LR-FHSS: Overview and Performance Analysis

Abstract

Long-range frequency hopping spread spectrum (LR-FHSS) is the new physical layer designed to address extremely long-range and large-scale communication scenarios, such as satellite IoT. At its core is a fast frequency hopping technique designed to offer higher network capacity while offering the same radio link budget as LoRa. Additionally, LR-FHSS finely manages packet transmission thanks to its design principles, enabling QoS policies on a per-packet basis. Given the notorious adoption of LoRaWAN in the IoT application landscape, this article is a reference for understanding how exactly LR-FHSS works, the performance it can offer, and its limitations and research opportunities.

Full text

THIS WORK HAS BEEN SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE FOR PEER REVIEW 1
LR-FHSS: Overview and Performance Analysis
Guillem Boquet, Pere Tuset-Peir´
o, Senior Member, IEEE, Ferran Adelantado, Senior Member, IEEE,
Thomas Watteyne, Senior Member, IEEE, Xavier Vilajosana, Senior Member, IEEE
Abstract—Long Range-Frequency Hopping Spread Spectrum
(LR-FHSS) is the new physical layer designed to address
extremely long-range and large-scale communication scenarios,
such as satellite IoT. At its core is a fast frequency hopping
technique designed to offer higher network capacity while
offering the same radio link budget as LoRa. Additionally,
LR-FHSS finely manages packet transmission thanks to its design
principles, enabling QoS policies on a per-packet basis. Given
the notorious adoption of LoRaWAN in the IoT application
landscape, this article is a reference for understanding how
exactly LR-FHSS works, the performance it can offer, and its
limitations and research opportunities.
Index Terms—LPWAN, LoRaWAN, LR-FHSS, Satellite
Networks
I. INTRODUCTION
The Internet of Things (IoT) relies on low-power wireless
communication technologies and protocols to enable reliable
wireless communication between distributed sensor and
actuator devices. Over the last decade, we have seen the rise of
Low-Power Wide Area Network (LPWAN) technologies [1],
with LoRaWAN becoming one of the most prominent players
in the market thanks to long-range and robust communications,
coupled with a simple network architecture that allows
for solutions that are easy to deploy and manage. At
the physical layer, LoRaWAN uses Long Range (LoRa), a
robust Chirp Spread Spectrum (CSS) modulation developed
by Cycleo, later acquired and commercialized by Semtech
under the LoRa trademark. The LoRa physical layer is
designed to prioritize uplink communication and ensure
low power operation, limited by a low data rate (250 bps
with spreading factor 12 and 125 kHz channel as the
most restrictive) and kilometer-scale communication range.
Different configurations of the physical layer are available
providing different levels of chirp redundancy and thus trading
off bandwidth utilization and robustness. LoRaWAN defines
different bandwidth configurations per channel, ranging from
125 kHz to 500 kHz, depending on the regional parameters
and available bandwidth. The access to these channels is based
on pure ALOHA, limited by regional duty cycle regulations
which constrain the time on air that can be utilized per device,
further limiting the maximum achievable throughput [2].
The technology has been widely adopted. However, there
are drawbacks and limitations to LoRaWAN, particularly in
dense deployments where performance, that is, the overall
network capacity, is severely limited by duty-cycle regulations
This research is co-financed by the EU Regional Development Fund within
the ERDF Operational Program of Catalonia 2014-2020 with a grant of 50%
of total cost eligible, the SPOTS project (RTI2018-095438-A-I00) funded by
the Spanish Ministry of Science, and the 2017 SGR 60 by the Generalitat de
Catalunya.
Fig. 1. Two LR-FHSS (DR8) 30-B and one LoRa (DR0/SF12) 10-B
packets transmitted simultaneously in the EU 868-870 MHz band (Channel 1).
A packet transmission using LoRa occupies the whole channel bandwidth
(125 kHz), whereas for LR-FHSS the fragments of a given packet are
distributed over time in randomly selected subchannels (488 Hz) within the
entire channel bandwidth (137 kHz). Despite several frame collisions, both
LR-FHSS packets will be decoded successfully with high probability thanks
to its redundancy.
and the use of simple Medium Access Control (MAC)
protocols [3]. In that sense, the research community has
issued during the last years several proposals to address
the fundamental operation of LoRaWAN at the physical
and data-link layers [4], [5], as well as the overall system
operation [6], using both analytic models and simulations [7].
