITS G5/IEEE802.11p – Testbed: CityLab/Smart Highway (imec)
The integration of these advanced algorithms with ALLBESMART’s network analytics framework UXPERT, constitutes an undeniable competitive advantage over our competitors, transforming UXPERT from reactive to proactive with significative added value and business impact.