Decoding Superposed LoRa Signals

Nancy El Rachkidy(1), Alexandre Guitton(1) , Megumi Kaneko(2)

(1) Université Clermont Auvergne, CNRS, LIMOS, F-63000 Clermont-Ferrand, France
(2) National Institute of Informatics, Hitotsubashi, 2-1-2, Chiyoda-ku, 101-8430 Tokyo, Japan
Emails: nancy.el rachkidy@uca.fr, alexandre.guitton@uca.fr, megkaneko@nii.ac.jp

Abstract—Long-range low-power wireless communications, two cases (slightly desynchronized and completely synchro-
such as LoRa, are used in many IoT and environmental mon- nized), and presents our two algorithms. Section IV gives our
itoring applications. They typically increase the communication
arXiv:1804.00503v1 [cs.NI] 22 Mar 2018

simulation results. Finally, Section V concludes the paper.

range to several kilometers, at the cost of reducing the bitrate
to a few bits per seconds. Collisions further reduce the perfor- II. S TATE OF THE ART
mance of these communications. In this paper, we propose two
algorithms to decode colliding signals: one algorithm requires In the following, we first describe the MAC protocol Lo-
the transmitters to be slightly desynchronized, and the other RaWAN, and then the physical layer LoRa. Note that our
requires the transmitters to be synchronized. To do so, we use paper proposes an improvement to LoRa, which can be used
the timing information to match the correct symbols to the
correct transmitters. We show that our algorithms are able to to improve the performance of any MAC protocol based on
significantly improve the overall throughput of LoRa. LoRa (including LoRaWAN).
Index Terms—LoRa, LoRaWAN, LPWAN, Interference can-
cellation, synchronized signals, desynchronized signals. A. LoRaWAN
LoRaWAN (in version 1.0 [9] or in version 1.1 [7]) is
I. I NTRODUCTION a MAC protocol based on LoRa. Three classes are defined,
depending on the communication paradigm: Class A is for
Long-range low-power communication technologies such as low-power uplink communications, Class B is for delay-
LoRa [1], Sigfox [2], and Ingenu [3], are becoming widely guaranteed downlink communications, and Class C is for EDs
used in Low-Power Wide Area Networks (LPWANs). These without energy constraints. In Class A, the only mandatory
technologies are suitable to cover large zones and are thus class of LoRaWAN, an ED can transmit at any time. It
becoming attractive technologies for Internet of Things (IoT) chooses a channel randomly, sends the frame, and waits for
and monitoring applications [4], [5], [6]. an acknowledgment during two successive receive windows.
LoRa [1] is a recent physical layer for LPWANs, which After its transmission, the ED is forbidden to transmit for a
extends the communication range by reducing the throughput. delay equal to 99 times the duration of the frame transmission.
LoRaWAN [7] is a simple MAC protocol based on LoRa, In this way, the transmission time of EDs does not exceed 1%.
which allows end-devices (ED) to communicate with a small LoRaWAN adapts the bitrate to the quality of links by
duty-cycle (1%) to a network server through gateways. Thus, implementing a trade-off between the signal robustness and
EDs can save energy, and the network lifetime is increased. the bitrate, through the use of the Spreading Factor (SF) of
The main issue in LoRa and LoRaWAN is the limited the signal: when an ED experiences a low signal quality, it
throughput: the indicative physical bitrate varies between increases its SF, which results into lower bitrate, but better
250 and 11000 bps [8]. Moreover, when two EDs transmit decoding capabilities of the signal. This modification is con-
simultaneously using the same parameters, a collision occurs trolled by the datarate (DR) of LoRaWAN, which is a value
and none of the signals are decoded by LoRa. Thus, both EDs ranging from DR0 (large SF, small bitrate) to DR6 (small SF,
have to retransmit, which further reduces the throughput. large bitrate).
In this paper, we show that it is possible to retrieve the European regional settings of LoRaWAN [8] define most
frames from superposed signals. For the case where super- LoRa parameters. The bandwidth of channels, BW , is equal
posed signals are slightly desynchronized, we propose a linear to 125 kHz for DR0 to DR5, and 250 kHz for DR6. The SF
algorithm based on timing information that attempts to decode varies from 12 down to 7 for DR0 to DR5, and is equal to
all frames. This algorithm always succeeds when there are 7 for DR6. The preamble length is equal to 8 symbols. The
two signals. We prove that for three or more signals, it is not physical bitrate varies between 250 bps for DR0, to 11000
always possible to decode each signal. Next, for the case where bps for DR6. The maximum MAC payload of a frame varies
superposed signals are completely synchronized, we propose a between 59 bytes for DR0 and 230 bytes for DR6.
simple algorithm requiring only one retransmission to deduce
the other colliding frame. To the best of our knowledge, this B. LoRa
