You are on page 1of 9

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCE.2019.2962605, IEEE
Transactions on Consumer Electronics
1

‘Smart’ Is Not Free: Energy Consumption of


Consumer Home Automation Systems
Chrispin Gray, Member, IEEE, Robert Ayre, Kerry Hinton, and Leith Campbell, Member, IEEE,

Abstract—The proliferation of consumer Home Automation The system aims to automate consumer device control while
Systems (HAS), which is increasingly based on the Internet providing timely energy usage data to consumers. Han et al.
of Things (IoT) architecture, comes with added costs for the [3] proposed a ZigBee and PLC-based SHEMS. Their system
additional electrical energy required to power the automation
interfaces and the standby energy consumption required to simultaneously monitors energy consumption and renewable
maintain connectivity and/or “smartness”. In this paper, we energy generation to automatically schedule device usage,
use a bottom-up approach to develop novel system-level energy using a home server, with the goal of cost minimisation. Chen
consumption models for consumer HAS devices and quantify and Lin [4] employed smart plugs to monitor appliance (non-
the energy consumption for a typical HAS. We then assess the smart) load characteristics and applied analytics to determine
potential impact on global Information and Communications
Technology (ICT) energy use. We show that, on average, HAS power-down opportunities to minimise energy usage. A study
may consume over one-third of the annual energy used in a mid- by Friedli et. al. [5] for the International Energy Agency
sized home, with non-trivial impact on the global ICT energy (IEA) estimated the global energy consumption of selected
footprint. IoT use-case applications using a combination of measurement
Index Terms—Energy consumption; power models; smart and publicly available manufacturer datasheets. Mehdi and
home; home automation; Internet of Things (IoT). Roshchin [6] studied different home electric appliances based
on their load characteristics. By applying task-scheduling
techniques, Mehdi and Roshchin presented models that are
I. I NTRODUCTION
optimised to minimise energy consumption. Langhammer and
Home Automation System (HAS), which consists of
A sensors and actuators and the network that connects
them, basically enables a “Smart Home”, which will also in-
Kays [7] presented energy models for HAS networks and
proposed a performance evaluation metric for a host of indoor,
HAS-enabling wireless network technologies, including BLE,
clude consumer “smart” devices like TV sets and refrigerators. KNX-RF and IEEE 802.15.4 variants. In both [6] and [7],
Here, we focus just on the HAS itself. A recent report however, critical enabling devices like the gateways are not
estimates that home automation would make up nearly 50% considered. The authors of this paper have further studied and
of all Internet of Things (IoT) devices by 2020 [1]. While compared the power consumption design characteristics of five
the deployment of billions of devices through IoT services commonly employed wireless technology options for HAS,
could offer many benefits, it comes with the need for additional including BLE, ZigBee and Wi-Fi as detailed in [8].
electrical energy. In this paper, we estimate the energy consumption of
Research work in the area of consumer HAS energy consumer HAS from the household to global level, using a
consumption has, for the most part, been focused on the bottom-up approach. We use a combination of direct measure-
monitoring, design and development of Smart Home Energy ment and modelling to estimate energy consumption of each
Management Systems (SHEMS) using a number of wire- component of a typical modern consumer HAS in the context
less network technology options (e.g. Bluetooth Low Energy of the IoT. We then estimate the global energy impact of
(BLE), ZigBee, Wi-Fi), while few have focused on developing these systems if current IoT device installation projections [1],
and optimising energy consumption models for HAS. Han and [5] come to pass. Modern consumer HAS services generally
Lim [2] proposed a context-aware SHEMS, using a ZigBee include novel, network-enabled IoT devices, a gateway device
sensor network, with a new on-demand routing protocol. and an attached cloud service. Hence, traditional network-
enabled devices like Smart TVs and home entertainment
Manuscripts received August 2, 2019. This work was supported by Alcatel-
Lucent Bell Labs, The State Government of Victoria, Australia, and The systems are not included. For smart consumer appliances,
University of Melbourne. only the added IoT module (e.g. sensor and communication
Chrispin Gray is with Department of Electrical and Electronic Engineer- modules) is considered.
ing, University of Melbourne, Parkville, Victoria 3010, Australia (e-mail:
chrispin.gray@unimelb.edu.au) The paper is organised as follows. The network architecture
Robert Ayre is with the Department of Electrical and Electronic Engi- of a consumer HAS is described in Section II and the device
neering, University of Melbourne, Parkville, Victoria 3010, Australia (e-mail: measurements described in Section III. Simplified energy mod-
rayre@unimelb.edu.au)
Kerry Hinton is with the Department of Electrical and Electronic Engi- els are detailed in Section IV, with network traffic measure-
neering, University of Melbourne, Parkville, Victoria 3010, Australia (e-mail: ments from an example HAS application given in Section V.
k.hinton@unimelb.edu.au) The energy consumption estimates for an average household
Leith Campbell is with the Department of Electrical and Electronic Engi-
neering, University of Melbourne, Parkville, Victoria 3010, Australia (e-mail: fitted with a fully-equipped HAS are given in Section VI.
leith.campbell@unimelb.edu.au) Published deployment scenarios are used to estimate the

0098-3063 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCE.2019.2962605, IEEE
Transactions on Consumer Electronics
2

To
IoT Gateway
Power
(Mains / Battery)

Home

(Communication Module)
Memory
(e.g. SRAM)
UART
Smart Hub
Microcontroller

Radio
SPI, I2C
(MCU)

ADC DAC
xDSL, PON, Internet

Sensor or Actuator
HFC
IoT Gateway

Unit (SAU)
Home Gateway
Sensor Actuator

Data Centres
IoT Device
IoT Devices
PC, Laptop, Smartphone Acquisition Actuation
Wi-Fi
(External Physical Quantity)

