You are on page 1of 6

2018 Global Internet of Things Summit (GIoTS)

Comparison of CoAP and MQTT Performance


Over Capillary Radios
Anna Larmo, Felipe Del Carpio Pontus Arvidson, Roman Chirikov
Ericsson Research, Ericsson Research,
Ericsson Oy, Ericsson AB,
Jorvas, Finland Stockholm, Sweden
{anna.larmo, felipe.del.carpio}@ericsson.com {pontus.arvidson, roman.chirikov}@ericsson.com

Abstract—The IoT protocols used in the application layer, for statistical purposes and traffic is mostly from the sensor
namely the Constraint Application Protocol (CoAP) and Message to the cloud, it may be acceptable to lose a report here and
Queue Telemetry Transport (MQTT) have dependencies to the there. Even in such scenarios, however, reliable transport may
transport layer. The choice of transport, Transmission Control
Protocol (TCP) or the User Datagram Protocol (UDP), on the be required on occasion when firmware updates or security
other hand, has an impact on the Internet of Things (IoT) patches are done on devices.
application level performance, especially over a wireless medium. The physical layer and the medium access control, i.e., L1
The motivation of this work is to look at the impact of the protocol and L2, may differ greatly depending what kind of access
stack on performance over two different wireless medium real- technology is used. In the end to end path of an IoT application
izations, namely Bluetooth Low Energy and Wi-Fi. The use case
studied is infrequent small reports sent from the sensor device data flow, there may be fast fiber links, Ethernet cabling, as
to a central cloud storage over a last mile radio access link. We well as radio access links over a cellular network, and finally
find that while CoAP/UDP based transport performs consistently the last mile over a short range radio access technology in
better both in terms of latency and power consumption over both the capillary domain such as Bluetooth Low Energy (BLE) or
links, MQTT/TCP may also work when the use requirements Wi-Fi. In this paper, we focus on the aforementioned capillary
allow for longer latency providing better reliability. All in all, the
full connectivity stack needs to be considered when designing an domain technologies.
IoT deployment.
II. P REVIOUS WORK
Keywords-IoT, CoAP, MQTT, TCP, UDP, BLE, 802.11ah.
Previous work on comparing CoAP and MQTT has been
done in the form of small scale trials over an unnamed
I. I NTRODUCTION
radio technology [2] and by analytical means [3]. In [2], a
The idea of IoT is to use the Internet Protocol, mainly IPv6, healthcare use case was realized with prototype hardware. The
to communicate between connected devices and the cloud. A performance of the two protocols was then measured when a
plethora of standards addressing the IoT on different levels of set of medical sensor data was transferred over an unspecified
the protocol stack are available today and may be combined wireless link. It was found that unpredictable packet loss is
in different ways. handled poorly by CoAP but better by MQTT due to use
The IoT umbrella covers a large number of use cases. Here of TCP. In [3], the performance of the two protocols was
we focus on use cases where small data transmissions are studied over very lossy links. It was found that by tuning
expected from and to a large number of sensors and actuators. CoAP parameters, the performance can be greatly improved
Such use cases are for example connected lighting, connected and CoAP often performs better than MQTT due to the lower
home sensors and actuators, smart metering, smart agriculture, overhead, although some level of packet loss needs to be
etc. accepted when UDP is used. Our contribution to the discussion
Most of the popular IoT applications today use either the is to study the performance over realistically modeled actual
Internet Engineering Task Force (IETF) standardized Con- PHY and MAC protocols and see how the combination of the
straint Application Protocol (CoAP) or the OASIS specified protocols all the way from the application layer down to the
Message Queue Telemetry Transport (MQTT) as the data physical layer impacts the application performance.
transfer of choice. Other examples of IoT application layer Both BLE and Wi-Fi are traditionally deployed in a star
transfer protocols are the Hyper Text Transfer Protocol, Data or point to point topology and designed to handle two-way
Distribution Service and Advanced Message Queuing Protocol communication. Also, both technologies have recently evolved
[1]. for longer range. Thus, they provide an interesting comparison
The requirements of the use case set boundary conditions [4], [5]. IEEE 802.15.4 systems are typically designed mainly
to the reliability and latency of the communication link. If for one-way data streams and to function in a mesh topology.
the use case involves collecting large amounts of sensor data While it is true that IEEE 802.15.4 systems may also be
978-1-5090-5873-0/17/$31.00
c 2018 IEEE

