BOM - 08078360 Bom

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

This full text paper was peer-reviewed at the direction of IEEE Instrumentation and Measurement Society prior to the

acceptance and publication.

Practical Limitations for Deployment of LoRa


Gateways
Albert Pötsch Florian Haslhofer
Linz Center of Mechatronics GmbH Institute for Communications Engineering
4040 Linz, Austria and RF-Systems
Email: [email protected] Johannes Kepler University, 4040 Linz, Austria

Abstract—In this paper, we discuss the upcoming network communicates over the air with (multiple) gateways which
stack LoRaWAN and its underlying modulation technology LoRa, relay the data messages to a central network server as can
which is an implementation of a LPWAN (Low Power Wide Area seen in Fig. 1. While the LoRa modulation is proprietary, the
Network) and consider practical limitations that can occur when
restricted Machine-to-Machine (M2M) data agreements for the LoRaWAN is an open standard being developed by the LoRa
connection of LoRa gateways to a cellular network are involved. Alliance.
We provide a theoretical estimate of the maximum possible In this paper, we focus on the usability of LoRa networks
communication range of a LoRa link depending on output that depend on gateways that are connected to the cloud using
power and spreading factor and illustrate a LoRa-transceiver’s cellular networks as IP-backhaul-interface and which are oper-
contrary behaviour of energy consumption and range versus the
achievable data rate by means of current consumption and RF ated with small and cost-effective Machine-to-Machine (M2M)
output power measurement results for different combinations rates for the network connection. Typical M2M contracts
of modulation parameters. Our analysis of the data overhead include data-volumes of less than 100 MB per month 4,5,6 , so
introduced by LoRa gateways identifies that the amount of the amount of data-overhead produced by a LoRa-gateway can
generated backhaul network traffic can cause substantial costs be a relevant issue, especially for large number of appliances
for IoT service providers in case of cellular-connected gateways
or limit the number of supported sensor nodes. when e.g. (municipal) service providers require thousands of
SIM cards to operate those gateways.
I. M OTIVATION AND I NTRODUCTION After giving an introduction into the basics of LoRa tech-
nology we provide a theoretical estimate of the maximum
The Internet of Things (IoT) is an upcoming topic in our
possible communication range of a LoRa link depending on
society. Daily used things such as washing machines, street-
output power and spreading factor. Followed by a compact
lamps or public dustbins in a city will be connected to the
overview about LoRaWAN, an open LPWAN network stack,
Internet and will send their status (e.g. completed washing
we further illustrate a LoRa-transceiver’s behavior of energy
process or full dustbin) into a cloud, indicating to other things
consumption and communication range versus the achievable
(machines, robots etc.) or humans that some action is required.
data rate by means of measurement results. Finally we analyze
Therefore, wireless networks, which can serve a huge amount
the packet overhead introduced by the LoRa gateway to create
of clients, are required. In smart homes, the most common
awareness on the importance of the amount of generated
short range communication technologies for connecting things
backhaul network traffic.
are WiFi, ZigBee or Bluetooth. In contrast, using these radio
standards would not be appropriate in smart cities or other II. R ELATED W ORK
domains like sport and wellness because of their short commu- Since LoRa was developed, several papers evaluated the new
nication range or a too high energy consumption [1]. For this technology with respect to performance, energy consumption,
area of applications Low Power Wide Area Network (LPWAN) range or size of the network. The maximum number of nodes
technologies such as LoRa 1 , SigFox 2 and Weightless 3 have supported by a single gateway is discussed by [2]. Furthermore
been developed. They all operate at sub-GHz-frequencies in this work shows a dependence of distance and capacity in the
the unlicensed ISM bands. Depending on the region, such uplink channel. In [3] the LoRaWAN indoor performance with
networks typically operate at 433, 868 or 915 MHz. nodes close to human bodies is investigated. A packet success
LoRa, which stands for Long Range, is a long-range wire- delivery ratio of 96.7% could be reached without acknowl-
less communications system, promoted by the LoRa Alliance. edgments and retransmissions. Also the work in [4] takes a
LoRaWAN (a trademark of the LoRa Alliance) involves a look at LoRaWAN traffic, such as packet error rate (PER),
protocol stack with the LoRa wireless technology as the received signal strength indication (RSSI) or signal to noise
physical layer. Within LoRaWAN a single sensor devices ratio (SNR), under various conditions. The evaluation of LoRa
1 www.lora-alliance.org 4 www.embeddedworks.net/m2m-data.html
2 www.sigfox.com 5 www.m2m-mobil.de
3 www.weightless.org 6 www.simfonymobile.com/global-mobile-connectivity