Fig. 1. Simple diagram of a home automation system supported by a cloud


service that is hosted at a data centre. Fig. 2. Schematic diagram of a generic IoT device.

energy consumption of these systems in the top two markets in 2) Actuator Device: Conversely, an actuator device initially
Section VII, and for a global scale in Section VIII. Section IX receives a signal via its radio interface. This is interpreted by
discusses possible options for mitigating the significant energy the MCU, which in turn issues a command to the physical
consumption increment. actuator element.
Each of these processes of sensing, processing, and com-
munication consumes some energy, often operating many
II. N ETWORK A RCHITECTURE OF C ONSUMER H OME times per day. Some devices operate autonomously, so that,
AUTOMATION S YSTEMS for example, the sense-process-communicate sequence repeats
periodically. These are referred to as time-based devices.
Figure 1 shows a simplified network architecture of a Others, labelled event-driven, operate when triggered by an
consumer HAS linked to a cloud computing service. external event (e.g. a button press, door opening). In the
The HAS architecture includes: end-devices (IoT devices) at intervals between operations, the device continues to consume
the customer premises; an IoT gateway (IGW), which essen- a (smaller) idle power. Further, the HAS may employ protocols
tially is an aggregator for IoT devices and provides application that impose additional requirements (e.g. connectivity moni-
management and control capabilities; a home gateway/modem toring, “keep alive”) which in turn add to energy consumption.
(HGW) for access to the Internet, and data centres hosting
cloud services. In operation, the IGW and IoT devices form a B. IoT Device Network
local network referred to as an IoT Device Network (IDN). The IDN refers to a network of IoT devices within a
In a “typical” consumer HAS installation, the IGW connects consumer HAS with a common gateway (i.e. the IGW). The
to the HGW either via Ethernet or Wi-Fi access [9]. A recent smart hub is an extension of the IGW for voice commands.
popular addition to the consumer home automation eco-system In its most common form, communication between the IoT
is the voice-activated Smart Hub. devices and IGW is via a wireless link. A range of short-
range wireless network protocols is used in HAS, including
Bluetooth (Classic and Smart/Low-Energy), ZigBee, Wi-Fi
A. Defining IoT Devices and RF 433 MHz [8]. While many of these protocols are
designed for point-to-point communication, employing a star
The range of IoT devices in a HAS can be broadly
topology, a few can operate in a mesh topology (e.g. BLE,
categorised as “sensors” or “actuators”, the latter including
ZigBee) forming a daisy chain of links towards the gateway.
controlled devices. In either case, the devices include a number
A basic point-to-point single-hop, star-topology connection
of functional elements, as shown in Fig. 2. These elements
between the IoT devices and the IGW is modelled here.
include a sensor or an actuator unit (SAU), a microcontroller
unit (MCU), and a radio interface module. Although shown as
III. E NERGY M EASUREMENT OF I OT D EVICES
discrete elements, it is more common today for these functions
(apart from the SAU transducers) to be integrated as a System- We have conducted measurements on a range of IoT devices
on-a-Chip. A brief description of the workflow processes of and report examples here. These will be used in Section IV to
sensor/actuator devices is given below. create more generic power consumption models for the HAS.
1) Sensor Device: In a sensor device, some physical param-
eter is initially measured by an appropriate transducer, and A. Measurement Method and Setup
the result passed to the MCU. The latter performs analog- For a number of IoT devices proposed for home use, we
digital conversion, may undertake some simple processing, and have studied their characteristics, including the device’s bit
prepares the measurement data for transmission by the radio rate, transmit sequence, protocol stack, energy consumption
module to the IoT Gateway. states (e.g. active, idle, standby) and trigger mechanisms.

0098-3063 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCE.2019.2962605, IEEE
Transactions on Consumer Electronics
3

0 60 120 180 240


80 80
100

Power Consumption (mW)


Power Consumption (mW)
70
60 Tx 100

Power Consumption (mW)


Power Consumption (mW)
70 50 70 90 90 tTx
40 80

60 30 PTx 60 80 70
20 PMCU 60
50
70
50
10
0
MCU 50
40
30
tMCU PTx
0 100 200 300 400 500 600 700 60 20
Time (ms) 10
40 40 50 0
126.2 126.5 126.8 127.1 127.4 127.7

Power Consumption (mW)


3
Sensing 40 PMCU
30 2
30 Time (sec)

PSEN 30
20 1 20
0 20
10 0 20 40
Time (ms)
60 80 100 120 10
10
0 0 0
0 60 120 180 240 50 100 150 200 250 300
Time (sec) Time (sec)
(a) (b)

300 300

Power Consumption (mW)


Power Consumption (mW)

300

Power Consumption (mW)


300
Event
Power Consumption (mW)

250 Transmit 250 Tx


Event
200
250 150 250 200

100 150
50
100
0
200 49.2 49.4 49.6 49.8
Time (sec)
50 50.2 200 50

0
5200 5400 5600 5800 6000 6200

150 150 PMCU Time (ms)

50 51 52 53 54 55
1.0 1
Power Consumption (mW)

0.8 Lockout Time 0.9


0.8
0.7
100 0.6 0.6
100
PTx
0.5
0.4 0.4
0.3
0.2 0.2
0.1
0.0
50
0
50 51 52 53
Time (sec)
54 55 50 Sensing

