Low-Power Wide-Area technologies as building block for smart sensors in air quality measurements

Currently, air pollution monitoring is done at coarse spatial resolution. This is mainly because of the high costs which come along with high accurate measurement equipment and missing network technologies. In Graz and Vienna, the two biggest cities in Austria, air pollution monitoring is done with 7 and 17 measurement stations, respectively. In Vienna for example, only one measurement station per 100.000 inhabitants exists. Additionally, in many stations, not all types of pollutants are measured. Therefore, to enable air pollution measurements at much denser spatial resolution, a different concept is required. Different concepts exist for enhancing resolution, which includes WSN – in specific LPWAN. These technologies enable high resolution sensor meshes, but also bring along disadvantages, in particular the utilization of cheap sensors with unacceptable large measurement uncertainties. The focus of this paper will be laid on network technologies. An extensive discussion on low-cost air pollution sensors can be found in [3].

Several projects and publications exist where WSN have been used to enhance spatial resolution in air pollution monitoring. In Zurich, Hasenfratz et al. [4] created a mobile sensor network by mounting sensors on trams and busses. Data collection was implemented using the GSM network. Additionally, ozone sensors have been mounted on bicycles. Using this approach it was possible to cover an area of about 100 km². The CITI-SENSE project (2012–2016) [5] deployed fixed, mobile and personal nodes in eight cities in Europe. Different sensor principles and solutions for fixed and mobile nodes have been compared. Sensors showed good results in laboratory measurements, but much degraded performance in field tests, where differences in humidity and temperature had strong impacts on the measurement results. They suggested that proper sensor technology development is the requirement for spatial dense sensor networks.

In the USA, the Community Air Sensor Network (CAIRSENSE) project [6] monitored air pollution in a 2 km² area, using fixed stations and four wireless sensor nodes, which have been supplied by solar power and partly by AC power. Xbee radios have been used for communication, which facilitated a maximum communication range of 500 m in a suburban area. Two nodes with a distance of about 1.5 km from the base station were unable to report data via wireless communication, as messages were not received. Furthermore, strong sensor dependency on humidity and temperature has been reported with most sensors providing unreliable results. Among others, these projects show possibilities of enhancing the spatial density, whereas still problems exist in bringing together spatial density with accurate sensor resolution.

LoRa, which stands for “long range” is a chirp spread spectrum (CSS) modulation technology [7] from Semtech Corporation. It features on the one hand the modulation scheme and on the other hand a media access control (MAC) protocol called LoRaWAN, which is promoted by the LoRa Alliance^™. LoRa operates in the Industrial, Scientific and Medical (ISM) bands of 433-, 868- or 915-MHz depending on the region where it is used. Depending on the used transceiver, it is able to provide a constant RF output power of 20 dBm and data rates of up to 37.5 kbps, with the latter strongly dependent on the chosen parameters. Among others, these parameters are spreading factor (SF), bandwidth (BW) and code rate (CR). They influence data rate, communication distance and its robustness to interference. A higher spreading factor and a lower bandwidth result in higher sensitivity, but in lower data rates. The maximum sensitivity of end nodes is −146 dBm. Several publications [8‐10] tested parameter settings and the resulting effects on reliability and communication distance. For instance, Augustin et al. [8] reached communication distances of up to 3.400 m in a suburban area in Paris.

As mentioned before, LoRaWAN is the MAC protocol developed by the LoRa Alliance^™. The specification defines three different types of end devices, which are listed in the following. The power consumption increases from class A to class C.

Additionally, LoRaWAN features a security stack using the Advanced Encryption Standard (AES) for secure communication and message authentication. LoRa networks can be custom built without paying substantial network fees, but the technology is protected and only the company Semtech is manufacturing LoRa ICs and distributing licenses.

Sigfox is an Ultra-Narrow Band (UNB) technology with binary phase shift keying (BPSK), developed by a company with the same name. In Europe, Sigfox uses the 868 MHz ISM band, while in the rest of the world the band in between 902–920 MHz. Transmission power and power consumption are similar to LoRa, with a maximum RF output of 14 dBm. The maximum data rate is specified with 600 bps, which is sufficient for most sensor applications. The network structure of Sigfox and LoRa is the same, whereby end nodes are not bound to specific gateways. Therefore, several gateways are able to receive and forward the messages, with a server or backend software responsible for filtering equal messages forwarded by multiple gateways. This approach extends the reliability of the network. The business model differs from that of the LoRa Alliance^™, in which the transceiver technology is provided to silicon manufacturers. In contrast to LoRa, the network and software is sold as a service with maximum number of messages per node and day dependent on subscription level. Both uplink and downlink messages are possible, whereby downlink slots for end devices specified directly after sending message to the gateway. Currently Sigfox is covering 45 countries and an area of 3.8 million km² with their network.

Narrow-Band Internet of Things (NB-IoT) has been developed by the 3rd Generation Partnership Project (3GPP). In contradiction to LoRa and Sigfox it is deployed in authorized spectrum and is employing existing cellular (mobile) networks, which are up to now, used for human-oriented communication. The required bandwidth is 200 kHz and current GSM bands, guard bands of the LTE carrier or unused blocks in the LTE band are frequencies at which NB-IoT could be deployed. Uplink and downlink communication use different frequencies, whereas data rates are limited for uplink to 20 kbps and for downlink to 200 kbps. NB-IoT reuses and enhances the LTE protocol, while reducing the functionalities to a minimum for IoT usage. The first NB-IoT transceiver (SARA-N2) which is on the market features a maximum output power of 23 dBm and input sensitivity of −135 dBm, requiring for this setting 220 mA for transmission. Lauridsen et al. [11] compared the coverage of GPRS, NB-IoT, LoRa and Sigfox in a 7.800 km² area in Denmark, using radio coverage simulations, resulting in the best outcome for Sigfox and NB-IoT, especially for indoor connectivity. Apart from these simulations, real distance measurements using Sigfox or NB-IoT are not published until now.