is the first work on LoRa interference cancellation. LoRa is a physical layer technology for LPWAN, based on a
The remainder of this paper is as follows. Section II Chirp-Spread Spectrum (CSS) modulation. In this modulation,
describes the modulation of LoRa. Section III describes our each LoRa chirp consists of a linear frequency sweep. The
 duration of the sweep is called symbol duration (SD), and III. C ANCELLATION OF L O R A SIGNALS
depends on the value of SF. The sweep is performed over
LoRa gateways are able to decode superposed LoRa signals
a frequency range of size BW . Chirps are divided into up-
as long as they are sent on different SFs. Notice however that
chirps, where the frequency sweep is increasing, and down-
some researchers have shown that signals on different SFs are
chirps, where the frequency sweep is decreasing.
not completely orthogonal [11], [12].
Each chirp can encode 2SF symbol values. To do this, LoRa
When several signals are received on the same channel and
shifts the sweep by the symbol value, as shown on Fig. 1 for
with the same SF, a difference of received power might cause
a down-chirp. The receiver is able to detect the sharp edge
the strongest signal to be captured [10], [13]. When several
in the instantaneous frequency trajectory [10]. The symbol
signals have a similar receive power, a collision occurs and all
value is equal to the shift in the frequency at the beginning
signals are considered lost.
of the symbol. It is also proportional to the time between the
beginning of the symbol and the sharp frequency edge. For In this paper, we focus on decoding superposed LoRa
up-chirps, it is proportional to the remaining time between the signals of similar receive power, on the same channel, with
sharp frequency edge and the end of the symbol. the same SF. To do so, we show that we can use timing
information to match the correct symbols to the correct ED.
In Subsection III-A, we describe our assumptions. In Sub-
frequency
PSfrag replacements section III-B, we provide our main algorithm, and we describe
how it can decode two signals that are slightly desynchronized.
BW value In Subsection III-C, we extend the discussion for the case
time of three signals (or more) that are slightly desynchronized.
Finally, in Subsection III-D, we present a simple algorithm
value
for the case of signals that are completely synchronized.
SD Note that our algorithms cannot be applied directly on
Figure 1. Example of a single LoRa down-chirp. Computing the symbol value LoRaWAN, as most communications in LoRaWAN are not
requires identifying the sharp frequency edge. synchronized. However, our algorithms could enable the de-
sign of a novel synchronized MAC layer based on LoRa,
To decode the value of a symbol, the receiver needs to know tailored to the star like topology of LoRaWAN, to reach better
the frontier of the symbol. Thus, LoRa introduces a preamble performances than the basic LoRaWAN.
of a few symbols (typically, eight). In uplink communications,
the preamble consists of up-chirps and the data consists A. Assumptions
of down-chirps. In downlink communications, the preamble
consists of down-chirps and the data consists of up-chirps. We assume that there are no non-linearity effects between
Figure 2 shows an example of an uplink communication down-chirps (respectively up-chirps). In other words, if two
with a short preamble (three symbols) and a few data symbols down-chirps (resp. up-chirps) c1 and c2 overlap at a given
(four symbols). We chose SF = 2 for the sake of simplicity, time t at the receiver side, the two observed frequencies are
leading to 2SF = 4 possible values per symbol. Let us assume the frequency of c1 (at time t) and the frequency of c2 (at
that a desynchronized node starts receiving the preamble, not time t). Without additional information, it is not possible to
necessarily at the exact beginning of the preamble. The node identify the correct frequency of each transmitter. We assume
detects a sharp frequency edge, which indicates the frontier of that when an up-chirp is superposed with a down-chirp, it is
a symbol. From this information, the receiver can synchronize not possible to detect any of the frequencies.
itself according to the transmitter. The end of the preamble is We also assume that it is possible for the hardware of
detected by the inversion of the chirps. In this example, the the receiver to detect all frequencies of overlapping down-
data symbols are 3, 0, 2, 2. chirps (resp. up-chirps) within δ time-units. In the following
eplacements examples, we will use δ = SD/4 unless stated otherwise1 .
We assume that when several frequencies overlap at a
given time, only one frequency is detected by the receiver.
For instance, if there are three nodes transmitting at a given
sender
time, but only two frequencies f1 and f2 are detected, we
preamble 3 0 2 2
assume that it is not possible to know whether two nodes