0 0
35 40 45 50 55 60 0 2 4 6 8 10 12
Time (sec) Time (sec)
(c) (d)
Fig. 3. Power consumption measurement of 433 MHz sensor devices. (a) Type I temperature and humidity (T&H) sensor device, acquiring data samples
(small spikes) every 7−8 s and transmitting (large spike) said data once per minute; (b) Type II T&H sensor device transmitting collected data once per
minute. (c) Passive Infrared (PIR) sensor device triggered at 5 s intervals; (d) a reed switch-based door/window sensor device triggered at 3 s intervals.
Power consumption measurements have been made in each is limited to 1 second sampling rate. A representative off-
phase of the device operation cycle and averaged over at the-shelf consumer HAS is used as our testbed (referred to
least 10 iterations. We used custom-built USB 3V, 5V and as Test-HAS). The Test-HAS consists of wireless sensors,
TABLE I actuators and cameras (i.e. IoT devices), a gateway unit (i.e.
M EASUREMENT OF A T IME -BASED T&H S ENSOR OVER 1 C YCLE (1 IoT gateway) and a cloud service hosted at a data centre,
MINUTE )
through which control and management of all connected IoT
devices and their data are achieved.
Instance Duration Power Energy
Phase Wireless communication between end-devices and the Test-
(N ) (ms) (mW) (mJ)
Sensing (SEN) 16 82 0.87 1.1 HAS gateway is via the ISM band 433 MHz connectionless
Transmit (Tx) 3 72 21.95 4.6 RF protocol (cf. Fig. 1).
Processing (MCU) 1 500 11.20 5.6
Standby (SBY) - 60 s 0.06 3.6
B. Sensor Devices
TABLE II
M EASUREMENT OF E VENT-D RIVEN S ENSOR D EVICES
Figure 3 shows power consumption traces for 433 MHz
time-based and event-driven sensor devices. Shown in Fig. 3a
PIR Sensor Door/Window Sensor is a Type I temperature and humidity sensor (T&H) collecting
Duration Power Duration Power samples at regular intervals (≈7−8 s) followed by 3 transmit
Phase (ms) (mW) (ms) (mW)
Sensing (SEN) - n/a - 0.07 bursts (extra 2 for redundancy) of a 5-Byte data packet within
Transmit (Tx) 870 146.9 630 86.2 a 60 s cycle. The average power and duration of each phase
Processing (MCU) 3650* 0.5 850 9.8 is given in Table I. A similar T&H device trace (Type II)
* Lockout time set to 5 s, which includes the duration for data Tx. in Fig. 3b shows non-distinct data sampling spikes but with
9V DC power meters, each with sampling rate of 5 ms (re- one continuous transmit burst, also within a 60 s cycle.
programmed to 1 ms for some tests) and accuracy of 10 or The repetitive behaviour of time-based sensors makes for a
100 µW. For AC mains power measurements, an AC power simple energy consumption estimate. From Table I, the energy
meter was employed, which provides a 10 mW accuracy but consumption of Type I is ≈15 mJ per cycle or ≈22 J per day.

0098-3063 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCE.2019.2962605, IEEE
Transactions on Consumer Electronics
4

PDEV Standby (SBY) PDEV Standby (SBY)


Sensing (SEN)
PDEV Actuate (ACT)
Event Processing (MCU)
Processing (MCU)
Communications (COM) 1 Cycle (TC) NR Standby (SBY) Communications (COM)
tCOM Sensing (SEN)
tCOM NR Listening Event Listening
Processing (MCU)
Trigger Communications (COM) tACT
PCOM Command
tSEN PCOM PMCU PMCU PACT
tSEN tCOM
NSEN
PMCU PSEN
PSBY PSBY PSBY
0 0 0
tMCU PSEN tMCU t PCOM tMCU t
t
(a) (b) (c)
Fig. 4. Power consumption time evolution model as a generalization of IoT device power measurements. In the figure, tSEN and tACT are the duration of the
sensing and actuating phases, tMCU is that for analyzing or interpreting sample data/commands, and tCOM , the time for Tx & Rx with NR retransmissions; (a)
A time-based sensor device acquires and transmits NSEN samples; (b) An event-driven sensor device reacts to an externally observed phenomenon (trigger);
(c) An actuator device responding to a received command via its radio in “listening mode” with the receiver always-on.
Traces for two event-driven sensor devices, a Passive In- The total energy consumption of the IoT devices in a HAS,
frared (PIR) and a Door/Window (DW) sensor device, are EIoT , is the sum of the energy consumed by each individual
given in Fig. 3c and Fig. 3d, respectively, with their measured IoT device i, such that:
values reported in Table II. Both devices transmit multiple N
bursts of a predefined 4-Byte data block when triggered. The
X
EIoT = EDEVi (2)
PIR prevents excessive re-triggering for the same event by i=1
employing a selectable lockout timer. Fig. 3c shows the power
for i = 1, 2,..., N , where N = number of devices in IDN.
trace for a 5 s lockout time. With the sensor of the PIR being
passive, negligible power dissipation was observed during the
sensing phase. We can then calculate the energy consumption B. IoT Device Energy Modelling
of the PIR and DW as 130 mJ and 63 mJ per event or 12 J Typically, a device operation includes a sequence of tasks
and 13 J for 100 events per day, respectively. Similar mains- that consume similar amounts of energy at each instance [10],
powered sensor devices were reported in [5] to consume about [11]. The factors that influence its energy consumption include
0.6 W on average. the volume of data transmitted, the type of wireless technology
and protocol used, the volume of data processing involved
C. Actuator Devices and any redundancy mechanisms [10]–[12]. To model an IoT
Four 230V controlled mains-power sockets were used as device, considering Fig. 2, the total power consumed PDEV (t)
examples of an actuator device. The controlled socket contains at time t can be characterised by the power consumed by its
a 433 MHz RF wireless receiver that receives commands from individual elements and their processes:
the gateway to open or close the power circuit. We recorded its
PDEV (t) = PSAU (t) + PMCU (t) + PCOM (t) + PSBY (3)
no-load average power consumption as 0.70 W in its “Socket
OFF” state and 0.67 W in the “Socket ON” state. On average, such that PSBY is the power dissipated when the device
similar mains-powered actuator devices have been reported to is in a standby or inactive state, PSAU (t), PMCU (t) and
consume about 1 W [5]. PCOM (t) represent the additional power consumed by the
sensor/actuator unit, the micro-controller unit and its interfaces
D. IP Camera in handling sample data or commands, and the radio for data
In the Test-HAS, an example IP camera (IPcam) was used communication with the gateway, respectively.
as an IoT device for video surveillance, configured to provide
a high definition image of 1280x720 pixels at 25 frames/s. The C. Modelling Sensor Device Energy
IPcam connects via Wi-Fi to the HGW. It consumed 3.75 W 1) Time-based Sensor Device: Figure 4 depicts the time
when streaming captured frames with a bit rate of 1.9 Mb/s. evolution of the power consumption of sensor and actuator
devices, as a generalisation of the examples in Fig. 3. For
IV. E NERGY M ODELLING the time-based devices, depicted in Fig. 4a, in addition to the
A. Home Automation System Energy Model three main tasks of sensing, processing, and communication,
The total Home Automation System energy consumption there are certain basic tasks running continuously, and lead
can be subdivided into 4 segments, modelled separately: (i) to the standby power consumption PSBY . For a given sensor
the energy consumed by the IoT devices, EIoT ; (ii) the IoT device, the energy consumed for a given task is the same in
gateway energy, EIGW ; (iii) the home gateway modem energy, each operational cycle. From (3), the energy consumption of
EHGW ; and, when available, (iv) the energy consumed by the a time-based sensor device, for 1 operational cycle in time Tc ,
Smart Hub, EHUB . The total energy consumption of a HAS is therefore given by:
over a diurnal cycle (time T = 24 hours) is: (m)
EDEVC = NSEN ESEN +NMCU EMCU +NCOM (NR ECOM )+ESBYC
EHAS = EIoT + EIGW + EHUB + EHGW (1) (4)

