1

Monitoring of Large-Area IoT Sensors using a LoRa

Wireless Mesh Network System: Design and Evaluation

Huang-Chen Lee, Senior Member, IEEE; Kai-Hsiang Ke

Abstract—Although many techniques exist to transfer data transfer data over long distances to remote gateways.
from the widely distributed sensors that make up the internet of LoRaWAN uses a star-network topology for communication
things (IoT; e.g., using 3G/4G networks or cables), these methods between LoRa gateways and IoT devices; only one hop is
are associated with prohibitively high costs, making them allowed between a gateway and a LoRa device. Some
impractical for real-life applications. Recently, several emerging experiments have shown that a LoRa device can transmit data
wireless technologies have been proposed to provide long-range
15 km in an open area, which is sufficient for most current
communication for IoT sensors. Among these, LoRa has been
examined for long-range performance. Although LoRa shows long-range IoT applications. However, indoor LoRa devices
good performance for long-range transmission in the countryside, may still be unable to communicate wirelessly with a nearby
its radio signals can be attenuated over distance and buildings, gateway, due to obstacles between sensors, which can attenuate
trees, and other radio signal sources may interfere with the signals. wireless signal strength and result in data losses and
Our observations show that in urban areas, LoRa requires dense communication errors.
deployment of LoRa gateways to ensure that indoor LoRa devices Increasing the spreading factor (SF) of LoRa PHY to 12 can
can successfully transfer data back to remote gateways. Wireless significantly extend the communication range by increasing the
mesh networking is a solution for increasing communication receiver sensitivity [9], but the drawback is that doing so can
range and packet delivery ratio without the need to install
lower the data throughput rate and cause more severe data
additional LoRa gateways. This study presents a LoRa mesh
networking system for large-area monitoring of IoT applications. collision due to the longer times required for transmission. This
We deployed 19 LoRa mesh networking devices over an 800m × issue can be exacerbated when a large number of high-density
600m area on our university campus and installed a gateway that LoRa devices send wireless data and receive
collected data at 1-minute intervals. The proposed LoRa mesh acknowledgements at the same time. Some studies [11] have
networking system achieved an average 88.49% packet delivery shown that these conditions can cause the packet delivery ratio
ratio, whereas the star-network topology used by LoRa achieved (PDR) to drop significantly. To solve this issue, telecom
only 58.7% under the same settings. To our knowledge, this is the operators need to deploy more LoRaWAN gateways to increase
first academic study discussing LoRa mesh networking in detail the probability of receiving the wireless data from the
and evaluating its performance via real experiments.
LoRaWAN devices; this solution increases the cost of the
Index Terms—LoRa, PHY, wireless mesh network, Internet of Things,
network infrastructure.
IoT, LoRaWAN, long-range wireless area network, multi-hop, data Wireless mesh networks are one feasible solution to increase
collection, sensor. the communication performance of devices in indoor networks.
This approach allows all LoRa devices on the network to act as
I. INTRODUCTION routers and relay data from other devices. In this study, which is
based on a previous study [10], we explore the use of wireless
he internet of things (IoT) aims to enable conventional
T sensing devices to communicate with other devices and to
cooperatively provide intelligent service. For example, IoT
mesh networks based on LoRa PHY with the aim of collecting
reliable data on the performance of this technique.
Fig. 1 shows a snapshot of the network topology of the
can be used to monitor air quality in a city to provide real-time
proposed LoRa mesh network on our university campus. In this
information and warnings to inhabitants as well as to control air
figure, GW denotes the gateway that manages the network and
pollution [1],[2]. Another example is monitoring of solar panels
collects data from the IoT sensors, which are indicated by
[3]; smart panels can be installed on the roofs of several circles with numbers. Nineteen sensors were installed in
buildings to optimize their collective efficiency. As the number different buildings across the 600m × 800m university campus.
of sensors in an IoT system grows, however, the issue of how to The yellow lines connecting the circles indicate the topological
transfer data among those devices becomes increasingly
paths generated by the data being relayed to the gateways at the
complex, and data transfer needs must be balanced with
moment of the snapshot.
operating considerations and infrastructure costs. In contrast to
The example in Fig. 1 shows that data generated by sensor 4
existing short-range wireless communication technologies were relayed by sensors 1 and 3 before finally arriving at the
currently used for indoor IoT sensors [4], several techniques gateway. Note that the network topology is formed
have been proposed for providing long-range and outdoor
automatically and may change due to environmental changes.
wireless communication, such as Sigfox [5], LoRa/LoRaWAN
Each sensor decides locally which sensor is the best one to help
[6], NB-IoT, and LTE-M [7]. Among these, LoRaWAN, which
relay its data. Using this approach, a sensor that cannot directly
is based on LoRa physical layer (PHY) [8] implementation, is a communicate with the gateway may find other sensors to help it
network standard for telecom operators. It allows them to to transfer data to the gateway, thereby raising the packet
provide network service and enable devices to wirelessly delivery performance (PDR) of each sensor.
 2

800m
19
16 GW
2
Science Building
11 20
5 13
12 9
14 Eng. Building
3 General Classrom ：RF
15 ：8F
：5F
Library ：4F
10 ：2F
：1F
600m
Literature Building Management Building
4
7
1
8
Sport FIeld
Activity Center
17
18

Fig. 1. A network topology snapshot of the LoRa mesh network devices distributed around a 600m × 800m university campus.