978-1-5090-5679-8/17/$31.00 ©2017 IEEE


A

AES-128 NWKSKEY

AES-128 APPSKEY

Source: https://www.zerynth.com/blog/getting-started-with-lorawan-and-python-using-zerynth-and-the-things-network

Fig. 1. Structure of a global LoRaWAN network

across a whole district was done by [5]. The authors of [6] are the customization of the LoRa modulation: Bandwidth (BW),
dealing with the interaction and impact of various nodes. It was Spreading Factor (SF) and Code Rate (CR). The SF describes
shown, that the coverage probability drops exponentially as the how many bits per symbol are encoded. Its number ranges
number of end-devices grows due to interfering signals using from SF6 (64 chips/symbol) to SF12 (4096 chips/symbol). Two
the same spreading sequence. Interference Measurements in signals with different spreading factors are orthogonal.
the European 868 MHz ISM band with focus on LoRa and By properly setting SF and BW, the time on air Tpacket
SigFox were performed by the authors of [7]. The interference (or packet duration) can be reduced or the range of the
of LoRa and IEEE 802.15.4g PHY layers is investigated by transmission can be optimized. According to the LoRa Modem
measurements found in [8]. The authors of [9] present an Designer’s Guide 7 Tpacket consist of Tpreamble plus Tpayload ,
extensive overview of what the capabilities and the limitations both are based on the symbol duration Tsym given by (1).
of LoRaWAN are and study the network scale with respect to
data rate and duty-cycle regulations. 2SF
Tsym = (1)
BW
III. L O R A BASICS LoRa modulation includes a variable error correction
The LoRa LPWAN solution includes two major compo- scheme that improves the robustness of the transmitted signal
nents. The first one, describing the physical layer, is the LoRa at the expense of redundancy. Introducing the code rate CR,
modulation, which is a CSS (Chirp Spread Spectrum) fre- the relationship between the effective data bit rate Rb , BW and
quency modulated technology developed by company Semtech SF for LoRa modulation can be expressed using Equation (2),
that uses wideband linear frequency modulated pulses whose further details can be found in the LoRa Modulation Basics
frequency increases or decreases based on the encoded infor- Application Note 8 and [10]. For a given BW of e.g. 125 kHz
mation. The main advantages of this approach are twofold: a and CR set to 54 datarates from 0.29 kbit/s (SF 12) up to
substantial increase in receiver sensitivity due to the processing 5.47 kbit/s (SF 7) can be obtained.
gain of the spread spectrum technique and a high tolerance
BW
to frequency misalignment between receiver and transmitter 7 . CR Rb = SF (2)
2SF
The second component is the LoRaWAN network protocol,
The public available documentation provided by SemTech
described in more detail in chapter IV.
does not reveal all the necessary information how the LoRa
In LoRa modulation the spreading of the spectrum is
RF modulation format is setup, in fact the coding of a LoRa
achieved by generating a chirp signal that continuously varies
packet is closed source. Attempts to decode the LoRa RF
in frequency. The frequency bandwidth of this chirp is equiv-
modulation format are carried out by open communities and
alent to the spectral bandwidth of the signal. The wanted
other companies 9,10 .
data signal is chipped at a higher data rate and modulated
onto the chirp signal. Several parameters are available for 8 www.semtech.com/images/datasheet/an1200.22.pdf
9 https://revspace.nl/DecodingLora
7 www.semtech.com/images/datasheet/LoraDesignGuide STD.pdf 10 www.link-labs.com/blog/what-is-lora
IV. L O R AWAN
LoRa defines only the physical layer (PHY) (layer 1) of
the Open Systems Interconnection (OSI) model. In contrast to
the LoRa modulation LoRaWAN is an open LPWAN standard
defining the medium access (MAC) and network (NWK) layer
(OSI layer 2 and 3) and is maintained by the LoRa Alliance 1 .
It is designed mainly for low data rate sensor networks,
wherein sensors exchange packets with the network server.
It also manages security, adaptive data rate (ADR) and de-
duplicates the received data which is potentially received by
multiple gateways.