In addition, the scalability and reliability issues of LoRaWAN
have been the focus of recent research efforts towards ensuring
fairness [8], improving channel usage [9], or even scheduling
LoRa transmissions [10].
As a novelty, Semtech has announced an extension of
the LoRa physical layer called LR-FHSS, which stands
for Long Range-Frequency Hopping Spread Spectrum [11].
The extension is motivated by emerging use cases with
increasingly larger and denser network deployments, including
satellite-scale LoRaWAN networks. Its main goal is to
increase network capacity and robustness by adopting the
basis of FHSS modulation technique, while keeping the same
communication range than LoRa and meeting the European
Telecommunications Standards Institute (ETSI) [12], the
Federal Communications Commission (FCC) [13] and the
Association of Radio Industries and Businesses (ARIB) [14]
regulations. Additionally, LR-FHSS has been designed to
introduce higher levels of network flexibility, targeting
applications that require differentiated service levels.
arXiv:2010.00491v3 [cs.NI] 10 Dec 2020

THIS WORK HAS BEEN SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE FOR PEER REVIEW 2
We expect LR-FHSS to have a big impact on LPWAN and
enable viable satellite IoT solutions. Therefore, in this article,
we provide an overview of the technology and its performance,
comparing it to today’s LoRa, and validate the scalability gain
announced by Semtech. We also identify open research issues
and directions for this new physical layer.
The remainder of this article is organized as follows.
Section II presents an overview of LR-FHSS. Section III
studies the scalability of LR-FHSS networks and compares it
to LoRa networks. Section IV outlines open research questions
for LR-FHSS networks. Finally, Section V concludes the
article.
II. LR-FHSS OPE RATI ON
LoRaWAN prioritizes uplink capacity and limits downlink
transmissions to sporadic data or control packets. LR-FHSS
is a fast FHSS modulation used for uplink only; downlink
communication is achieved with current LoRa, since the same
radios can switch between modulations.
LR-FHSS relies on two bit rates (162 and 325 bps),
according with the new LoRaWAN Data Rate (DR) modes
(Section II-A). To initiate transmission of a packet (Section
II-B), end-devices randomly select one of the LR-FHSS
channels available and use a pure ALOHA access mechanism.
As illustrated in Fig. 1, the channels are divided into several
subchannels (Section II-C), which the transmitter uses to
change the carrier frequency according to a certain hopping
pattern at each transmission (Section II-D). First, several
replicas of the header are transmitted. The number of replicas
is defined by each LR-FHSS DR. The packet payload is then
split into fragments with a duration of ∼50 ms. Contrary
to the packet header, only a single copy of each fragment
is transmitted. All headers and payload fragments are sent
consecutively on each subchannel determined by the frequency
hopping sequence of the transmitter. At the other end, the
gateway reassembles the packet payload using the information
within a header.
Unlike for LoRa channels, as long as LR-FHSS
transmissions fall entirely in the gateway processed bandwidth,
the packets can be demodulated. This means that the
gateway does not need to know in advance any channel
hopping sequence, nor the exact channels frequencies and
bandwidths. Therefore, different devices can use a different
spreading bandwidth and may not all have the channel at
the same frequency. It also implies that multiple transmitters
can operate at the same time, provided they use different
channel hopping sequences and the gateway is able to
listen to the whole channel bandwidth at the same time.
This increases the complexity of signal detection at the
receiver, compared to LoRa. However, it allows hundreds of
packets to be received simultaneously, which is appropriate
for satellite-scale networks where the number of devices
interfering within the coverage area of a gateway located at
space is much greater than the number of devices found in
current LoRaWAN use cases.
LR-FHSS can be used with limited support (no intrapacket
hopping) for current SX1272/76 devices through a firmware
update. In contrast, the newer SX1261, SX1262 and SX1268
modems are fully LR-FHSS compatible and can take
advantage of its FHSS modulation. Lastly, demodulation of
LR-FHSS signals can be achieved by all gateways using the
SX1302 modem with a firmware upgrade.
A. Data Rates
LR-FHSS (LR-FHSS) DR modes are shown in Table I.
In the EU863-870 band, those are DR8/10 (slower, higher
robustness) and DR9/11 (faster, lower robustness). DR8 and
DR10 use a coding rate of 1/3, 3 header repetitions, and a
physical bit rate of 162 bps. DR9 and DR11 use a coding
rate of 2/3, 2 header repetitions, and a physical bit rate
of 325 bps. The aforementioned coding rates are used to
convolutionally encode the payload bits that are decoded
using a Viterbi decoder. Although the specification allows
for header repetitions ranging from 1 to 4, only the two
configurations above are used. This redundancy is key to
ensuring the robustness of LR-FHSS when fragments are
spread in frequency. For example, the gateway device is able
to reassemble a packet transmitted using DR9 with high
probability, even if 1 of the 2 headers and 1/3 of the bits within
the payload fragments are lost. These mechanisms mitigate the
impact of interference from other LoRa and LR-FHSS devices
operating in the same band, as well as devices using other
wireless technologies.
B. Packet Format
The LR-FHSS packet, Fig. 2, is composed of a SyncWord,
a PHY Header (including CRC) and the payload (including
CRC). The first transmitted fragments (2 or 3 replicas)
contain the entire header (SyncWord, Header, Header CRC).
The header is transmitted at a fixed bit rate for a duration
of 0.233 s, containing information about channel hopping
sequence, payload length, data rate, number of header replicas
and coding rate. Each packet header contains the necessary
information such that the gateway can compute the exact list
of frequencies that the packet will be using. Thus, the gateway
must receive at least one copy of the header to be able to detect
and reassemble the packet. Note in Fig. 1 the transmission of
3 consecutive header replicas with the same duration, which
are not split across multiple fragments despite being longer
than ∼50 ms. The next fragments only contain the payload
and are transmitted at the configured data rate. Recall that its
information is encoded in such a way that even if 1/3 of the
fragments are lost, the information can still be recovered and
the packet can be reassembled with high probability.
Additionally, LR-FHSS enables to select the modulation,
number of header repetitions, coding rate and frequency
spreading bandwidth, on a packet per packet basis. This allows
a fine grained network resource management based on the link
quality that allows to reach the required QoS for each device.
C. Channels and subcarriers
Similar to LoRaWAN, the frequency band is split into
different Operating Channel Width (OCW) channels. The

THIS WORK HAS BEEN SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE FOR PEER REVIEW 3
TABLE I
LR-FHSS MA IN SP EC IFIC ATIO NS AN D PARA ME TER S FO R EU AN D US REGIONS.
Region European Union (ETSI, 863-870 MHz) United States (FCC, 902-928 MHz)
LoRaWAN data rate alias DR8 DR9 DR10 DR11 DR5 DR6
LR-FHSS number of channels 7 4 7 4 8
LR-FHSS OCW (kHz) 137 336 1523
LR-FHSS OBW (Hz) 488
Minimum separation between
LR-FHSS hopping carriers (kHz) 3.9 25.4
Number of physical carriers available
for frequency hopping in each OCW channel 280 (8x35) 688 (8x86) 3120 (52x60)
Number of physical carriers usable for
frequency hopping per end-device transmission 35 86 60
Coding rate 1/3 2/3 1/3 2/3 1/3 2/3
Physical bit rate (bits/s) 162 325 162 325 162 325
Max. MAC payload size (bytes) 58 123 58 123 125 125
Max. MAC payload fragments 61 64 61 64 130 65
Header replicas 3 2 3 2 3 2
PHY header duration per replica (seconds) 0.233 0.233 0.233 0.233 0.233 0.233
PHY time on air (seconds) 0.70 + 3.06 0.47 + 3.19 0.70 + 3.06 0.47 + 3.19 0.70 + 6.48 0.47 + 3.24
1
SyncWord Header Header
CRC Payload CRC
Payload length Data
rate
Coding
rate Grid Hop Band-
width Hopping sequence SyncWord
index RFU
4 bytes 4 bytes 1 byte 2 bytes
2 bits
8 bits 3 bits 2 bits 1 bit 1 bit 4 bits 9 bits 2 bits 2 bits
Fig. 2. LR-FHSS packet structure. While the header is 5 bytes long (including header-CRC), the radio transmits a total of 80 bits as it uses a convolutional
code with a 1/2 code rate. Between the header and the payload, the transceiver waits a time equal to 2 bits (shaded gray) to allow the receiver to decode and
process the header. The header is 114 bits long, the payload can be up to 125 bytes.
number of channels available depends on the region of
the world in which the network operates, thus a white-list
is used to mark those channels. In North America (FCC
902-928 MHz), the specification defines 8 LR-FHSS OCW
channels with 1.523 MHz bandwidth and center frequencies
equal to (903 + 1.6n)MHz, n= 0,...,7. In Europe
(ETSI 863-870 MHz), the LR-FHSS OCW channels are
defined to have a bandwidth equal to 137 kHz or 336 kHz
depending on the selected DR. The number of channels that
can be used depends on the specific gateway technology
used by the network operator. The number of LR-FHSS
channels supported by a gateway depends on the number of
Digital Signal Processors (DSPs) used, each DSP covering
1.523 MHz. For example, in a gateway with one DSP and
200 kHz channel spacing, end-devices can use up to 7
LR-FHSS channels when using a 137 kHz OCW channel for
the uplink.
Each LR-FHSS OCW channel is divided into several
Occupied Band Width (OBW) physical channel subcarriers
with a bandwidth of 488 Hz. For example, the 137 kHz
channel bandwidth in the EU 863-870 MHz regional
parameters enables 280 LR-FHSS OBW subcarriers in
a single LR-FHSS OCW channel, therefore supporting
simultaneous transmission. Because the regional regulations
impose restrictions on the time and bandwidth occupation,
subcarrier hopping policies must have a minimum frequency
hop. That is, two consecutive subcarriers of a hopping
sequence must be separated by a minimum distance. In the
EU band, the minimum frequency shift is 3.9 kHz for each
fragment [12], thus creating 8 simultaneous grids of 35 usable
subcarriers in each of the 137 kHz LR-FHSS OCW channels
for DR8/9. In the US band, the minimum separation between
LoRa–E hopping subcarriers is 25.4 kHz. Similarly, this results
in 52 simultaneous groups of 60 usable subcarriers for each
particular 1.523 MHz OCW channel.
Table I summarizes the LR-FHSS physical layer both
for ETSI and FCC regulations. In some configurations,
the relatively small number of subcarriers in a group (35
subcarriers in each of the 8 groups in an EU 137 kHz
OCW channel) may cause fragment collisions, as these are
sent in a pseudo-random frequency hopping pattern. However,
the probability of concurrent transmissions, that is, two end
devices selecting the same frequency hopping pattern at
approximately the same time, is very low.
D. Frequency Hopping Policy
For each uplink packet, the device randomly selects a
channel amongst the ones enabled that can support the data

THIS WORK HAS BEEN SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE FOR PEER REVIEW 4
rate. The transmission starts on a random physical subchannel
(grid) inside the LR-FHSS spreading bandwidth, then follows
a pseudo-random frequency hopping pattern computed by the
device. Given a specific OCW frequency offset, the hopping
pattern is obtained as the outcome of a simple 32-bit hash
function executed by the LR-FHSS device, modulo the number
of available channels in the grid or physical subcarriers usable
for channel hopping per end-device transmission (35 or 86 in
EU and 60 in USA, Table I). In particular, the input of the hash
function is determined by the result of a device-specific 9-bit
random number (header’s hopping sequence field in Fig. 2)
plus the product of the current fragment number and 216.
Finally, the hash is multiplied by the minimum frequency
separation between LR-FHSS hopping carriers to comply with
the regulations of the transmission band. All of this ensures
a pseudo-random pattern where all physical channels are
statistically used equally.
III. KEY PERFORMANCE AS PE CT S
Understanding the performance of LR-FHSS under
saturation conditions and how it compares to LoRa are
essential questions to any LR-FHSS prospective user. In
this section, we evaluate and compare its performance from
the end-device and the network scalability perspective. The
software used in this article is publicly available at https:
//github.com/wine-uoc/.