We analyzed the communication performance of the wireless technologies to become commercially available; therefore,
mesh network and the result shows that mesh networks can many academic studies have been devoted to discussing and
deliver better performance without incurring the cost of analyzing LoRa’s performance. LoRa (specifically, LoRa PHY)
installing additional gateways. We also analyzed the is a long-range wireless transmission technique, and
performance tradeoffs of different configurations. LoRaWAN is a networking design that enables telecom
Briefly, the contributions of this study are as follows: operators to provide subscription services based on LoRa PHY.
1. Whereas other studies [11]–[16] have continued to focus LoRa PHY is a proprietary chirp spread spectrum (CSS)
on analysis of the standard LoRaWAN protocols, scheme that uses the sub-1 GHz wireless frequency band. It
discussing how to alleviate the issues caused by signal features extremely high sensitivity, down to −137 dBm, and a
attenuation and collision in high-density wireless devices, maximal link budget of up to 157 dB [23]. Several parameters
this study proposes a LoRa mesh networking system and of LoRa PHY can be adjusted for different performance goals,
evaluates its performance. We thereby identify a new including power level, SF, bandwidth (BW), and coding rate
approach for fulfilling the practical need to collect data (CR). Changes in power level, SF, BW, and CR involve
from IoT sensors scattered over a campus-sized area multiple tradeoffs among transmission time on air (data
without using expensive 3G/4G networks or laying cables throughput), power consumption, and transmission range, and
for data transmission. different tradeoffs result in varying receiver sensitivities,
2. This study is one of the few works to empirically evaluate transmission time on air, and signal-to-noise ratios (SNRs).
the performance of LoRa mesh networking. The evaluation In a study of how SF affects receiver sensitivity in LoRa and
illustrates that LoRa signals can be attenuated or blocked LoRaWAN [13], researchers used a Semtech SX1276 Mbed
by buildings and other obstacles and that the proposed shield as the end device and a Cisco 910 industrial router as the
design can mitigate this problem by using mesh gateway. The BW setting of the LoRa device was 125 kHz, the
networking to increase PDR without deploying additional CR was 4/5, and the power output level was set to the minimum,
gateways. 2 dBm, with a 3-dBi antenna. A packet transmission test
3. The proposed LoRa mesh network module can be showed that the receiver sensitivity (the minimum signal
integrated with other IoT applications to help collect data strength needed to decode the incoming packet correctly) was
from distributed sensors, bypassing complex data −123 dBm with an SF of 7. This improved to −135 dBm when
transmission and collection issues. Users do not need to the SF was increased to 12, that is, larger SF resulted in better
know how to build a network or route the data because all receiver sensitivity and longer transmission range. The
the communication protocols are contained in the module. researchers also tested the transmission range under different
This significantly simplifies the use of long-range wireless SF settings in an outdoor, urban area; the results indicated that,
communication and facilitates IoT applications over large with an SF of 12, the transmission range could reach 3500 m
regions. with a 40% PDR; the PDR collapsed to 0% at SF 7 or 9.
However, the larger SF also resulted in slow data throughput.
II. RELATED WORK AND DESIGN GOALS The researchers tested the throughput of LoRaWAN at BW =
Several long-range wireless communication technologies 125 kHz, and the results illustrated that the throughput was
have recently been proposed. LoRa was one of the first such about 780 bytes/s at an SF of 7 and dropped to only 28 bytes/s at
 3

an SF of 12. This shows that the channel data capacity was environment. ISA100.11a implements its routing mechanism in
limited if the SF was large. a data link layer for data forwarding based on routing graphs,
The design and drawbacks of LoRaWAN are similar to those which are created by the System Manager in the ISA100.11a
of the ALOHA [17] protocol, as there is no mechanism to architecture. Both WirelessHART and ISA100.11a can provide
arbitrate access to a shared wireless frequency. Transmissions better communication reliability as the network is centrally
by more than one transmitter at the same time may result in data managed, but they are not specifically designed for monitoring
collision. LoRaWAN uses this approach to simplify the design large areas like a university campus. Therefore, it may be
of the media access control (MAC) layer for battery-powered inappropriate to make a direct comparison between these
LoRa device, which is required for saving valuable energy, thus standards and the design proposed here.
prolonging its lifetime. Therefore, a LoRa device can transmit The conclusion based on these previous studies, focusing on
data to the gateway at any time, causing significant packet evaluating the performance of LoRa/LoRaWAN, is clear: to
collision if many devices transmit signal simultaneously [11]. serve more high-density and indoor LoRa devices requires
To test long-range outdoor transmission, Petäjäjärvi and deployment of more LoRa gateways. However, this approach
colleagues [14] installed a LoRa node on the roof rack of a car defeats the original purpose of LoRa/LoRaWAN, that is, to
and another on the radio mast of a boat. They tested the provide an inexpensive and efficient IoT communication
communication range of their system using a frequency of 868 service. Another issue is that users may need to pay telecom
MHz and an output power level of 14 dBm, with an SF of 12. operators for subscriptions to a LoRaWAN service, and, even
They observed that the maximal communication range in an as subscribers, they might not receive good communication
open area was more than 15 km on land and 30 km over water. quality if the operator’s gateway is far away.
However, this good performance is mainly due to lack of In response to these issues, this study takes a different
obstructions and use of only a few LoRa devices transmitting approach by building a wireless mesh network system based on
data on the same frequency simultaneously. LoRa PHY rather than LoRaWAN. The proposed design can
When a gateway needs to serve a large number of LoRa collect data from IoT sensors distributed across a large region,
devices at the same time, the capacity of the uplink channel such as a tall building or a university campus, using only a
available to a LoRaWAN node strongly depends on the single LoRa gateway. The design allows all LoRa devices to act
distance from the base station [15]. With so many devices, the as data routers, thereby enhancing the communication
capacity could drop to a mere 100 bits/s on average, especially performance of indoor devices that may have difficulty
for LoRa channels with high SF and for the most distant nodes. communicating with the gateway directly. The design goals of
Additionally, in LoRaWAN, there is no clear channel the proposed system are as follows:
assessment (CCA) mechanism, which increases the probability
of packet collision. The PDR can drop to just 25% when node 1. The campus-wide wireless mesh network must be able to
density is very high [11]. Deploying additional gateways can collect data from IoT sensors, even indoor sensors. The
increase the performance in terms of receiving data, but at network should be able to function in environments with
added cost. Directional antennas have also been considered as a physical obstacles such as buildings, and it should provide
means of alleviating interference in LoRa networks [12], but automatic routing of data back to the gateway. This is very
only simulations have been evaluated to date. The time-related important if nonprofessional users are to manage the
performance of LoRa and LoRaWAN was also evaluated [16] system.
to determine the uncertainty of schedule to transmission and 2. The deployment procedure for the IoT sensor should not
long-term clock stability. require any prior configuration. Users should be able to
One commercial device, from NiceRF [18] can supposedly simply install their IoT sensors without considering
help form a LoRa mesh network, but very limited information communication issues. Unlike LoRaWAN, there should
is provided in its datasheet, including information about how to be no need to add gateways to increase performance; only
form a mesh network, how to relay data, and the one gateway should be needed.
communication performance of the device. Page 8 of the device 3. The wireless mesh network must be manageable.
datasheet [18] indicates that it is simply based on repeating the Information about the network routing path, data delivery
received data in the same network ID. Therefore, this approach ratio, received signal strength (RSS), and SNR of each
is similar to the ad hoc on-demand distance vector routing path should be collected and displayed on the gateway.
(AODV) method of Zigbee and the managed flood method of This information should help the manager understand the
Bluetooth 5.0 mesh networks [19]. This approach should work status of the system and correct communication problems
when network traffic is low, but it may be difficult to maintain in case of weak data links.
communication quality as traffic increases, and it may not be 4. The design should be empirically verified in an outdoor
able to manage the network topology. In contrast, the industrial environment to ensure that the system works in practical
application standard time-slotted channel hopping (TSCH) scenarios.
wireless mesh network systems such as WirelessHART [20] These goals guided the design of our system, which is
and the ISA100.11a [21] standard both support graph routing as described in the next section.
well as source routing; both standards are based on IEEE
802.15.4. In the architecture of WirelessHART, a special
computer Network Manager manages the network topology III. SYSTEM ARCHITECTURE AND DESIGN
and communication schedule. It also manages communication In this section, we describe the design of the system.
quality and adapts to different types of traffic and changes in its
 4

