Content uploaded by Krzysztof Grochla
Author content
All content in this area was uploaded by Krzysztof Grochla on Nov 09, 2021
Content may be subject to copyright.
MITIGATION OF LORA INTERFERENCES VIA DYNAMIC CHANNEL WEIGHTS
Rafał Marjasz
Anna Strzoda
Konrad Połys
Krzysztof Grochla
Institute of Theoretical and Applied Informatics, PAS
Bałtycka 5, 44-100 Gliwice, Poland
email: {rmarjasz, astrzoda, kpolys, kgrochla}@iitis.pl
KEYWORDS
LoRa, LP WAN, radio interferences, simulation
ABSTRACT
The Low Power Wide Area Networks, such as LoRa (Long
Range), allow communication over a large area with limited
bandwidth and small data rates, using an unlicensed band.
LoRa uses multiple channels, but the number of channels
available is relatively narrow - e.g. 8 in most countries in
Europe. The transmission is subject to interferences due to
many other devices transmitting within the same frequen-
cies. In this paper, we evaluate how strong the interferences’
inuence is by analysing the packet delivery rate in a large
scale LoRa network. The measurements have been made
over more than seven months and show noticeable variabil-
ity in packet delivery rate per channel both in space and
in time. We propose to estimate the scale of interferences
through continuous observation of the number of delivered
packets for every channel per LoRa gateway or per selected
area. We also discuss how this estimation may mitigate the
interferences by assigning the dynamic weights to the LoRa
channel selection procedure. We created an algorithm that
periodically updates the weights, and a simulation model to
evaluate the algorithm. A signicant improvement in the ef-
ciency of package delivery is visible, from 12% to 49% less
packet loss.
INTRODUCTION
The Low Power Wide Area Network (LPWAN) is a set of
novel wireless communication technologies, providing large
coverage areas, low bandwidth, small data rates and used to
transmit limited data sizes, but providing very low energy
usage and long battery life operation (S. Farrell (2018)). The
LoRa (Long Range) is a proprietary low-power wide-area
network modulation technique and one of the most com-
monly used LP WAN technologies, using license-free sub-
gigahertz radio frequency bands. The LoRaWAN protocol
denes the transmission frames and MAC layer for LoRa
networks. The LoRa modulation is used together with the
LoRaWAN protocol (Sornin et al. (Jan. 2015)), dening the
packet format, and a star-of-stars topology forms a network
in which gateways exchange messages with end devices and
forward them to a central network server.
The transmission in LoRa is subject to interferences due to
many other devices transmitting within the same frequen-
cies. Due to the use of ISM bands, there are many devices
using the same set of frequencies. Although quite a few pa-
pers describe the analysis of radio signal propagation mea-
surements in LP WAN, the interferences variability in time
and per channel has not been profoundly analysed. E.g. the
authors of (Augustin et al. (2016)) measure the coverage of
the LoRa for end nodes located on the ground and water,
showing an estimation of the channel attenuation model.
In (Blenn and Kuipers (2017)), the analysis of the quality of
the received signal is presented based on a 7-months set of
experiments. However, there is limited information on the
variability of the interferences in space and time.
We evaluate how strong the interferences’ inuence is by
analysing the packet delivery rate in a large scale LoRa net-
work. We present the analysis of measurements made over
more than seven months and show noticeable variability in
packet delivery rate per channel. We also discuss how the
LoRaWAN network may be modied with minimal changes
to the conguration and behaviour of devices to allow the
mitigation of the interferences. The measurements are used
as an input for the simulation model evaluating the pro-
posed method of interference mitigation.
The rest of the paper is organized as follows: the second
section describes the state of the art regarding the measure-
ments of LoRa networks and the interference mitigation.
Next, we describe the results of the signal and interference
variability analysis. In the next section we propose an al-
gorithm to mitigate the interferences and we continue with
the performance evaluation of the proposed algorithm. We
nish the paper with a short conclusion.