/18/$31.00 ©2018 IEEE


2018 Global Internet of Things Summit (GIoTS)

brings more overhead but also reliability with its connection


oriented confirmed delivery of packets. A TCP connection
requires a three-way handshake between the sensor and the
central node at the setup phase. On the contrary, a UDP data
transmission does not require any setup or bi-directionality; the
packet is simply sent from the transmitter end to the receiver
end at a suitable time. TCP fast open is a variant of the TCP
protocol, which allows data transmission along with TCP SYN
and SYN-ACK transmissions [9]. Although TCP fast open will
Fig. 1. Typical IoT stack options for massive MTC use cases.
simplify TCP usage for IoT, we have chosen to limit our study
on the legacy TCP as it still is widely used.
Security is a key part of any IoT stack. There are different
deployed in a star topology, we focus on the two newer ways to provide security and typical options for the discussed
technologies in this evaluation. stack are to use either Transport Layer Security (TLS) for TCP
When good battery performance is required from the access or Datagram Transport Layer Security (DTLS) for UDP. Secu-
link, the use of different kinds of sleep mechanisms often rity solutions typically bring overhead and may introduce extra
becomes a necessity, as one of the most power consuming handshakes to the procedure as well. However, in this work
modules in the radio device is the receiver. The total power we assume that the security handshakes have already been
budget of the IoT device may need to be shared between performed at an earlier time, e.g., out-of-bound at deployment
the radio module, the actual sensor and/or actuator, and a or manufacturing phase of the sensor life cycle.
processor.
A. Short introduction to CoAP and MQTT
In the remainder of this paper, we study how different
protocol stacks perform over BLE and IEEE 802.11ah last CoAP is a RESTful application layer transfer protocol
mile short range radios. The paper is organized as follows: specified by the IETF [6]. CoAP typically is run over UDP
first we present an overview of the IoT protocol stack with connections, although work to adapt to TCP is also ongoing
different protocol options in the different layers, followed by [7]. CoAP may be run in two different modes, namely the con-
a description on basics of the CoAP and MQTT protocols. We firmed delivery mode and the non-confirmed delivery mode.
discuss the potential of the two chosen IoT radio connectivity For sensor reporting with relaxed reliability requirement the
solutions, BLE and 802.11ah. Further, we explain the IoT use non-confirmed mode is sufficient. Confirmations may be also
case scenario and system level assumptions for our simulation requested only for some packets.
study. Finally, we analyze the IoT performance over BLE and MQTT [8] is an application layer publish/subscribe transfer
802.11ah in terms of uplink packet transmission delays and protocol specified by OASIS. In MQTT, topics are set up
battery levels of connected sensors. between the client and the server which are then subscribed
and published to. For sensor reporting, the sensor client
III. I OT S TACK FOR M ASSIVE MTC registers at a server to publish sensor data, and vice versa
Several realizations of the full IoT stack have been discussed register for topics to receive updates when needed.
in the industry and academia. However, it is likely that
convergence to a single option will not happen in the near B. Power saving on the radio interface
future due to different characteristics of the large variety of Low power consumption is a key requirement for massive
use cases grouped under IoT. Figure 1 depicts different options IoT radio modules. Turning off the receiver is often referred to
for the IoT stack on different layers for the massive Machine as sleeping or discontinuous reception (DRX). Turning off the
Type Communication (MTC) use cases. transmitter may need to be done as well and this is referred
When CoAP [6] is used in the transfer layer, UDP is the to as discontinuous transmission (DTX). When the radio is
default transport assumed. It should be noted though that asleep, the device will not be reachable. If sleep intervals are
in the IETF, TCP transport for CoAP is currently discussed short, in the order of tens of milliseconds, this may not be a
[7]. MQTT [8], relies on TCP on the transport layer, while problem. However, to bring down the power consumption the
a version for UDP called MQTT-SN has been discussed as sleep periods will need to be longer [10].
well. More broadly, Data Distribution Service for Real-Time Sleeping mechanisms are typically part of the Medium
Systems typically utilizes UDP, while with Hypertext Transfer Access Control (MAC) protocol. This functionality can be
Protocol and Advanced Message Queuing Protocol, TCP is realized in different ways depending on spectrum type used
used instead by default. To understand how different stacks and whether network control can be used.
may impact application level performance and system design, In unlicensed spectrum, configuring sleeping cycles is not
the differences between the two different transports need to be straight forward as the network cannot guarantee a transmis-
understood. sion opportunity for the time window where the device should
The key differences between TCP and UDP are that UDP be available. This is the case for example in carrier sense
is connectionless, light weight, and unreliable, whereas TCP multiple access (CSMA) used for channel access in Wi-Fi. In
2018 Global Internet of Things Summit (GIoTS)