Received JOIN? Add

the node to the
child_list.
START

USB UART to USB IC No

CP2102
Is child_list
Application empty?
Processor
UART
MCU RF Transceiver Yes
Nano100LE3BN SX1278
Start to gather data
from child_list
sequentially
Fig. 2. The architecture of the proposed LoRa wireless mesh Broadcast beacon
periodically Sub-function
network module. Node_query

The basic design of the LoRa wireless mesh network module

is shown in Fig. 2. It is based on a Nuvoton Nano100LE3BN Fig. 4. Basic flowchart of a gateway.
[22], an ARM M0 microprocessor, and a Semtech SX1278 [23]
Sub-function
LoRa RF 430 MHz transceiver with a 1.9-dBi gain helical Node_query
antenna. The current design focuses on wireless
communication performance rather than, for instance, low Send QUERY request to
power requirements; therefore, all nodes and the gateway are the target node
wall-powered via 5V USB adapters.
The microprocessor communicates with the LoRa Got Timeout Timeout_count
transceiver via a serial peripheral interface (SPI) to send and data/Ack
+1
receive data wirelessly. The microprocessor is also integrated or Timeout?
with a USB-UART converter Silicon Lab CP2102 to convert
Got data
UART to USB for communicating with the application Timeout_count No
processor. The nodes and the gateway use the same hardware Process packet >5 ?
platform. All mesh networking protocols are run on an ARM
M0 microprocessor. This module can act as a radio module, Yes
similar to XBee [12], to accept AT-commands from an Timeout_count = 0
Set target node
application processor with USB or UART. An example of an missing
AT-command list can be found in [24]; therefore, the
application processor can use this module as a network End
interface, so users do not need to worry about how the wireless Fig. 5. Flowchart of sub-function node_query of a gateway.
network is formed.
We introduce a network formation scenario, schematically
shown in Fig. 4. Child_list is a data structure that is initially
empty. It is held by the gateway and will store the list of LoRa
nodes joined to its network. In the initial stage, the gateway will
broadcast beacons periodically (i.e., every 60 seconds) inviting
other LoRa nodes (hereafter called nodes) to join its network, as
Management shown in the flowchart. A LoRa node hearing the beacon from
LoRa Wireless Mesh Gateway User Interface Application the nearby gateway may join this network by sending a JOIN
Network request and setting the gateway as its parent.
Fig. 3. System architecture. Fig. 4 shows how the gateway adds the joined node to its
child_list and then begins to gather data by calling the
Fig. 3 illustrates the architecture of the proposed LoRa sub-function node_query, as illustrated in Fig. 5. This function
wireless mesh network system. The integration of the IoT sends a QUERY request to the target node to gather data from
sensors with the LoRa wireless mesh network module forms the the node in the child_list. If the QUERY request times out, then
mesh network automatically. The gateway of this system will the function adds one to the counter timeout_count, marking the
sequentially collect the data accumulated in the IoT sensors. target node as missing if timeout_count > 5. The gateway will
The gateway outputs the data to the management user interface, remove the missing node from its child_list and will not attempt
displaying the status of the wireless mesh network and showing to gather data from it in the next node query. Once the gateway
the data in the IoT sensors. The user’s application can read the has the first child, and child_list is not empty, the gateway will
data from the user interface, which is useful in applications stop broadcasting the periodical beacon. Instead, the gateway
such as electricity/water–meter monitoring or solar panel will use the data packets sent to its children as beacons for
management. notifying other nodes to join it if necessary, as shown in Fig. 4.
 5