were transmitting with f1 and one with f2 , or one node was
transmitting with f1 and two with f2 .
receiver
desynchronization information 1 Please note that on real LoRa hardware, the decoding of signals is not
carried out by directly detecting the sharp frequency edges, but instead by
Figure 2. Example of a LoRa uplink frame, with a short preamble and four
computing a fast Fourier transform and detecting the peak in the frequency
symbols of data, with SF = 2. The receiver synchronizes itself with the
domain [10]. With our proposition, this translates into either detecting the two
sender during the preamble.
sharp frequency edges in the time domain, or the two peaks in the frequency
domain.
 We also assume some properties on the frames: all nodes Data decoding: We define the sequence of decoded data for
transmit with the same preamble duration, the frame length is n1 by d1 and the sequence of decoded data for n2 by d2 . At
included at the beginning of the frame, and there is at least frontier t2 , the receiver stores the current frequencies, which
one symbol change during the whole frame (that is, a data correspond to Flim− (t2 ) = {0, 3}. At frontier t3 , the receiver
payload does not consist of a sequence of identical symbols). computes Flim+ (t3 ) by updating the previous frequencies
Finally, in the following, we consider two cases: the case Flim− (t2 ) = {0, 3}, and obtains Flim+ (t3 ) = {0, 1} (each
where nodes are slightly desynchronized, and the case where frequency is reduced by 3 since 3δ time units have passed
nodes are fully synchronized. In the case where nodes are since t2 , as δ = SD/4). It detects the current frequencies
slightly desynchronized, we assume that all nodes start their Flim− (t3 ) = {0, 1}. There is no change in the frequencies
transmission within SD −δ time units, and that any two nodes (Flim+ (t3 ) = Flim− (t3 )), since the beginning of the data
start their transmission with a delay of δ time units or more. In of n1 starts with the repeated symbol 1. Thus, the algorithm
the following examples, if there are three transmitting nodes, leaves ∗ for the first symbol of n1 (to be decoded later), so
we consider that node n1 starts at time t0 , node n2 at time d1 = (∗, 1). At frontier t4 , the receiver computes Flim+ (t4 )
t0 + δ, and node n3 at time t0 + 2δ. by updating the previous frequencies Flim− (t3 ) = {0, 1},
and obtains Flim+ (t4 ) = {0, 3} (since δ time units have
B. Case of two slightly desynchronized signals
passed). It detects the current frequencies Flim− (t4 ), and
In this subsection, we consider the superposition of two obtains Flim− (t4 ) = {0}, which is equivalent to {0, 0}.
signals from two transmitters that are slightly desynchronized Thus, one frequency changed from 3 to 0, which is that of
replacements
(by at least δ time units, and at most SD − δ time units). n2 , since it is a frontier of n2 , hence, d2 = (3, 0). Thus,
Figure 3 shows an example of the reception of two slightly the current symbol of n1 corresponds to frequency 0 too
desynchronized signals. The preamble length is two symbols, (which is translated into 1 at the beginning of the symbol
and SF = 2. The figure shows the signal of the first transmitter frontier of n1 , which was t3 ). At frontier t5 , the receiver
n1 starting at t0 , the signal of the second transmitter n2 starting computes Flim+ (t5 ) by updating the previous frequencies
at t0 +δ, and the superposed signal at the receiver. Note that the Flim− (t4 ) = {0, 0}, and obtains Flim+ (t5 ) = {1, 1}. It detects
data transmitted by n1 is (1, 1, 3, 2, 2), and the data transmitted the current frequencies Flim− (t5 ) = {1, 3}. The frequency of
by n2 is (3, 0, 2, 3, 1). We will first explain our algorithm on n1 changed from 1 to 3, hence d1 = (∗, 1, 3). So, the current
this example, and then proceed with a more formal description. symbol of n2 corresponds to frequency 1 (which is translated
to 0 at the beginning of the symbol frontier of n2 , which is
t4 ). The algorithm continues until t12 , where no frequency
is received. Thus, the algorithm knows that all nodes have
n1
stopped their transmissions. The algorithm removes the last
1 1 3 2 2 predicted symbol of n1 (indeed, at t11 , it considered that
n1 was transmitting a symbol with the same frequency as
the frequency of n2 ). At this step, the decoded frames are
n2 d1 = (∗, 1, 3, 2, 2) for n1 and d2 = (3, 0, 2, 3, 1) for n2 .
δ 3 0 2 3 1 Then, the algorithm replaces all special values ∗ with the
first known value of the frame. The algorithm uses the frame
length present in each frame to truncate the frames to their
receiver correct length. Finally, the algorithm outputs are (1, 1, 3, 2, 2)
t0 t1 t2 t3 t4 t5 t6 t7 t8 t9t10 t11t12 and (3, 0, 2, 3, 1), as expected.
Figure 3. Superposition of two slightly desynchronized signals. Generalization of Preamble detection and data decoding
In this paragraph, we generalize the example given above
Example of Preamble detection and data decoding and we formulate our proposition in Algorithm 1.
