Long-range reliable IoT networks. What is the cost to pay?
Current status of the technology
The IoT networks capable of providing long range are known as LPWANs: Low Power Wide Area Networks. LPWANs are star-topology networks composed of battery operated devices mostly deployed in harsh environments, where the battery replacement is costly. Moreover, the data sent by these devices consists in a few packets/day, most of the times without being acknowledged, as a way to save energy and to satisfy the 1% duty-cycle restrictions imposed by ETSI in the license- free bands.
The LoRa technology developed by Semtech it is currently one of the most adopted technologies for the industrial IoT applications. LoRa networks operate at very low data rates, ranging from 0.3 kbps to 27 kbps depending on the Spreading Factor (SF) in use, but are able to transmit at up to 15km (if using the smallest data rate).
The trend is towards massive connectivity
These low-cost energy and duty-cycle restricted networks are being challenged to offer high quality of service while connecting a massive number of industrial assets.
When increasing the number of devices in the network, the number of packet collisions increases. If the network is operating at the fastest data rate (SF7 for LoRa), these collisions decrease the network performance, but still, the network is alive. When a long range is needed, the network operates at the lowest data rate possible (SF12 for LoRa). For this case, an increased number of devices kills the network: none of the sent packets can arrive at the gateway.
Packet fragmentation: a possible solution
LPWAN networks are Aloha-based networks, so that collisions can happen
at anytime during packet transmission and so, the larger the packet, the higher the probability of collision during its transmission. Any collision causes the loss of the whole packet. Fragmenting the packet will reduce this loss to the interfering sections of the colliding packets, leaving some of the fragments of the packet unaffected by this collision. This can be seen in the following figure.
Running network simulations confirms the gain in performance when using fragmentation. The performance is measured by the network goodput: the percentage of correct packets received by the gateway relative to the number of packets sent in the network. Results show that at SF12 the network goodput can raise from 10% to more than 80% by fragmenting in more than 25 each packet sent in a 5-nodes network. The denser the network, the more important the gains obtained.
Packet fragmentation: costs of implementation
When packet fragmentation is used, the payload will be divided to the amount of fragments needed and one extra Byte of header will be added to each fragment. This additional header for each fragment causes the increase of the network energy consumption and of the end to end delay for sending the data (compared to the case when no packet fragmentation is used).
What has a high impact on the energy consumption of the sensor nodes is the spreading factor used, as the higher the spreading factor, the higher the duration of the packet and implicitly, the energy consumption for sending it.
When operating in duty-cycle restricted networks, the end to end delay increases with almost 120% when fragmenting in 50 fragments compared to the case when no fragmentation is used, for networks operating at SF7. For networks operating at SF12, the end to end delay increase with the number of fragments is much higher, reaching an overhead of 260%, independent of the number of nodes in the network. The delay increase is caused exclusively by the headers that create the need of extra time off (imposed by the 1% duty cycle restriction).
In what concerns the optimal number of fragments/packet to be used, there is a trade-off between the goodput performance that can be obtained and the extra costs in terms of energy consumption and latency.