If a node can hear the beacon from the gateway, it must be able If not receive any
to overhear the data packet from the same gateway. So this packet from GW for
START k minutes, find a new
approach saves time by eliminating unnecessary beacon
parent
broadcasting by the gateway.
Listen/Overhear to beacon or
As shown in Fig. 6, a power-on node m will listen for a data packet from nearby
beacon from a nearby gateway or will overhear a data packet gateway or nodes
(functioning as a beacon for m) from a nearby gateway or node. Receive a beacon or
If m has not yet joined the network, it will choose a suitable packet
network parent (either the gateway or a node). The choice of a Check the received
Already joined Yes command packet or
suitable network parent is based on multiple factors. In this a network? data package and
design, the node stores the received signal strength indicator respond properly
No
(e.g., RSSI) of recently received beacons/packets from different
Choose a suitable
gateways or nodes in a candidate list. The node’s hop-count parent and send Sub-function
distance to the gateway is also recorded in the list. The node m JOIN request Process_packet
selects the best parent from the list based on its own subjective
decision, i.e., the node with the strongest RSSI and with the
smallest hop-count distance to the gateway in the present Wait for
moment. The reason behind this is that a node will try to find a confirmation
parent that is the closest to the gateway but that also has good
communication quality.
Therefore, considering buildings, obstacles, interferences in
the field, and the location of the node, that is, whether it is Fig. 6. The basic flowchart of a LoRa node.
indoors or outdoors or on the first floor or the top floor, the
geographically nearest node based on the two-dimensional Sub-function
Process_packet
top-down view is not always the best parent candidate. The
various methods for deciding on a suitable parent candidate are
not discussed deeply in this manuscript. In the practical Determine
packet type
implementation, the gateway has a global-view topological
Receive a JOIN i.e., Respond to
map of the mesh network after the network is formed, and it can Receive a
request GW’s data request or
data/command packet
force assignment of a new parent to a node according to its command
comprehensive information. This approach is reasonable, as it
is very difficult for a node to find the best parent with its own
Yes
local-view and limited knowledge. Process and
Am I the packet
After the node m decides on a suitable parent, it sends a JOIN destination?
update child_list
request and waits for confirmation before joining the network. if accept this No
Process the packet
If the node m receives a confirmation, then it will join the JOIN request and respond
network. Afterward, if node m receives a data or command Is the packet
source from No
packet, it will process this packet and respond as appropriate by
my parent or
calling the sub-function process_packet. child_list?
As shown in Fig. 7, process_packet determines what type of
packet has been received and processes the packet as Respond to this Yes Drop the packet
appropriate. If a node n receives a JOIN request from another request
Relay the packet
node, then n may accept or reject this request based on its
current status (for example, n may reject the JOIN request if n
already has too many children) and respond accordingly. If the
Return
incoming packet is a data or command packet, and if the packet
destination is n itself, then n will process the packet and Fig. 7. The flowchart of the sub-function process_packet of a
respond accordingly; for example, n will respond to a data LoRa node.
request or command from the GW. If the destination is not n
itself but the packet source is n’s parent or child, then n will
help to relay this packet to the destination; otherwise, it will
drop the packet.
 6

Query 100% 95% 98% 98% 91%

M1
Node 1 Node 1 Node 3 90%
80% 94% 75%
Gateway M2 70%
56% 58%
54%

PDR(%)
60%
Node 2 50% 40%
40% 28%
30%
M4
Query M3 20%
Node 3 Node 1 Node 3 10% 0% 0% 0% 0%
M5 0%
Gateway 300 400 500 700 800 1000
M6
SF6 56% 40% 0% 0% 0% 0%
Node 2 SF9 95% 75% 28% 0% 0% 0%
SF12 94% 98% 98% 91% 58% 54%
Fig. 8. Procedure for a node joining the network and querying Distance (m)
the gateway.
Fig. 9. PDR at different distances under different SF settings.
We present an example scenario, illustrated in Fig. 8, in
which node 1 joins the network and sets the gateway as its First, we tested the proposed LoRa module at different
parent. The gateway adds node 1 to its child_list. Then, the transmission distances by setting the output power level to 20
gateway queries nodes in its child_list for data. In this design, dBm and the BW to 125 kHz. Two nodes were placed 1 m
the gateway initiates the query procedure. As shown in Fig. 8, above the ground. The transmitter sends 200 packets to the
the gateway sends query M1 to node 1, and node 1 reports its receiver, which is placed in the transmitter’s line of sight on a
data M2 back to the gateway. Meanwhile, the nearby node 3 street in an urban area, with moving and stationary obstacles
overhears M2. Note that M2 for node 3 is a beacon. Thus, node 3 (i.e., pedestrians, cars, and buildings) between them. This
joins the network and sets node 1 as its parent. Node 1 notifies experiment tested the PDR of two nodes under different SF
the gateway of the newly joined node 3. The mesh network is settings, including SF = 6, 9, and 12. According to the design of
formed and all nodes are connected to the gateway based on the the LoRa PHY layer, larger SF paired with higher receiver
procedure. Therefore, in the next query round, the gateway will sensitivity should result in higher PDR. Fig. 9 shows that SF =
query node 3 to gather its data using the message sequence M3, 12 yielded the best PDR among all the tested distances. With
M4, M5, and M6. Based on this sequence, all the nodes can join SF = 6, the PDR was 56% at 300 m and dropped to 0% at 500 m.
this multi-hop wireless mesh network, and the gateway can With SF = 12, the PDR was 98% at 500 m, observably better
query all the nodes. than SF = 6. Even when the distance was extended to 1000 m,
In addition, because the gateway has a global-view the PDR remained at 54%. This experiment confirmed that
topological map of the mesh network and has all the routing large SF can increase the transmission range and data delivery
path information, in this design, the node can ask the gateway performance.
for the routing path to another node; therefore, the routing is
implemented. However, in the real-world application, in most B. In-Building Range Test
cases, the gateway will initiate the query procedure for
collecting data from the nodes; direct communication between In order to further investigate the communication
nodes is used less frequently. performance of the designed LoRa module, we conducted an
experiment in a five-story classroom building on our university
IV. EVALUATION AND DISCUSSION campus to test how the LoRa signal travelled vertically across
We evaluated the proposed system to verify the applicability the building’s floors. The receiver node was installed on the
of the design. In this section, we describe and discuss our fifth floor (5F), the top floor of this building, and transmitter
evaluation. nodes were installed on 1F, 2F, 3F, 4F, and 5F. For two nodes,
we used the same test procedure as in the previous test.
Different LoRa PHY settings were tested, including various SF
A. Preliminary Experiments
and BW values. The RSSI, PDR, and time on air were collected
for evaluation.
 7

SF=6 SF=8 SF=9 SF=10 SF=11 SF=12 particularly at SF = 6 on 1F. However, in both tests, the PDR of
100% SF = 6 was still low compared with PDR at other SF settings.
90% While the PDRs were all almost 100% at SF settings from 7 to
80% 12 (Fig. 11), the PDRs with SF = 6 ranged from 80% to 93%.
70% We conclude that, if we want to increase the PDR, we should
PDR(%)

60% set the SF to a larger number and the BW to a smaller number.

