Development of a Smart Metering Microservice

Based on Fast Fourier Transform (FFT) for

Edge/Internet of Things Environments
Alina Buzachis∗ , Student Member, IEEE, Antonino Galletta∗ , Student Member, IEEE,
Antonio Celesti∗ , Member, IEEE, Maria Fazio∗ and Massimo Villari∗ , Member, IEEE.
∗ Department of Mathematics, Computer Science, Physics and Earth Sciences, University of Messina,
Viale F. Stagno D’Alcontres 31, 98166 Messina, Italy
{abuzachis, angalletta, acelesti, mfazio, mvillari}@unime.it

Abstract—In recent years, great attention has been given to centers into an intermediate layer, closer to users, defined as
new Internet of Things (IoT) technologies. The IoT concept is Edge. In this regard, Gartner also predicted that around 50% of
nowadays intrinsic to traditional products and services. With generated sensing data will be processed outside of traditional
its rapid development, more and more small smart devices are
connected over the Internet in order to monitor, collect and Cloud data centers in an Edge layer by 2022 [2].
exchange data in real-time to provide smart IoT-as-a-Services In this scenario, the concept of microservice or microservice
(IoTaaS). A few years ago, IoT devices exclusively sent data to architecture is gaining more and more consensus among
a centralized Cloud data center; today it is possible to perform both the industrial and academic communities because it
“on board” processing tasks at the Edge of the network and allows the development of emerging cutting-edge IoTaaS [3].
subsequently share or use the obtained results closer to users.
This paper, focusing on a smart grid scenario, investigates the Specifically, such an architectural style allows to foster the
possibility of creating an IoTaaS for smart metering, including a separation among components and it is tailored to the needs
microservice for IoT devices capable of acquiring and processing of today’s software which requires a high degree of flexibility
electrical data using the Fast Fourier Transform (FFT) algorithm. and dynamism simplifying the IoTaaS development within a
In particular, we experimentally use the smart metering IoTaaS flexible and dynamic hybrid Cloud/Edge/IoT environment well
running on a Raspberry Pi 3 device to perform a harmonic
analysis of a frequency signal of the domestic electrical grid fitting their requirements.
in order to characterize the non-linear loads associated to Power Quality (PQ) is one of the major topics that has been
the electronic devices (e.g., smart TV, computers, etc) with associated to IoT from the beginning in both industrial and
the purpose of monitoring their status and preventing possible domestic fields. It determines the fitness of electric power to
malfunctions and faults. consumer devices. PQ management is possible by means of a
Keywords—Smart metering, smart grid, energy, power, FFT,
Cloud Computing, Internet of Things, Edge Computing. smart grid, i.e., an electrical grid which includes a variety of
operation and energy measures including smart meters, smart
I. I NTRODUCTION appliances and energy efficiency resources.
Nowadays, we are observing an explosion of the number of In this paper, we focus on a smart home scenario that
Internet of Things (IoT) devices spread all around the world proposes a smart meter IoTaaS consisting of an Fast Fourier
along with an increasing demand of efficient cutting-edge Transform (FFT)-based microservice running on a Raspberry
pervasive services and applications. In this context, Gartner Pi 3 device able to coordinate the acquisition and perform
aforesaid that the number of connected IoT devices will reach harmonic analysis of the frequency signal of the domestic
20.4 billion [1] by 2020. Therefore, the emerging business electric grid frequency signal to characterize the non-linear
perspectives coming from IoT are pushing private, public, loads originated by the current absorption of electronic devices
and hybrid Cloud providers to integrate their systems with (e.g., smart TV, computers, etc) connected on it in order to
IoT devices equipped with sensors and actuators in order to monitor their status and prevent possible malfunctions and
provide together with the traditional Infrastructure, Platform breakdowns.
and Software-as-a-Service (IaaS, PaaS, SaaS) even a new type The rest of the paper is organized as follows. A brief
of service level, that is called IoT-as-a-Service (IoTaaS). At overview of both background and state of the art is summa-
the same time, with the advent of IoT the ongoing evolution rized in Section II. Motivations along with an overview of our
of Cloud Computing has led to a drastic change in applica- system prototype are described in Section III. Implementation
tion design, deployment and delivery. Therefore, the need of details along with a reference scenario are provided in Sections
efficient services and applications has pushed ICT operators IV and V respectively. A performance analysis is discussed in
to move part of their services from the “central” Cloud data Section VI. Conclusions are summarized in Section VII.