0098-3063 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCE.2019.2962605, IEEE
Transactions on Consumer Electronics
5

where NSEN is the number of samples collected, NMCU and be significant [5]. For these consumer appliances, only the
NCOM are the number of processing and transmission in- additional energy required to maintain the network-enabled
stances, NR is the number of message retransmissions, for communication module is considered. We used manufacturers’
NR > 1 and ESBYc = PSBY × Tc (see Fig. 4). The total energy datasheet values for communication modules, as well as re-
consumption can then be calculated for Nc cycles of operation ported measurements in the literature.
over time T (e.g. Nc = 1440 for T = 24 hrs).
2) Event-Driven Sensor: An event-driven sensor device is F. Modelling an IoT Gateway and Smart Hub
depicted in Fig. 4b, which similarly performs 3 main tasks
The measured power consumption profile of many network
but at irregular intervals in response to a detected external
elements (NE), like the IoT gateway, can be modelled by two
trigger. The sensing task (tSEN ) runs continuously, so that
components; a component linearly dependent on the device
PSEN and PSBY are indistinguishable. Alternatively, for some
workload, rising above a constant base power consumption
other event-driven sensor devices, the sensing task may be
(second component), as in [13]. However, the power con-
a continuous low-level communication process (i.e. a radio
sumption is in many cases almost constant with load. In [14],
receiver in listening mode) that receives a trigger event from
for example, a mini PC had a power variation of < 5%
its parent gateway. The sensing task can be active with low
(70 mW from 2 W idle power) for a load up to 90 Mb/s.
power draw or passive with negligible power draw.
Our measurement of a different mini PC showed similar
To model an event-driven sensor, consider a device having
results. Thus, we take the energy consumption of an IGW
NE events detected in a total operating time T . In general,
over time T (diurnal cycle) to be based on constant power:
the energy consumed over time T is as given in (5) but,
EIGW = Pidle × T . For a Smart Hub, we also consider only its
depending on the device, ESEN may be approximately equal
idle power, 3 W.
to ECOM . Assuming the events are repetitive with a similar
amount of data transmitted during each operation, the total
energy consumption of an event-driven sensor device over time G. Modelling a Home Gateway Modem
T is: The HGW is dedicated to a single customer’s Internet
(e)   service, with traffic from multiple end-user services [15]. To
EDEV ≈ ESEN + NE (T ) EMCU + NR ECOM (5) account for all services gaining network access, a shared
for NE > 0 and NR > 1, where ESEN = PSEN T . power model [15] is adopted. This model allocates to each
service a fraction of the no-load/idle power, which is in
proportion to the service’s contribution to the total traffic. This
D. Modelling Actuator Device Energy is in addition to the incremental power consumption for the
In general, the operation of an actuator contains a pe- service, which is a function of the service’s data rate. The
riod of communication, a period of processing followed by combined power consumption P (R) for a service with data
the excitation of the actuating unit. More advanced devices rate R has the form:
may also include status reporting, ACK and ”keep-alive”
!
Pidle Pmax − Pidle
functions. Fig. 4c shows a conceptual model for the power P (R) = + R (7)
ρRmax Rmax
consumption of an actuator device: such as the AC actuator
devices measured in Section III. A command received by the where Pidle and Pmax are the no-load and maximum load power
radio incurs a power draw PCOM during time tCOM and is consumption of the shared HGW, Rmax its maximum data rate
processed by the MCU with power PMCU over time tMCU , and ρ is the utilisation such that 0 < ρ 6 1.
which in turn triggers the actuator with power PACT for a As with the IGW, the power consumption of our example
period tACT . For NE events triggered during an operating time HGW (an ADSL2+ wireless modem) was found to be largely
T , each leading to a processing event and an actuation event, independent of traffic load, up to the maximum rate on the
NMCU = NACT = NE , where NACT is the number of times test line, 10 Mb/s. An incremental power consumption of only
the actuator unit is energised. We further assume that, in most ≈4% was observed; specifically, an increase of 300 mW from
cases, PCOM = PSBY or alternatively, PCOM  PSBY such that an idle power of 7 W. We therefore neglect the load-dependent
PCOM + PSBY ≈ PCOM . Thus, the total energy consumption of term in (7) and focus on the allocation of the idle power
an actuator device for a period T is given by: between IoT and other traffic.
(a)   In (7), the term ρRmax represents the load on the gateway
EDEV ≈ ECOM + NE (T ) EMCU + EACT (6)
(e.g. a 1 Gb/s router with a utilisation of 0.5 has a load =
where NE > 0 and ECOM = PCOM T . 500 Mb/s). It comprises the load due to background traffic,
Rbgd , and the load due to the IoT traffic, RIoT . For the IoT,
then, (7) can be simplified to:
E. Modelling Smart Consumer Appliances  
Pidle
Smart consumer appliances include intelligence and network P (RIoT ) ≈ RIoT (8)
connectivity (mostly wireless) to control, actuate, report or Rbgd + RIoT
manage their internal functions. These appliances are largely The background traffic varies by time of day. To estimate the
tethered to mains power, with their radio modules consuming background traffic of a single-household, we obtained diurnal
a constant standby/idle power; the energy consumed can traffic statistics for a tier-2 ISP with a predominantly retail

