Optimal LoRaWAN Gateway Deployment for Advanced Metering Infrastructure: A Greedy Capacity-Coverage Approach with Class-C Bidirectional Capacity Analysis

Advanced Metering Infrastructure (AMI) systems require bidirectional wireless communication for remote meter reading, disconnection, and demand–response management across large numbers of smart meters. LoRaWAN Class-C is a strong candidate for large-scale AMI due to its long range, low infrastructure cost, and native downlink support, but gateway placement must simultaneously satisfy uplink Pure-ALOHA capacity, downlink duty-cycle limits, and geographic coverage constraints in non-uniform device distributions. This paper proposes a greedy capacity-coverage gateway placement algorithm for a LoRaWAN AMI deployment serving 89,350 smart meters in Lam Luk Ka district, Pathum Thani Province, Thailand, based on real GIS building-footprint data. The algorithm seeds each gateway at the densest unserved 300 m grid cell, assigns devices within the planning radius R up to the tighter of the uplink Pure-ALOHA capacity and the downlink duty-cycle capacity, both computed for a 98% packet delivery target, and continues placing gateways until 98% of devices are geographically covered. For the 15 min reporting interval, the algorithm yields K* = 137 gateways, only 7% above the arithmetic lower bound of 128, and requires 2.9× fewer gateways than iterative K-means under the same placement constraints. An interval-based scenario analysis across five reporting periods of 5, 10, 15, 30, and 60 min reveals that 15 min is the crossover design point where longer intervals are limited by the downlink duty-cycle and shorter intervals are limited by uplink Pure-ALOHA collision, making 15 min the point at which both constraints are simultaneously near-binding. RX2 reconfiguration from SF10 to SF7 is shown to be essential: the default configuration yields only 121 devices/gateway at the 15 min rate, requiring 739 gateways, while SF7 reconfiguration raises capacity 5.8x to 702 devices/gateway. A Monte Carlo simulation with Urban Okumura–Hata path loss and log-normal shadowing validates the placement, achieving a mean uplink PDR of 98.0% and downlink PDR of 99.9%, both meeting the 98% design target. Class-C energy consumption is 112,787 mJ per 900 s reporting cycle, 434x that of Class-A (260 mJ), confirming that Class-C bidirectional AMI is feasible only for mains-powered smart meters.