A. Components of a LoRaWAN Network


Several components of the network are defined in the
LoRaWAN specification and are required to form a LoRaWAN
network: End-devices, gateways and the network server, as can
seen in Fig. 1.
Fig. 2. Required output power for a required communication range of a LoRa • End-devices are the low-power sensor nodes that com-
link, depending on the spreading factor SF and assuming a path-loss exponent municate with gateways using LoRa-links.
of 3 for urban area
• Gateways are the intermediate devices that forward pack-
ets coming from end-devices to a network server over an
For the communication range of LoRa different statements IP-based backhaul network allowing a higher throughput,
can be found reaching from multiple kilometers 11 to a maxi- such as Ethernet or cellular network standards. There can
mum of 15 km 12 . To estimate the communication range d of be multiple gateways in a LoRa deployment, and the same
LoRa we consider equation (3) for the path loss L, where f data packet can be received and forwarded by more than
is the LoRa-frequency and c the speed of light. Assuming one gateway.
a path-loss exponent n (PLE) of 3 corresponds to signal- • The Network server is responsible for the management
barriers in an urban area (the PLE can be set from 2 (free of the whole network. It knows about active nodes and
space) to 6 (high obstruction)). The link budget LBudget for when a new node (end-device) joins the network, it will
the transmission path is calculated by (4) with the receiver ask the application-server if the node is allowed to join
sensitivity SRX (SF, BW ) and transmitted power PT X . the network and if so, which settings to use for this node.
It de-duplicates the received data which can possibly be
2 !
received by multiple gateways, it authenticates this data

4πdf
L = 10 log10 dn (3) (to make sure that these are no replay-attacks), it forwards
c
this (encrypted) data to the application-server and it will
ask the application-server if it should send anything back.
LBudget (SF, BW ) = PT X − SRX (SF, BW ) (4)
Besides managing the data-flows, it also manages the
To obtain a theoretical estimate of the maximum possible state of the node through so called MAC-commands (e.g.
communication range of a LoRa link depending on output to change the data-rate, channels etc.).
power and spreading factor, we assumed there are no antenna A special feature of LoRaWAN is the Adaptive Data Rate
gains to get a worst case result and set the path loss L (ADR), which is also managed by the network server. The end-
equal to LBudget . Now, if the path loss is equal to the device decides if a default data rate should be used or if the
link budget, then the receiver can just receive the signal. network server should adjust the data rate. If ADR is activated,
Values of SRX (SF, BW ) are taken from the SX1301 LoRa the server optimizes the data rate for a minimum time on
concentrator/gateway transceiver datasheet 13 . air with the least possible transmission power to save battery
According to Fig. 2 the theoretical maximum range can be power. The data rate is calculated according to the maximum
achieved with SF 12 and highest transmission power (23 dBm). signal-to-noise ratio of the last 20 received messages. ADR
In an urban area with little obstructions, a gateway located is implemented such that only the network can increase and
23 km away can - under ideal conditions - successfully re- only the node can decrease the data rate. As long as there is a
ceive the sent signal. This calculation confirms the range- reliable link, the server sets a higher data rate. If the connection
information given by TheThingsNetwork 11 and DigiKey 12 . is lost, the node reduces the data rate until the message reaches
11 www.thethingsnetwork.org
the server again.
12 www.digikey.com/en/articles/techzone/2016/nov/lorawan-part-1-15-km- Unlike traditional cellular networks, end-devices are not
wireless-10-year-battery-life-iot associated with a particular gateway in order to have access
13 www.semtech.com/images/datasheet/sx1301.pdf to the network. The gateways serve simply as a link layer
relay and forward the packet received from the end-devices Class A
Transmit Rx1 Rx2
to the network server after adding information regarding the (Uplink) RX
(Downlink) (Downlink) battery-powered
delay 1
TX time on air end-device initiated activity
reception quality. Thus, an end-device is associated with a (Tpacket) RX delay 2

network server, which is responsible for detecting duplicate


Beacon Ping Slots Transmit Rx1 Rx2
packets and choosing the appropriate gateway for sending back from Gateway
additional
(Downlink) (Uplink) (Downlink) (Downlink)
Beacon
from Gateway