0098-3063 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCE.2019.2962605, IEEE
Transactions on Consumer Electronics
6

180%
ISP Traffic Profile
160% Load Level
Relative Load Level (μl)

140%
120%
100%
80%
60%
40%
20%
0%
0 2 4 6 8 10 12 14 16 18 20 22 24
Time of day (Hours)

Fig. 5. Average diurnal traffic profile of a tier-2 ISP network with step load
levels, which are percentages of the average load over a day.
TABLE III Fig. 6. Power consumption of HGW attributable to the NB traffic as a function
D IURNAL T RAFFIC L OAD L EVELS FOR A T IER -2 ISP. T HE H OURLY of its data rate for 6 different background traffic levels profiles.
AVERAGE DATA R ATES FOR THE HGW ARE C ALCULATED U SING ITS TABLE IV
E STIMATED AVERAGE DAILY T RAFFIC AND L OAD L EVELS (µl ). AVERAGE DATA R ATE AND P OWER C ONSUMPTION OF THE G ATEWAY.

Traffic Profile Image Data Rate (kb/s) Power


Load Level Duration Hourly Average Data Attached Devices
l Resolution DS US (Watts)
(µl ) (Tl ) Rate in Mb/s Rbgdl Sensors & Actuators - 2 5 2
1 40% 2h 0.49 IP Camera (HD) 1280 x 720 13 216 2
2 50% 4h 0.61 Webcam (SD) 640 x 480 17 126 3.4*
3 90% 4h 1.10 * Difference of 1.4 W due to the bus-powered USB webcam.
4 110% 8h 1.35
5 140% 4h 1.71 traffic generated by the Test-HAS gateway and its power
6 160% 2h 1.96
draw were observed and recorded, as shown in Table IV.
customer base, over 1 week in May 2017 [16]. This traffic The Test-HAS gateway consumes 1.2 W when powered via
over time, relative to the daily average, is used as a proxy a 5V USB 2.0 port or 2 W when the unit is mains-powered.
for the diurnal traffic variation of an “average” customer as The upstream (US) and downstream (DS) traffic were low
given in Fig. 5. The data is quantised into 2-hour time blocks when only sensors and actuators were connected. Unlike more
(dash lines), and into one of six relative traffic levels, as recent multi-function HASs with in-built video streaming, the
listed in Table III. These relative load levels are used as an Test-HAS is limited to an image streaming application, which
approximation of a single-household hourly background data transmits still images from the camera to its cloud servers at
rate (Rbgd ) relative to the household’s average daily data rate. 3 s intervals. The average DS and US data rates when the
From Australian Bureau of Statistics, the average data Test-HAS gateway is individually connected to two different
volume usage of all Australian fixed-line households was cameras are given in Table IV. We can thus deduce that the
168 GB per month (assuming 10% upload volume) for the last traffic generated by the cameras alone is about 20−33 times
quarter of 2017, and is doubling every 2 years [17]. Hence, we more than that of other devices combined.
assume a household monthly data usage of 400 GB for early
2019. For this usage volume, the calculated daily average data VI. E STIMATING THE E NERGY I MPACT OF C ONSUMER
rate is 1.22 Mb/s (for 13.2 GB daily usage and 30.4 days per HAS ON H OMES
calendar month). Based on this daily usage and the diurnal As an illustrative example of the additional energy cost
cycle, the hourly average data rate for background traffic is of a consumer HAS, we assume a simple consumer HAS
given in Table III. From (8), we can then calculate the hourly installation in a home. Consider a mid-size 3-bedroom double-
average power consumption of the HGW attributable to the storey suburban home fitted with a whole-home consumer
IoT, for a range of RIoT data rates, as plotted in Fig. 6. HAS as given in Table V. There may be 200 connectable
devices per person or more than 50 connected devices per
V. T RAFFIC M EASUREMENT – T HE T EST-HAS C ASE household by 2020 [18]. Thus, the system consists of a
S TUDY plethora of IoT devices, including smart light bulbs, smart ap-
The Test-HAS gateway consists of a mini PC main pro- pliances, controlled socket actuators, surveillance cameras,an
cessor and a microcomputer daughter-board as a unit. The IoT gateway, a smart hub and a HGW. For smart light bulbs,
gateway was set up with 3 T&H sensors, 2 PIR sensors, a the average number of lamps per household is 24 in EU [19],
wireless electronic button (like a door bell), and 3 controlled 36.6 in Australia [20] and 67.4 in the US [21]. We assume 40
socket actuators. (Cameras were added later.) Using the packet smart light bulbs for a mid-size home. A first-generation LED
sniffing application “Wireshark” and the power meter, the light bulb (Wi-Fi−enabled) may consume 2 W in standby.

