Academia.eduAcademia.edu

Figure 3 – uploaded by MUHAMMAD USAMA

See full PDFdownloadDownload figure
Fig. 3: Proposed mechanism for implementing our proposed algorithm in real LoRa networks. layer parameters for the transmission. This mechanism has also been shown in Fig. 3. Our algorithm also works well on the reduced action space by allowing the agent to choose from a specific subset of actions. This reduced subset can be made according to the data-rate requirements of different applications by fixing a certain SF, CR, or transmission channel.

Figure 3 Proposed mechanism for implementing our proposed algorithm in real LoRa networks. layer parameters for the transmission. This mechanism has also been shown in Fig. 3. Our algorithm also works well on the reduced action space by allowing the agent to choose from a specific subset of actions. This reduced subset can be made according to the data-rate requirements of different applications by fixing a certain SF, CR, or transmission channel.

Related Figures (9)

Fig. 1: Architecture of LoRaWAN consisting of LoRa EDs, LoRa gateway, network server, and end-user. The EDs serve different purposes and transmit the data to the gateway based on the application requirements. The received data packets are forwarded to the network server, which in turn forwards them to the end-user.
Fig. 1: Architecture of LoRaWAN consisting of LoRa EDs, LoRa gateway, network server, and end-user. The EDs serve different purposes and transmit the data to the gateway based on the application requirements. The received data packets are forwarded to the network server, which in turn forwards them to the end-user.
The basic mapping of our algorithm on the workflow of DRL has been given in Fig. 2. The agent takes a specific action (choosing a PHY-layer parameters combi- nation for the new LoRa ED) at a time-step and receives a reward based on the achieved packet delivery ratio (PDR) and power-usage based on that chosen action. Fig. 2: Mapping of our problem setup on DRL. The gateway is considered the agent and the LoRa network represents the environment.
The basic mapping of our algorithm on the workflow of DRL has been given in Fig. 2. The agent takes a specific action (choosing a PHY-layer parameters combi- nation for the new LoRa ED) at a time-step and receives a reward based on the achieved packet delivery ratio (PDR) and power-usage based on that chosen action. Fig. 2: Mapping of our problem setup on DRL. The gateway is considered the agent and the LoRa network represents the environment.
TABLE III: Specifications of the DDQN in LoRaDRL
TABLE III: Specifications of the DDQN in LoRaDRL
TABLE II: Specifications of the LoRa Network Simulations
TABLE II: Specifications of the LoRa Network Simulations
TABLE IV: Table showing the performance of an 8-channel LoRaDRL in a dense LoRa network consisting of a single base-station. The values are presented with 95% confidence interval. TABLE V: (a) Table showing the percentage of SF values of an 8-channel LoRaDRL in a dense LoRa network consisting of a single base-station. (b) Table showing the per-SF PDR performance of an 8-channel LoRaDRL in a dense LoRa network consisting of a single base-station.
TABLE IV: Table showing the performance of an 8-channel LoRaDRL in a dense LoRa network consisting of a single base-station. The values are presented with 95% confidence interval. TABLE V: (a) Table showing the percentage of SF values of an 8-channel LoRaDRL in a dense LoRa network consisting of a single base-station. (b) Table showing the per-SF PDR performance of an 8-channel LoRaDRL in a dense LoRa network consisting of a single base-station.
Fig. 5: Figure showing the performance of LoRaDRL with different MAC protocols. It can be seen that there is a minor performance difference while using different MAC protocols with LoRaDRL. The bars are plotted with 95% confidence interval.
Fig. 5: Figure showing the performance of LoRaDRL with different MAC protocols. It can be seen that there is a minor performance difference while using different MAC protocols with LoRaDRL. The bars are plotted with 95% confidence interval.
Fig. 6: Figure showing the performance of a multi-channel LoRaDRL and LoRaSim scheme under frequency jamming attack. A small drop of performance is seen in the case of LoRaDRL as it shifts to the other available frequency. While in the case of LoRaSim, a sudden drop of performance can be seen because of the absence of a feedback loop.
Fig. 6: Figure showing the performance of a multi-channel LoRaDRL and LoRaSim scheme under frequency jamming attack. A small drop of performance is seen in the case of LoRaDRL as it shifts to the other available frequency. While in the case of LoRaSim, a sudden drop of performance can be seen because of the absence of a feedback loop.
Related topics:
Computer ScienceReinforcement Learning
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved