"PowerBlade: A Low-Profile, True-Power, Plug

PowerBlade: A Low-Profile, True-Power,
Plug-Through Energy Meter
Samuel DeBruin, Branden Ghena, Ye-Sheng Kuo, and Prabal Dutta
Electrical Engineering and Computer Science Department
University of Michigan
Ann Arbor, MI 48109
{sdebruin, brghena, samkuo, prabal}@umich.edu
We present PowerBlade, the smallest, lowest cost, and lowest power
AC plug-load meter that measures real, reactive and apparent power,
and reports this data, along with cumulative energy consumption,
over an industry-standard Bluetooth Low Energy radio. Achieving
this design point requires revisiting every aspect of conventional
power meters: a new method of acquiring voltage; a non-invasive,
planar method of current measurement; an efficient and accurate
method of computing power from the voltage and current channels;
a radio interface that leverages nearby smart phones to display data
and report it to the cloud; and a retro power supply re-imagined
with vastly lower current draw, allowing extreme miniaturization.
PowerBlade occupies a mere 1" × 1" footprint, offers a 1/16" profile,
draws less than 180 mW itself, offers 1.13% error on unity power
factor loads in the 2-1200 W range and slightly worse for non-linear
and reactive loads, and costs $11 in modest quantities of about 1,000
units. This new design point enables affordable large-scale studies
of plug-load energy usage—an area of growing national importance.
Categories and Subject Descriptors
B.4.2 [HARDWARE]: Input/Output and Data Communications—
NETWORKS]: Special-Purpose and Application-Based Systems
General Terms
Design, Experimentation, Measurement, Performance
AC meter, Smart meter, Energy metering, Power metering, Plug-load
metering, Data aggregation, Intermittent power, Wireless sensor
Permission to make digital or hard copies of part or all of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for third-party components of this work must be honored.
For all other uses, contact the authors.
Copyright is held by the authors.
SenSys’15, November 01–04, 2015, Seoul, Republic of Korea
ACM 987-1-4503-3631-4/15/11.
Figure 1: Profile, front, and perspective views of PowerBlade, a wireless power meter that measures real, reactive, and apparent power.
PowerBlade’s low profile and plug through form factor allow it to sit
inconspicuously and unobtrusively between an electrical plug and
an outlet, and its square-inch footprint spares adjacent outlets from
being blocked. PowerBlade stores accumulated energy data locally
and reports both instantaneous power and accumulated energy to a
nearby smartphone or gateway over a Bluetooth Low Energy radio.
Residential and commercial buildings in the US used 2,760 TWh
of electrical energy in 2014 [19]. The majority of that usage comes
from clearly obvious loads including HVAC, lighting, and appliances, however, approximately a 20% (and growing) share of electricity usage is due to “plug-loads,” often called miscellaneous electrical loads (MELs) in industry terms [27]. These diverse loads, from
televisions and computers to vending machines and box fans, represent the long-tail of electricity use. Understanding the characteristics
of these loads requires insight into each device’s individual consumption but the methods today are limited. As a result, ratepayers,
regulators, and researchers lack the tools to unobtrusively monitor
plug-load energy use with high fidelity and low cost.
To help address this problem, we present PowerBlade, a new
power/energy meter that achieves a vastly smaller form factor than
prior systems. PowerBlade occupies an unobtrusive and essentially
two-dimensional volume, 1" × 1" × 1/16", and meters loads plugged
through it and into an outlet, as Figure 1 shows. Despite its small
size, PowerBlade is a wireless true power meter, capable of metering
real, reactive, and apparent power at kHz frequencies, aggregating
these measurements into cumulative energy, and transmitting these
data several times per second using a Bluetooth Low Energy (BLE)
radio to a nearby smartphone or gateway. At a cost of $11 in modest
quantities of 1,000 units, PowerBlade is the smallest and lowest cost
AC plug-load meter with 1.13% accuracy over a 2-1200 W range
for unity power factor loads, and slightly worse for non-linear and
reactive loads. This small form factor, coupled with easy access to
and transport of the meter data, may enable new applications.
Metering Device
Power Supply
True Power?
Data Output
Kill-A-Watt [6]
Watts Up [17]
Belkin Conserve Insight [3]
ACme-A [26]
ACme-B [26]
Monjolo [23]
Gemini [20]
Capacitor fed
Capacitor fed
Capacitor fed
Capacitor fed
Energy harvest
Energy harvest
Hall effect
Current Transformer
Current Transformer
PowerBlade (this work)
Resistor fed
Static Power
450 mW 14.0 in3
590 mW 31.9 in3
440 mW 21.8 in3
1000 mW 13.7 in3
100 mW 13.7 in3
4 mW
7.8 in3
Not Published
80-176 mW†
0.07 in3
Table 1: Comparison of various power meters. PowerBlade is the smallest, lowest power, wireless true power meter. †Depends on data rate.
PowerBlade provides an opportunity to think about power metering in a new way. Rather than metering every outlet in a building,
PowerBlade is capable of metering every load, following the plug
to any outlet with which it mates. This tight binding allows data to
be collected about the plug-loads themselves rather than whatever
happens to be plugged into an outlet, representing a paradigm shift
from prior work that employed bulky power meters. Moreover, existing power meters do not make their data easily accessible to users,
but using BLE allows PowerBlade’s measurements to be received
by unmodified smartphones. Section 2 provides a more detailed
comparison of PowerBlade with existing power meters.
PowerBlade’s small profile and literally plug-through design require revisiting nearly every aspect of power meter design. Since
PowerBlade has no outlet of its own, it utilizes exposed conductive
tabs built from rigid-flex PCB to make contact with the plug-through
AC load. Traditional AC-DC power supplies require substantial volume to hold charge or dissipate power, owing to high load power, but
PowerBlade draws less than 6 mW average power itself, allowing
it to utilize a small and simple power supply. Conventional current
sensing requires breaking the AC path with a sense resistor, encircling it with a current transformer, or monitoring the Hall effect
in a plane parallel to the current flow, but these methods are not
viable in the PowerBlade design. Rather, PowerBlade measures the
magnetic field generated from the current passing through it using
a wire-wound inductor optimally placed in the plane perpendicular
to the flow of current. This magnetometer design allows the current
waveform to be measured using commodity electronic components.
These and other design concerns are described in detail in Section 3
and specific implementation choices are presented in Section 4.