0098-3063 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCE.2019.2962605, IEEE
Transactions on Consumer Electronics
7

TABLE V TABLE VI
E NERGY C ONSUMPTION OF D EVICES IN A M ULTI -F UNCTION HAS. E STIMATED N UMBER OF S MART H OMES UP TO 2025.

No. of Events Power Energy Smart Homes (millions)


IoT Device Region
Units /day (W) /day (kJ) 2015 2020 2025
T&H Sensor 6 1440 - 0.13 North America 12.7 46.2 92
Door Sensor 3 10 - 0.02 Europe (EU28+2) 5.3 44.9 146
Window Sensor 20 2 - 0.12
PIR/Motion Sensor 12 288 - 0.98 TABLE VII
Smoke Detector 2 - 1 172.8 D EPLOYMENT S CENARIOS OF C ONSUMER HAS IN S MART H OMES .
Door Lock Actuator 3 - 1 259.2
Controlled Socket Actuator 15 - 1 1296.0 Devices Large Medium Small
Garage Door Actuator 1 - 1 86.4 Sensors 43 25 10
Smart Light Bulb 40 - 2 6912.0 Actuators 19 10 4
Smart Appliance 16 - 1 1382.4 Smart Light Bulb 40 20 5
Surveillance Camera 4 - 3.75 1296.0 Smart Appliance 16 6 3
IoT Gateway 1 - 2 172.8 IP Camera 4 2 1
Smart Hub 1 - 3 259.2 IoT Gateway 1 1 1
Home Gateway Modem 1 - 7 268.1* Smart Hub 1 1 1
Home Gateway Modem 1 1 1
*This value represents a share of the daily energy consumption of the HGW
allocated to the HAS as a fraction of its average data rate, calculated using
(8). with smartphone apps or web portal as a user interface; thus,
the report does not include legacy systems. The report further
1% 2% 2% focuses on 2 major markets (North America (NA) and Europe
Energy Consumption (kWh)

800
700
Sensors
9% 17%
600 (EU28+2)) with forecasts up to 2020, given in Table VI. The
Actuators 500

9% Smart Light Bulbs 400 projections of the numbers of equipped households in [24] end
300
12% Smart Appliances 200 in 2020. We have extrapolated the forecast to 2025, but used
IP Camera 100
IoT Gateway 0 half of the quoted CAGR: the projected proportion of equipped
48%
Smart Hub households would have exceeded 100% at the quoted CAGR
Home Gateway
on the extended period. The modified 10-year forecast has an
(a) (b) effective CAGR of ≈22% and ≈39% for NA and EU28+2,
Fig. 7. Annual energy consumption of a mains-powered devices in a simple respectively.
HAS. (a) Percentage share by device type. (b) Energy consumption in kWh. To estimate the Smart Home energy impact, an estimate of
Consumer smart appliances are assumed to be designed with the number of devices per household is required. The above
802.11n Wi-Fi modules with an idle power of ≈1 W [22], Smart Home report indicated that between 12−17% of the
including power supply inefficiencies. 2015 installed Smart Home systems were multi-function sys-
The annual energy consumption in kWh is expressed as: tems, while the remainder were function-specific. To cater for
! this adoption pattern, we consider three deployment scenarios
365 × daily energy (kJ) (large, medium and small), as given in Table VII. The large-
Annual Energy (kWh) = (9)
3600 scale deployment is effectively a multi-function, whole-home
system similar to that in Table V. A small-scale deployment
From (9), we calculate the annual additional energy con- refers to point solutions, as described in [24], while a medium-
sumption of a HAS installed at a mid-size suburban home scale is a nearly mid-point between the other two. Using
to be about 1232−1460 kWh (depending on whether or not values from Table V and deployment estimates in Table VII,
the sensors are mains-powered). To put this in perspective, the energy consumption of large-, medium-, and small-scale
the current annual energy use of a similarly sized house in systems is calculated to be 1460 kWh, 756 kWh, and 289
Victoria, Australia is reported as approximately 4067 kWh kWh, respectively.
[23]. Thus a whole-home HAS would add approximately Assuming a distribution of 20% of homes with large-
30−36% to the annual household electricity consumption. scale systems, 40% with medium-scale systems and 40% with
The contribution from each device type is given in Fig. 7a small scale systems, together with Smart Home forecasts in
and Fig. 7b. Smart light bulbs consume nearly 50% of the Table VI, the estimated total energy consumption for Smart
total additional energy, with sensors being the next major Homes in NA and EU28+2 is given in Fig. 8. The figure
contributor at nearly a fifth of the total. foreshadows an increase from about 9 TWh in 2015 to 65 TWh
in 2025 for NA, and from 4 TWh to 104 TWh in EU28+2 over
VII. E STIMATING THE E NERGY C ONSUMPTION OF S MART the decade, assuming the installation rates are maintained.
H OMES IN S ELECTED M AJOR M ARKETS To put this in perspective, about 9 large, ∼2 GW power
In this section, we broaden the model to include a spread of plants will be required to support Smart Homes, in just the
smart home installation types. We estimate the future energy two regions considered.
consumption of Smart Homes based on their installed base and
market forecast. As stated earlier, we consider a “Smart Home” VIII. E STIMATING THE G LOBAL E NERGY I MPACT OF HAS
in this context as having a fully-equipped consumer HAS but D EVICES
we do not include traditional appliances like PCs and TV Whilst the report in [24] discussed numbers of HAS-
sets. A market-based report in [24] focuses on Smart Homes equipped dwellings, two other reports [1], [5] have discussed

