Developing a congestion management scheme to reduce the impact of congestion in mixed traffic LoRaWANs

Abstract

The development of wireless technologies such as the Long Range Wide Area Network (LoRaWAN) protocol has enabled connectivity for a wide range of devices for optimisation and automation purposes. These types of technologies support the rapid expansion of the Internet of Things (IoT) deployments which frequently consist of large quantities of low-cost connected devices which require a network protocol suitable to their needs. These devices are frequently energy-constrained, and communication is a major energy consumer. These connected devices typically generate a lot of uplink traffic, i.e., packets sent from a device to a Gateway (GW). Additionally, minimal downlink traffic, i.e., packets sent from the GW to a device would also be required. A significant source of downlink traffic is the transmission of Acknowledgements (ACKs) to confirm the reception of packets sent by devices.

The advantages of LoRaWANs are excellent scalability if the network is uplink-dominated (no or minimal packets require downlink traffic in response). Traffic in these networks can be defined as being one of two types: unconfirmed (no ACK required) and confirmed (ACK is required). A network’s scalability is quickly reduced by the presence of even low amounts of downlink traffic as this causes congestion. The impact of supporting mixed-traffic (unconfirmed and confirmed) on scalability is minimal in small networks but increases quickly in large networks, especially if they only contain one gateway.

The contributions of this research consist of three objectives. The first objective was to study the cause and impact of congestion in mixed-traffic LoRaWANs operating with only one gateway. By studying the root causes of congestion, new solutions were developed to improve network performance (in the form of improved packet delivery). These new solutions were the focus of the second objective and led to a congestion management scheme and an algorithm that reduces congestion. The third objective was to study how congestion should be managed in battery-less capacitor-based devices that obtain energy through energy harvesting.

The current research literature revealed that the two main reasons confirmed traffic leads to congestion are that LoRaWAN gateways are half-duplex and that Duty Cycle (DC) restrictions limit the number of sent ACKs. Thanks to their half-duplex design, a gateway cannot receive any transmissions whilst transmitting downlink messages, causing it to miss packets belonging to both traffic types. DC restrictions may prevent the transmission of an ACK, resulting in the retransmission of packets a GW has successfully received. Furthermore, an ACK must be sent during one of only two timing windows (called receive windows), with the second window using a very low data rate (increasing transmission time).

The literature survey revealed that the LoRaWAN protocol does not have a congestion monitoring and response mechanism. In response to this, a novel method to monitor congestion as well as a response mechanism was developed for LoRaWANs. The response mechanism is a novel algorithm tasked with reducing congestion, which is done through requests for the aggregation of confirmed packets by nodes. A commonly used open-source ns-3 based LoRaWAN simulator (“lorawan") served as a suitable research platform. Network performance was studied via packet reliability, by examining the success probability of unconfirmed traffic (ULPDR) and confirmed traffic (CPSR). The developed algorithm was then evaluated using a series of simulations to determine its impact on network performance across congestion levels and how several LoRaWAN parameters influence this impact. Congestion was created by increasing the packet arrival rate (traffic volume) in the network by adjusting the transmission interval of nodes.

Congestion was monitored through the newly developed Adaptive Congestion Scheme (ACS), which is similar in design to the protocol’s Adaptive Data Rate (ADR) scheme. The developed scheme is split into two parts, with one operating as part of the Network Server (NS) and another on the device, capable of communicating with each other. When congestion is detected, the developed algorithm called groupedPackets is activated to improve packet reliability.

The groupedPackets algorithm was designed to reduce the traffic volumes generated by confirmed nodes by requesting that they aggregate their application packets. This causes nodes to send larger but less frequent confirmed packets. A major cause of congestion was the number of ACKs required, which is now reduced due to fewer confirmed packets being sent. This change enabled GWs to be able to send more ACKs before DC restrictions were met.

This improvement in the ratio between required and transmitted ACKs caused the number of retrans- missions of confirmed packets to reduce, which resulted in less interference thanks to fewer packet collisions. Finally, this also mitigated the half-duplex nature of GWs, as the number of packets lost during GW transmissions was reduced as fewer transmissions were required.

When groupedPackets was evaluated, it enabled improved CPSR and ULPDR values when compared to identical networks operating using the standard protocol. The algorithm showcased considerable improvements in CPSR in congested networks (high packet arrival rates) with only minimal impact at lower rates as CPSR was already above 98 %. Performance was further increased when the number of retransmissions was set to eight, allowing aggregated payloads multiple chances at success. Increasing the base packet size from 10 B to 20 B slightly reduced groupedPackets’ performance as this hampered the number of packets it could aggregate. Using groupedPackets to aggregate unconfirmed packets led to marginal improvements in a best-case scenario and no improvements in less favourable scenarios.

In battery-less LoRaWANs, the radio is the primary energy user. The standard groupedPackets algorithm underperforms in these networks as it is unaware of energy constraints. An improved version was developed, in which the node informs the NS side of the algorithm of energy constraints after optimising the amount of sent payloads. This version allowed confirmed nodes whose capacitance is in a “transition“ range of values to increase the number of sent payloads, which resulted in an increase in CPSR for this range.

Enabling retransmissions did result in improved packet delivery in both standard LoRaWANs and the improved groupedPackets networks. The effectiveness of this is dependent on sufficient energy being available for retransmissions. Confirmed nodes require more energy than unconfirmed nodes, and as a result, the minimum capacitance to support multiple transmissions differed between the two types. Matching the transmission settings used by receive window two to those used by window one resulted in a significant increase in CPSR in standard networks and also offered a slight improvement in groupedPackets networks. By matching the settings, more ACKs can be sent, which reduces the number of retransmissions. Energy consumption was also reduced due to the higher data rate used.

Compared with a standard network, combining groupedPackets, enabling retransmissions and matching the window settings were effective. The combination improved ULPDR by ≈ ten percentage points, increased CPSR by ≈ 40 percentage points and can be achieved by increasing the node’s capacitance from 6 mF to 7.5 mF.

This work contributed to an improved version of the protocol in which the viability of the use of confirmed traffic was increased in mixed-traffic LoRaWANs. The developed congestion monitoring scheme proved to be effective in detecting congestion and taking action. This research found that the developed algorithm can significantly improve packet reliability, at the cost of an increased delay and node capacitance in the case of battery-less devices.