Smart city IoT deployments are driving innovations and research in long range low power wireless communication networks. Previous Wireless Sensor Network (WSN) deployments would have used custom hardware and protocols to facilitate communication. The developments in this area have led to a new type of wireless communication network, LPWANs.
The city of Southampton, UK was used as a test bed to evaluate LoRaWAN, one of the LPWAN technologies. This evaluation has required deploying the necessary gateway infrastructure, and assessing its performance. LoRaWAN is used as communication means for the air quality monitors which are currently being deployed in and around Southampton [
4,
5,
6,
7]. These air quality monitors log data continuously to local storage and transmit average Particulate Matter (PM) concentrations at regular intervals. These averages allow the air quality in the city to be monitored in near real time. The transmission and receiving times of these messages have been logged and compared. This has enabled the calculation of the percentage of packets successfully received over the duration of the deployment, investigations into the end-to-end delays observed within the network, and any atmospheric effects to be considered.
1.1. Low-Power Wide Area Network (LPWAN)
Bardyn et al. [
8] state the main characteristics of a LPWAN are: ultra low-power operation, low–cost, no need to wake an end device to maintain network connectivity, ease of deployment of infrastructure nationwide, and secure data transfer. While not included in this list, the long range is also a defining feature of the networks. This means that these LPWAN technologies are not competitors to Bluetooth [
9], WiFi [
10], Zigbee [
11], or other short range wireless communication technologies. A detailed comparison between LoRaWAN, Sigfox and NB-IoT is presented by Mekki et al. [
12], and summarised in
Table 1. Despite NB-IoT using licensed frequencies compared to LoRaWAN and Sigfox which use the license free Industrial, Scientific and Medical (ISM) band, all technologies have the same problem that the frequencies available in each region differ. This regulatory complexity creates additional challenges when moving devices internationally. All three networks also offer encryption of the payload to prevent eavesdropping of traffic. Security analysis of the LoRaWAN protocol found multiple weaknesses in the LoRaWAN V1.0 specification [
13], many of which have been addressed in version 1.1 [
14], which is available and soon to be adopted.
LoRaWAN is built on the lower level LoRa protocol, which can be used on its own, but previous work using LoRa for a smart city environment concluded that more robust communication could be achieved by using LoRaWAN on the LoRa physical layer [
15]. LoRaWAN is the only network for which it is easy and simple to deploy your own gateway. Both Sigfox and NB-IoT are operated by infrastructure companies, and any additional gateways have to fit within the national operators deployment plan. A personal Sigfox gateway has been announced but distribution is managed by the local network operators who have to be contacted for information [
16]. There are multiple vendors offering pre-built LoRaWAN gateways for sale, as well as instructions to make your own custom gateway from a kit of parts.
The ability for users to deploy gateways makes LoRaWAN suitable for city-scale IoT deployments, especially when combined with the localisation and bandwidth capabilities.
LoRaWAN
A LoRaWAN deployment can be run totally independently from all other LoRaWAN networks, and this may be beneficial in some commercial or defence use cases; other more open deployments can be build around the existing LoRaWAN community. This community is based around The Things Network (TTN) [
17,
18], a large LoRaWAN development community and a global deployment community which is rapidly expanding. It is centred on an open and collaborative network providing solutions to facilitate the use of LoRaWAN that enable users to easily use existing gateways to transmit their messages or to add gateways to the network. Data received by TTN is published as an MQTT [
19] topic which can be subscribed to by consumers. TTN handles the de-duplication of messages that have been received by multiple gateways simultaneously, further reducing the complexity of implementation. TTN currently has
96,000 members, providing
10,000 gateways across
countries.
There is no standard gateway hardware in use on TTN. Any LoRaWAN gateway can be connected to the network. This includes commercially made gateways or those made by users. The different gateway types have different features and make use of different backhaul networks, which is discussed further in
Section 2.1.2. Some gateways support a satellite backhaul and work is ongoing to transmit LoRaWAN messages direct to satellites [
20]. Although this technology is not currently used in this deployment, it is of interest for future rural deployments.
Localisation of devices based on multilateration of signals is a well established technique [
21]. Both Sigfox and LoRaWAN offer support for localisation using different methods. LoRaWAN supports both Timed Difference of Arrival (TDOA) and Received Signal Strength Indication (RSSI) for multilateration of transmissions. This enables devices without Global Navigation Satellite System (GNSS) receivers to provide location aware data streams. The accuracy of the calculated location is dependent on the gateway hardware, the number of gateways that receive the transmission, and the type of the gateway that receives the transmission. All gateways can be used to provide signal strength measurements which can be used for RSSI-based location calculations. For TDOA localisation calculations a (
) fine grained time stamp is needed for the message. This fine grained time stamp is not available on all gateway nodes because of the specific hardware requirements needed to record the message arrival with the required accuracy. This data can then be fed into the LoRa Cloud location service [
22] (previously known as Collos) which uses this data to calculate a position. RSSI gives accuracy of 1000 m to 2000 m compared to TDOA which is in the range 20
to 200
[
23]. An evaluation of LoRaWAN localisation is presented by Fargas and Petersen [
24].