realized for example with a 3GPP radio link such as LTE. In


this paper, we study two different options for the last mile
radio, namely BLE and IEEE 802.11ah aka low power Wi-Fi.

A. Bluetooth Low Energy


BLE, also known as Bluetooth Smart [12], is a low power
radio technology developed for the IoT. The technology is
based on the Bluetooth technology, but has been further
developed to allow for operation on devices powered only
by coin cell batteries or even through energy harvesting.
The protocols of BLE have been specifically designed with
low energy consumption in mind and allow for very low
duty cycles. BLE operates in the 2.4 GHz ISM band and
utilizes an adaptive frequency hopping scheme to co-exist with
Fig. 2. TCP three-way handshake for uplink sensor report. other technologies, for example, Wi-Fi. The BLE standard is
developed by the Bluetooth special interest group and BLE is
part of the Bluetooth specification versions 4.0 and higher.
frequency hopping systems, such as BLE, interference may BLE supports both connection oriented data transfers,
happen just when the device is awake and thus a similar as well as a connection-less mode. For connection-oriented
problem as with CSMA exists. Channel access may also be mode, a connection is first established and each subsequent
limited with duty cycles and depending on time window over over-the-air transmission is explicitly acknowledged. For the
which duty cycles are calculated, communication in either connection-less mode, transmissions are performed in a broad-
direction may be constrained. casting manner.
Massive IoT use cases in IoT are often expected to be uplink Sleeping in BLE may happen both when the device is
oriented and carrying just small amounts of data. However, if connected and when it is not connected. In the simplest
TCP is used, the communication will flow in both directions case, all time outside of a connection is sleeping. Within a
and furthermore, the TCP protocol puts time constraints on connection the device may sleep between connection events.
how fast the next packet needs to be delivered for congestion The interval between connection events may vary from a few
control features to kick in. milliseconds to a few seconds. A connection is closed when
there is no more data to exchange.
C. TCP over sleepy MAC
The TCP three-way handshake begins with sending a SYN B. IEEE 802.11ah aka low power Wi-Fi
message to open a connection. This is a followed with a SYN- IEEE 802.11ah amendment [13] is intended to operate in
ACK from the other end, after which the second SYN is sub 1 GHz bands to provide longer transmission range com-
acknowledged and data may be sent over the newly established pared to legacy Wi-Fi and support large number of associated
connection. Analyzing the three-way handshake for sensor stations (STAs) with a single Access Point (AP). 802.11ah
type uplink oriented traffic shows the points in time when uses Carrier-Sense Multiple Access with Collision Avoidance
the device should be awake as depicted in Figure 2. (CSMA/CA) like other members of the 802.11 family with
The communication is initiated by the device. We assume no improved energy saving mechanisms. 802.11ah introduces
DTX mechanism is used and thus the device may initiate data Traffic Indication Map (TIM) segmentation, extension of the
transmission immediately when the packet is available and sleep cycle up to 5 years, a new class of sensor type STAs,
channel access scheme used in the system allows transmission. target wake time, and support of shorter operating channel
The problematic points in the communication become the bandwidths of 1 MHz and 2 MHz. Access point may divide
instants in time, when DL messages arrive. In case a DRX large number of STAs to different TIM groups, thereby effi-
mechanism is in place in the MAC layer, the packets will be ciently indicating the availability of downlink buffered traffic
queued in the radio node on the network side until device is to a specific group at a time. A DTX mechanism is specified
available for DL reception. for 802.11ah in the form of the restricted access windows
(RAW). This mechanism will impact the latency negatively
IV. C APILLARY R ADIO ACCESS for any user that is blocked from transmitting in the system
Capillary networks are IoT networks where the last mile when the load is low. When the load is high, on the other
access is realized typically with an unlicensed low power radio hand, the latency may benefit from RAW. RAW is not part of
technology [11]. Typical use cases addressed with capillary the current evaluation as we focus on low load.
networks fall into the massive IoT category. Other technical
solutions for massive IoT are low power wide area IoT V. S IMULATOR M ODEL
technologies such as NB-IoT and LoRa. A capillary gateway is The evaluations discussed in the following were performed
used as a connection point towards the Internet with backhaul using an Ericsson internal event based radio network system
2018 Global Internet of Things Summit (GIoTS)