STATE OF THE ART
The LPWAN technologies, and LoRa in particular, are in-
creasingly used in modern smart cities. Existing research
focuses greatly on LoRa’s performance, in particular its
range, capacity, scalability, the interaction and interference
between its transmissions. These include (Bor and Roedig
(2017)) and (Lavric and Popa (2017)), where the authors eval-
uate LoRa performance under various sets of congurations
and conditions. In the topic of radio disturbances (Orfanidis
et al. (2017)) introduces the research on interference interac-
tions between LoRa and IEEE 802.15.4g networks. Measure-
ments presented in (Lauridsen et al. (2017)) show that there
is a high probability of interference between signals above -
105 dBm within the mandatory LoRa 868.0-868.6 MHz band.
This research was conducted in a shopping area and a busi-
ness park in downtown Aalborg. Another interference type
study was presented in (Stoynov et al. (2018)) where the fo-
cus is put on the so-called Co-Spreading Factor Interference
(CSFI) and Inter-Spreading Factor Interference (ISFI).
Many concepts how to minimize the inuence of interfer-
ences in long range wireless networks were researched. In
(Voigt et al. (2016)), the use of directional antennas and mul-
tiple base stations as methods of dealing with inter-network
interference were compared and investigated. The authors
of (Almeida et al. (2017)) propose a multi-technology oppor-
tunistic platform for environmental data gathering. This ap-
proach addresses the heterogeneity of IoT devices through
the employment of multi-technology communications WiFi
and LoRa, and serves as an infrastructure for the network.
Another interference mitigation has been proposed in (Zhou
et al. (2019)), where a new Data Rate and Channel Control
(DRCC) scheme is proposed for the allocation of wireless re-
sources in LoRa networks. DRCC opportunistically adjusts
the SF based on the short-term Data Extraction Rate to deal
with random loss. It can also balance the link load of all
available channels to moderate the number of collisions and
increase the throughput. (Kerkouche et al. (2018)) demon-
strates the ability to optimize LoRaWAN technology per-
formance applying lightweight learning methods, namely
multi-armed bandit algorithms, to select more appropriate
communication parameter values SF and emission power.
MEASUREMENTS OF INTERFERENCE VARIABILITY
This chapter focuses on the analysis performed to identify
the channel interference for a network end device. The
study was carried out based on data set consists of mea-
surements from the sample LoRa network infrastructure in
a typical Polish city. The data set includes radio packets col-
lected over about half a year from several thousand end de-
vices distributed with dierent densities in space. The net-
work topology covers areas with a high density of nodes in
length, typical for urban and suburban regions character-
ized by lower density distribution of nodes. We have ana-
lyzed over four million data points representing a deliver or
not delivered packet and the received signal strength (RSSI)
metric.
In detecting anomalies on channels, i.e. interferences, we
use the approach that is based on three-sigma statistical tool
(Pukelsheim (1994)) applied to two measures; packet loss
probability and mean RSSI level. We consider pairs of net-
work devices, i.e., end device - access point, which received
a signicant part (to increase the accuracy of the calcula-
tions; at least 70%) of all packets from this node. Network
redundancy allows identifying radio packets not delivered
to the best access point (but received by the other gateways)
and the channel on which the lost packets were transmitted.
Therefore, it is possible to determine a value for packet loss
probability on a given frequency channel in communication
between an end node and a given access point. An anomaly
on the channel is understood as the value of a packet loss
probability outside the calculated three-sigma range deter-
mined as presented below.
Each set of packets lost and delivered from a single end node
to a specic access point on a given channel constitute a
separate population. For each such population, an interval
of three-sigma was determined for packet loss probability
and mean RSSI level. Eventually, for each end node, it is
checked whether the value of the packet loss probability or
the average RSSI level (calculated from the observed data)
on each of the eight channels exceeds the respective three-
sigma range.
The three-sigma limits parameters; mean µand standard
deviation σfor the packet loss probability on the specied
frequency channel for a single end device are calculated as
follows:
Let Xbe a discrete random variable, and suppose that the
possible values it can assume are given as follows:
X=(0,radio packet has been delivered
1,radio packet has not been delivered (1)
For each possible value xof the random variable X P (X=
x)can be obtained from the probability function p(x)given
by the following formula
p(x) = (n
n+m, x = 0
m
n+m, x = 1 (2)
where nis the number of radio packets delivered from an
end device to the specied gateway on the single channel,
and mis the number of lost radio packets in the same pop-
ulation.