a reply (if any). Logically, gateways are transparent to the end- Beacon Period

devices. Class B low-latency server initiated


downlink messages using extra Ping-slots
Fig. 3 shows the composition of the three LoRaWAN classes Class C
of end-devices to address the various needs of applications: Transmit Continuous RX no-latency server initiated
(Uplink) until next Uplink downlink messages
• Class A end-devices can initiate an uplink transmission
based on their own needs. This class of devices allows
Fig. 3. Composition of the three LoRaWAN classes of end-devices
bi-directional communication, whereby each uplink trans-
mission is followed by two short downlink receive win-
dows. Downlink transmission from the server at any other
time has to wait until the next uplink transmission occurs.
Class A devices have the lowest power consumption,
but also offer less flexibility on downlink transmissions.
The support of class A features is mandatory for all
LoRaWAN end-devices.
• Class B end-devices open additional receive windows
(ping slots) at scheduled fixed time intervals for the
purpose of enabling server initiated downlink messages.
A periodic beacon from the gateway is required for
synchronization, so that the network server is able to
know when the end-device is listening.
• Class C end-devices have almost continuous receive win-
dows and therefore have maximum power consumption.
The server can initiate a transmission at any time. Further
details can be found in the LoRaWAN specification 1 .
LoRaWAN does not enable device-to-device communica-
tions, therefore packets can only be transmitted from an end-
device to the network server, or vice-versa. Device-to-device
communication can only be established by way of two gateway
transmissions.
Since LoRa Class A end-devices uses the ALOHA protocol,
there is no collision avoiding mechanism like Carrier Sense
Multiple Access/Collision Avoidance (CSMA/CA) in WiFi.
Class A does not allow for a known reaction time when the Fig. 4. LoRaWAN system architecture of the open source LoRa Server project
provided by www.brocaar.com
customer application or the server wants to address the end-
device. However, two signals can be received at the same This implementation splits the architecture in three main
time by a gateway under the condition that one signal is 6 components: The Gateway bridge, the LoRa (Network-)server
dB stronger than the other one. This was shown by [2]. and the Application-server as can be seen in Fig. 4. All these
LoRaWAN messages are encypted with two keys as can be systems are required for a complete setup of a LoRa-network.
seen in Fig. 1. One for the MAC commands and application The services can run on distributed computers as well as on
payload (NWkSKey), and one for end-to-end encryption of the the same machine, since its a pure software solution.
application payload (AppSKey). The NWSKey is only known The Gateway Bridge is responsible for the communication
by the network to prove data-integrity and the AppSKey with a gateway. It translates the UDP protocol from the
is distributed to the application-server, for decrypting the gateway to a JSON message and publishes it to the MQTT
application payload 14 . implementation at the server. The entire data traffic has to go
B. LoRaWAN software solutions through the bridge. The application-server is the graphical user
interface for configuring a gateway, creating new applications
Several software solutions exists that implement the Lo-
or editing the authentication mode for a node.
RaWAN network stack. For our evaluation we chose the open-
source LoRa Server project provided by www.brocaar.com. C. Global IoT networks
14 www.lora-alliance.org/portals/0/documents/whitepapers/ Several communities strive for providing a world-wide solu-
LoRaWAN Security-Whitepaper V6 Digital.pdf tion for a common network for “things” by using LoRaWAN-
enabled networks. Examples are companies like Inmarsat
(Britain), KPN (Netherlands) or the open and crowd-sourced
community-driven TheThingsNetwork 11 . A setup including a
public server and several gateways all over the world makes
it possible to build up an extensive global network. The basic
idea is that the community or company provides the server
and cloud services, whereas the gateways come from private
people. Thereby a LoRa node from third parties can be easily
attached (with credentials) to the ‘big community’. It is also
possible to run all services on the same device as the gateway
does, to form an enclosed system, which is usually used for
private LoRa networks with only one gateway. This option
makes it possible for everyone to build up his own network