PowerBlade offers accurate metering over a wide range of load
powers. For resistive loads, such as incandescent bulbs, PowerBlade
has an average error of 1.13% over the range of 2 W to 1200 W. We
also test the meter on a range of common household loads including
a toaster, refrigerator, Xbox, and WiFi router. Across these loads,
PowerBlade has an average error of 6.5%. Section 5 provides a
detailed evaluation of system accuracy, usability, and safety.
Despite PowerBlade’s small size and high accuracy, we identify
several opportunities for future improvements to the system. These
include better synchronization between current and voltage channel
acquisition, better calculation of power factor, securing the metered
data, and providing timestamped and synchronized interval data at
the 1, 15, or 60 minute intervals. We discuss these future improvements in Section 6.
With PowerBlade in hand, ratepayers, regulators, and researchers
have an increased ability to understand plug-load usage patterns.
Large-scale, long-term deployments in residential and commercial
settings should lead to greater awareness of plug-load usage and
plug-load trends—particularly growing energy waste due to high idle
power—providing critical data to regulators for efficiency standards.
A true-power meter performs five distinct functions: (i) reduce
the AC mains voltage to low DC voltages to power the meter itself,
(ii) measure voltage, (iii) measure current, (iv) calculate power and
energy, and (v) communicate these measurements. Scaling a power
meter to PowerBlade’s form factor requires revisiting each of these
functions as they are all intimately tied to system form factor.
Table 1 presents a sampling of commercial and research meters,
and key design choices along the five dimensions that these earlier
devices embody. In this section, we survey these systems, noting
the effect that these choices have on overall size and power draw.
In particular, we survey three commercial meters—Kill-A-Watt [6],
Watts Up [17], and Belkin Conserve Insight [3]—and three research
meters—ACme [26] (two different versions), Monjolo [23], and
Gemini [20]—alongside PowerBlade (this work).
AC-DC Power Supply
For a power meter to operate from the AC mains, it must rectify
and step down the AC voltage to provide itself with low voltage
DC. Half- or full-wave rectifiers are typically used for this purpose,
and they can occupy a small volume using a single or multiple
low-profile diodes. Voltage step-down, in contrast, requires more
volume and is not blindly amenable to scaling. Moreover, step-down
techniques often do not scale as DC power is reduced, requiring
a minimum volume regardless of the DC power supplied, while
energy harvesters often require bulky current transformers that are
fundamentally unsuited to PowerBlade’s form factor [20, 23].
Transformers are the most common step-down technique in “wall
wart” power supplies and they are used in the ACme-B design [26].
They are inexpensive, efficient, and isolated devices but their volume
does not scale linearly to the form factors needed in our case—the
smallest AC transformer available from Coilcraft, for example, is
approximately 0.31 in3 [4], which would dominate our volume.
In capacitor-fed power supplies, a high voltage series capacitor
drops the line voltage and limits current. Capacitor-fed power supplies are common in many AC power meters [3, 6, 17, 26] even
though they do not provide isolation. Fortunately, however, their
volume scales more directly with the load current they can supply.
To source 10 to 100 mA, the capacitor must be high-valued (1-2 µF),
and for high voltage rated capacitors this high value is only found
in film capacitors. Such capacitors are often 0.5" on a side or larger.
If, however, the supply must only source 10s to 100s of µA, the
capacitor can be much smaller—high voltage 10-50 nF capacitors
can be found in small, surface-mount ceramic packages.
PowerBlade eschews energy harvesters and transformers, due
to scaling challenges, and instead embraces a capacitor-fed, Zenerregulated power supply. The key to making this design point viable is
scaling the electronics power draw down to meet the limited supply.
Voltage Measurement
A true power meter must acquire time-synchronized voltage and
current measurements and multiply them together to obtain power.
The voltage channel is frequently obtained by intercepting the plug’s
prongs and using a voltage divider to obtain a scaled-down version
of the voltage signal [3, 6, 17, 26]. Unfortunately, intercepting the
power lines to obtain the voltage is not possible in a planar design.
Other designs distribute the voltage and current measurements, and
wirelessly recombine them, to obtain power [20, 28], while others
do not use the voltage channel signal at all [23]. None of these
these approaches are ideal for a plug-through meter. Taking distributed measurements requires at least two different devices which
increases cost and makes deployment cumbersome, while only using
the current channel leads to errors for non-unity power factors [20].
In contrast with the prior work, PowerBlade borrows an idea from
the FlipIt plug-through USB charger [22] for its voltage acquisition.
FlipIt uses spring-loaded wire contacts molded into a 0.066” thick
piece of plastic through which a plug’s prongs pass. PowerBlade
improves upon this design by integrating the contacts directly into
the PCB by employing a rigid-flex material which eliminates the
molding, enabling the only meter design point that makes contact
with the prongs without the need for an AC receptacle.
Current Measurement
Among the meters we survey, the most common methods for
measuring current employ a sense resistor placed in series with the
electrical path [3,6,17,26], a Hall effect sensor placed co-planar to a
current carrying conductor trace [26], and a current transformer (CT)
that encircles the current carrying conductor [20, 23]. Unfortunately,
none of these designs are suited to an essentially planar, plug-through
form factor. Current sense resistors are inexpensive, accurate, and
small, but they require the electrical path to be broken and an AC
receptacle and prongs be used, making them unsuitable for our
application. Hall effect sensors work by measuring the deflection
of electrons in a conductor exposed to a magnetic field (like the
one generated by a current). However, they require the magnetic
field lines to be perpendicular to the plane of the sensing element
which, in our case, is challenging since the magnetic field lines are
co-planar with the circuit board; hence a Hall effect sensor would
require a non-trivial third dimension.
In contrast with these methods of current sensing, PowerBlade
uses an optimally-placed surface mount inductor to measure the
variation in magnetic flux produced by a current carrying conductor, detectable as a voltage across the inductor’s terminals. Used
in this way, the inductor functions as a search coil (or inductive
sensor) whose terminal voltage is proportional to the rate of change
of the current over time. This approach requires signal integration
to recover the original current signal. Using a small, surface mount
inductor in this manner enables PowerBlade to maintain an essentially two-dimensional form factor—something that is difficult using
conventional current sensing methods.
Power Calculation
There exist three common options for calculating power. The
first is to use a power metering integrated circuit like the Analog
Devices ADE7753 [1]. Three of the surveyed meters employ this
or a similar chip [3, 17, 26]. These metering chips take as inputs
current and voltage signals and provide as outputs real, reactive, and
apparent power, as well as power factor, zero crossings, and other
measurements. They provide accurate measurements but are costly
in terms of area and power draw, consuming 0.3" × 0.3" × 0.08"
PCB area and drawing 25 mW, respectively, placing them outside
of acceptable area and power budgets for our application.