0098-3063 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCE.2019.2962605, IEEE
Transactions on Consumer Electronics
8

20152016201720182019202020212022202320242025
120 120.00
numbers of HAS devices. Whilst the number of dwellings will
North America be moderated by population growth, the number of devices
Energy Consumption (TWh)

100
EU28+2 may follow a different trajectory. Global forecasts of IoT
100.00
device deployments tend to be coarse with very wide variations
[1], [5], [18].
80 80.00To estimate the energy impact of HAS deployment from
a device-count perspective, market forecasts up to 2020 from
60 two different organisations were used. Forecast 1 (used by IEA
60.00
in [5]) and Forecast 2 in [1] are given in Table VIII. Forecast 1
can be considered more conservative than Forecast 2, which
40 40.00
is a potential worst-case scenario or upper bound.
To estimate the global energy impact over a longer time
20 20.00
span, we extrapolated both forecasts to 2025, assuming their
respective CAGR, giving a 10-year projection. Applying the
0 values in Table V, for all mains-powered devices, with the
0.00
same 20/40/40 distribution of device installations as given in
the previous section, we estimate the global energy consump-
tion for HAS to rise from 8 TWh in 2015 to 58 TWh in 2025
Fig. 8. Annual energy consumption estimate for smart home systems in NA
and EU28+2. for the Forecast 1 projection or from 27 TWh to 157 TWh
TABLE VIII for the Forecast 2 projection. The disparity is considerable as
HAS D EVICE D EPLOYMENT E STIMATE UP TO 2025. the Forecast 2 projection is nearly 3-fold higher than that of
Forecast 1, which further shows the coarse nature of today’s
HAS Devices (billions)
Entity
2015 2020 2025
estimates. For illustrative purposes, Fig. 9a and Fig. 9b show
Forecast 1 (19% CAGR) 0.7 1.9 5.1 plots of the estimated HAS device growth distribution, and the
Forecast 2 (22% CAGR) 2.4 5.7 14 corresponding energy consumption estimate over the 10-year
period, using the Forecast 2 projection.
16
Installed IoT Devices (Billions)

Smart Hub
IoT Gateway
14 IX. C ONCLUSION
Smart Appliances
12 IP Camera This paper has presented detailed measurements of devices
10 Smart Light Bulbs in consumer home automation systems and developed novel
Actuators
8
system-level energy models. The models account for key
Sensors
functional components of the devices and their operational
6 characteristics over time. Using this bottom-up approach, a
4 first-order estimate of the potential energy impact of consumer
HAS on the average household has been developed and
2 extended, using ICT industry projections, to estimate energy
0 consumption on a global scale.
2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 Our estimate indicates a household energy consumption
(a) increment of more than one-third of the current household
180
annual consumption for a simple off-the-shelf HAS deploy-
Smart Hub ment. This could be much higher if augmented by full-
Energy Consumption (TWh)

160 IoT Gateway scale video surveillance applications with higher throughput
Smart Appliances demand. Smart light bulbs were the least energy efficient of
140
IP Camera
120
the range of IoT devices considered and contribute the most to
Smart Light Bulbs
that additional consumption. Smarter design of these products
100 Actuators
should be the prime focus for product development. At a more
Sensors
80 general level, the balance between the potential benefits and
60
energy savings through “smartness”, as against the energy cost
of their “smartness”, should be considered.
40
On a global scale, the potential energy impact of HAS
20 deployment is non-trivial and could reach about 157 TWh
0 by 2025, nearly 14% of the global ICT industry energy
2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 consumption forecast for that year (estimated at 1140 TWh
(b) [5]). It indicates that there is a significant price to pay if the
Fig. 9. (a) Global forecast of installed IoT devices for HAS [1]; (b) Global more optimistic market growth projections eventuate, while
energy consumption estimate for HAS devices. energy efficiency improvements do not match that growth.

0098-3063 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCE.2019.2962605, IEEE
Transactions on Consumer Electronics
9

R EFERENCES [22] P. Bertoldi, “Code of Conduct on Energy Consumption of Broadband