for connecting sensors in his environment. This allows for
completely different approaches with regard to LoRaWAN, in
which each method has its benefits according to its field of Fig. 5. Current consumption and radiated RF power in transmit mode of a
application. RF96 LoRa-module using following LoRa-parameters: BW = 125 kHz, CR =
4/5, SF = 7.
V. R ESULTS AND DISCUSSION
A. Current and RF power measurement
An important goal when using LoRa technology for trans-
mitting IoT data is low power consumption to achieve long
battery lifetime. Fig. 5 presents results from current and RF
power measurements on 50 Ω of a RF96 transceiver-module 15
for different output power settings. The current into the
transceivers supply pin is measured, the leakage current over
the SPI-connections to the microcontroller is neglected for
this evaluation. The output power on the RF96 module can
be specified from 5 dBm to 23 dBm in 1 dB steps. For every
single step a current measurement was taken. The transmitted
data payload contained 12 bytes, with LoRa parameters BW =
125 kHz, CR = 4/5 and SF = 7. The current values fluctuate
in a range of 2 mA during transmission, the standby current
with enabled transceiver in idle mode is about 4 mA at a
supply voltage of 3.3 V. The programmed RF output power Fig. 6. Measured transmission time and energy for transmitting a 12 byte
is verified by measuring the conducted power of the module payload @ 13dBm TX power and CR = 4/5.
with a Rhode & Schwarz ZVL spectrum analyzer. The results
show the expected behaviour.
BW and SF influence the symbol duration given by (1) and transmission times also higher sensitivity of the receiver can
therefore the time on air of the data packet and, in turn, the be achieved, which results in a larger communication range.
energy consumption. To verify the influence, we measured the The data rates and sensitivities depend only on the modulation
power consumption over time for different combinations of bandwidth and the spreading factor but not on the carrier
these two settings, while the RF output power setting was kept frequency.
constant at 13 dBm. The control of the transceiver registers As a consequence, for the lowest energy consumption and
were performed using the software library RH RF95.h from a high data-rate the spreading factor should be as small as
RadioHead 16 . In Fig. 6 it can be seen, that doubling the BW possible, whereas the bandwidth should be as big as possible.
reduces the transmission time by half, while the larger the SF If the major concern is a long communication range between
is, the longer a transmission lasts, which also verifies (1). A LoRa end-devices and the gateways, the data rate must be
longer transmission time obviously results in more consumed lowered to increase RF sensitivity of the receiver.
energy, in fact the measured energy per transmission is only B. Analyzing LoRaWAN packet overhead
a scaled version of the TX time because of a constant power
consumption during transmitter activity. But, as stated in the After considering the transmission between sensor node
datasheet of Semtech’s gateway-chip SX1301 13 , with longer and the gateway (also called concentrator), now we take a
closer look on the data overhead that is introduced by the
15 www.hoperf.com/rf transceiver/lora/RFM96W.html generation of UDP packets containing the original LoRa-
16 www.airspayce.com/mikem/arduino/RadioHead payload. For the reception of LoRa-modulated packets, a
special concentrator chip, e.g. SX130114 , is used. This chip
can demodulate 8 signals, with different spreading factors,
in parallel. The end devices (nodes) are connected in a star
topology to the gateway. One gateway can handle thousands
of LoRa nodes, though with increasing number of nodes the
amount of data-collision will also rise as shown in [11]. The
main task of the gateway is to convert data received by LoRa
into a UDP packet. Therefore a JSON string, including the
payload and other information about the gateway as well
as the connection-quality will be generated. This string is
transferred over a high-data-rate network, such as Ethernet,
WiFi or Cellular. So the low-power concept of LoRaWAN
ends at the gateway. In case of a cellular backhaul network
for the gateways, the generated data volume can be highly
relevant here, because common M2M contracts often include
data volumes of less than 100 MB per month or are even billed
per megabyte4,5,6 . For IoT service providers operating a large Fig. 7. Size of the UDP packet generated by the gateway vs. the payload
amount of gateways it is essential to keep the transmitted data sent by LoRa end-devices