To evaluate its key performance aspects we developed
a packet-based network simulator that implements the
basic functionality of both protocols. The simulator uses
an exponential distribution to model the random time
between independent packet arrivals, where the arrival rate is
established at the maximum transmission rate allowed by the
duty cycle regulations. Packets are allocated in a discrete time
and frequency space with a millisecond time resolution and
OBW subcarrier granularity. Simultaneous transmissions only
cause a collision if they both select the same OBW subcarrier
and overlap in time. For each LR-FHSS device, a channel
hopping sequence is created at the beginning, according to
the procedure described in Section II and the parameters
presented in Table I. That is, packets are divided into several
fragments and spread following the device’s hopping sequence
at transmission time. At reception, a LR-FHSS packet is
successfully decoded if minimum one of the headers and 1/3
(or 2/3) of the payload have not collided. For LoRa, packets
are transmitted using the entire available channel bandwidth,
and decoded only if they have not collided.
We performed extensive simulations assuming ideal channel
conditions and the EU 863-870 MHz band, which imposes
the most restrictive per-channel duty cycle of 1%. Only
non-colliding packets are successfully received, hence, the
results presented should be considered as a lower performance
bound of both LoRa and LR-FHSS.
A. End-device Capacity
Fig. 3 shows the time on air (top) and the number of
packets per hour per end-device (bottom) for different MAC
payload sizes (10 and 50 B) for LoRa DR0 (SF12), DR5 (SF7
Fig. 3. Relationship between the packet duration (seconds) and the maximum
per end-device transmission rate (packets/hour) when abiding to the EU
868-870 MHz band regulation, which imposes a 1% duty cycle per device
and channel.
125 kHz), and the LR-FHSS data rates available in Europe
(DR8/10 and DR9/11). As depicted, LoRa DR5 provides the
highest number of packets per hour per end-device (369.1 to
873.5), whereas LR-FHSS DR8/10 provides the least (10.7 to
27.5). In contrast, LoRa DR0 provides between 15.6 and 36.3
packets per hour per end-device, whereas LR-FHSS DR9/11
provides similar capacity regardless of the MAC payload size,
having between 20.1 and 46.6 packets per hour per end-device.
As expected, the lower LR-FHSS DR combined with the
imposed duty cycle regulations limit the number of packets
that can be transmitted. Observing these results, one could
conclude that it does not provide any real benefit in terms
of individual end-device capacity. However, LR-FHSS is
not designed to increase the number of packets that can
be transmitted by each end-device independently. Rather,
its strength lays in the overall network capacity increase
provided by the statistical multiplexing of combining time and
frequency diversity.
B. Network Capacity
Fig. 4 compares a 125 kHz LoRa channel with a
137 kHz LR-FHSS channel (supporting 8 grids of 35 physical
subcarriers), and shows how LR-FHSS scales with the number
of end-devices transmitting at the maximum possible rate
using a payload of 10 B. The number of supported end-devices
peaks at 50 when using LoRa regardless of the DR used. This
can be explained by the fact that the network is limited at
the data-link layer by the ALOHA MAC protocol, and at
the physical layer by the radio duty cycle regulation (1%
at the EU 868-870 MHz band). Higher DRs provide higher
goodput because each DR (almost) doubles the data rate of the
previous one, hence the transmission time on air of the packet
decreases proportionally and more packets (information) can
be fitted, offering the same channel load. In other words,
the total offered load remains constant despite the DR used
because devices’ maximum transmission rate is fixed by

THIS WORK HAS BEEN SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE FOR PEER REVIEW 5
Fig. 4. Total useful bytes received per hour (goodput) when a given number
of LoRa and LR-FHSS end-devices transmit 10 B of payload at the maximum
1% duty cycle allowed by the EU 868-870 MHz regulation. Both LoRa and
LR-FHSS results are obtained considering 125 kHz and 137 kHz channels,
respectively.
the duty-cycle regulation. A higher DR provides additional
capacity, resulting in an unchanged collision probability in this
scenario. Simulations performed with payload sizes up to 50 B
showed the same behavior.
In contrast, the number of end-devices that can transmit
simultaneously in LR-FHSS increases, leading to a significant
increase in the goodput of the entire network as it scales.