978-1-7281-2365-3/19/$31.00 ©2019 IEEE

 II. BACKGROUND AND R ELATED W ORK printers, and other electronic devices representing an important
In recent years, a great attention has been given to smart grid issue in the power system.
applications in Cloud, Edge and IoT scenarios. In particular, Generally, a linear electrical load draws a sinusoidal current
as far as smart metering in IoT, several initiatives have been proportional to the sinusoidal voltage. The reason for such
proposed up to now. A smart IoT based energy metering a behavior is that the linear loads do not depend on the
system for microgrids with a load management algorithm is voltage to determine their impedance at a given frequency.
proposed in [4]. A smart metering platform using big data and These loads do not cause any problem to the network to
IoT technologies is proposed in [5]. In particular, a platform which they are connected or other consumers of a utility. They
prototype was developed and tested for detailed profiling of always follow the Ohm’s law. Instead, power electronics loads
energy consumption in buildings that allows measuring and do not always follow the Ohms law. Unlike the linear loads
monitoring the electric power consumption of individual appli- they do not consume power continuously. When a sinusoidal
ances. An emerging paradigm based on Low Power Wide Area voltage is applied to a non-linear electrical load, it does not
Networks (LPWAN) - LoRa, i.e., a straightforward solution draw a sinusoidal current. The non-sinusoidal current is due
for IoT networks is discussed in [6]. The deployment analysis to the device impedance changing over a complete voltage
of a Narrowband IoT (NB-IoT) system for smart metering is cycle. These loads have the potential of distorting the supply
discussed in [7]. A combination of an advanced IoT metering voltage waveform and might cause problems to other loads
infrastructure with Cloud-based analytics to optimize smart as well. In fact, harmonic distortions could produce Power
electricity distribution in residential areas is discussed in [8], Quality (PQ) problems such as voltage distortion, equipment
whereas, an IoT based smart home with real time e-metering malfunction, poor quality of power factor and component
using e-controller is discussed in [9]. failure. Moreover, harmonic distortion also causes financial
Regarding FFT applied for smart metering in IoT a few loss of the customers and electric power companies. Harmonic
initiatives have been proposed so far. A grid frequency esti- disturbances encountered are, for instance, harmonics, voltage
mation algorithm based on the fractional FFT for IoT nodes sags, rapid amplitude variations (flicker), and transients. At
time stamps is proposed in [10]. Specifically, a maximum the same time, manufacturing processes and various equipment
merging model in fractional spectrum domain for dealing become increasingly sensitive to a distorted voltage waveform.
with complicated power grid channel interference in China In this situation, accurate measurement techniques to detect,
grids with data measured by the wireless sensor networks classify and assess different related PQ problems in distorted
is discussed. An ultra-low-power variable-accuracy bit-serial environments need to be developed and applied. Therefore, the
FFT butterfly processing element for IoT devices is proposed development of a proper real-time power quality measurement
in [11]. Specifically, an ultra-low power Decimation-in-Time system based on Digital Signal Processors (DSPs) is essential.
(DIT) radix-2 butterfly arithmetic block for an FFT processor, DSPs are well suited for such mathematical techniques and
designed using a bit-serial architecture and operating at a near- real-time waveform processing, and form a powerful tool for
threshold voltage (0.6 V) is discussed. power quality assessment.