In this paragraph, we give an example in order to explain Preamble detection: The superposition of preambles will
how the receiver, using our proposition, can detect preambles result in the superposition of up-chirp symbols, except for the
and decode data sent by two transmitters. end of the last preamble. Thus, the receiver will detect two
Preamble detection: During [t0 ; t0 + δ], the receiver detects sharp frequency edges for most preamble symbols. Each of
the preamble of n1 . During [t0 + δ; t0 + 2δ], the receiver is this sharp edge will allow the receiver to know the symbol
able to detect that two slightly desynchronized signals are frontier of a transmitter. The beginning of the first data symbol
transmitted, and is able to deduce the symbol frontiers of of the first node is not decodable, as it corresponds to a down-
both transmitters. At frontier t1 , or more precisely, during chirp superposed with the up-chirp of the end of the preamble
[t1 ; t1 + δ], the receiver is not able to detect the superposition of the second node.
of preambles anymore (mixed up and down chirps). Thus, it Data decoding: From the first data symbol of the second
knows that the transmission of the first data symbol of n1 node, only down-chirps are superposed, and thus it is possible
has started. This symbol is currently undecodable due to the to detect all sharp edges. The difficulty relies in correlating
overlapping of an up-chirp with a down-chirp. each frequency with the symbols of each node. To do so, we
use the following property: sharp edges can occur only at the
beginning of a symbol, when the symbol changes, or once
during a symbol. When the sharp edge occurs during a symbol, Algorithm 1: Decoding of two slightly desynchronized
it can be predicted if the symbol value is known. superposed LoRa signals.
Algorithm 1 describes our proposed algorithm. It starts for each frontier ti of a data chirp do
after the superposed preambles have been received, and thus compute currentSymbol and currentNode
considers that the symbol frontier of each transmitter is known. if currentSymbol=0 and currentNode=1 then
The algorithm considers the frontiers of all data symbols skip (frequencies cannot be detected)
sequentially, apart from the first frontier of the first node which else
cannot be decoded. At each frontier, the receiver updates Flim− (ti ) ←detect current frequencies
the previous frequencies (since frequencies change over time if currentSymbol=0 and currentNode=2 then
in LoRa chirps, and time has passed since the detection of skip (Flim− (ti ) is already computed)
the previous frequencies). Then, the receiver compares these
else
(updated) previous frequencies with the current frequencies2. compute Flim+ (ti ) by updating Flim− (ti−1 )
The following two cases are the only possible cases. changedF ← Flim− (ti ) − Flim+ (ti )
Case 1: One frequency has changed. This can only happen if |changedF | = 0 then
when a new symbol starts, which can only occur at the symbol the new symbol in symb[currentNode] is
frontier. Since the receiver knows if the current frontier is for equal to the previous (or ∗)
the first or the second transmitter, it knows the new symbol for
else
the current node (based on the new frequency), the previous
the previous symb. in symb[currentNode]
symbol for the current node (based on the frequency that has
is equal to the value of
changed), and the current symbol for the other node (based
Flim+ (ti ) − Flim− (ti )
on the frequency that did not change).
the new symbol in symb[currentNode] is
Case 2: No frequency has changed. This can only happen
equal to the value of changedF
when the new symbol is equal to the previous symbol (this
was the case on Fig. 3 at times t3 and t9 ). If the receiver knows
the previous symbol of the current node (time t9 of Fig. 3), for each node n do
the new symbol can be deduced. Note that at the beginning replace in symb[n] all the leading ∗ values to the
of the algorithm, however, the first symbol value cannot be first defined value
deduced when it is repeated (time t3 of Fig. 3). In this case, truncate the frame according to its length
the algorithm leaves a special value (denoted by ∗ here). As
soon as one symbol changes, the receiver is able to deduce the
values of all these repeated symbols. This is why we assumed
at least one symbol change per frame.
The time complexity of our algorithm is linear with the
number of symbols of the longest frame. Most of the symbols
are decoded on the fly, δ time units after the beginning of the
symbol, except for the symbols repeated initially (see the last n1
loop of the algorithm). The space complexity of our algorithm
0 0 0 0 0 0
is O(1), since the storage requirement is limited to the value of
the first non-special symbol for each node. Thus, the algorithm
is optimal in time and space, for two nodes3 . PSfrag replacements
n2
C. Case of three slightly desynchronized signals 1 1 1 1 1 1