Notably, LR-FHSS ensures a goodput proportional to the
throughput that is one order of magnitude larger than
LoRa for the fastest of the DRs (DR5) and two orders of
magnitude for the slowest (DR0). The maximum goodput
while using DR9/8 and a payload size of 10 B occurs when
8,000/18,000 end-devices transmit at the maximum allowed
rate (given the payload size, Fig. 3) or, in other words,
370,000/500,000 packets with 10-B payload are generated per
hour within the network. Above that number of end-devices
or packet generation, the performance of DR9 decreases faster
than DR8, causing DR8 to scale better. This is because
DR8 repeats the header 3 times (2 times in DR9) and
transmits more redundant information for error correction,
which increases robustness against collisions. Thus, we can
extrapolate that DR8 performs better than DR9 in terms
of goodput on channels with significant interference. For
example, in Fig. 4 when 16,000 or more end-devices are
transmitting simultaneously at the maximum allowed rate, the
LR-FHSS DR8 network outperforms DR9.
Fig. 5 shows the crossover points between LoRa and
LR-FHSS goodput as a function of the number of generated
packets per hour for the different LoRa DR configurations
and MAC payload sizes (10 and 50 B). If we compare
LR-FHSS with devices transmitting 10 B of payload size
against the lowest LoRa data rate configuration (DR0), the
LR-FHSS network requires a network load greater than 2,800
and 985 packets/hour for DR8 and DR9, respectively, to
provide a higher goodput than the LoRa DR0 network. At
Fig. 5. Comparison between LoRa and LR-FHSS DRs performance. Markers
show the crossover points in number of generated packets per hour required
for LR-FHSS to provide a better goodput than LoRa, as a function of payload
size.
the other end, LR-FHSS requires around 68,000 packets/hour
for DR8 and 23,000 packets/hour for DR9 to provide higher
goodput compared to the LoRa DR5 configuration. This is an
exponential growth with a growth rate of almost 2, similar to
the increase in the available data rate when the LoRa DR is
increased from DR0 to DR5. The same growth rate is found
when devices transmit 50 B of payload size. However, the
crossover points occur for lower values of generated packets
per hour, meaning that LR-FHSS is more efficient in terms
of goodput when the ratio between payload and overhead
increases. Interestingly, the required number of packets/hour
needed for LR-FHSS to be greater than LoRa is divided
by 2 if devices increase the payload size from 10 to 50-B.
For example, regarding LR-FHSS DR8 and LoRa DR0, the
crossover point for 10 B is at 2,800 packets/hour while for
50 B is at 1,400 packets/hour.
Finally, notice that the results presented are for a single
LoRa and LR-FHSS channel and, hence, the overall network
capacity has to be multiplied by the number of available
channels and DR configurations in the band of interest. For
example, transmitting a 10-B payload in the EU 868-870 MHz
band, a legacy LoRa DR0 network could ideally support a
total load of 96,000 packets/hour at its peak efficiency, when
considering the use of 8 independent 125 kHz channels and
6 DRs simultaneously. In contrast, the total capacity of a
EU LR-FHSS network would be around 3.5M packets/hour
when using DR8, and 1.48M packets/hour when using DR9.
This assumes 7 and 4 channels with 8 frequency grids each,
as summarized in Table I, which represents a 36×and 15×
capacity increase for LR-FHSS when compared to LoRa.
C. Use Cases
With a lower data rate than DR0 (SF12), the time on air
for a LR-FHSS packet and the number of packets that an
end-device can transmit during a period of time is smaller
than LoRa. Yet, this nominal under-performance does not

THIS WORK HAS BEEN SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE FOR PEER REVIEW 6
translate into an effective reduction of the goodput because
fragment spreading reduces collision probability. This mode
of operation can also be exploited to scale the network in
terms of the number of end-devices supported concurrently.
That is, LR-FHSS slices the spectrum so that multiple
end-devices can use the same channel with a insignificantly
low collision probability. Hence, LR-FHSS offers higher
network scalability when compared to LoRa DR0. Considering
the presented results, we can conclude that LR-FHSS provides
better network scalability than LoRa thanks to the adoption
of FHSS modulation technique. In particular, the 162 bps
LR-FHSS DRs provide two orders of magnitude more network
capacity, while offering the same radio link budget than LoRa
DR0 (SF12). In addition, LR-FHSS retains the long-range
communication characteristics of LoRa. This makes LR-FHSS
suitable for terrestrial and satellite deployments, where a high
end-device density or a large gateway coverage area leads to an
increased level of interference. Summarizing, LR-FHSS allows
the necessary >155 dB link margin for low Earth orbit (LEO)
satellite IoT plus the capacity to receive hundreds of packets
simultaneously.