A second option employed by several meters in our survey is to
calculate power from the acquired current and voltage in software
running on an embedded microcontroller. This approach can be
quite accurate but the accuracy is bounded by the sampling rate of
the ADCs and the performance of the signal processing pipeline.
PowerBlade employs this approach and uses a low power microcontroller to achieve an acceptably lower power draw.
In contrast with these approaches—hardware and software—the
Monjolo design does not report true power [23]. A Monjolo power
meter does not actually sample either the current or the voltage
waveform. Rather, it uses a fixed VRM S value and an approximation
of the IRM S value based on the activation interval of an energy harvesting sensor. With these data, Monjolo estimates apparent power,
which is the per-cycle product of VRM S and IRM S , but not real or
reactive power. Moreover, Monjolo exhibits high error for certain
loads with significant harmonic content.
Data Communication
To be useful, a power meter must communicate its data to the
outside world. How these data are communicated to users affects the
meter size. All of the commercial meters we survey communicate
their data to the user with an LCD display on the unit [3, 6, 17]. Of
these three, the Kill-A-Watt and Watts Up have a readout physically
located at the outlet while the Conserve Insight offers a readout on
a five foot tether. Although a display improves data visibility, it also
couples the size of the meter with the size of the screen, leading to a
tension between making the meter smaller and making the screen
larger. This tensions affects usability as more readable screens take
up more space. In addition, data displayed solely on a screen cannot
be easily recorded by the user for later use. Thus, LCD screens are
not well-suited to the challenge of pervasive power metering.
Another communications option is wired, like USB or Ethernet,
and one of the commercial meters we survey uses this method in
addition to its LCD readout [17]. Although this offers an attractive
low power, high-bandwidth link, the form factor ramifications are
severe. The USB connection must be isolated from AC mains to be
connected to a host computer, and this isolation requires substantial
additional space that is unavailable in an essentially planar design.
Low power wireless radios allow the system to display or process
data remotely, using a smartphone or computer. Unlike an LCD,
radios require physically very little space in the meter. The radioenabled meters we survey use 802.15.4, and ACme [26] uses a
multi-hop mesh network, allowing data to propagate over potentially
long distances. However, none of these meters allow users to connect
directly using a smartphone. PowerBlade also employs a radio, but
we chose a Bluetooth Low Energy radio which can directly leverage
the rich interface available on nearby smartphones and will soon
support IP connectivity and end-to-end networking [25].
Alternate Power Metering Methods
Fine-grained power data does not require individual load metering. This analysis explores devices that, similar to PowerBlade,
meter consumption at the outlet where power is being drawn, but
methods exist for acquiring this data without physical metering. Wu
et al. demonstrate the ability to determine appliance on/off state
through deployed sensors, and they use preexisting knowledge of
the draw of each of these states to determine total power [30]. Further, ElectriSense [24] is able to determine device state simply by
monitoring AC voltage at a single point and measuring the EMI
generated by switched mode power supplies and propagated by the
wiring throughout the building. Both options provide consumption
information without the overhead of plug-load meters, but they also
require prior knowledge of the systems to be measured.
Figure 2: Zener regulated, half-wave rectified power supply. ZIN ,
either a resistor or capacitor, is used to shunt the line voltage. D1
is a rectifying diode. COU T is used to store charge between each
positive half cycle of the AC wave. The maximum voltage exposed
to the load is controlled by the Zener voltage of DZ . On very low
power loads, this supply can be made in a very small volume.
PowerBlade performs the same five functions as other power meters, but in a planar, plug-through form factor. This section examines
the design space and tradeoffs in achieving this design point.
Power Supply
PowerBlade optimizes for size by using a power supply design
that does not require an IC [21]. Figure 2 shows a Zener regulated,
half-wave rectified power supply. This circuit offers a low component count, and only the shunt impedance, ZIN , must be rated for
AC voltage. The choice of ZIN affects the final system’s available
power, overall volume, and idle power draw. In this study, we only
examine components rated for the application: leads separated by at
least the AC mains spark gap of 1.25 mm (package 1206 and larger).
Resistors are further rated for the idle power dissipated (VAC
and capacitors must be class X (AC rated “across the line”) or above.
We note that this circuit is not isolated. Its ground is tied directly
to the neutral line, which could be a possible safety issue when interacting with the circuit. Since the system is wireless, and is intended
to be entirely packaged, this does not present a risk for the end user.
More discussion on packaging is included in Section 5.2.4. One
aspect of safety that does pertain here, however, is component count.
In order to limit inrush current if ZIN is a capacitor, it must be in
series with a resistor. Further, when the system is de-powered the
capacitor maintains its voltage. In order to prevent shock, an additional high-value bleed resistor must be placed in parallel with the
capacitor. If ZIN is a resistor, neither extra component is required.
Although this is not explicitly included in the volume discussion
below, it must be considered when selecting the final design.
Figure 3 shows the design space for ZIN based on component
volume and supplied current. A third parameter, not pictured, is
idle power: the resistors add an idle power even when the load is
not applied. This idle power scales directly with current supplied.
Capacitors, however, do not affect idle power. The current draw
follows Ohm’s law (IM AX ≈ VAC /ZIN ). Volume is calculated
using components from DigiKey’s electronic component database,
and for resistors the smallest available component with the required
power rating is shown.
This figure illustrates the fundamental tradeoffs for this simple
supply: the volume occupied by the supply scales directly with the
provided current. In other words, current is not free even though
the system is attached to AC. Whether for resistors or capacitors,
accommodating more current means moving to a larger package.
Even within a given package, small variations are introduced, while
small variations in value result in quantum jumps in volume.
Maximum Volume
Our selection
Flip It Charger
Current Supplied (µA)
Minimum Current
Volume (mm3)
1206 Res
1808 Cap
2512 Res
2220 Cap
J-Lead Res
Film Cap
Figure 3: Selecting ZIN based on volume occupied and current
that can be supplied. If ZIN is a resistor, lower resistance provides
more current but also requires a larger size/package to dissipate the
additional power. A capacitor does not increase idle power, and high
capacitance (low impedance) provides more current but the extra
capacitance requires a larger size/package. The resistors marked
are the smallest available at the required power rating. There is a
maximum volume that the system can occupy, as well as a minimum
current required. The upper left quadrant, shaded, represents the
viable design space for PowerBlade. Although any component in
this quadrant is viable, there exist pareto-optimal points for volume,
current, and idle power. We select a 1210 resistor, marked, back
from the frontier to allow for tolerance as we develop the design.