Apart from electric energy monitoring, security is one Advanced signals processing techniques and quantities have
of the major issues in IoT smart metering. In this regard, to be able to correctly describe changes in all their aspects,
a dynamic group authentication and key exchange scheme whereas it must be possible to assess them accurately and
based on threshold secret sharing for IoT smart metering quickly with a limited amount of data. With this identifica-
environments is proposed in [12]. Even regarding security, tion, power system operators can decide a strategy to reduce
an efficient certificate revocation management schemes for harmonic distortion level with filter placement. Fast Fourier
IoT-based advanced metering infrastructures in smart cities is Transform (FFT) allows to analyze the current waveform input
discussed in [13], whereas privacy-preserving protocols for in the presence of multiple devices. Moreover, it allows to
secure and reliable data aggregation in IoT-enabled smart identify the devices from the current harmonics.
metering systems are discussed in [14]. The objective that we aim to achieve in this paper, is to
In this paper, differently from the aforementioned scientific distinguish different loads and electronic devices based on the
initiatives available in literature, we will focus on the design spectral variations of the current, generated by their connection
and implementation of a smart metering IoTaaS-based on FFT to the electric grid. To give a clarifying example, we intend
for domestic environments. to distinguish a smart TV from a computer according to the
analysis of their harmonic characteristics. In order to achieve
III. P ROBLEM A NALYSIS AND O BJECTIVES D EFINITION such a goal, it is necessary to develop a smart metering IoTaaS
Nowadays, electric energy is apparently the most impor- including a microservice able to:
tant and necessary factor for the development in different
• acquire the analog electrical signal coming from domestic
application domains. In modern electrical energy systems,
grid and to perform A/D conversion.
voltages and especially currents become less sinusoidal and
periodical and even steady state behaviour may be completely • process the sampled data with FFT algorithms and provide
lost due to the increasing of non-linear loads such as smart an interpretation of the obtained results in order to characterize
TV, computers, LHE lamps, variable speed drives, fax, laser characterize the non-linear loads;
 IV. S MART M ETERING I OTAA S P ROTOTYPE
A. Hardware Design
We opted to use Raspberry Pi 3, a small, powerful and
lightweight ARM based computer, to implement the metering
system due to its ease of use and cost (only 35$). Raspberry
Pi does not have a ability to read analog inputs. Analog inputs
are useful because many sensors produce analog outputs, so
we need a way to make the Pi analog-friendly. Nevertheless, Fig. 2. Smart metering device.
this can be easily carried out by wiring up an ADC chip on it.
In order to find the best trade-off in terms of cost and sapling
(Clock pin), DIN (Data In from Raspberry Pi), and /CS (Chip
frequency range for the ADC, we tested two different chips:
Select), AGND (Analog Ground, used sometimes in precision
1) ADS1115 : 16bit, 860SPS, I 2C, PGA, cost ' 5 C; circuitry), GND and V REF (Analog Voltage Reference, used
2) MCP3008 : 10bit, 75KSPSat2.7V, SPI, cost ' 3 C. for changing the “scale”).
Even though the ADS1115 chip guarantees excellent reso-
lution and more advanced programming possibilities, unfortu- B. Software Design
nately, it has a low sampling rate. This, for practical purposes, The software side of this system was developed by means
would limit the sampling band to about 400Hz to comply with of a container-based microservice that runs on Raspberry Pi
Shannon-Nyquist sampling theorem. Such a narrow band may 3 (Raspbian OS) and manages the acquisition, processing and
not be enough, in fact there could be harmonics at higher output phases of the smart metering IoTaaS. The container-
frequencies that our system would not be able to capture. based approach was adopted in order to simplify service
Instead, the MCP3008 chip, even though has a resolution deployment according to an IoT-Cloud scenario. The mis-
limited to only 10bits, allows us to evaluate a wider band croservice was developed according to the producer-consumer
of frequencies. According to the sampling theorem we could paradigm. In order to process sampled electrical data two Java
go to sample signals with maximum frequency up to about libraries have been identified and used: Pi4J and JTransforms.