That is, P(X=x) = p(x)∀x∈ {0,1}. Than, the popula-
tion mean and standard deviation are given by the following
formulas
µ=E(X) = X
x
xp(x)(3)
σ=v
u
u
tX
x
(x−µ)2p(x)
N(4)
where Nis the total number of delivered (to specied
gateway) radio packets on the specied frequency channel
by single end device.
In our considerations, we also take into account the RSSI
level analysis because large uctuations, a gradual decrease
in the RSSI level, or the presence of low outliers may result
in loss of connectivity between devices. The three-sigma
e
i
k
LoRa gateway
end node - group A
end node - group B
Figure 1: Arrangement of end nodes belonging to groups A
or B in space.
limits parameters; mean µand standard deviation σfor the
mean RSSI level on the specied frequency channel for a
single end device are calculated as follows:
µ=1
M
M
X
i=1
RSS Ii(5)
where Mis the total number of packets delivered from sin-
gle end device to specied gateway, and
σ=v
u
u
t
1
N
N
X
j=1
(RSS Ij−1
N
N
X
j=1
RSS Ij)2(6)
where Nis the total number of radio packets delivered by
single end device on the specied frequency channel to gate-
way.
Figure 1 illustrates the map of end nodes for which the co-
existence of anomalies for the packet loss probability and
the mean RSSI level on the same (at least one) channel was
detected by the presented analysis. Considered nodes ac-
count for 1.78% of the network topology. For the sake of
simplicity, they are marked as nodes from group A. The red
points denote the nodes that are a specic subset of set A,
namely the devices for which the coexistence of anomalies
was detected for the probability of packet loss and the aver-
age level of RSSI on the 4th channel (group B). This group of
nodes constitutes 73% of all nodes for which the coexistence
of outliers of the packet loss probability and the mean RSSI
was detected (packet loss probability and mean RSSI values
are outside the three-sigma range). In the gure 2 the CDF
and PDF of packet loss probability for group B are presented.
As a result of the presented analysis, the nodes belonging to
group B are located in a distinct area with potential interfer-
ence on channel 4. Access points marked in the gure 1 with
names e, i and k have a low percentage of received packets
on channel 4 (compared to the other channels), as shown in
gure 3. Thus there is strong potential in reconguring the
part of the network to mitigate the interferences.
0.0 0.2 0.4 0.6 0.8 1.0
PDF and CDF of packet loss
probability for group B
0.0
0.2
0.4
0.6
0.8
1.0CDF
PDF
Figure 2: PDF and CDF of packet loss probability for nodes
belonging to group B.
PROPOSED INTERFERENCE MITIGATION ALGO-
RITHM
LoRaWAN uses all channels with the same probability in a
uniform manner. The results shown in gure 3 show that
in a real-life scenario, this leads to sending a similar share
of packets on channels with a large and small number of in-
terferences. There is a potential to improve packet delivery
probability by sending more packets on channels experienc-
ing more negligible interferences.
We propose a continuous observation of the number of de-
livered packets for every channel per LoRa gateway to es-
timate the level of interferences. When the proportion of
received packets on a specic channel is not consistent with
the probability of this channel usage for the transmission,
this channel is subjected to heavier interferences than other
channels and should be used less intensively. Thus we pro-
pose that each transmitting device uses weights assigned
to the channels, dening the probability of its usage. The
network is divided into a set of smaller areas, grouping ge-
ographically the end nodes. The weights are periodically
updated per area and transmitted to the end devices, based
on the statistics for every channel, using the following algo-
rithm, executed for each network fragment:
•Store the current probabilities as previous probabilities
•Count the number of received messages for each chan-
nel
•Reset message counters
•For each channel i
–NewPi=currentPi+ (currentPi-previousPi)
–If NewPi> (1 / nChannels) * 2 then N ewPi= (1 /
nChannels) * 2
–If NewPi<0.05 then N ewPi= 0.05
•Normalize the sum of NewPito 1
where: NewP is newly calculated expected probability of
delivery, previousP is previously calculated expected prob-
12345678
0%
5%
10%
15%
20% a
12345678
0%
5%
10%
15%
20% b
12345678
0%
5%
10%
15%
20% c
12345678
0%
5%
10%
15%
20% d
12345678
0%
5%
10%
15%
20% e
12345678
0%
5%
10%
15%
20% f
12345678
0%
5%
10%
15%
20% g
12345678
0%
5%
10%
15%
20% h
12345678
0%
5%
10%
15%
20% i
12345678
0%
5%
10%
15%
20% j
12345678
0%
5%
10%
15%
20% k
12345678
0%
5%
10%
15%
20% l
12345678
0%
5%
10%
15%
20% m
12345678
0%
5%
10%
15%
20% n
12345678
0%
5%
10%
15%
20% o
channel
percentage of packets received on channel
Figure 3: Distributions of the number of delivered radio packets on channel per each LoRa gateway.
ability, currentP is actual probability, and nChannels is a
number of channels in use.