Figure 4: Low volume power supplies. Each supply is built within a
0.875 inch square. Large volume components are necessary for each
supply and cannot be reduced significantly as load power decreases.
This relationship holds true for larger systems with increased
power requirements. Figure 4 shows three low volume, direct rectification power supplies: the SR086 [13], the SR10 [14], and the
LNK302 [9]. Each is dominated by one or more large components
whose sizes scale with the current that can be supplied. The relationship is not limited to capacitors and resistors: although the
SR10 uses a capacitor shunt similar to PowerBlade, the SR086 and
LNK302 use a high-voltage transistor and inductor, respectively. All
three of these supplies would exist in the far upper right of Figure 3,
providing significantly higher current but requiring greater volume.
Figure 3 can be used to determine an approximate power point
for the system. This can be used to select the remainder of the components, which in turn will yield a more precise minimum required
current. The maximum volume and minimum current shown on the
figure are for the final PowerBlade system. The range of acceptable
volumes is determined experimentally on several NEMA outlets,
with an additional dashed line representing an approximation of the
Flip It USB charger’s volume which also utilizes a very similar form
factor and usage model.
The upper left quadrant of Figure 3, shaded, is the viable design
space for PowerBlade. The pareto-optimal point for current is a
10 kΩ 2512 resistor that can supply 5.5 mA and requires 13.1 mm3 .
The pareto-optimal point for volume is a 47 kΩ 1206 resistor that
can supply 1.1 mA and requires 3.2 mm3 . These would add an idle
power of 1.4 W and 300 mW, respectively. The pareto-optimal point
for idle power is a 33 nF 2220 ceramic capacitor that can supply
684 µA and requires 42.8 mm3 , but adds no idle power. Our selection
is also marked, a 1210 resistor selected away from the frontier to
provide ample tolerance in power dissipation and design flexibility
as we continue to develop the system. We selected a resistor just
over our minimum current threshold to minimize idle power draw.
Field Strength (µT)
Y Coordinate (in)
Voltage Sensing
Voltage sensing in this form factor requires a planar contact
method, a voltage divider, and an ADC to acquire the voltage signal digitally. Of these, the method to contact AC voltage is the
only requirement that cannot be solved with small, readily available
components. We have identified two possibilities for contact: small
spring loaded pins and flexible tabs built into the PCB itself.
A flexible tab built into the PCB bends to provide contact as the
AC plug is inserted (this is the contact method shown in Figure 1).
The benefit is easy manufacturing: the process of printing the circuit
board provides the contact method. The tradeoff is in longevity. The
flexible materials are not designed for elasticity, and after several
insertions the flexible tab no longer provides a strong contact.
Another option is to mount a spring loaded pin sideways in the
plane of the PCB. If the tip of the pin is rounded, the AC plug can
slide past it as the spring-loaded pin applies contact. Such components are available off the shelf [10], and since they are designed
to be compressed, they will maintain contact over more insertions
than flexible tabs. The tradeoff is in the difficulty in manufacturing:
mounting the pins is currently a manual procedure. We have yet to
identify a process to reliably and repeatably mount the pin.
X Coordinate (in)
X Coordinate (in)
Y Coordinate (in)
Figure 5: Superposition of magnetic fields around the prongs of an
AC load plugged through PowerBlade. The top figure shows the
relative magnitude of the field and the bottom shows the direction
and relative magnitude, both with a 1 A constant current. The current
flows in opposite directions in the prongs, so the fields add constructively between the prongs and destructively outside. Between the
prongs, the field is largely oriented vertically. This model allows us
to optimally position and orient a wirewound inductor in the field to
maximize induced voltage.
Current Sensing
Many existing current sensing techniques, like shunt resistors, are
planar, but rely on interrupting an AC conductor, which PowerBlade
cannot do. Instead, PowerBlade senses current non-intrusively by
detecting the magnetic field surrounding an alternating current. As
current passes through the AC prongs inserted through PowerBlade,
the charge moving through each prong generates a magnetic field.
Those fields add constructively between the prongs and destructively
outside the prongs. Figure 5, generated by applying the Biot-Savart
law, shows the relative strength and orientation of this magnetic field
in the plane of PCB surrounding the prongs.
Two aspects of this figure guide the optimal placement of a sensor
to measure this field. First, the magnitude of the field is on the order
of 10-100 nT, which establishes bounds on the required transducer
sensitivity. Second, although intuition might suggest that the constructive fields would be maximized directly between the prongs,
the rapid decrease in field strength with distance from the conductor
means the strongest signal is closest a prong.
Hall effect sensors and other vector magnetometers are capable of sensing this magnetic field, but packaged units cannot meet
PowerBlade’s form factor needs or power requirements. Many such
devices also fail to meet the requirements of this system’s design
due to low sensitivity or low sampling rates. Instead, we observe
that a surface mount wirewound inductor placed in the field can act
as an inductive sensor or magnetometer. The alternating current in
the wire causes a changing magnetic field which passes through the
coils of the inductor and generates a voltage. A coil magnetometer
used in this way is also known as a search coil [29].
The equation governing the voltage generated in the coil can
be determined from Faraday’s law of induction and is shown in
Equation (1), where µ is the magnetic permeability, N is the number
of turns in the coil, and A is the cross-sectional area of the coil. Note
that voltage is proportional to the change in magnetic field strength
over time. This means that voltage on the inductor is proportional to
the change in current over time, rather than the current itself. The
signal must be integrated to recover the original current waveform.
V = −µ N A
Power Calculation
If the form factor or power requirements prevent the use of a
dedicated metering chip, as is the case for PowerBlade, then the only
remaining option is to implement custom measurement software in a
low power microcontroller. However, the choice of microcontroller
must balance the fidelity of measurements with the availability of
power. For example, a higher sampling rate will improve measurement accuracy, but it will also draw more power due to increased
data conversion rate and more frequent processor wakeups. Similarly, the measurements themselves must be scaled from raw “ADC
counts” to power statistics (W, VA, etc.) through various transfer
functions that may require floating point arithmetic. The floating
point operations could be performed on the PowerBlade unit itself,
power and performance permitting, or they may be applied in the
receiver. The particular operating point depends on a balance of
many variables.
Energy Quanta (µJ)
Idle Power Optimal
Volume Optimal
Recharge Rate (Packets/second)
Figure 6: Maximum energy quanta vs recharge rate for two options
for ZIN . VZ is fixed at 10 V. Both optimal volume and optimal idle
power draw are shown, and increasing COU T results in a higher
energy quanta at a slower recharge rate (upper left corner). The
minimum energy to boot and send a packet is shown for three radio
transceivers. For a given ZIN , the maximum packet transmission
rate for each transceiver lies at the intersection of the dotted line
with the given curve.
