Table 1. LoRa Modulation Parameters at BW = 125 kHz. Table 2. RSSI from each location. Fig. 3. (a) LoRa Device. (b) GateWay Table 3. Parameters for performance analysis in SF selection. Figure 5 summarizes the FSR for each selected approach when the number of LoRa devices is varied from 3 to 30 (3, 9, 15, 30). The number of devices at each location is equal, i.e., one device at each location for 3 devices and 10 devices at each location for 30 devices. From Fig. 5, it can be seen that the FSR decreases as the number of LoRa devices increases for all approaches. This is due to the increase in packet collisions as the number of LoRa devices increases. In addition, the method using the MAB algorithm has a higher FSR, indicating that the autonomous distributed reinforcement learning approach is effective. Among them, the proposed method obtains higher FSR than other methods using reinforcement learning, which makes it suitable for LORaWAN systems. Figure 6 shows the ratio of SF selection per deployment location when there are 30 LoRa devices using ToW-dynamics approach. Packets from LoRa devices located at positions 2 and 3, which are far away from the GW, have weak reception strength at the GW, so a high SF should be selected. In other words, the proposed method does not require prior information on the distance to the GW, and the SF selection is appropriate for each LoRa device.