50% Fig. 12 illustrates how time on air (the amount of time
40% needed to transmit data wirelessly) was affected by different
30%
BW and SF settings for a packet with a 20-byte payload. The
20%
time on air is measured by using the GPIO interrupt of the
10%
0%
node's microprocessor; the GPIO interrupt was triggered when
1F 2F 3F 4F 5F the LoRa transceiver to send data and receiving the notification
Floor of data sent out from the transceiver, and to measure the time
interval between two interrupt triggering with timer. As the SPI
Fig. 10. PDR on different floors under different SF settings
between the microprocessor and the LoRa transceiver is SPI,
(BW = 250 kHz).
and the time measurement is not very accurate, some
uncertainties may be introduced, varying estimations of the
SF=6 SF=8 SF=9 SF=10 SF=11 SF=12 time on air. However, this might be acceptable as this result is
100% for understanding how SF and BW affect the time on air. We
90% can see that larger SF and smaller BW resulted in longer time
80%
on air. The time on air was only 16ms for SF = 6 and BW = 250
70%
kHz, but it increased to 2881 ms for SF = 12 and BW = 62.5
PDR (%)

60%
50%
kHz. Therefore, although larger SF and smaller BW settings
40% increase the PDR, these settings also significantly prolong the
30% time on air and may cause serious packet collision if multiple
20% transmitters send data simultaneously. We can compare this
10% drawback to single-hop LoRaWAN systems, which suffer from
0% low PDR with a high density of LoRa devices, as mentioned in
1F 2F 3F 4F 5F [13]. In the proposed system, the GW initiates all
Floor
communications and queries devices. Therefore, the LoRa node
Fig. 11. PDR on different floors under different SF settings cannot actively send out data, thus avoiding packet collision at
(BW = 125 kHz). the cost of long transmission delays. Although this delay is a
drawback, in the context of IoT data collecting, this approach
3000 prevents packet collision without introducing complications
2500 into the simplified design.
Time on Air (ms)

2000

1500 C. Campus-Scale Experiment: 19 LoRa Nodes on a University

1000 Campus
500

0
SF=12 SF=11 SF=10 SF=9 SF=8 SF=6
In the campus-scale experiment, we placed 19 LoRa nodes
BW=62.5 kHz 2881 1438 799 400 220 61 (originally, we installed 20 nodes, but node 6 was lost during
BW=125 kHz 1441 719 400 200 110 31 the experiment), distributed in an 800m×600m area of our
BW=250 kHz 721 360 201 101 56 16
university campus. We allowed the gateway to collect data at
BW=62.5 kHz BW=125 kHz BW=250 kHz
1-minute intervals for 8 days. In this setting, for each node the
Fig. 12. Time on air (in ms) under different BW and SF maximum latency is 60 seconds, depending on when the node is
settings. queried by the gateway. Each node is queried by the gateway
one time every minute, so it will be queried 60 (minutes) × 24
Fig. 10 shows the PDR at BW = 250 kHz. When the (hours) × 8 (days) = 11520 times (packets) if a node is
transmitter was placed on 1F, the LoRa radio signal had to connected to the network during all 8 days. The LoRa PHY
travel through four concrete stories to 5F. Therefore, all the parameters were set to SF = 12 and BW = 250 kHz. In this
PDRs for 1F are low compared with those of the other floors. experiment, we focused only on communication performance;
For the LoRa transceiver, the larger bandwidth (BW) can the content of data packets was not considered. The nodes form
increase the PHY rate and reduce the time on air, but it also a wireless mesh network automatically and the gateway
lowers the PDR as the output power is distributed across a collects data from these nodes periodically based on a
larger bandwidth. Setting SF = 6 further lowers the receiver multi-hop network.
sensitivity and causes even lower PDR. As a result, the PDRs at Referring back to the snapshot of the network topology
SF = 6 are the lowest among the SF settings in this experiment. shown in Fig. 1, the color of the circles with node numbers
Fig. 11 shows the PDRs at BW = 125 kHz. All PDRs at BW indicates the floor of the building on which each node was
= 125 kHz were higher than the PDRs at BW = 250 kHz, installed (e.g., 1F or RF). Nodes at 1F, like node 12, need the
 8

help of other nodes to relay data to the gateway (denoted by requests from the gateway, then report data only when queried,
GW in Fig. 1) because they cannot communicate with the GW avoiding the problem of intra-network packet collision.
directly. In this case, node 14 at 5F and node 3 at 8F assisted
node 12 in relaying its data to the GW (the path is denoted by
12-14-3-GW).

800m
GW
19
16 GW

Science Building
5
11
2

20
Node
13
12 9
14 Eng. Building
3 General Classrom ：RF
15 ：8F
：5F
Library ：4F
10 ：2F
：1F
600m
Literature Building Management Building
4
7
1
8
Sport Field
Activity Center