simulator with detailed models of application, IP, transport, TBTT interval for beacon transmission is modelled as 100
network and link layers for both BLE and Wi-Fi. This in- ms in the current simulation. The channel bandwidth for
ternal simulator is similar in structure and capabilities to the the Wi-Fi transmissions is assumed to be 2 MHz with the
commonly used NS3 simulator. carrier frequency of 864 MHz. The antenna is assumed to be
Sensor reports were randomly generated by means of a omnidirectional.
Poisson process. For TCP, the session is terminated after each
sensor report was received in the other end of the connection, C. Simulation scenario
however, the termination procedure of a TCP connection has We simulate a scenario in which wireless sensors are
not been explicitly modelled. A simplified physical layer connected to the central device or gateway in a star topology.
was used for both technologies, with a detailed model of We assume that all the devices are within communication
the MAC layer, as well as the IP, TCP, and UDP layers. range from the central device.
The CoAP and MQTT layers were simplified with a simple The sensors communicate with a cloud server in the Internet
payload size model on top of the transport. More details on and we assume Internet delays to be 40 ms one way. This value
the technology specific models are discussed in sub-sections. is chosen to reflect a server that is deployed reasonably close
IP header compression was not modelled in the simulation. to the sensor.
Varying the payload is used instead. In the scenario devices generate sensor reports with packet
Buffer limitations of the devices have also not been mod- sizes of 25, 150 and 400 bytes as transport layer payload.
elled. The devices are also assumed to be in close vicinity of These values are chosen to show the impact of increasing
one another so that no, or at least very few packets, should be payload and to show that, e.g., header compression would
lost over the air, except for those lost due to collisions. have very little impact. We assume the transfer layer protocol
headers for CoAP and MQTT respectively to be included in
A. BLE modeling this payload as well any needed object representation overhead
The simulator contains detailed models of all parts relevant from, e.g., Light Weight Machine to Machine and security. The
to the evaluation according to revision 4.2 of the Bluetooth average report intensity is 5 packets per second. This means
specification. that in the network 5 packets are generated in total in a second.
In these evaluations the connection-oriented mode is utilized
as two-way communication is necessary, at least for the TCP VI. P ERFORMANCE E VALUATION
parts. A new BLE connection will be established for each new Simulations of CoAP with UDP and MQTT with TCP trans-
data transfer. This means that for a typical transaction, the fer layer over both BLE and IEEE 802.11ah were performed to
sensor will wake up upon generation of data to transmit, try study the impact of the transport layer protocol over different
to establish a new connection to a central device and transmit MAC layers. The main aspects studied are the extra delay
the data within the connection. and the extra power consumption caused by TCP usage when
Connection events are configured to happen every 7.5 ms compared to using UDP. We focus on simple sensor report
thus giving the sleeping cycle for a BLE connection. The traffic in the uplink direction. When using CoAP, this type
connection is initiated by the sensor who enters the connection of traffic can be sent non-confirmed, while with the MQTT
as slave. Each connection event is initiated by the master, stack TCP will provide reliability. We simulate also the CoAP
which in this case is the central device. confirmed case to show it has very little if no impact. By
utilizing two different link layers, we also gain understanding
B. IEEE 802.11ah modeling on how different sleeping mechanisms on the MAC layer
The simulator contains detailed models of all parts relevant impact the performance.
to the evaluation according to draft 8.0 of the IEEE 802.11ah
specification. A. Packet delay evaluation
Stations (STAs) are assumed to be in power save mode, The packet delays on the application layer from the time
and may wake up whenever there is uplink data to transmit. instant of packet creation and packet reception from the TCP
In power saving mode, DL data transmission from the Access and UDP layers in the receiver end are shown in Figure 3 for
Point (AP) to STA works such that the AP generates beacon both BLE and 802.11ah in cumulative distribution function
frames at periodic times given by the Target Beacon Transmis- format (CDF).
sion Time (TBTT). A Traffic Indication Map (TIM) frame in As can be seen in the BLE results, the difference between
the beacon is used to indicate the availability of the downlink CoAP (UDP) and MQTT (TCP) over the BLE connection ba-
data. If the beacon indicates availability of downlink data for sically corresponds to the three-way handshake that increases
a particular STA, then this STA transmits a Power-Save poll the number of messages that need to be exchanged. With the
(PS-poll) to the AP to inform the AP it is available in DL Internet delay of 40 ms one-way, the difference to UDP is at
to retrieve the data. In return, the AP transmits the DL data least 80 ms. The connection event interval used is 7.5 ms,
to the corresponding STA. Also, as STAs are in power-save adding an average delay of 2.75 ms to each DL oriented
mode, a STA may go to sleep, if the TIM element within the message. The spread in the delays for the packets that do not
beacon frame does not indicate that DL data is available. make it across at the first attempt depends on the BLE central
2018 Global Internet of Things Summit (GIoTS)