In Table 2, LoRa, Sigfox and NB-IoT are compared with the respect to different factors. LoRa and Sigfox are deployed in license free bands in contrary to NB-IoT, which is using licensed LTE frequency bands. Communication range cannot be directly specified, as it is very dependent on the environment. However, it is expected that all three technologies are capable of reaching several kilometers, at least in suburban areas. The maximum data rate of Sigfox (600 bps) lags behind the other two competitors, whereas the bit rate of LoRa depends on the selected PHY settings. Comparing similar output power, the SARA-N2 NB-IoT transceiver [12] requires 100 mA for 13 dBm TX power and 46 mA for RX. This is in case of transmission and listening up to five times the required power consumption of LoRa and Sigfox. Sigfox and LoRa are using star network topologies with nodes not directly linked to a specific gateway in contrast to NB-IoT. Mekki et al. [13] compared LPWAN technologies on different factors like range, battery lifetime, scalability or payload length, which showed the similarities between LoRa and Sigfox in contrary to NB-IoT.

One main limitation of LoRa and Sigfox provides the ERC Recommendation 70-03 [14] which limits the duty cycling to 1 percent (exception is one band with 10 percent) and the maximum output to 14 dBm (exception is one band with 27 dBm) in the unlicensed ISM bands.

The measured results are shown in Fig. 1. It was possible to receive messages from all positions. As expected, the packet delivery ratio (PDR), in which only messages without any payload error were counted, is strongly depended on the communication distance. A PDR of more than 95 percent could be achieved from four positions (1, 4, 7, 8), with distances between 710 m and 1.380 m, while the worst PDR were measured at positions 3 and 6, with 15 and 16 percent respectively. At position 5, having a distance of 1.740 m from the sink, 43 percent of the messages could be properly received. Generally speaking, the PDR correlates with the node distances, but dependency on topography and the number of buildings in between cannot be neglected.

Furthermore, the averaged RSSI values are shown in the bottom plot of Fig. 1. It becomes apparent, that the RSSI values at position 1 and 8 are significantly higher than from other positions. That can be explained by the descending terrain, which was leading to a close to line of sight (LOS) positioning. In addition to that, results a higher RSSI magnitude in a weaker signal, which indicates a higher distance. This increasing distance was leading to more packets with wrongly delivered payload.

Comparing these results to the measurements from Augustin et al. [8], only smaller distances in a more urban area have been measured. Furthermore, it has to be mentioned, that by using a more sensitive receiver like the SX1276, or a custom gateway as message sink, better results can be achieved. The reliability of the communication link can additionally be extended, by resending lost messages.

In this section the measured path loss is compared to the Okumura–Hata Model [15]. This relatively simple model evaluates the path loss for a certain distance between base station and a mobile sensor node. It can be seen as an extension of the free space model. The path loss in dB is thereby given as:

Whereby factors \(A\), \(B\) are depending on the frequency and the antenna height (of the base station and the mobile node). The distance between the base station and mobile node is specified by \(d\) in km. Factors \(A\) and \(B\) are given using the following formulas:

With \(f_{c}\) represents the used frequency, \(h_{b}\) is the base station height and \(h_{m}\) the height of the mobile node.

Factors C and \(a( h_{m} )\) are additionally depending on the deployed environment. For small and medium sized cities they can be calculated as followed:

As a result, the evaluated model for distances up to 5.000 m is depicted in Fig. 3. The red shaded area represents the communication range between base station and mobile nodes, which is not usable due to insufficient link budget in case of the used LoRa SX1272 transceiver. This would result in a maximum communication distance of about 4000 m in small and medium sized cities. Figure 4 shows the model compared to the measured path loss. The measured path loss scatters due to the terrain and the obstacles in between. This cannot be represented by the simple Okumura–Hata Model, which does not include terrain profile or dominant obstacles. However it can be seen, that the measured path loss values at positions 1, 7 and 8 are well modeled by the Okumura–Hata Model. This is due to the descending area between sink and sensor position, which makes the path loss depending on the distance only. On the other hand, the modeling quality of all other positions is worse. Especially sensor position 3 shows the impact of dominant obstacles. It is directly placed behind the main building of the technical university, resulting in a weak communication link and the highest path loss. It can be concluded, that the Okumura–Hata Model is a rough indicator for the communication distance and the density of required base stations. Furthermore needs to be mentioned, that other models exist, which adjust to the area of interest (cf. [16]).

Benefits of WSNs for low cost air pollution monitoring have already been discussed in Sect. 1.2. The main limiting factors of LPWAN technologies are packet size, data rate and the limitation of maximum transmissions per hour and day. As described in the previous section, the duty cycle and therefore the transmissions per day for the ISM bands, used by Sigfox and LoRa, are limited by the EU. Sigfox additionally confines the number of transmissions by a fee-dependent subscription level. Whether these restrictions are constraining a possible deployment of LPWAN depends on the chosen sensor principle and the data evaluation mechanism. As a consequence, the following factors affect the required data transfer:

As the discussion shows, plenty of factors affect the data transfer requirements, which is also shown by Clements et al. [17]. If the principal data evaluation can be done locally, data rate limitations should not confine the utilization of LoRa or NB-IoT. However, the previously discussed restrictions of Sigfox might limit multi-sensor platforms.

Furthermore, additional logic is required at the backend of the network for filtering redundant packets or request wrongly or lost packets. Finally, the evaluation and presentation of the received information should not be underestimated with a vast amount of data to be expected.