Note that with our hypotheses, decoding three or more

signals is not always possible. For instance, Fig. 4 shows two n3
sets of different signals that produce the same superposition
0 2 0 2 0 0
of frequencies, and thus cannot be decoded.
Our algorithm is able to decode many cases of slightly
desynchronized signals for n transmitters, when n ≥ 3. It
3
receiv.
2 In practice, it may take up to δ time units to obtain the current frequencies, Figure 4. When three nodes that are slightly desynchronized transmit frames,
so the receiver might have to update the current frequencies based on the it is not always possible to decode them: these two sets of frames produce
detection time. the same superposition of frequencies.
3 As we will see in Subsection III-C, our algorithm is not able to decode
all frames for three nodes or more, so it cannot be considered optimal in this
case.
only fails to do so when the number of received frequencies B. Case of slightly desynchronized signals
is within [2; n − 1] (which never occurs when n = 2). In Figure 5 shows the percentage of successful decoding of
this case, even if the algorithm knows that the frequency of colliding signals, as a function of the number of colliding
the current node has changed, it cannot determine what is the signals, in the case where signals are slightly desynchronized.
new value, as it has n − 1 > 1 possibilities. It can still deduce When there are two or more colliding signals, LoRa is not able
the value of the previous symbol for this node. At the next to decode any signal. When there are exactly two colliding
frontier for this node, though, the value of this symbol might signals, our algorithm is always able to decode both of them.
be deduced, depending on the number of other frequencies. When there are three or more colliding signals, our algorithm
is not able to decode some signals: the proportion of signals
D. Case of two synchronized signals that can be decoded depends on SF and on the frame length.
Indeed, when SF is large, the number of possible values
We now consider the case of two synchronized signals.
for each symbol is large, and the probability that several
When the two transmitters are completely synchronized, it
transmitters use the same frequency is low. For SF = 7 and
can be noticed that (at most) two values for each symbol
n = 3 colliding frames, our algorithm is able to decode about
duration are obtained, one for each transmitter. With our
80% of the frames. This number drops rapidly as the number
assumptions, though, it is not possible to match each value
of transmitters increases.
to the correct transmitter. The uncertainty of two values for
each symbol might seem large, but it is quite small compared
100 Conv.LoRa
to the fact that each symbol carries in fact SF bits of data. Prop.Alg.1 (SF7)

