Experimental Evaluation of Urban Points-of-Interest as Predictors of I2V 802.11 Data Transfers

Smart Cities will leverage the Internet-of-Things (IoT) paradigm to enable cyber-physical loops over urban processes. Vehicular backhauls contribute to IoT platforms by allowing sensor/actuator nodes near roads to explore opportunistic connections to passing vehicles when other communication backhauls are unavailable. A placement process of nodes that includes vehicular networks as a connectivity backhaul requires estimates of infrastructure-to-vehicle (I2V) wireless service at potential deployment sites. However, carrying out I2V measurement campaigns at all potential locations can be very expensive; so, predictive models are necessary. To this end, qualitative characteristics of a potential site, such as infrastructural points-of-interest (POI) relating to traffic (i.e., traffic lights, crosswalks) and fleet activities (i.e., bus stops, garbage bins) can inform about the vehicles’ mobility patterns and quality of the I2V service. In this paper, we show the contribution of POI (and site-specific information) to I2V transfers, leveraging a real-world dataset of geo-referenced I2V WiFi link measurements in urban settings. We present the distributions of throughput with respect to distance per POI class and site, and apply exponential regression to obtain practical throughput/distance models. We then use these models to compare I2V transfer estimation methodologies with different levels of POI-specific data and data resolution. We observe that I2V transfer estimate accuracy can improve from an average over-estimation of 18.3% with respect to measured values, if site or POI-specific information metrics are not used, to 9.3% in case such information is used.