IPSN19 杨景 2019.4.12

Background Success of Internet of Things (IoT) depend highly on reliable communication of billions devices LoRa (and LPWAN) is at forefront of commercial interest long coverage, simplifying deployment and management, etc.

Problem and Motivation Performance of LoRa is intrinsically dependent on the environment Channel models and empirical evidence are still largely lacking in the literature Challenge traverse multiple types of environments spatial extension and sequence are highly variable in the large area

Goal Model LoRa links and estimate their quality in a simple and low-cost way use remote sensing techniques to obtain environment-specific information in a low-cost manner 长距离 遥感 覆盖信息

Remote Sensing exploiting the propagation and reflection properties of electromagnetic waves the target scene is illuminated by a source of electromagnetic radiation sensors mounted on satellites, airplanes or UAVs, measure the radiation reflected by the objects in the scene enabling the automatic extraction of detailed information over a large- scale area

LoRa Link in Theory high sensitivity: −140 dBm free-space path loss (FSPL) model upper bounds 1400km in practice ≥30km the range of LoRa links can vary by a factor of 100 or more unique to low-power long-range links Wifi -90dbm 10^5 weaker than lora Factor 5

The Baseline: Open-space Scenarios Prx expected received power (RSSI) Ptx transmission power Gtx , Grx are transmitting and receiving antenna gains FSPL free-space path loss Expected Signal Power (ESP) to obtain the energy of the signal 141 TTN gateways 每种颜色一个网关

The Baseline: Open-space Scenarios not uncommon for LoRa to attain very wide coverages at shorter distances (<50 km) the error is quite high

The Baseline: Open-space Scenarios 环境 衰减 高度 考量 不同网关高度不一样 the need to classify the attenuation of the surrounding environment the need to use a model that considers height as an integral parameter of the propagation model

Assessing the Impact of Obstacles A long “transitional region” any distance under 6 km, we get a mix of link quality, from good to poor

Assessing the Impact of Obstacles 植被覆盖以及环境的影响 R2 3km 在$6解释 R1 through buildings R2, R3, R4 through farming fields R4 is bad as GA is placed indoors and near the NW side of the building(towards R1)

Extracting Land Cover Images acquired via the Sentinel-2 satellite constellation, a last-generation remote sensing system of the European Space Agency (ESA).

high ( 92%) overall accuracy for all three tiles Extracting Land Cover high ( 92%) overall accuracy for all three tiles 提取质量 评估 Overall Accuracy (OA): percentage of test pixels correctly classified Producer’s Accuracy (PA): the percentage of correctly classified pixels for the given class User’s Accuracy (UA): the percentage of correctly classified pixels computed w.r.t. the overall number of pixels associated to the given class

Exploiting Land-cover Knowledge Land Cover makes different

Exploiting Land-cover Knowledge R2 3km处 为 0 the height of obstacles belonging to nlos classes plays a role in determining link quality

Model the Impact of Land Cover A model that needs measurements requires empirical data log-normal shadowing model d distance from transmitter PL(d0) path loss at d0 n path loss exponent of the environment σ is the standard deviation of a zero-mean Gaussian random variable X

A General Model A model that needs measurements using the Okumura-Hata model Hm, hb height of transmitter and receiver d distance from transmitter f transmission frequency a(hm) depending on the environment Urban model Suburban model

Validation in the Wild 判断视距和非视距 全路线 靠近节点附近 node in LOS or NLOS

Estimation Accuracy of ESP (a)(b) ≤ 9dB (c)(d) ≥ 32dB

Predicting the Gateway Coverage the gateway coverage is significant variability in space depending on both the direction and the distance Intersection performance more better

Contribution analyze empirically the maximum range of LoRa in free space with a weather balloon, and use that evaluation as a baseline to gain insights about the negative effects of the environment at ground level propose a dedicated processing toolchain to automatically analyze the types of environment traversed by LoRa links analyze and compare models in the literature for LoRa validate the combined use of the Okumura-Hata model and our toolchain on a large real-world dataset from ‘The Things Network’

Thanks~