LoRaWAN supports three different modes of operation known as classes:
A,
B, and
C. Each class has different priorities in terms of performance and energy consumption which have been analysed by Cheong et al. [
25]. The default for all LoRaWAN devices is to operate in class
A, meaning that data can only be received by the end device in a short window after transmission. If the gateway has a Global Positioning System (GPS) receiver then it can be used to provide a beacon broadcast which enables accurate time alignment between end devices, thereby enabling class
B which includes scheduled receive windows [
14]. Class
C requires the end-node to be listening continuously and is designed for either mains–powered devices or transmission of firmware updates to end–nodes during scheduled windows. Operation in any mode requires support from the full hardware and software stack.
The LoRaWAN community is continuously developing, with version 3 of the TTN network stack having over 20 releases in the last year [
26]. The rapid development of the network stack means that once an area has gateway coverage additional features can be added through upgrades to the network stack.
1.2. LoRa and LoRaWAN Test Beds
Multiple cities have been used to test LoRa and LoRaWAN; some of these deployments use the LoRa Physical (PHY) layer on its own without LoRaWAN on top. Tzortzakis et al. [
27] present one such work in which two nodes were deployed at the National Technical University of Athens campus and reported environmental parameters back over a LoRa network. These nodes were 800
and 500
away from the gateway. Both the end nodes and the gateway (with General Packet Radio Services (GPRS) backhaul) in this network are solar powered. During the 10
deployment, 100% of transmitted packets were received.
Lee and Ke [
28] have deployed a system that has two major differences to the deployment used in this paper; the network uses the 433
LoRa band, and is a mesh network, unlike the star used in LoRaWAN. By using a mesh rather than a star a node can forward messages received by other nodes onwards to the gateway. This has the advantage of offering greater coverage than can be achieved using a single gateway, but comes at the cost of: increased protocol complexity, increased energy usage on nodes, and less efficient use of available radio bandwidth. During the course of an 8
deployment consisting of 18 nodes being queried at a 1
interval, an average of
% packet reception was achieved using a mesh compared to the
% achieved when using a star network. Given the retransmissions required to implement a mesh network, it is not clear how this approach would scale.
Other test beds have used the LoRaWAN protocol on top of the LoRa PHY layer. Pasolini et al. [
29] considered both scenarios. During their range tests using LoRa, a maximum range of 2390 m was achieved using Spreading Factor (SF) 12, this poor performance was suggested to be caused by the low height above ground (
) for the transmitting node. The results from this range experiment were then used as inputs for a simulation to model a planned large scale LoRaWAN deployment to optimise the choice of SF [
30]. The results for observed range are significantly below those observed by Basford et al. [
7] and Petäjäjärvi et al. [
31], but are substantially better than the
observed by Loriot et al. [
32]. Kulkarni et al. [
33] concluded that their tests at a location
away from the gateway reached the limit of their deployment; a likely reason for this is the gateway being installed “on a desk in a faculty office”. No details are given as to the elevation of the office. By installing the gateway on a desk it will provide representative data for the indoor experiments performed, but the outdoor measurements should not be compared with data from other deployments with outdoor gateway locations.
Doğan [
34] tested LoRaWAN in diverse conditions, in both indoor and outdoor environments, including a tunnel. During the outdoor experiments different power and SF settings were used at four locations across the city with distance from 0.5
to 3.3
. For each combination of parameters, 1000
were transmitted over the course of 10
. The performance of the network for each power and SF combination was very location dependent with two locations achieving 100% delivery rate for 78% of the combinations tested. Increasing the SF does not always lead to an increase in packet reception.
Marais [
35] deployed a LoRaWAN network for two research projects: a test bed and a water usage monitoring system. The test bed consisted of 18 nodes and the water monitoring project consisted of 34 nodes transmitting every 10
with Adaptive Data Rate (ADR) enabled. The test bed nodes are located between 0.1
to 5.2
away from the gateway. As well as looking at packet delivery rates, the performance of ADR was analysed, with a higher delivery rate being achieved when ADR was disabled [
36].
The deployments considered to this point only used a single gateway node. Wixted et al. [
37] deployed three gateways across Glasgow. These gateways were used for both coverage mapping and reliability monitoring. The reliability monitoring was performed using acknowledged transmissions over a
link. Once initial technical problems were addressed, 98% of messages were successfully received by the gateways.
As well as looking at delivery rates for LoRaWAN networks, there have been studies into the end-to-end delays of LoRaWAN messages. Fernandes Carvalho et al. [
38] performed a test using a LoRaWAN transmit node connected to a PC and four separate devices listening to the MQTT application data stream. This experiment was performed as part of the Brescia Smart Living project which covers and area of 80
using over 100 gateways. These gateways then forward the messages to a Patavina NetSuite which manages the LoRaWAN network. Over the course of a day, 1440 messages were transmitted at 1
intervals with the overall average end-to-end delay being 400
to 700
, but delays of several seconds were observed.
Pötsch and Hammer [
39] performed an analysis of the end-to-end latency of a LoRaWAN network. When the entire LoRaWAN stack was running on a single node, end-to-end latencies of
were observed for SF 7 and 9, increasing to
for SF 12. When the gateway was separated out and using a Universal Mobile Telecommunications Service (UMTS) connection to the network server, the latency increased to
for SF 7 and 9, and nearly 3000
for SF 12. The change to a UMTS connection also dramatically increased the standard deviation of the latency of received messages.
The remainder of this paper is structured as follows.
Section 2 describes the test bed developed in Southampton; an analysis of the dataset gathered is presented in
Section 3; finally, conclusions are presented and areas for future work highlighted in
Section 4.