Data Communication
Data communication often requires a continuous burst of power.
For a wireless radio, this is the energy required to send a single
packet, but even a display screen must display for a minimum duration to allow a human to read it. In some cases, the current required
may be more than what is available, and if so, the power supply must
store at least the energy required for a single such event (referred to
here as energy quanta). How much energy is available is determined
by three components from Figure 2: ZIN , VZ , and COU T .
Increasing COU T will increase the energy available, but at the
expense of a slower recharge rate. Decreasing ZIN results in greater
current supply, which will increase both recharge rate and energy
quanta (as additional current is available during the discharge event).
Increasing VZ will increase both energy quanta and recharge time,
but will increase energy quanta more due to the quadratic term in energy calculation. Higher VZ values are therefore preferred, but VZ is
also constrained by other components. Small options for COU T are
commonly limited to 16 V, and many of the commercially available
miniature buck converters offer significantly higher efficiencies in
the 10-15 V range.
Figure 6 shows the recharge rate and energy quanta for two of the
three pareto-optimal options for ZIN from Section 3.1. The current
optimal option is not shown; it could supply sufficient current to
continuously operate certain radios but is undesirable due to its
high idle power draw. VZ is fixed at 10 V, and for a given curve,
increasing COU T results in moving up and to the left. The range of
capacitance shown for each curve (up to 22 µF) is readily available
in a variety of small ceramic packages.
Also shown is the minimum energy required to boot and send a
packet for three possible radios, the CC2420 [15], LTC5800 [8], and
nRF51822 [11]. The measurement of the CC2420 was performed
by Yerva et al. [31], the figure for the LTC5800 is available in its
datasheet, and we measure the energy for the nRF51822 ourselves.
The energy quanta figures need not be exact; their purpose is to drive
the selection of a radio.
For a given radio and selection of ZIN , the maximum recharge
rate, in packets per second, is the intersection of the two lines. This
can be used to determine the possible data rate or, if the required
recharge rate is lower, ZIN can be reduced from the optimal to
reduce idle power. The latter is the case in PowerBlade.
Figure 7: PowerBlade PCB layout with key features labeled. Exact
board dimensions are 1" × 1" × 0.066". The position of the sense
inductor, placed close to the neutral prong, is marked.
Idle Power
22 nF (120 kΩ)
33 nF (80 kΩ)
56 nF (47 kΩ)
100 kΩ
80.6 kΩ
75 kΩ
456 µA
705 µA
1161 µA
550 µA
687 µA
733 µA
451 µA
740 µA
1137 µA
538 µA
658 µA
716 µA
43.0 mm3
43.0 mm3
95.0 mm3
2.6 mm3
3.0 mm3
3.0 mm3
0 mW
0 mW
0 mW
140 mW
170 mW
190 mW
Table 2: Expected and actual maximum current for various possible
component choices for ZIN in the supply in Figure 2. Also shown
are the volume required and cost of each component, as well as its
idle power draw when the system is connected to AC.
This section describes how the components identified in Section 3
are integrated into a system in PowerBlade. This section also covers
the steps required to operate the system as a true power meter,
including how the wireless system is used and how the meter is
Power Supply
We explore several components for ZIN to evaluate their performance. Table 2 shows the expected maximum current from the
supply and our experimentally measured actual current, as well as
the volume, cost, and idle power that result from that component
being selected. Resistors are able to deliver comparable current to
that of capacitor shunts for a smaller volume and lower cost, but
they increase idle power draw even when the load is powered off.
Although early versions of PowerBlade had footprints for both a
2220 package capacitor and 1210 package resistor, current designs
optimize for size and cost by only providing space for the resistor.
In addition to the significantly smaller size, the low cost of a resistor
outweighs the cost added by its idle power. At $0.12 per kilowatthour, the 170 mW added by an 80 kΩ resistor would outweigh
the cost of a 33 nF capacitor only after nine years of continuous
operation. Current implementations use an 80.6 kΩ resistor, which
leads to a design constraint for the rest of the system in that it
must operate below a maximum average current of about 658 µA.
Summing the power draw of the system and the idle power draw
of the supply, this leads to an overall power draw of 176 mW for
PowerBlade. Our present choices reflect the desire for frequent data
transmissions supporting interactive use, but we note that higher
ZIN values are also possible with a concomitant reduction in data
transmission rate.
Voltage Sense
R I1
R I2
Current Sense
Figure 8: Cross-sectional layer stack-up in PowerBlade. Shown are
the entire stack-up (right), as well as the stack-up of the tab itself
(left) that contacts the AC plug. Using a rigid-flex PCB allows for
easy manufacturing with multiple insertions.
The remainder of the components from Figure 2 are a 10 V Zener
diode and 250 mA rectifier diode, both in a small SC-79 package,
and a combined 55 µF from two parallel capacitors for COU T . We
select a larger capacitance than the minimum required from Figure 6
to allow tolerance as we continue the design, and the added delay is
only experienced when the unit is first powering on. The load will
nominally see 8.9 V: the Zener voltage minus the rectifier forward
voltage of 1.1 V. To regulate this output to 3.3 V we use a 3.3 V buck
regulator, the Texas Instruments TPS62122 [16]. At the nominal
input voltage of 8.9 V this regulator has an efficiency of 85-90%.
The single-PCB system shown in Figure 1 is a rigid-flex PCB—a
stack-up of layers of both rigid and flexible materials with conductive copper layers sandwiched in between. For our proof of concept
we select flexible tabs over spring loaded pins due to the ease in
manufacturing. This method of circuit board construction maintains
good contact for approximately 10-30 insertions, providing a pathway to a proof of concept, but we intend to continue to explore
methods using spring loaded pins in future work.
Figure 8 shows a cross section of the flexible tabs. The innermost
layers of the PCB are all flexible, the inner layers 2 and 3 of copper
are fixed to a flexible polyimide core. These layers extend into the
tabs to make contact with AC. They continue throughout the entire
system, but are rigidized with FR4 material between this flexible
core and the outer copper layers.
Extending only 5 layers into the tab provides insufficient support;
the tabs bend on the first insertion and never return to form. Instead,
we add an additional polyimide/adhesive pair on each side of the
tab, with only the tip remaining exposed for contact. After exploring
several options for dover , we find that fully covering one side of the
tab while exposing only the tip on the other side allows for maximum
insertions. This requires loads be plugged through PowerBlade from
back to front, but NEMA polarization means most loads already
plug from this direction.