Packet delays for sensor reports over BLE, 5 reports per second Packet delays for sensor over 802.11AH, 5 reports per second
100 100

90 90

80 80

70 70

60 60
CDF [%]

CDF [%]
50 50
25 B, MQTT 25 B, MQTT
40 25 B, CoAP 40 25 B, CoAP
25 B, CoAP + ACK 25 B, CoAP + ACK
30 150 B, MQTT 30 150 B, MQTT
150 B, CoAP 150 B, CoAP
20 150 B, CoAP + ACK 20 150 B, CoAP + ACK
400 B, MQTT 400 B, MQTT
10 400 B, CoAP 10 400 B, CoAP
400 B, CoAP + ACK 400 B, CoAP + ACK
0 0
10-2 10-1 10-1
Packet delay [s] Packet delay [s]
(a) (b)
Fig. 3. Packet delays for BLE and 802.11ah for varying packet sizes and a fixed arrival rate of 2 packets per second.

device being busy serving another user. In the TCP case, this device. However, the sensor devices becomes unavailable to
delay may in turn cause TCP timeouts introducing further the network, whenever a connection is not open. Thus, this
delays. It should be noted that the BLE central device that model does not work well when the packet route has long
is modelled only supports one concurrent connection, which latency on the end to end path or when actuation is performed,
is significantly less than for a typical device. These results unless the connection would be opened periodically to poll for
may thus be somewhat pessimistic. DL data in the other end of the radio connection. In case such
The Wi-Fi results look rather different. Due to the 100 periodic polling would be performed, a similar degradation in
ms TIM interval, the first DL oriented transmission will take power consumption should be expected as is seen in the case
on average 50 ms longer than an uplink transmission with a of 802.11ah as the polling would introduce a function similar
uniform distribution over the 100 ms TIM period. The delay to the beacon reception in Wi-Fi.
spread from 150 ms to 250 ms can be clearly seen in the Adding header compression to the evaluation would have a
graph. very small impact on the performance due to reduced number
of bits transmitted over the air, as can be seen from the
B. Power consumption performance
difference between the different payloads.
Power consumption is affected by the choice of transfer and
thus transport layer protocol. The results are consistent and VII. C ONCLUSIONS AND FUTURE WORK
reflect the time needed to perform the actual data transfer,
which in case of MQTT (TCP) requires more messaging over In this paper we studied the performance of sensor reporting
the air (Figure 4). over CoAP and MQTT with UDP and TCP transport protocols
The larger difference between ah and BLE comes from the respectively running on top of two different radio interfaces,