37kHz, covering widely the hypothetical band of interest. For In particular, these Java libraries allow us to manage the Rasp-
this reason, we decide to use the MCP3008 chip. berry Pi device and execute the FFT algorithm respectively on
the acquired sampled electric grid signal.
a) Producer: In our structure, the producer is represented
by the ADC chip, which produces data at each performed
conversion. In this paper, we used the com.pi4j.io.spi package
because the MCP3008 chip uses SPI to communicate with
the Raspberry Pi device. The package also includes: SpiDe-
vice interface, SpiFactory, SpiChannel and SpiMode classes.
Therefore, Producer class creates a reference of type SpiDevice
(interface) that will be used to manipulate the object of
type SpiDeviceImpl necessary for the SPI communication.
The DEFAULT SPI SPEED parameter is used to set the
clock frequency to 1MHz. An infinite loop repeats the read()
Fig. 1. Metering system’s real architecture. method, passing as a parameter the channel number (pin) on
which the input the need to be digitized is connected.
The MCP3008 chip uses the SPI bus protocol which is b) Consumer: The consumer is represented by the FFT
configured by the Raspberry Pi’s GPIO header as shown in algorithm which consumes the data produced by the MCP3008
Figure 1. The MCP3008 chip is electrically powered with 3.3V chip. The JTransforms library encompasses several pack-
since the power supply with 5V would imply output pulses of ages, each of one containing classes that deal with different
the same amplitude that would damage Raspberry Pi 3. By transformations, e.g, DCT, DHT, DST, FFT. Thus, we used
choosing V DD = 3.3V , we are able to guarantee the correct org.jtransforms.fft package. The packages includes two sub-
output voltage. In order to understand how to interface the classes which could perform the FFT calculation with either
MCP3008 chip with the Italian electrical grid providing an single or double precision as well as for the calculation
alternating voltage of 220V in effective value at a frequency of mono/multidimensional transforms. Given the limited re-
of 50Hz, we used the LEM HAIS 50-P current transducer with sources of Raspberry Pi 3, we chose to use the single-
low pass filter with cut-off frequency fT = 3KHz as depicted precision FloatFFT 1D class that consumes less memory and
in Figure 2. compute resources. Although the JTransforms library allows
In order to read analog data, the following pins were used: the processing of any array size, it is important to emphasize
V DD (power) and DGND (Digital Ground) to power the that the performance is better when the array has a power of
MCP3008 chip, DOUT (Data Out from MCP3008), CLK 2.
 c) Buffer for writing/reading data: Producer and con- The scenarios marked as 2 and 4 consider the presence
sumer must be executed concurrently by two threads. Thus, a of forbidden devices. In particular, the scenario 2 considers
shared data structure is required and at the same time the con- the presence of cryptocurrency mining [15], instead scenario
current access to the resource must be managed. The ideal data 4 represents the charging of an electric car. In both cases,
structure is a FIFO queue. The java.util.concurrent package the smart metering system is able to identify the forbidden
provides a series of interfaces and classes for multithreading devices and to disconnect them from the electrical grid.
and concurrent access. Scenario 3 , instead, shows the normal usage of a personal
The BlockingQueue < E > interface defines blocking meth- computer. In this case, no action has to be done because there
ods for inserting and extracting data from the queue. The is a normal usage of the electrical grid. Apart from a smart
put() method inserts an element in the queue, while the
take() one extracts an item from the queue. Among the
classes that implement the BlockingQueue < E > interface, the
LinkedBlockingQueue < E > class, creates a dynamic queue
that is optionally limited in size, in which the elements are
arranged according to the FIFO logic. The put() and take()
methods allow us to insert and retrieve data from the structure
respectively. We chose to use this class for our microservice.
d) A Solution for Calculating fs Through the Sample
Class: The acquisition at a specific frequency from the
MCP3008 chip requires real-time I/O management. Our Rasp-
berry Pi device, equipped with a Raspbian Operating System
(OS), does not handle real-time applications. In fact, all
the running processes and threads need to be scheduled in
advance, whereas the miscroservice has to manage the variable
reading frequency from the MCP3008 chip. In order to handle
this inconvenient the Sample class was created. This class
conceptually represents a single sample returned by the ADC
conversion performed by the MCP3008 chip. It contains two
instance variables: a float variable named voltage and a double
variable named time. For each conversion, the value of the
sampled signal coming from the domestic electric grid will be
Fig. 3. Reference scenario.
saved within voltage, while time will save the chronological
instant (nanoseconds) in which the data have been stored.
university campus, the smart metering IoTaaS can be used in
The consumer, through this artifice, will be able to know the
any smart building.
time interval elapsed for the sampling of the elements taken
from the FIFO queue. By saving the time instants relative VI. M EASUREMENTS AND O BSERVATIONS OF D IFFERENT
to the following samples, it will be possible to calculate the N ON -L INEAR L OADS
frequency fs with a reasonably reliable approximation.
The experiments were conducted according to two phases,
V. C ASE S TUDY: S MART U NIVERSITY C AMPUS namely calibration and validation. Both phases were per-
In an university campus, several actors, i.e., professors, formed in the Laboratory of Electronics, Department of Engi-
students, employees and so on, can connect their own devices neering, University of Messina.
to the grid. These devices could be dangerous for others
or forbidden by internal faculty regulations. Checking every A. Calibration Phase
device connected to the grid is very difficult and expensive. Before using the smart meter IoTaaS with real loads, it
Therefore, the idea is to create a smart metering IoTaaS system was necessary to perform calibration tests to ascertain the
that can monitor a specific part of the campus electricity grid correct functioning of the system, i.e., if it is able to correctly
and disconnect whether a problem occurs. sample an analog electrical signal and detect its frequency
Figure 3 shows the proposed scenario. As the reader can components. To do this, we used a signal generator that
observe, the scenario labeled with 1 represents the case allowed us to test the behaviour of the meter by varying the
in which a device (e.g, microwave oven) is damaged and input signal at different frequencies as well as using signals
generates on the grid undesired harmonics. The latter could of different shapes (e.g., triangular, sinusoidal, square wave).
damage other devices potentially connected on the same part Furthermore, we have also used an oscilloscope capable of
of the domestic grid. Therefore, the smart metering IoTaaS is calculating FFT, connected in parallel to the realized system,
able to identify that device and eventually disconnect it from in order to verify if the output of the meter corresponds to the
the grid. spectrum of the input signal.
 The FFT algorithm executed on the Raspberry Pi 3, in this b) Drill: Respect to computer monitor which has a very
configuration, was able to detect with high approximation the low current absorption in stand-by mode, the drill presents a
frequency components up to about 8KHz; data processed by slight current absorption. In particular, we have two peaks cor-
the meter deviate slightly from those of the equipment used responding to 48Hz and 144Hz as shown in Figure 5. As well
for the test. Thus, the transform of a pure sine wave with as in the active mode with 100% of velocity, the FFT transform
f = 100Hz, has a spectral component exactly at 100Hz. The calculated with the smart metering system is illustrated in
implemented system places the component at a frequency Figure 6. By turning on the drill, we notice the presence of
equal to 100Hz±3%. As previously explained, this error is
due to the sample windows as well as spectral leakage than
can occur. This result meets our expectations, as the system
allows us to distinguish the fundamental harmonic of the
electrical network (50Hz) from the higher-order replicas (e.g.,
100Hz, 150Hz, 200Hz, an so on) that must be detected to
characterize the load. This result meets our expectations, as the
system allows us to distinguish the fundamental harmonic of
the electrical network (50Hz) from higher order replicas (e.g.,
100Hz, 150Hz, 200Hz, and so on) which must be detected to
characterize the load.
B. Validation Phase
The aim of the experimentation is to characterize non-linear
loads generated by specific electrical devices by looking for Fig. 5. Drill stand-by: FFT spectrum obtained with the metering system.
an identifying “pattern” in the spectrum of each of them. The
IoTaaS has been tested with: (i) a computer monitor, (ii) a two significant harmonics (Figure 6) corresponding to 51Hz
drill, and (iii) monitor and drill together. and respectively 152Hz.