Measurement Amplifiers
Figure 9 shows the amplifier circuits used to measure voltage
and current. PowerBlade measures line voltage directly through a
voltage divider with a VCC /2 offset to measure both positive and
negative phases. RF = 4.99 kΩ and RI1 = RI2 = 953 kΩ, so
VSEN SE can be approximated as Equation (2).
Figure 9: Voltage and current sensing circuits. Voltage sensing requires little volume, consisting of a voltage divider and amplifier
with a VCC /2 offset. Current sensing uses a horizontally wirewound
inductor, amplified and filtered. The resulting signals VSEN SE and
ISEN SE are measured by the MSP430.
− 5.24 × 10−3 VAC
Based on these configurations, the voltage signal has a peak to
peak amplitude of 1.79 V where the AC is 120 VRM S (United States)
and 3.28 V where the AC is 220 VRM S (much of Europe and China).
PowerBlade measures the signal from the sense inductor in multiple stages. The inductor is referenced to 250 µV and amplified in
two stages with a combined gain of about 6100x. Low frequency
noise is removed with a high pass filter between the first and second
stages, and this filter is referenced to 54 mV so the final signal is
centered around VCC /2.
Equation (3) describes the output of the current sense stage as a
function of the derivative of the AC current, where α is a lumped
parameter consisting of the characteristics of the coil, gain, signal
distortions, and general uncertainty. After integration the current is
represented by Equation (4), where β accounts for DC offsets in the
system and integration offsets.
Z Current ≈
dt ≈ α I + β
Calibrating PowerBlade
PowerBlade calibration requires two steps. The first step is to
measure the scaling and offset values α and β, which must be done
once per design. The second step is a device-specific calibration that
accounts for slight variations between units.
To determine the scaling and offset values α and β, respectively,
we measure the reported RMS current from PowerBlade for a range
of resistive (unity power factor) loads. Figure 10 shows the RMS
values of current reported as raw values. For the current system, these
measurements are linear with an R2 value of 0.999, and indicate
an α of 40.85 and β of 25.0. Subtracting the offset β happens in
PowerBlade, and for further testing, all units comprising a batch are
programmed with this value.
Uncalibrated RMS current
(arb. units)
Actual RMS current (A)
Figure 10: PowerBlade reported uncalibrated RMS current. These
samples are used to calculate α and β from Equation (4) and allow
us to refine our power calculations.
To reduce the computational burden on PowerBlade, we divide by
α in the receiver, and for increased accuracy, each unit is calibrated
again. We connect each PowerBlade in a batch to a 200 W load to
compute a device-specific α, store that value in FRAM, and transmit
it with each packet. We have observed a mean value for α across
multiple units of 41.79, with a 95% confidence interval of 1.87.
Finally, after calibrating α for each device, we notice an exponential error not accounted for by the model. We experimentally
determine the following correction P = PRAW −6.6e−0.015PRAW ,
where PRAW is the power (in watts) after applying the α and β corrections, removes this error and yields the final power figures. The
values of 6.6 and -0.015 are determined once per design.
Power Calculation
We save the overhead of a power metering IC by performing
our power metering calculations on a MSP430FR5738. This chip
is the master controller of the system and is the only component
not automatically power gated at startup. It is selected due to its
small size, power efficiency of 81.4 µA/MHz, and integrated 16 kB
non-volatile FRAM, eliminating an external component.
To measure power, the MSP430 samples VSEN SE and ISEN SE
at 2.52 kHz (42 samples per AC cycle). This frequency is both an
even divisor of our timing clock (32,768 Hz) and an even multiple
of the frequency to be sampled (60 Hz). Because ISEN SE is proportional to the derivative of current, the second measurement step
integrates ISEN SE to obtain current (VSEN SE is a good representation of voltage). The integration is performed in software, but could
be performed by hardware in the future. The final calculation step
involves calculating power from voltage and current.
Because the MSP430 has real-time access to both voltage and
current waveforms, it can function as a true power meter. Real power
is determined by multiplying voltage and current at each point, and
then averaging over the number of samples. Apparent power is determined by first calculating the root mean square voltage and current
over a cycle, VRM S and IRM S , respectively, and then multiplying
them. Knowledge of both real and apparent power allows the system to determine reactive power as well as the power factor of the
load. Real power is also aggregated in the MSP430 to compute total
watt-hours measured over time, and this number is stored in FRAM.
BLE Communication
Data are transmitted in broadcast-style BLE advertisement messages. The MSP430 first communicates to the nRF51822 via UART
at 38,400 baud, and the nRF51822 repeats this data in the advertisement. The MSP430 sends UART data nominally at 1 Hz, and the
nRF51822 sends advertisements at 5 Hz, so 4-5 identical packets
are transmitted each second. This greatly increases the likelihood of
reception, and does not dramatically affect the power draw.
In addition to a sequence identifier and information regarding
versioning and scaling, each PowerBlade packet contains four fields:
line voltage, instantaneous real power, instantaneous apparent power,
and watt-hours. Real power and apparent power are 1-second averages, and can be used together to calculate power factor. Watt-hours
is an over-time total, and in the event of zero packet loss watt-hours
will, once scaled, also equal the integral of real power.
PowerBlade is robust against packet loss. The intended recipient
of broadcast advertisements is either a smartphone or a fixed BLE
receiver, but in the event of no receiver, only the resolution of the
missed packet is lost. The overall watt-hours total remains an accurate reading in any received packet. Watt-hours is stored as a 32-bit
number and can overflow. In the worst case with present calibration
scaling values, measuring an 1,800 W load will lead to an overflow
every 29 days. A 100 W load will overflow after 523 days of continuous measurement. Overflows are signaled in the advertisements
so the true watt-hours reading can be recovered. If a receiver is not
present for long periods of time, potential for data loss exists.
System Operation
Figure 11 shows the startup phase of PowerBlade and 7 s of
steady-state operation. When the system first starts, there is only
power to boot the MSP430; if more components are drawing power,
the 3.3 V power rail will never enable and the system will lock up.
Instead, MOSFETs separately power gate the sensing circuits and
BLE radio. With only the MSP430 running, the capacitor charges,
enables the 3.3 V rail, and eventually reaches a nominal 8.9 V.
When the MSP430 detects that the capacitor has charged to the
nominal voltage it enables the sensing circuits, and these remain
powered for as long as PowerBlade is powered. The MSP430 spends
1 s collecting measurements before enabling the nRF51822, which
also remains powered for the duration of the trace. At this point the
device has entered steady-state operation.