18
17
Node Node Node
Fig. 13. Another network topology snapshot of the LoRa mesh
network devices.
Fig. 14. Pictures of the GW and nodes deployed in this
Another snapshot of the network topology was taken and is experiment.
shown in Fig. 13. In this snapshot, node 2 has changed its
parent to node 3 and has skipped node 14, communicating with For these low-PDR nodes, with the exception of node 4, the
the gateway using the path 12-3-GW. These results confirm PDRs increased dramatically if the hop-count constraint was
that our LoRa mesh network module can dynamically form a relaxed to 3Hop, allowing a node to use, at most, three hops to
wireless mesh network and adapt to environmental changes. deliver data to the GW. For example, the PDR of node 12 was
Note that, in both snapshots, node 18 was not connected to the 0% at 1Hop, but it increased to 94.94% at 3Hop. For all nodes,
network at the moment the snapshot was taken. Further the average 1Hop PDR was 58.7% and the average 3Hop PDR
examination revealed that it was located indoors and was not in was 88.49%. These results support our assertion that the
a good position to contact a potential parent. Node 18 still proposed LoRa mesh networking solution can increase PDR
achieved at most an average PDR of 69.74%, as show in Fig. without deploying an additional GW, avoiding additional
15. infrastructure costs associated with collecting IoT sensor data.
Fig. 14 shows how the GW and the nodes were deployed in With regard to the relationship between transmission
this experiment. The GW uses a 3-dBi omni-antenna installed distance and PDR, multiple factors affect the PDR, including
on the roof of the building. All nodes were installed at locations buildings, obstacles, interferences in the field, the location of
that were not in the line of sight of the GW. the node, such as whether it is indoors or outdoors, or whether it
Fig. 15 shows the end-to-end PDR of this campus-scale is on a first floor or top floor. Thus, the geographical distance
experiment. The PDR was counted on the gateway side, based based on the top-down view does not always correlate with the
on the data log collected in this experiment. For ease of PDR. Transmission distance is not a good factor for estimating
evaluation, data were analyzed by applying different PDR, which is very difficult to model without considering these
limited-hop counts. In the figure, 1Hop indicates that all nodes factors.
were restricted to sending data to the GW with only one hop, Fig. 16 shows the parent candidates for node 10 in the
that is, with no relay from other nodes. If a data packet was experiment. Note that this does not include the case in which
delivered back to the GW with more than two hops, this packet node 10 sends data directly to the GW. In this design, a node
was not counted in the 1Hop PDR. Therefore, the performance chooses its parent according to its current local information. If
of 1Hop includes only packets delivered with one hop, so it can the environment changes or the node loses contact with its
be seen as a LoRaWAN simulation. We can see that the 1Hop current parent for a period of time (i.e., 5 minutes pass without
PDRs are relatively low. Some nodes did not successfully receiving any command or data request from its parent), then
deliver any data to the GW, resulting in PDRs of 0% (i.e., nodes the node will try to find a new parent by sending a JOIN request.
1, 4, 8, 12, 17, 18, and 19 in Fig. 15). This result indicates that Fig. 16 shows that if node 10 cannot join the GW, then it has
the one-hop LoRa network does not perform well for indoor four parent candidates: nodes 3, 9, 14, and 15. Referring to the
nodes, which again confirmed the analysis of previous studies pie chart on the left side of the figure, we can see that node 9
[11],[12],[13]. The PDR could fall even further if nodes are was node 10’s parent 74% of time and transferred 7,085 packets.
allowed to send data actively (such as in the ALOHA protocol The bar chart on the right side of the figure indicates that node 9
in LoRaWAN approach). The proposed design, by contrast, had the highest average PDR, almost 100%, and the strongest
uses a polling-based approach whereby all the nodes listen for RSSI among all the parent candidates.
 9

Fig. 17 shows the relationship between RSSI and PDR. send out data only if it has received a command to do so from
These signal quality statistics are based on delivery of all the gateway. If nodes are allowed to send data actively, without
packets to the gateway. We can see that the average PDR was waiting for a data request from the gateway, as in the ALOHA
above 96.1% when RSSI was larger than −110 dBm, and the LoRaWAN approach, packet collision may decrease PDR in
average PDR dropped to 5.8% when the RSSI was smaller than the event that many nodes send data simultaneously. Striking a
−130 dBm. In the range from −110 dBm to −130 dBm, the balance between transmission delay and packet collision is a
RSSIs decreased as the standard deviation of PDR increased, complex task.
indicating unreliable communication. The maximum standard Regarding the timing performance degradation, as the
deviation of PDR occurred when RSSI was in the range of −120 proposed method uses a mesh network, the number of nodes
dBm to −130 dBm. In this range, packet delivery was unreliable, that a network can serve is smaller than for conventional
so nodes should avoid low RSSI parents if a stable star-topology networks. Referring to Fig. 8, for a one-hop node
communication path is to be established. (i.e., node 1 in Fig. 8) the message transmission time is 2t,
Fig. 18 illustrates the relationship between PDR and SNR. where t is the time to transfer a message M. For a two-hop node
We can see that larger SNR accompanies higher PDR and (i.e., node 3 in the figure), the time is 4t, and for a three-hop
smaller standard deviation of SNR. For all SNR values larger node it is 6t. Considering a given period time p, the number of
than −5, the PDRs are greater than 94.2%. A significant cliff nodes from which a gateway can collect data in a star-topology
appears in the PDR trend at SNR below -20, dropping the network is p2t, but this drops to only one-third of that (p6t)
average PDR to only 5.4%; by contrast, the PDR was still 65% when all the nodes are three-hop nodes in the mesh network.
when the SNR was in the range of −15 to −20. The mesh network is a solution for trading off long range with
The above information in combination with the results multiple hops, and it can reduce the number of nodes that can
suggests that a stable LoRa PHY link with a PDR > 90% can be be served in given period of time. Therefore, this degradation of
achieved with RSSI > −110 dBm and SNR > −5. By contrast, timing performance should be considered in designing the IoT
this would be impractical for LoRaWAN systems, which data collecting system. Meanwhile, in this design, in order to
require deployment of more gateways near the nodes to achieve avoid multiple nodes simultaneously uploading their data over
the same RSSI and SNR. This further supports the proposed the wireless network, causing data collisions, the gateway is
design, which can increase PDR while circumventing the responsible for polling data from the nodes and the nodes are
requirement for additional gateways. not allowed to actively upload data. The drawback of this
approach is that the latency between generating data in a node
to uploading that data to the gateway is high. However, the
V. DISCUSSION AND CONCLUSION polling approach has greater flexibility by allowing the
This study presents the design of a LoRa wireless mesh gateway to decide when to query which node, in contrast to the
network system for collecting data from IoT sensors distributed approach in which communication is initiated by nodes or
across a large geographical area. Our preliminary and centralized scheduling.
in-building experiments confirmed the findings of previous In addition, questions remain about the security of such a
studies regarding the effect of various LoRa PHY parameters network. As this study focuses on data transmission, security
(e.g., SF, BW) on communication performance, including PDR issues have not been discussed in this paper. However, we can
and time on air. In our campus-scale experiment, 19 LoRa directly apply LoRaWAN's security solution, which uses
nodes were distributed over an 800m × 600m area. Data Application Server to provide application session keys for
analysis comparing the LoRa PHY one-hop wireless network end-to-end security and Network Server to provide network
(i.e., simulated LoRaWAN) with our proposed LoRa mesh session keys for network session security in a multi-hop
network shows that the mesh network can significantly increase network.
PDR without installing additional gateways. In addition, parent Finally, the cost in terms of power consumption may also be
selection, PDR vs. RSSI, and PDR vs. SNR were an issue if this system is to be implemented with, for instance,
comprehensively analyzed. The results suggest thresholds for battery-powered nodes. Regarding the issue of low power at
RSSI and SNR for maintaining communication quality. These nodes, the router node needs to help relay data from other nodes,
thresholds may be impractical for LoRaWAN architecture, so its wireless transceiver must be on at all times; thus, high
where achieving the ideal RSSI and SNR would require power consumption is unavoidable. Therefore, in the current
deployment of many additional gateways, thus raising the cost design, all the router nodes are wall-powered. If some nodes
of the infrastructure. have limited energy, a straightforward approach can be used to
As far as we know, this is the first academic study extending disable the data relaying function of these nodes, similar to
LoRa PHY to mesh networking. It is also the first to evaluate LoRa Class A End Devices [25]. Thus, energy used in wireless
such a mesh network in practical experiments over a large communication can be saved as a simple node, that is, one with
geographical area. This study has explored the potential of IoT its data relaying function disabled, only enables its wireless
sensor deployment and monitoring and collecting of data from transceiver while it needs to send or receive data. However, this
those IoT sensors in an area requiring long-range transmissions. requires further modifying the protocol and introducing friend
Further study on this topic will focus on adjusting the SF/BW nodes as defined in the Bluetooth 5.0 mesh network [19] to
parameters dynamically to raise PDR. Further study will also assist the simple nodes in buffering and relaying data. As the
be required to understand how to minimize delays in data current design has been demonstrated the correctness and
transmission from the nodes. In the current design, the node can performance of the system, such modifications of specific
requirements can be pursued in future work.
 10