beacon model used in Wi-Fi. There is no periodic wake up to namely BLE and IEEE 802.11ah. Unlike UDP, TCP includes
check for DL data in BLE, whereas in Wi-Fi this is inbuilt a handshake procedure to establish a session. This leads to
to the MAC protocol. The ah results are thus affected by the additional exchange of packets and therefore affects the delay
periodic wake up to listen the beacon every 100 ms. For a of packets and battery lifetime of IoT devices. Furthermore,
fairer comparison in the UDP case, the sensor devices could TCP acknowledgements used to ensure reliability need to be
be configured to non-TIM devices meaning that they would later received in the transmitter to avoid retransmissions.
not need to wake up to listen to the beacon. This would yield Delays are comparable for UDP between BLE and
a more similar power consumption to the BLE case. For TCP 802.11ah. They are dictated mainly by the time it takes to
use, however, TIM operation is required, as the device would transfer data over the air. When TCP is used, however, the
otherwise not be capable of two-way communication. TIM model in 802.11ah brings a built in minimum delay cor-
responding to the TIM configuration to the communication. In
C. Discussion these results, the delay spread further reflects the chosen TIM
In the evaluated scenario, BLE and 802.11ah have compa- interval of 100 ms, a typical interval configured for beacons in
rable performance. However, all disadvantages are not clear Wi-Fi networks today. However, to optimize for device battery
from the simulation results. When the end device acts as a lifetime in IoT deployments, longer TIM intervals may be
master in the BLE connection, we optimize sleep for the sensor configured.
2018 Global Internet of Things Summit (GIoTS)

Sensor battery lifetime [years], BLE Sensor battery lifetime [years], AH


25 1.2
1 minute report interval 1 minute report interval
10 minutes report interval 10 minutes report interval
1
20

0.8
15

0.6

10
0.4

5
0.2

0 0
P

K
TT

TT

P
K

TT

TT

TT

TT
oA

AC

AC

oA

AC

AC
oA

oA

oA

oA
AC

AC
Q

Q
Q

Q
C

C
C

C
M

M
M

M
+

+
+

+
B,

B,
B,

B,

B,

B,
B,

B,
B,

B,

B,

B,
P

P
P

P
oA

oA

oA

oA
oA

oA
25

25
0

0
25

25
0

0
15

40

15

40
15

40

15

40
C

C
C

C
B,

B,

B,

B,
B,

B,
0

0
25

25
15

40

15

40
(a) (b)
Fig. 4. Battery lifetime for BLE and 802.11ah for packets being generated with an inter-arrival time of 1 and 10 minutes for different packet sizes and
transport protocols. Please note different y-axis scales showing battery lifetime in years.