For each characterized load, a bar graph illustrating the
representation of the frequency components will be shown.
During this phase, windows of different sizes were used. The
window of 2048 samples with fs = 20KHz allowed us to
obtain fairly faithful spectra and acceptable processing times.
The consumer’s threshold was set to 1, for instance, data
saved and displayed in bar graphs do not contain frequency
components with an amplitude less than 1mV . By doing so,
we tried to “isolate” the most significant components for the
characterization of loads.
a) Computer Monitor: Figure 4 shows the FFT transform
of the computer monitor active obtained from the smart
metering system. It can be noticed that when the monitor is
active there is the presence of many significant harmonics,
Fig. 6. Drill active at 100% of velocity: FFT spectrum obtained with the
for instance, at 48Hz, 148Hz, 249Hz and 340Hz respectively. metering system.
Such harmonics characterize the computer monitor load.
c) Monitor and Drill Together: In the last experiment of
the validation phase, we considered both computer monitor
and drill active. Therefore, we illustrate the FFT transform
in Figure 7. By analyzing the Figure 7, we can notice that
the obtained spectrum presents two significant harmonics in
the neighborhood of 50Hz and respectively 157Hz. Therefore,
these two harmonics are characteristics for both monitor
and drill; the amplitude of these two peaks is significantly
increased due to the presence of both loads active.
C. Discussion
The proposed metering system is able to correctly place
the harmonics detected in a neighborhood of the correct
frequencies to which they should be located. As explained
Fig. 4. Monitor active: FFT spectrum obtained with the metering system. above, the phenomenon is linked to the slight variability of the
 the harmonic spectral content originating from their current
absorption calculated using the FFT transform. This has been
pursued using the Raspberry Pi 3 device which, suitably pro-
grammed using a container-based microservice, coordinates
the acquisition operations of an ADC chip and processes the
data returned by the FFT algorithm in order to monitor the
status of the connected electrical devices.
During the experimental phases, it was possible to identify
the characteristic “patterns” for each examined load, which
allowed to recognize the individual power loads connected to
the network. The simultaneous use of multiple loads entails
difficulties in distinguishing them due to the overlap between
the harmonics. In order to solve this, in future work, we are
Fig. 7. Monitor and Drill active: spectrum obtained with the metering system.
planning to use convolutional neural networks (CNNs). More-
50HZ 100Hz 150Hz 250Hz 350Hz 450Hz over, we would like to extend our efforts to more complicated
Monitor active
6.35 4.73 2.93 2.43 grid models to get a more complex view of the usefulness of
[mV]
Drill stand-by
these new evaluation methods mainly from the point of view
3.01 1.53
[mV] of distribution network operators.
Drill active 100%
[mV]
9.81 3.37 R EFERENCES
Drill + Monitor
active [mV]
15.41 6.67 4.41 2.46 2.24 [1] Gartner Says 8.4 Billion Connected ”Things” Will Be
TABLE I. S IGNIFICANT HARMONIC COMPONENTS OF THE ANALYZED in Use in 2017, Up 31 Percent From 2016, 2017.
LOADS . https://www.gartner.com/newsroom/id/3598917.
[2] What Edge Computing Means for Infrastructure and Operations
Leaders, 2017. https://www.gartner.com/smarterwithgartner/what-edge-
fS , and the size of the window. However, despite this margin computing-means-for-infrastructure-and-operations-leaders/.
of imprecision, it is possible to distinguish between them, the [3] E. Ahmed, A. Ahmed, I. Yaqoob, J. Shuja, A. Gani, M. Imran, and
significant harmonics and the correct frequency value to which M. Shoaib, “Bringing computation closer toward the user network:
Is edge computing the solution?,” IEEE Communications Magazine,
they refer. vol. 55, no. 11, pp. 138–144, 2017.