PERFORMANCE EVALUATION
We have created a simulation model using OMNeT++, FloRa
model Slabicki et al. (2018), and our previous work Marjasz
et al. (2019), taking into account packets collisions. They
occur when both interfering frames possessing the same
spreading factor are transmitted on the same frequency. We
have implemented a detailed interference model according
to the results presented in Rahmadhani and Kuipers (2018).
It has been expanded by implementing an algorithm that
detects interference on transmission channels and reduces
packet loss. Four scenarios are considered. The rst one
consists of 100 nodes uniformly distributed on the area of
0.168 km2with the division on square subareas with 80
meters side length. The second and third scenarios are
based on localization (Fig. ??) from actual deployment, and
they are consist of 4173 nodes on the area of 60 km2with
square subareas of 120 and 300 meters side length, accord-
ingly. In the fourth scenario, the real-life urban implemen-
tation of the LoRa network determines the loss of the pack-
ets. Data collected from 97 days was divided approximately
into three monthly periods. For each period, the probabil-
ities of packet non-delivery on each of 8 channels for ev-
ery end-node were determined (the ratio of lost packets to
packets sent on a given channel in a given period). The
city topology provided with the measurements was used
to carry out the simulation in a period of 18000 days (ap-
proximately 50 years). It was divided into monthly peri-
Scenario W/o algorithm With algorithm Dierence Loss less by
100 nodes 80m area 0.933933 0.966326 0.032398 49%
City 120m area 0.939590 0.965613 0.026024 43%
City 300m area 0.936381 0.964388 0.028006 44%
Table 1: Packet loss ratios for three simulations scenarios,
where Cal culateP robabilitiesI nterval = 5 days.
ods, for which the previously determined monthly distri-
butions of the probability of packet non-delivery were re-
peated periodically every three months (200 repetitions).
The purpose of such an operation is to reproduce period-
ically repetitive changes occurring during the entire sim-
ulation time. To determine the impact of the algorithm
on the quality of packet delivery in the entire network, a
number of simulations were conducted. We have analysed
the relation between the interval of probabilities calculation
Cal culateP robabilitiesI nterval ∈ {10,12,14,...,60}
[days] and the amount of packets required to calculate the
probabilities: P acketC ountMul tiplier ∈ {2,3,4,...,9}
[ppc] – packets per channel.
The Table 1 presents the statistics for delivery of packets,
collected as an average value from all square subareas in
scenarios 1-3. A signicant improvement in the eciency
of package delivery is visible, with over 40% less packet lost.
Thanks to the algorithm, the average number of messages
lost is decreased by up to 49%.
CONCLUSIONS
The paper proposes an interference mitigation scheme
based on weights assigned to the LoRa networks channels,
5.595 5.584
81.566 79.783
12.839 14.633
0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
75
80
85
with algorithm w/o algorithm
Percent [%]
Real life implementation - the highest improvement
Collisions Packet reception Packet rejection
Figure 4: Packet loss for a real-life scenario –
best pair of algorithm parameters simulation
(Cal culateP robabilitiesI nterval = 42 [days] and
P acketC ountMul tiplier = 5 [ppc]) versus no algorithm
in use simulation.
dening its usage probability. We show analysis of the mea-
surements of a large scale LoRa network, oering that the
interferences vary in time and frequency. Through peri-
odic updates of the weights based on the observed packet
delivery ratio, the proposed interference mitigation algo-
rithm signicantly improves the eciency of package deliv-
ery. We have created a discrete event simulation model fed
with interference data from the measurements. The perfor-
mance evaluation in random and real-life topology shows
that the interference mitigation scheme allows decreasing
the number of lost packets by up to 49%.
ACKNOWLEDGEMENTS
This research was partially funded by Polish National
Center for Research and Development grant number
POIR.04.01.04-00-0005/17.