Regarding the battery lifetime, TCP uses more energy due [10] T. Tirronen, A. Larmo, J. Sachs, B. Lindoff, and N. Wiberg, “Machine-
to more packets sent, than UDP. The proportion between TCP to-machine communication with long-term evolution with reduced de-
vice energy consumption,” Transactions on Emerging Telecommunica-
and UDP remains for BLE and 802.11ah, but BLE device’s tions Technologies, vol. 24, no. 4, pp. 413–426, 2013.
battery lifetime is 17 times longer compared to 802.11ah. This [11] J. Sachs, N. Beijar, P. Elmdahl, J. Melen, F. Militano, and P. Salmela,
is due to the recurring wake-up to listen to the Wi-Fi beacon “Capillary networks–a smart way to get things connected,” Ericsson
Review, September, vol. 9, 2014.
every 100 ms even when there is no traffic to be received. [12] “Bluetooth Core Specification 4.2,” Bluetooth SIG, Standard, 2014.
The study confirms, the previous results where CoAP out- [13] “P802.11ah, Draft 10.0, Wireless LAN Medium Access Control (MAC)
performs TCP in latency. The study also shows differences in and Physical Layer (PHY) Specifications,” IEEE, Draft, 2016.
the different radios. While Bluetooth air interface is faster in
general, the Wi-Fi air interface provides lower latency when
the network is loaded.
R EFERENCES
[1] A. Al-Fuqaha, A. Khreishah, M. Guizani, A. Rayes, and M. Moham-
madi, “Toward better horizontal integration among IoT services,” IEEE
Communications Magazine, vol. 53, no. 9, pp. 72–79, 2015.
[2] Y. Chen and T. Kunz, “Performance evaluation of IoT protocols under
a constrained wireless access network,” in Selected Topics in Mobile
& Wireless Networking (MoWNeT), 2016 International Conference on.
IEEE, 2016, pp. 1–7.
[3] M. Collina, M. Bartolucci, A. Vanelli-Coralli, and G. E. Corazza,
“Internet of Things application layer protocol analysis over error and
delay prone links,” in Advanced Satellite Multimedia Systems Conference
and the 13th Signal Processing for Space Communications Workshop
(ASMS/SPSC), 2014 7th. IEEE, 2014, pp. 398–404.
[4] P. D. Marco, R. Chirikov, P. Amin, and F. Militano, “Coverage analysis
of Bluetooth Low Energy and IEEE 802.11ah for office scenario,” in
Personal, Indoor, and Mobile Radio Communications (PIMRC), 2015
IEEE 26th Annual International Symposium on, Aug 2015, pp. 2283–
2287.
[5] F. D. Carpio, P. D. Marco, P. Skillermark, R. Chirikov, K. Lagergren, and
P. Amin, “Comparison of 802.11ah and BLE for a home automation use
case,” in Personal, Indoor, and Mobile Radio Communications (PIMRC),
2016 IEEE 27th Annual International Symposium on, Sep 2016.
[6] Z. Shelby, K. Hartke, and C. Bormann, “The constraint application
protocol (CoAP),” Internet Requests for Comments, RFC 7252, 2014.
[7] C. Bormann, S. Lemay, H. Tschofenig, K. Hartke, B. Silverajan, and
B. Raymor, “CoAP (Constrained Application Protocol) over TCP, TLS,
and WebSockets,” IETF draft, December 2017.
[8] “MQTT, version 3.1.1,” OASIS Standard, October 2014.
[9] Y. Cheng, J. Chu, S. Radhakrishnan, and A. Jain, “TCP fast open,”
Internet Requests for Comments, RFC 7413, 2014.

You might also like