IV. OPE N RESEARCH CHALLENGES
This section discusses two main open research challenges
presented by LR-FHSS: optimal selection of frequency
hopping sequences and coexistence with legacy LoRa
networks.
A. Optimal Frequency Hopping Sequences
The LR-FHSS specification uses a simple frequency
hopping pattern based on a 32-bit hash function. These
hopping sequences create grids on the OCW channels that
space subcarriers a minimum frequency distance according
to the region’s specification. While these schemes ensure
a spreading of the packets among the subcarriers, optimal
policies can be designed to address specific channel conditions.
In particular, schemes that dynamically adapt sequences
to continuously match a changing environment. Because
LR-FHSS networks can manage the spectrum used by each
device individually, adaptive frequency hopping [15] can either
aim to maintain data throughput or allow flexible fine-grained
network resource management to achieve the desired QoS
per device. Hence, an important research topic is to evaluate
different hopping sequences to maximize the properties of
the FHSS scheme. Also, it would be interesting to explore
sequence management policies that determine how and when
these dynamic sequences need to be updated.
B. Coexistence with Legacy LoRa Networks
As LoRaWAN deployments will integrate end-devices
using different LoRa DRs, including the novel LR-FHSS,
and will have to coexist with other LoRaWAN networks
using different combinations of DRs, coexistence is a key
aspect to investigate. In particular, the results evaluating the
coexistence between LoRa and LR-FHSS DRs will allow
to build recommendations and best-practice strategies for
deploying these technologies and ensuring their performance.
Such recommendations may be to dedicate some channels
to LR-FHSS, or to limit the proportion of different DRs in
each channel. Another topic related to coexistence is adaptive
data rate (ADR), for which the role of LR-FHSS needs to
be defined. In that regard, LoRaWAN may have to include
new functionalities to adapt to an increase in network load
by dynamically switching to LR-FHSS when the goodput
degrades.
V. CONCLUSIONS
LR-FHSS is a physical layer designed to offer flexibility
for differentiated services and increase the scalability of
LoRaWAN networks, retaining its long-range communication
characteristics. This article has presented LR-FHSS, evaluated
its performance and highlighted its limitations. Results showed
a significant increase in network scalability at the cost of
individual end-device capacity. This article has also discussed
open research challenges including mechanisms for optimizing
frequency hopping sequences, policies for correctly planning
and adapting the network to load, as well as LoRaWAN
networks coexistence.
REFERENCES
[1] U. Raza, P. Kulkarni, and M. Sooriyabandara, “Low Power Wide Area
Networks: An Overview,” IEEE Communications Surveys Tutorials,
vol. 19, no. 2, pp. 855–873, 2017.
[2] F. Adelantado, X. Vilajosana, P. Tuset-Peiro, B. Martinez,
J. Melia-Segui, and T. Watteyne, “Understanding the Limits of
LoRaWAN,” IEEE Communications Magazine, vol. 55, no. 9, pp.
34–40, 2017.
[3] J. C. Liando, A. Gamage, A. W. Tengourtius, and M. Li, “Known and
Unknown Facts of LoRa: Experiences from a Large-Scale Measurement
Study,” ACM Trans. Sen. Netw., vol. 15, no. 2, Feb. 2019.
[4] R. B. Sørensen, D. M. Kim, J. J. Nielsen, and P. Popovski, “Analysis of
Latency and MAC-Layer Performance for Class A LoRaWAN,” IEEE
Wireless Communications Letters, vol. 6, no. 5, pp. 566–569, 2017.
[5] D. Magrin, M. Capuzzo, and A. Zanella, “A Thorough Study of
LoRaWAN Performance Under Different Parameter Settings,” IEEE
Internet of Things Journal, vol. 7, no. 1, pp. 116–127, 2020.
[6] A. Mahmood, E. Sisinni, L. Guntupalli, R. Rond´
on, S. A. Hassan, and
M. Gidlund, “Scalability Analysis of a LoRa Network Under Imperfect
Orthogonality,” IEEE Transactions on Industrial Informatics, vol. 15,
no. 3, pp. 1425–1436, 2019.