REFERENCES
Almeida R.; Oliveira R.; Sousa D.; Luis M.; Senna C.; and Sar-
gento S., 2017. A Multi-Technology Opportunistic Platform
for Environmental Data Gathering on Smart Cities. In 2017
IEEE Globecom Workshops (GC Wkshps). 1–7.
Augustin A.; Yi J.; Clausen T.; and Townsley W.M., 2016. A
Study of LoRa: Long Range, Low Power Networks for the
Internet of Things.Sensors, 16, no. 9. ISSN 1424-8220.
doi:10.3390/s16091466. URL https://www.mdpi.
com/1424-8220/16/9/1466.
Blenn N. and Kuipers F., 2017. LoRaWAN in the wild:
Measurements from the things network.arXiv preprint
arXiv:170603086.
Bor M. and Roedig U., 2017. LoRa Transmission Parameter
Selection. In 2017 13th International Conference on Dis-
tributed Computing in Sensor Systems (DCOSS). 27–34.
Kerkouche R.; Alami R.; Féraud R.; Varsier N.; and Maillé P.,
2018. Node-based optimization of LoRa transmissions with
Multi-Armed Bandit algorithms. In 2018 25th International
Conference on Telecommunications (ICT). 521–526.
Lauridsen M.; Vejlgaard B.; Kovacs I.Z.; Nguyen H.; and Mo-
gensen P., 2017. Interference Measurements in the Euro-
pean 868 MHz ISM Band with Focus on LoRa and SigFox. In
2017 IEEE Wireless Communications and Networking Con-
ference (WCNC). 1–6.
Lavric A. and Popa V., 2017. A LoRaWAN: Long range wide
area networks study. In 2017 International Conference on
Electromechanical and Power Systems. 417–420.
Marjasz R.; Grochla K.; Strzoda A.; and Laskarzewski Z.,
2019. Simulation Analysis of Packet Delivery Probability
in LoRa Networks. In P. Gaj; M. Sawicki; and A. Kwiecień
(Eds.), Computer Networks. Springer. ISBN 978-3-030-
21952-9, 86–98.
Orfanidis C.; Feeney L.M.; Jacobsson M.; and Gunningberg
P., 2017. Investigating interference between LoRa and IEEE
802.15.4g networks. In 2017 IEEE 13th International Confer-
ence on Wireless and Mobile Computing, Networking and
Communications (WiMob). 1–8.
Pukelsheim F., 1994. The three sigma rule.The American
Statistician, 48, no. 2, 88–91.
Rahmadhani A. and Kuipers F., 2018. When lorawan frames
collide. In Proceedings of the 12th International Workshop
on Wireless Network Testbeds, Experimental Evaluation &
Characterization. 89–97.
S. Farrell E., 2018. Low-Power Wide Area Network (LP-
WAN) Overview. RFC 8376. doi:10.17487/RFC8376. URL
https://rfc-editor.org/rfc/rfc8376.
Slabicki M.; Premsankar G.; and Di Francesco M., 2018.
Adaptive conguration of LoRa networks for dense IoT de-
ployments. In NOMS 2018-2018 IEEE/IFIP Network Opera-
tions and Management Symposium. IEEE, 1–9.
Sornin N.; Luis M.; Eirich T.; Kramp T.; and
O.Hersent, Jan. 2015. LoRaWAN Specication.
LoRa Alliance, Standard V10. URL https:
//lora-alliance.org/resource_hub/
lorawan-specification-v1-0/.
Stoynov V.R.; Poulkov V.K.; and Valkova-Jarvis Z.V., 2018.
Interference Management in LoRaWANs — Overview and
Simulation Study. In 2018 Advances in Wireless and Optical
Communications (RTUWO). 251–256.
Voigt T.; Bor M.C.; Roedig U.; and Alonso J.M., 2016. Miti-
gating Inter-network Interference in LoRa Networks.CoRR,
abs/1611.00688. URL http://arxiv.org/abs/
1611.00688.
Zhou Q.; Xing J.; Hou L.; Xu R.; and Zheng K., 2019. A Novel
Rate and Channel Control Scheme Based on Data Extraction
Rate for LoRa Networks. In 2019 IEEE Wireless Communi-
cations and Networking Conference (WCNC). 1–6.