and in consequence also costs at a minimum.
To evaluate how much data overhead is generated by the
gateway, packets with a defined payload length from the end number of supported sensor nodes can be limited because of
device to the gateway were sent. Using a UDP sniffer these data-volume restrictions in common M2M contracts.
sensor-data packets were detected while as messages generated ACKNOWLEDGMENT
by the gateway itself, e.g. heartbeat-messages and data-statistic
This work has been supported by the Linz Center of
were filtered out. Fig. 7 shows that the generated data-overhead
Mechatronics in the framework of the Austrian COMET-K2
amounts to about substantial 190 bytes per message due to the
programme.
conversion into a JSON string and insertion of extra infor- R EFERENCES
mation. In case of sending a 10 byte payload every 5 minutes
[1] L. Li, J. Ren, and Q. Zhu, “On the application of LoRa LPWAN tech-
from a single sensor node, a data volume of about 1.7 MB nology in Sailing Monitoring System,” in 13th Annual Conf. Wireless
per month is needed, which does not include the heartbeats On-demand Network Systems and Services (WONS), 2017.
(12 byte uplink typically every 30 s) and statistic-info (about [2] K. Mikhaylov, . J. Petaejaejaervi, and T. Haenninen, “Analysis of
Capacity and Scalability of the LoRa Low Power Wide Area Network
110 bytes in e.g. a one-minute interval) from the gateway. With Technology,” in 22th European Wireless Conference 2016.
the before mentioned data limit of 100 MB, a maximum of [3] J. Petäjäjärvi, K. Mikhaylov, M. Hämäläinen, and J. Iinatti, “Evalu-
only 50 nodes per gateway can be served. ation of LoRa LPWAN technology for remote health and wellbeing
monitoring,” in 10th Int. Symposium on Medical Information and
This proves that the low-energy and low resources concept Communication Technology (ISMICT), 2016.
of LoRa includes only the connection between end-devices [4] T. Petric, M. Goessens, L. Nuaymi, L. Toutain, and A. Pelov, “Mea-
(sensor-nodes) and gateways. After the LoRa transmission, surements, performance and analysis of LoRa FABIAN, a real-world
implementation of LPWAN,” in IEEE 27th Annual Int. Symposium on
there is no economical use of data any more, which can be a Personal, Indoor, and Mobile Radio Communications (PIMRC), 2016.
problem in the aforementioned situation. [5] A. J. Wixted, P. Kinnaird, H. Larijani, A. Tait, A. Ahmadinia, and
N. Strachan, “Evaluation of LoRa and LoRaWAN for wireless sensor
VI. C ONCLUSION networks,” in 2016 IEEE SENSORS, Oct 2016, pp. 1–3.
[6] M. C. Bor, U. Roedig, T. Voigt, and J. M. Alonso, “Do LoRa Low-
In this paper we summarized the basics of the LoRa Power Wide-Area Networks Scale?” in Proc. 19th ACM Int. Conf. on
modulation and the open standard LoRaWAN that defines Modeling, Analysis and Simulation of Wireless and Mobile Systems, ser.
MSWiM ’16. New York, NY, USA: ACM, 2016.
the communication protocol and system architecture for a [7] M. Lauridsen, B. Vejlgaard, I. Z. Kovacs, H. Nguyen, and P. Mogensen,
Low Power Wide Area Network. We provided a theoretical “Interference Measurements in the European 868 MHz ISM Band with
estimate of the maximum possible communication range of Focus on LoRa and SigFox,” in 2017 IEEE Wireless Communications
and Networking Conference (WCNC), March 2017, pp. 1–6.
a LoRa link and illustrated a LoRa-transceiver’s behavior [8] C. Orfanidis, L. M. Feeney, and M. Jacobsson, “Poster: Measuring PHY
of energy consumption and communication range versus the layer interactions between LoRa and IEEE 802.15.4g networks,” in 2017
achievable data rate by means of current consumption and RF IFIP Networking Conference (IFIP Networking), 2017.
[9] F. Adelantado, X. Vilajosana, P. Tuset-Peiró, B. Martı́nez, and J. Melià,
output power measurement results. By analyzing the packet “Understanding the Limits of LoRaWAN,” CoRR, vol. abs/1607.08011,
overhead generated by the LoRa gateway we have shown that 2016. [Online]. Available: http://arxiv.org/abs/1607.08011
LoRaWAN realizes the low energy and low resources con- [10] A. Augustin, J. Yi, T. Clausen, and W. M. Townsley, “A Study of
LoRa: Long Range & Low Power Networks for the Internet of Things,”
cept only between end-device and gateways. After the LoRa Sensors, vol. 16, no. 9, 2016.
transmission to the gateway, there is no economical use of [11] O. Georgiou and U. Raza, “Low Power Wide Area Network Analysis:
data. In scenarios, where IoT service providers are dependent Can LoRa Scale?” IEEE Wireless Communications Letters, vol. 6, no. 2,
pp. 162–165, April 2017.
on a large amount of cellular-connected LoRa gateways, the

You might also like