%successful decoding
Prop.Alg.1 (SF12)
Thus, we propose a simple algorithm for this case. When 80
two such frames collide, the algorithm stores the possible
values for each symbol, and requests any of the transmitters 60
to retransmit its frame. When one frame is retransmitted, the
algorithm is able to decode it, and is able to deduce the values 40
of the colliding frame of the other node too, by elimination.
Thus, instead of having to retransmit two colliding frames, 20
only one retransmission is required. PSfrag replacements
0
1 2 3 4 5
IV. N UMERICAL RESULTS
Number of superposed signals
In this section, we evaluate, by simulation, the performance Figure 5. When the signals are slightly desynchronized, our algorithm is able
of our algorithms in terms of percentage of successful decod- to successfully decode some cases of colliding signals.
ing of colliding signals and throughput. We consider the two
cases independently. Figure 6 shows the percentage of successful decoding of
colliding signals, as a function of the SF, for n = 2 and
n = 3 superposed signals. We notice that LoRa is not able
A. Parameter Settings
to decode colliding signals for any SF. This is due to the
Simulations are carried out using our own simulator de- fact that in LoRa, a gateway cannot receive more than one
veloped in Perl. We model a network with a single gateway, signal on the same channel and with the same SF. However,
a single network server, and one hundred EDs. We assume we can see that our algorithm can decode both signals when
that all the EDs transmit on the same channel with the same n = 2. When n = 3, the performance of our algorithm varies
SF, and that their signals are received at roughly the same significantly with SF. This is due to the fact that with a large
power levels at the gateway, i.e., no capture conditions. We SF, the probability to detect a single frequency decreases.
assume that time is divided into slots, and each ED has a Thus, the gateway increases its chances to receive a number
probability p to transmit a frame during a slot, with p ≤ 0.01 of frequencies equal to the number of symbols to decode.
in order to be consistent with the duty-cycle of 1%. For our Thus, compared to LoRa, our algorithm achieves a gain of
algorithms, transmissions on the same slot are considered to 100% when the gateway received two colliding signals and a
be slightly desynchronized (in Subsection IV-B) or completely gain between 18% and 99% when the gateway received three
synchronized (in Subsection IV-C). We choose two values for colliding signals.
SF: SF 7 (which is the smallest SF in LoRaWAN) and SF 12 Figure 7 shows the throughput as a function of the duty-
(which is the largest SF in LoRaWAN). The frame length cycle. In this scenario, we consider a network of one hundred
is set to 50 bytes. We did not force frames to have at least EDs with a duty-cycle less or equal to 1%. We compute
one symbol change. However, the probability that a frame is throughput for LoRa and for our algorithm for two values of
generated with the same repeated symbol is very small, and we SF: SF7 and SF12. We notice that the throughput increases
did not observe it during our simulations. Simulation results when the duty-cycle increases, since each ED sends more
are obtained by averaging over ten thousand samples. frames. Our algorithm enables to achieve a much higher
 % Successful decoding 100 100
11
00
001
110
0
1
Conv.LoRa
Prop.Alg.2 00
11
11
00

% Successful decoding
80 80
001
110
001
110 11
00
001
110
60 60

00
11 00
11
001
110
0
1 00
11
40 40

001
110 00
11
00 00
11
20 20

replacements
0
Conv.LoRa
Prop.Alg.1 (n=2)
PSfrag
Prop.Alg.1 (n=3) replacements
0 001
110 11
00
00 11
11 1 00
0 11
1 0
0 10
1
4 6 8 10 12 1 2 3 4 5
SF Number of colliding signals
Figure 6. Our algorithm outperforms LoRa when a gateway receives n = 2 Figure 8. When the signals are synchronized, our algorithm is able to
or n = 3 colliding signals, with the same SF and on the same channel. decode one frame per collision of two frames, provided that the other one
is retransmitted.

throughput than LoRa, with a gain of up to 60% for a

duty-cycle of 1%. This shows that our algorithm provides computes a gain of up to 25% compared to LoRa. Compared
remarkable throughput gains, even at the system level. to the case where transmissions are slightly desynchronized,
we observe a decrease of 50% of the throughput.
40
Conv.LoRa (SF7) 40
35 Prop.Alg.1 (SF7) Conv.LoRa (SF7)
Conv.LoRa (SF12) 35 Prop.Alg.2 (SF7)
Throughput (bits/s)

30 Prop.Alg.1 (SF12) Conv.LoRa (SF12)

Throughput (bits/s)
30 Prop.Alg.2 (SF12)
25
25
20
20
15
15
10
10
5
replacements 5
0 PSfrag replacements
0.2% 0.3% 0.4% 0.5% 0.6% 0.7% 0.8% 0.9% 1% 0
Duty-cycle (in percentage) 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Duty-cycle (in percentage)
Figure 7. Colliding signals negatively impact the throughput in LoRa.
However, our algorithm is able to increase the throughput up to 60% when Figure 9. Colliding signals negatively impact the throughput in LoRa.
the duty-cycle is equal to 1% and the network is of 100 EDs. However, our algorithm is able to increase the throughput by up to 25%
even for synchronized signals and 100 EDs.