[7] F. Van den Abeele, J. Haxhibeqiri, I. Moerman, and J. Hoebeke,
“Scalability Analysis of Large-Scale LoRaWAN Networks in ns-3,”
IEEE Internet of Things Journal, vol. 4, no. 6, pp. 2186–2198, 2017.
[8] B. Reynders, W. Meert, and S. Pollin, “power and spreading factor
control in low power wide area networks,” in 2017 IEEE International
Conference on Communications (ICC).
[9] J. Ort´
ın, M. Cesana, and A. Redondi, “Augmenting LoRaWAN
Performance With Listen Before Talk,” IEEE Transactions on Wireless
Communications, vol. 18, no. 6, pp. 3113–3128, 2019.
[10] B. Reynders, Q. Wang, P. Tuset-Peir´
o, X. Vilajosana, and S. Pollin,
“Improving Reliability and Scalability of LoRaWANs Through
Lightweight Scheduling,” IEEE Internet of Things Journal, vol. 5, no. 3,
pp. 1830–1842, 2018.
[11] N. Sornin, M. Luis, T. Eirich, T. Kramp, and O. Hersent, “RP2-1.0.2
LoRaWAN Regional Parameters,” LoRa Alliance Technical Committee
and others, 2020.
[12] E. T. ETSI, “Electromagnetic compatibility and Radio spectrum Matters
(ERM); Short Range Devices (SRD); Radio equipment to be used in the
25 MHz to 1 000 MHz frequency range with power levels ranging up to
500 mW,” European harmonized standard EN, vol. 300, no. 220, p. v3,
2019.
[13] FCC, “FCC regulation. Title 47 Chapter I, Part 15, and Section 247:
Operation within the bands 902-928 MHz, 2400-2483.5 MHz, and
5725-5850 MHz,” Federal Communications Commission (FCC), 2015.

THIS WORK HAS BEEN SUBMITTED TO IEEE COMMUNICATIONS MAGAZINE FOR PEER REVIEW 7
[14] ARIB, “920MHz-Band Telemeter, Telecontrol and Data Transmission
Radio Equipment. Standard ARIB STD-T108,” Association of Radio
Industries and Business, p. v1.3, 2019.
[15] P. Popovski, H. Yomo, and R. Prasad, “Strategies for adaptive frequency
hopping in the unlicensed bands,” IEEE Wireless Communications,
vol. 13, no. 6, pp. 60–67, 2006.
Guillem Boquet received his M.Sc. and PhD. in Telecommunications
Engineering from Universitat Aut`
onoma de Barcelona (UAB) in 2014 and
2020, respectively. He is currently a Researcher at the Wireless Networks
(WiNe) group of the Universitat Oberta de Catalunya (UOC).
Pere Tuset-Peir´
o(M’12, SM’18) is Assistant Professor at the Universitat
Oberta de Catalunya (UOC) and Senior Researcher at the Wireless Networks
(WiNe) group. He received his M.Sc. in Telecommunications Engineering
from Universitat Polit`
ecnica de Catalunya (UPC), and his Ph.D. in Network
and Information Technologies from Universitat Oberta de Catalunya (UOC).
Ferran Adelantado (M’08, SM’19) is Associate Professor at the Universitat
Oberta de Catalunya (UOC) and Senior Researcher at the Wireless Networks
(WiNe) group. He holds a M.Sc. degree in Telecommunications Engineering
(2001) and a PhD (2007) from the Universitat Polit`
ecnica de Catalunya (UPC).
Thomas Watteyne (sM’06, M’09, SM’15) holds a PhD in Computer Science
(2008), an MSc in Networking (2005) and an MEng in Telecommunications
(2005) from INSA Lyon, France. He is a Research Director at Inria in Paris,
leading the EVA research team, and network designer at Analog Devices. He
previously worked at Orange Labs and UC Berkeley.
Xavier Vilajosana (M’09, SM’15) received his B.Sc. and M.Sc in Computer
Science from Universitat Polit`
ecnica de Catalunya (UPC) and his Ph.D. in
Computer Science from the Universitat Oberta de Catalunya (UOC). He has
been a researcher at Orange Labs, HP and UC Berkeley. He is now Professor
at UOC.