100%
90%
80%
PDR(%) 70%
60%
50%
40%
30%
20%
10%
0%
1 2 3 4 5 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1Hop 0.00 98.6 98.8 0.00 89.4 91.9 0.00 97.0 86.6 83.5 0.00 95.3 95.2 83.3 95.6 0% 0% 0.33 99.2
2Hop 6.94 98.6 98.8 0.00 98.6 91.9 71.1 97.0 94.6 83.5 30.2 95.1 96.9 91.4 96.4 70.8 0% 95.8 99.2
3Hop 65.6 99.3 98.8 8.3% 98.6 96.8 91.2 97.0 99.5 83.5 94.9 95.5 97.6 99.9 97.0 91.0 69.7 97.3 99.2
Node number

Fig. 15. PDR under different hop-count limitations.

Node 10
100% -140
15, 307, 80%

RSSI (dBm)
14, 10, -130
PDR (%)

3%
0% 3, 2180, 60%
23% -120
40%
20% -110
9, 7085,
0% -100
74%
3 9 14 15
Node 10
Node number
RSSI PDR

Fig. 16. The parent selections of node 10 in this experiment.

100%
90%
80%
70%
60%
PDR(%)

50%
40%
30%
20%
10%
0%
-80≦RSSI<- -90≦RSSI<- -100≦RSSI<- -110≦RSSI<- -120≦RSSI<-
RSSI≧-80 RSSI<-130
90 100 110 120 130
Avg 98.9% 98.6% 97.1% 96.1% 86.1% 69.3% 5.8%
Max 100% 100% 100% 100% 100% 100% 16.7%
Min 93.3% 85.0% 16.7% 0.0% 16.7% 0.0% 1.7%
Std 1.4% 2.3% 8.6% 10.0% 18.4% 34.1% 4.5%
RSSI level (dBm)

Fig. 17. The relationship between PDR and RSSI.

100%
90%
80%
70%
60%
PDR(%)

50%
40%
30%
20%
10%
0%
-15≦SNR<- -20≦SNR<-
SNR≧5 0≦SNR<5 -5≦SNR<0 -10≦SNR<-5 SNR<-20
10 15
Avg 98.8% 97.6% 94.2% 87.0% 74.6% 65.0% 5.4%
Max 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% 16.7%
Min 85.0% 16.7% 0.0% 1.7% 0.0% 0.0% 1.7%
Std 1.9% 6.9% 11.2% 17.4% 30.9% 41.4% 4.4%
SNR level

Fig. 18. The relationship between PDR and SNR.