C. Case of synchronized signals

V. C ONCLUSION
Figure 8 shows the percentage of successful decoding of
colliding signals, as a function of the number of colliding Collisions in LoRa negatively impact the throughput of the
signals, in the case where signals are synchronized. Since network, which is already very limited by definition. In this
LoRa is unable to decode any signal when there are several paper, we propose two novel collision resolution algorithms
EDs transmitting on the same channel and with the same SF, that enable to decode some cases of collisions in LoRa by
the percentage of successfully decoded signals is zero for two exploiting the specific properties of this physical layer. Our
and more colliding signals. With our algorithm, when there are first algorithm focuses on the case where nodes are slightly
exactly two simultaneous transmissions, one of them can be desynchronized. The second algorithm focuses on the case
decoded provided that one node retransmits its whole frame. where nodes are synchronized. Based on our simulation re-
Thus, for n = 2 colliding signals, our algorithm decodes 50% sults, we observe that the first algorithm is able to significantly
of the frames, accounting for the retransmission time. improve the throughput, by decoding all collisions of two
Figure 9 shows the throughput as a function of the duty- signals, and many collisions of three signals. The second
cycle. The throughput computed by Conv.LoRa shows the algorithm is also able to improve the throughput, but by
same performance as the one computed for the desynchronized decoding only one signal when two signals are colliding.
signals. However, with our second algorithm, the gateway is These results promote the investigations of a new MAC layer
able to decode one frame for each collision of two frames, pro- based on LoRa, leveraging on the proposed collision resolution
vided that the other frame is retransmitted. Thus, our algorithm algorithms and thereby outperforming LoRaWAN.
 R EFERENCES
[1] Semtech Corporation, “AN1200.22 LoRa Modulation Basics,” Semtech,
Application note Revision 2, 2015, accessed 2018-01-29. [Online].
Available: http://www.semtech.com/uploads/documents/an1200.22.pdf
[2] Sigfox, http//www.sigfox.com.
[3] Ingenu, http//www.ingenu.com.
[4] M. Centenaro, L. Vangelista, A. Zanella, and M. Zorzi, “Long-range
communications in unlicensed bands: the rising stars in the IoT and
smart city scenarios,” IEEE Wireless Communications, 2016.
[5] K. E. Nolan, W. Guibene, and M. Y. Kelly, “An evaluation of low power
wide area network technologies for the internet of things,” in Inter-
national Wireless Communications and Mobile Computing Conference
(IWCMC), pp. 439–444.
[6] J. Petäjäjärvi, K. Mikhaylov, R. Yasmin, M. Hämäläinen, and J. Iinatti,
“Evaluation of LoRa LPWAN technology for indoor remote health and
wellbeing monitoring,” International Journal of Wireless Information
Networks, vol. 24, no. 2, pp. 153–165, 2017.
[7] LoRa Alliance Technical Committee, “LoRaWAN 1.1 Specification,”
LoRa Alliance, Standard V1.1, 2017.
[8] ——, “LoRaWAN 1.1 Regional Parameters,” Standard V1.1, Revision
A, 2017.
[9] N. Sornin, M. Luis, T. Eirich, T. Kramp, and O. Hersent, “LoRaWAN
Specification,” LoRa Alliance, Standard V1.0, 2015.
[10] C. Goursaud and J.-M. Gorce, “Dedicated networks for IoT: PHY /
MAC state of the art and challenges,” EAI endorsed Transactions on
Internet of Things, 2015.
[11] D. Croce, M. Gucciardo, I. Tinnirello, D. Garlisi, and S. Mangione, “Im-
pact of spreading factor imperfect orthogonality in LoRa communica-
tions,” in International Tyrrhenian Workshop on Digital Communication
(TIWDC), ser. Communications in Computer and Information Science
(CCIS), vol. 766, 2017, pp. 165–179.
[12] G. Zhu, C.-H. Liao, M. Suzuki, Y. Narusue, and H. Morikawa, “Evalua-
tion of LoRa receiver performance under co-technology interference,” in
IEEE Consumer Communications and Networking Conference (CCNC),
2018.
[13] J. Haxhibeqiri, F. Van den Abeele, I. Moerman, and J. Hoebeke, “LoRa
scalability: a simulation model based on interference measurements,”
Sensors, vol. 17, no. 6, p. 1193, 2017.