We evaluate PowerBlade on the basis of accuracy in reporting
real power for both a calibrated resistive AC load and an assortment
of household loads. We also present benchmarks that affect the usability of the system, including PowerBlade’s volume, cost, wireless
performance, and safety of using the system.
Power Metering Accuracy
We evaluate PowerBlade’s accuracy in two parts. First, we explore
bench top accuracy, where we use PowerBlade to measure the power
draw of a programmable AC load—the APS 3B012-12 [2]—set
to unity power factor. This allows us to measure a large part of
the metering range (up to 1200 W) in defined increments and in
a controlled setting. Second, we use PowerBlade to measure the
power draw of several household items. Although not an exhaustive
list, this is representative of PowerBlade’s target usage.
For the bench top accuracy, ground truth is provided by the 3B01212 itself via its serial interface. For the household tests, ground truth
is taken from two sources. On the low range, we use a professionally
calibrated Power Line Meter (PLM) [5]. This device is limited to
480 W, however, so to measure larger loads we take ground truth
from a Watts Up [17]. We report the Watts Up measurements over
the low range as well, and it is clear that it is less accurate than the
PLM (typical error of about 1.72%). The need to use two different
meters for ground truth demonstrates the difficulty in creating a
highly accurate whole-range metering solution. For each test we
take 30 PowerBlade measurements, 30 ground truth measurements,
and report the arithmetic mean and 95% confidence interval.
Steady State
System Current (mA)
Advertisements (x13)
UART (x3)
Regulator Powered
MSP430 Boot
Capacitor Voltage (V)
Time (ms)
Figure 11: PowerBlade timing: storage capacitor voltage and 3.3 V regulator output current for the first 7 s of operation. At startup the
measurement circuits and the nRF51822 are automatically disabled. For approximately 2 s the capacitors charge gradually, and around 2.25 s
the MSP430 boots. When the MSP430 detects the capacitor has charged to the nominal voltage of 8.9 V it enables the voltage and current
measurement circuits. For the first 1 s of the amplifiers being powered there is no data to transmit, so the nRF51822 is kept disabled. After 1 s
of measurement the MSP430 enables the nRF51822 and sends data over UART. At this point the nRF51822 begins advertising data at 5 Hz for
the remainder of operation and the MSP430 updates with new information over UART at 1 Hz. This results in up to five identical packets
transmitted, which increases the likelihood of reception, and a sequence number transmitted with each packet denotes the duplicate.
Power Factor
Measured Power (W)
Actual Power
Measured Power
Measured Power Factor
Actual Power (W)
Figure 12: Metering accuracy for a variable resistive (power factor
equals 1) load: measured power vs actual power as well as measured
power factor. Also shown are the 95% confidence intervals for real
power. The minimum AC load for accurate metering is about 2 W
and over the range from 2 W to 1200 W the average accuracy in
real power is 1.13%. PowerBlade’s metering and reporting system
is accurate over a range of resistive loads.
150 W Bulb
Drill (Max)
Hot Air
TV (Normal)
50 W CFL
TV (Static Image)
Drill (Low)
Resistive Loads
Resistive loads with a unity power factor, which include incandescent lights and power-factor-corrected devices, exhibit a sinusoidal
current waveform in-phase with voltage. To measure PowerBlade’s
accuracy in this simple but common case, we use an APS 3B012-12
programmable AC load set to a fixed unity power factor. Figure 12
shows the end-to-end accuracy for PowerBlade metering this resistive load. Displayed are the reported real power and power factor
from PowerBlade, as well as the ground truth power up to the programmable load’s maximum of 1200 W. Note that the true power
factor is equal to one throughout the test.
We measure 29 wattages from 2.2 W to 1200 W: 50 W to 1200 W
in increments of 50 as well as 2.2 W, 5 W, 10 W, and 75 W. For these
measurements the average error is 2.3 W and the average percent
error is 1.13%. At 2.2 W the error is 0.21 W (9.5%) and at 1200 W
the error is 7.01 W (0.6%).
162.17 W
108.22 W
235.21 W
827.87 W
1246.96 W
1729.73 W
305.54 W
196.23 W
48.57 W
129.51 W
50.44 W
52.68 W
106.63 W
9.11 W
51.10 W
Watts Up Error
1.22 W
1.69 W
0.88 W
0.86 W
-1.08 W
-0.04 W
0.64 W
-0.41 W
1.95 W
0.22 W
4.18 W
PowerBlade Error
-0.99 W
-5.30 W
2.96 W
-22.11 W
15.24 W
16.01 W
-1.93 W
-9.03 W
-9.51 W
-4.00 W
-0.83 W
-4.49 W
36.97 W
-0.62 W
20.40 W
Table 3: Metering accuracy for a cross section of household devices.
On this selection, the average percent error in real power is 6.5%.
Although these devices produce more complex waveforms than a
fully resistive load, PowerBlade remains acceptably accurate.
Household Devices
Resistive loads constitute a large fraction of household devices,
but not all loads have sinusoidal current waveforms. Table 3 shows
PowerBlade’s accuracy for several devices found in a common
household. For these devices, we simultaneously take a PowerBlade
measurement, a Watts Up measurement and, if the load is below
480 W, a PLM measurement. If available, the PLM is used as ground
truth. If not, the Watts Up is used. Although the fridge draws about
100 W, its start-up power tripped the 4 A fuse in the PLM, preventing PLM measurements for that load. For each device, we report its
power and power factor, as well as PowerBlade’s error and, if not
used as ground truth, the Watts Up error for comparison.
This set of devices has a range of power from 9 W to 1730 W and
a range of power factors from 0.30 to 1. The average absolute error
for PowerBlade measurements of these devices is 10 W (4.3x higher
than the resistive load), and the average percent error is 6.5% (5.8x
higher). This error is dominated by two devices with highly inductive
(low power factor) draws: the blender (PF=0.49) and the drill set
Watt-hour Accuracy
PowerBlade is designed to measure and report both instantaneous
power and watt-hours, the sum over time that will be used by the util-
Hot Air Gun - 247W, PF=0.75
Drill (Low) - 49W, PF=0.28
Current Waveforms
Watts Up Power
PLM Power
PowerBlade Power
PLM Power Factor
Time (min)
Figure 14: Metering accuracy over time for a television in use.