 11

ACKNOWLEDGEMENTS [17] Abramson, N. The Aloha System: Final Technical Report; Hawaii
University: Honolulu, HI, USA, 1974.
The authors would like to thank ICP DAS Co. Ltd, Taiwan, for [18] "NiceRF LoRa-MESH MESH series Mesh network modulesmodules", 2017. [Online]
its financial and technology supports. Available: [Link]
[Link]
[19] "Bluetooth 5.0 Mesh Networking Specifications
Specifications", 2017, [Online]
REFERENCES Available:
[Link]
[Link] [Accessed:
[1] Yun Cheng, Xiucheng Li, Zhijun Li, Shouxu Jiang (Harbin Institute of 30- Dec- 2017].
Technology), Yilong Li, Ji Jia (Air Scientific), Xiaofan Jiang “AirCloud:
[20] IEC 62591, “Industrial communication networks - Wireless
A Cloud-based Air-Quality
Quality Monitoring System for Everyone,” SenSys
communication network and communication profiles -
2014, Nov.2014
WirelessHART™”, 2010.
[2] Jalpa Shah, Biswajit Mishra “IoT enabled Environmental Monitoring
[21] "The
The ISA100 Standards Overview & Status",Status 2017. [Online] Available:
System for Smart Cities’’, International Confere
Conference on Internet of Things
[Link]
[Link]/Documents/PDF/ISA100_Standards-Overview_WC
and Applications (IOTA), Jan.2016
I_Webinar_V3_12Apr10.aspx
[3] B. Andò, S. Baglio, A. Pistorio, G. M. Tina and C. Ventura, "Sentinella:
[22] [Link]. (2017). NUC100 Advanced Series. [online] Available at:
Smart Monitoring of Photovoltaic Systems at Panel Level," in IEEE
[Link]
[Link]
Transactions on Instrumentation and Measurement, vol. 64, no. 8, pp.
mcus/nuc100-200-advanced-series/series/ [Accessed 19 Sep. 2017].
2188-2199, Aug. 2015.
[23] [Link]. (2017). SX1278 137 MHz to 525 MHz Low Power Long
[4] G. Mois, S. Folea and T. Sanislav, "Analysis of Three IoT
IoT-Based Wireless
Range Transceiver [online] Available at:
Sensors for Environmental Monitoring," in IEEE Transactions on
[Link]
[Link] [Accessed
Instrumentation and Measurement, vol. 66, no. 8, pp. 20562056-2064, Aug.
19 Sep. 2017].
2017.
[24] " XBee/XBee-PRO PRO S1 802.15.4 (Legacy) RF Modules", 2017. [Online]
[5] "Sigfox - The Global Communications Service Provider for the Internet
Available:
of Things (IoT)", [Link], 2017. [Online]. Available:
[Link]
ources/documentation/Digidocs/90000982/Def
[Link] [Accessed: 02- Sep-
Sep 2017].
[Link] [Accessed: 30- Dec- 2017].
[6] "LoRa-alliance".
alliance". [Online]. Available: [Link]
[Link]
[25] "LoRaWAN Classes", 2017, [Online] Available:
[Accessed: 02- Sep- 2017].
[Link]
[Link]/technology [Accessed: 30- Dec- 2017].
[7] K. Flynn, "Standardization of NB-IOT IOT completed", [Link], 2017.
[Online]. Available:
[Link]
news/1785-nb_iot_complete. Huang-Cheng Lee (SM’12) received
Huang
[Accessed: 02- Sep- 2017]. his Ph.D. from the National Tsing-Hua
[8] "LoRa Modulation Basics", 2017. [Online]. Available: University, Hsin
Hsin-Chu, Taiwan, in 2010.
[Link]
.[Link]/images/datasheet/[Link]. [Accessed:
03- Sep- 2017].
He has been in the industry since 2000
[9] M. Centenaro, L. Vangelista, A. Zanella and M. Zorzi, “Long “Long-Range
and has a wide breadth of experience
Communications in Unlicensed Bands: the Rising Stars in the IoT and designing personal digital assistants,
Smart City Scenarios”. Fabulous 2015 conference, S Sept. 2015. cellular phones
phones, and low-power
[10] Kai-Hsiang Ke, Qi-Wen Liang, Guan-Jie Jie Zeng, Jun Jun-Han Lin, embedded systems. He joined the
Huang-Chen
Chen Lee, A LoRa Wireless Mesh Networking Module for Department of Communications
Campus-Scale
Scale Monitoring, The 16th International Conference on
Information Processing in Sensor Networks (IPSN 2017), Demo Session,
Session Engineering and Electrical Engineering,
Pittsburgh, Pennsylvania, April 2017. National Chung-Cheng University, Chiayi, Taiwan, in 2011,
[11] N. Varsier and J. Schwoerer, "Capacity limits of LoRaWAN technology where he has been an Associate Professor since 2015. He has a
for smart metering applications," 2017 IEEE International Conference on strong track record of collaborating with industry partners to
Communications (ICC), Paris, 2017, pp. 1-6.
transfer technologies from academ
academic research into practice. His
[12] Thiemo Voigt, Martin Bor, Utz Roedig,edig, and Juan Alonso,
Alonso "Mitigating
Inter-network
network Interference in LoRa Networks,"
Networks In International
current research interests include wireless sensors, mesh
Conference on Embedded Wireless Systems and Networks (EWSN ’17). networking, internet of things,
things and low-power embedded
Uppsala, Sweden, Feb, 2017. systems, especially for natural and industrial environment
[13] Augustin, A., Yi, J., Clausen, T., & Townsley, W. M. (2016). A study of monitoring. Dr. Lee has been an Associate Editor of the IEEE
LoRa: Long range & low power networks for the internet of things. Transactions on Instrumentation and Measurement since 2015
Sensors, 16(9), 1466.
and an Associate Editor of the IEEE Sensors Journal since
[14] Juha Petäjäjärvi, Konstantin Mikhaylov, Antti Roivainen, Tuomo
Hänninen, Marko Pettissalo “On the Coverage of LPWANs: Range 2017.
Evaluation and Channel Attenuation Model for LoRa Technology”
International Conference on ITS Telecommunications on 14th, 2015. Kai-Hsiang
Hsiang Ke received his B.S. and
[15] Mikhaylov, K., Petäjäjärvi, J., & Haenninen, T. (2016, May). “Analysis M.S. degrees from the Department of
of capacity and scalability of the LoRa low power wide area network Communications Engineering,
technology”. In European Wireless 2016; 22th European Wireless
Conference; Proceedings of (pp. 1-6). VDE. National Chung
Chung-Cheng University,
[16] M. Rizzi, P. Ferrari, A. Flammini and E. Sisinni, "Evaluation of the IoT Taiwan
Taiwan, in 2015 and 2017. He is
LoRaWAN Solution for Distributed Measurement Applications," in IEEE currently an engineer at Delta
Transactions on Instrumentation
tation and Measurement, accepted and to Electronics, Tainan, Taiwan
Taiwan.
appear, 2017.