During the validation phase we also tried to use wider [4] K. Roy, V. Banninthaya, S. Prabhu, A. Koomar, D. Karnataki, and
windows. This allowed to obtain a better frequency resolution G. Shankar, “Smart iot based energy metering system for microgrids
with load management algorithm,” pp. 252–256, 2018.
of the FFT at the expense of processing times. By setting [5] A. Berouine, F. Lachhab, Y. Nait Malek, M. Bakhouya, and R. Oulad-
n = 4096, the processing times of the consumer thread became sine, “A smart metering platform using big data and iot technologies,”
higher than the those needed to acquire the block of samples. vol. 2018-January, pp. 1–6, 2018.
[6] L. Vangelista, “Frequency shift chirp modulation: The lora modulation,”
It is clear that these conditions would not have allowed the IEEE Signal Processing Letters, vol. 24, pp. 1818–1821, Dec 2017.
system to process the information acquired in real-time. For [7] M. Pennacchioni, M.-G. Di Benedette, T. Pecorella, C. Carlini, and
instance, the plots realized with the measured values produced P. Obino, “Nb-iot system deployment for smart metering: Evaluation
of coverage and capacity performances,” vol. 2017-January, pp. 1–6,
by the smart metering IoTaaS are quite similar with the spectra 2017.
shown by the digital oscilloscope. This shows the presence of [8] R. Ramakrishnan and L. Gaur, “Smart electricity distribution in residen-
a disturbance to f ≈ 950Hz caused by the system itself. tial areas: Internet of things (iot) based advanced metering infrastructure
and cloud analytics,” pp. 46–51, 2016.
Table I summarizes the amplitudes associated with the [9] R. Bhilare and S. Mali, “Iot based smart home with real time e-metering
harmonics of various orders for each electrical load produced using e-controller,” 2016.
by the considered electronic devices. Each row of the table [10] Q.-L. Mao and M.-Y. Zhai, “A new grid frequency estimation algorithm
based on the fractional fft for iot nodes time stamps,” Cluster Computing,
(except for the last one) defines an unique “pattern” for the Jan 2018.
identification of the electronic devices. Only the computer [11] Y. Lu and T. J. Kazmierski, “An ultra-low-power variable-accuracy bit-
monitor presents the “pattern” (fund → 6.35mV , 3◦ → 4.73mV , serial fft butterfly processing element for iot sensors,” in 2016 IEEE
Asia Pacific Conference on Circuits and Systems (APCCAS), pp. 13–16,
5◦ → 2.93mV , 7◦ → 2.43mV ). A well as the drill is character- Oct 2016.
ized by (fund → 108.08mV , 2◦ → 10.86mV , 3◦ → 4.71mV ). [12] D.-H. Lee and I.-Y. Lee, “Dynamic group authentication and key ex-
The spectrum generated by several simultaneous loads does change scheme based on threshold secret sharing for iot smart metering
environments,” Sensors (Switzerland), vol. 18, no. 10, 2018.
not allow the immediate distinction of the devices due to the [13] M. Cebe and K. Akkaya, “Efficient certificate revocation management
overlap between the harmonics, but however, it can be noticed schemes for iot-based advanced metering infrastructures in smart cities,”
a significantly increase in the amplitude of harmonics placed at Ad Hoc Networks, 2018.
[14] S. Tonyali, K. Akkaya, N. Saputro, A. Uluagac, and M. Nojoumian,
50Hz and respectively 157Hz which characterize both loads. “Privacy-preserving protocols for secure and reliable data aggregation
in iot-enabled smart metering systems,” Future Generation Computer
VII. C ONCLUSIONS AND F UTURE W ORK Systems, vol. 78, pp. 547–557, 2018.
In this paper, we developed a smart metering IoTaaS that [15] “Students are mining bitcoin in university dorms, cam-
puses are alert.” https://www.newsbtc.com/2018/08/12/
is able to recognize non-linear loads generated by electronic students-are-mining-bitcoin-in-university-dorms-campuses-are-alert/.
devices connected over on a common electrical grid, based on [Online]; Accessed: 2019-03-28.