Equipment - Version 6.0,” European Commission - DG JRC, Ispra, Tech.
[1] Cisco, “The Zettabyte Era: Trends and Analysis,” June 2016. Rep., Feb. 2017.
[Online]. Available: https://webobjects.cdw.com/webobjects/media/pdf/ [23] Australian Energy Regulator, “Energy made easy,” [Online]. Available:
Solutions/Networking/White-Paper-Cisco-The-Zettabyte-Era-Trends- https://www.energymadeeasy.gov.au/benchmark.
and-Analysis.pdf [24] J. Svanberg, “Smart Homes and Home Automation,” BERG Insight,
[2] D. Han and J. Lim, “Design and implementation of smart home energy Gothenburg, Sweden, Tech. Rep. ed.4, May 2016. [Online]. Available:
management systems based on zigbee,” IEEE Trans. Consum. Electron., http://www.berginsight.com/ReportPDF/ProductSheet/bi-sh4-ps.pdf
vol. 56, no. 3, pp. 1417–1425, Aug. 2010.
[3] J. Han, C. Choi, W. Park, I. Lee, and S. Kim, “Smart home energy
management system including renewable energy based on zigbee and
plc,” IEEE Trans. Consum. Electron., vol. 60, no. 2, pp. 198–202, May
2014.
[4] M. Chen and C. Lin, “Standby power management of a smart home Chrispin Gray (S’13–M’18) completed his PhD
appliance by using energy saving system with active loading feature in electrical and electronic engineering at the Uni-
identification,” IEEE Trans. Consum. Electron., vol. 65, no. 1, pp. 11– versity of Melbourne, Australia, in 2018; his thesis
17, Feb. 2019. focused on developing models and energy-efficient
[5] M. Friedli, L. Kaufmann, F. Paganini, and R. Kyburz, “Energy Efficiency architectures for Internet-of-Things (IoT) networks
of the Internet of Things,” IEA, Paris, France, Tech. Rep. no.1, Apr. and services. He has held a number of positions in
2016. the telecommunications industry, including RF and
[6] G. Mehdi and M. Roshchin, “Electricity consumption constraints for transmission network design and optimisation engi-
smart-home automation: An overview of models and applications,” in neer roles. His research interests include Internet of
Sustain. in Energy and Buildings: Proc.7th Int. Conf. SEB-15, vol. 83, Things (IoT), Edge Computing, Machine Learning
2015, pp. 60–68. and access network technologies.
[7] N. Langhammer and R. Kays, “Performance Evaluation of Wireless
Home Automation Networks in Indoor Scenarios,” IEEE Trans. Smart
Grid, vol. 3, no. 4, pp. 2252–2261, Dec 2012.
[8] C. Gray and L. Campbell, “Should my toaster be polled? Towards an Robert Ayre received his B.Sc. degree in electronic
energy-efficient Internet of Things,” in 26th Int. Telecommun. Netw. engineering from George Washington University,
Appl. Conf. (ITNAC), Dec. 2016, pp. 26–31. Washington, DC, in 1967, and B.E. and M.Eng.Sc.
[9] B. Andrés, F. Alejandra, J. Miguel, S. Augusto, and W. Pedro, “Towards degrees from Monash University, Melbourne, Aus-
the evolution of smart home environments: A survey,” Int. J. Autom. tralia, in 1970 and 1972, respectively. In 1972, he
Smart Technol., vol. 6, no. 3, pp. 105–136, Sept. 2016. joined Telstra Research Laboratories, working in a
[10] G. Bag, Z. Pang, M. E. Johansson, X. Min, and S. Zhu, “Engineering number of roles, primarily in the areas of optical
friendly tool to estimate battery life of a wireless sensor node,” J. Ind. transmission for core and access networks, and in
Inf. Integr., vol. 4, pp. 8–14, Sept. 2016. broadband networking. In 2007, he joined the Uni-
[11] B. Martinez, M. Montón, I. Vilajosana, and J. D. Prades, “The Power of versity of Melbourne, Melbourne, Australia, where
Models: Modeling Power Consumption for IoT Devices,” IEEE Sensors he has been working on the energy efficiency of a
J., vol. 15, no. 10, pp. 5777–5789, Oct 2015. range of networking technologies.
[12] V. Konstantakos, A. Chatzigeorgiou, S. Nikolaidis, and T. Laopoulos,
“Energy Consumption Estimation in Embedded Systems,” IEEE Trans.
Instrum. Meas., vol. 57, no. 4, pp. 797–804, Apr. 2008.
[13] C. Gray, R. Ayre, K. Hinton, and R. S. Tucker, “Power consumption Kerry Hinton received an Honours B.E. in 1978, an
of IoT access network technologies,” in IEEE Int. Conf. Commun. Honours B.Sc. in 1980 and M.Sc. in Mathematical
Workshop (ICCW), June 2015, pp. 2818–2823. Sciences in 1982, and a Ph.D. in 1984. In 1984,
[14] F. Kaup, P. Gottschling, and D. Hausheer, “PowerPi: Measuring and he joined Telstra Research Laboratories, Australia.
modeling the power consumption of the Raspberry Pi,” in 39th Annu. In 2006, Dr. Hinton joined the Centre for Ultra
IEEE Conf. Local Comput. Netw., Sept. 2014, pp. 236–243. Broadband Information Networks at the University
[15] K. Hinton, F. Jalali, and A. Matin, “Energy Consumption Modelling of of Melbourne, Australia. In 2011 he joined the Cen-
Optical Networks,” Photon. Netw. Commun., vol. 30, no. 1, pp. 4–16, tre for Energy Efficient Telecommunications, also at
Mar. 2015. the University of Melbourne. From 2013 to 2016,
[16] IX Australia, ISP Peering Traffic, accessed: May 24, 2017. [Online]. Dr. Hinton was Director of the CEET. Dr. Hinton
Available: http://monitor.nsw.ix.asn.au/cacti/graph view.php. has authored over 130 papers, all in peer reviewed
[17] Australian Bureau of Statistics, “Internet Activity Report,” Dec. 2016, journals and conferences.
[Online]. Available: http://www.abs.gov.au/ausstats/abs@.nsf/mf/8153.0.
[18] J. Bradley, J. Barbier, and D. Handler, “Embracing the Internet of
everything to capture your share of 14.4 trillion,” White Paper, 2013.
[19] B. Lapillonne, K. Pollier, and N. Samci, “Energy efficiency Leith Campbell (M’87) is an Honorary Fellow,
trends for households in the EU,” Enerdata, Tech. Rep., May Melbourne School of Engineering, University of
2015. [Online]. Available: http://www.odyssee-mure.eu/publications/ Melbourne. He has spent over 35 years in the
efficiency-by-sector/household/household-eu.pdf telecommunications industry, both in research and
[20] P. Bertoldi, “2016 Residential Lighting Report,” Department of Industry, development and as an international consultant on
Innovation and Science, Australia, Tech. Rep., Jul. 2016. telecommunications regulation. He serves on the
[21] W. R. Gifford, M. L. Goldberg, P. M. Tanimoto, D. R. Celnicker, and Editorial Advisory Board of the Journal of Telecom-
M. E. Poplawski, “Residential Lighting End-Use Consumption Study: munications and the Digital Economy.
Estimation Framework and Initial Estimates,” Pacific Northwest National
Laboratory (PNNL), Richland, WA (US), Tech. Rep., 2012.

0098-3063 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like