Shown on the figure are reported power from the PLM (ground
truth), Watts Up, and PowerBlade, as well as power factor reported
from the PLM. At the end of 15 minutes the PLM reported 49.07 Wh,
the WattsUp 49.28 Wh (0.42% error), and PowerBlade 46.80 Wh
(4.62% error). This is consistent with PowerBlade’s instantaneous
error of 4.60% for the television in full use.
ity company to levy charges. The figure of watt-hours also accounts
for the possibility that one or multiple packets are not received:
resolution is lost but watt-hours remains an accurate long-term measure. To evaluate PowerBlade’s accuracy in reporting watt-hours,
we take simultaneous measurements from the PLM, Watts Up, and
PowerBlade for a television in normal viewing use. Figure 14 shows
the measurements over time from Watts Up and PowerBlade as
compared to the PLM.
After 15 minutes of normal television use, the PLM reported
49.07 Wh, the Watts Up reported 49.28 Wh (0.42% error), and
PowerBlade reported 46.80 Wh (4.62% error). In instantaneous
measurement trials the PowerBlade measurements for the television
in full use were off by an average of 4.60%, the watt-hours figure of
4.62% error is consistent with the instantaneous readings.
Drill (Max) - 229W, PF=0.98
Power Factor
Power Measurement (W)
to low power (PF=0.30). Each of these devices has an error of over
20 W and a percent error of over 30%, and excluding these two, the
average absolute error drops to 7.2 W (3.1x the resistive load) and
the average percent error drops to 4.3% (3.8x the resistive load).
Correctly metering such devices is an area of future development.
The difficulty in measuring highly inductive loads, and further
the difference in accuracy between the programmable load and the
household devices, can be partially explained by examining the current waveforms. Figure 13 visualizes the current measurement process in PowerBlade for a few loads from Table 3. The known current
signal, measured by a commercial current transformer [7], is shown
along with ISEN SE , the signal output from PowerBlade’s current
amplifier (which, as described in Section 3.3, represents the derivative of the current waveform). Also shown is the post-integration
representation of current. Voltage for each load is synchronized, and
zero crossings of the common voltage are denoted by vertical lines.
Visible on the figure is the integral/derivative relationship between
ISEN SE and known current, as well as the fidelity of the integrated
signal to that known current. For devices with sinusoidal or otherwise smooth current waveforms, the integrated signal tracks well
with known current. For other devices, however, high frequency components in the current waveform are suppressed by the integrator,
resulting in increased error.
Usability Benchmarks
PowerBlade’s accuracy makes it comparable to other power metering systems, but it is the usability of the system, and in particular
the size, cost, and wireless communications, that most distinguish
it. We also show through standard safety testing that PowerBlade is
safe to use with the addition of an enclosure.
Blender - 115W, PF=0.50
Resistive Load - 250W, PF=1.00
The defining characteristic of PowerBlade is its volume: the entire
system is a single PCB. This circuit board is 1.0" on a side, and the
PCB itself is 0.023" thick. The thickest component on the surface is
the antenna at 0.043", so the combined total thickness of the system
is 0.066". This is the same thickness as the pass-through section of
the FlipIt charger, which is a certified commercial product.
Actual Current
Measured Current
Absolute Error
Figure 13: Current sensing fidelity. Output of current sense amplifier, ISEN SE , along with the internal representation of current, are
shown for several household devices. Also shown is the true current
waveform as measured by a commercial current transformer [7].
Visible is the derivative relationship between actual current and
ISEN SE , as well as the distortions introduced by this sensing and
integrating technique. This shows that PowerBlade’s current sensing
method reasonably captures the current waveform.
The component breakdown in PowerBlade with costs is listed
in Table 4. Prior to consumer use, PowerBlade needs an enclosure,
but the system could be largely assembled for $10-$15 per unit.
Although it is important to note the distinction between the cost
of PowerBlade and the price of other systems, this is slightly less
than the price of Kill-A-Watt ($23.99) and significantly less than the
price of Watts Up ($130.95). The cost of $10-$15 for PowerBlade is
also an un-optimized reporting of DigiKey pricing; the minimum
viable cost would likely be much lower.
Sense inductor
Other passives
Table 4: Cost for the PowerBlade system. The total system is roughly
$11 in quantities of 1,000. We believe this represents an acceptable
price point for effective plug-load metering deployments.
Wireless Range
We test the effectiveness of PowerBlade communications by measuring packet reception rates in three configurations of PowerBlade
units. First we deploy and evaluate a single PowerBlade as a baseline. Next we place three PowerBlade units throughout a room as
a more typical deployment case. Finally, we place three units on
a single power strip and activate them simultaneously to test for
possible packet collisions. We record both unique and total packets
received per second. PowerBlade updates data at 1 Hz, and BLE
packets are sent at 5 Hz, so nominally we should receive 1 unique
and 5 total packets per second.
For each configuration we evaluate packet reception rate at three
transmission distances, and perform the entire experiment both in
an apartment and in a lab. The apartment consists of three rooms,
with measurements taken in the same room, adjacent room, and two
rooms away. Only a single other BLE device is active in the apartment. The lab consists of one room and hallways, and measurements
are taken in the same room, immediately outside of the room in
the hallway, and 20 m down the hallway from the room. The lab
environment includes 16 non-PowerBlade BLE devices as well as
numerous other devices active in the 2.4 GHz band.
Figure 15 shows the reception rate for each of these trials. In all
cases, the unique reception rate is at or close to the nominal of 1 per
second when in the same or adjacent rooms, but the total reception
rate decreases from the same to the adjacent room. Further, the total
reception rate is higher in the apartment than in the lab for both
distances. Taken together these three results suggest that range and
interference do effect BLE transmissions, but the redundancy in
PowerBlade helps ensure reliable data communication.
The distant measurements show continued decline in total packets,
but also a decrease in unique packets: 20% to 50% of the unique
packets are not received at all (all five redundant packets were all
dropped). This indicates that this distance—whether two rooms
separated in a residential setting or 20 m down a hallaway in a
university building—exceeds PowerBlade’s usable range. We note
that RF designs require some degree of lumped-parameter tuning to
achieve maximum performance but PowerBlade’s RF circuitry has
not be tuned yet, so that may help explain these results.
PowerBlade is a prototype not designed for consumer use in its
current form, but safety when using and deploying the system is
still important. The UL safety standard that covers PowerBlade
is UL 2735: Standard for Safety in Electric Utility Meters [18].
Although we do not have the resources at this time to perform an
exhaustive safety evaluation, the enclosure requirements and single
fault and mechanical tests in UL 2735 are good initial benchmarks
for exploring the overall safety of the system.
Unique Packets per Second
Antenna, balun, & crystals
Buck converter
Single, Apartment
Separate, Apartment
Close, Apartment
Single, Lab
Separate, Lab
Close, Lab
Same Room
Adjacent Room
(a) Unique packets received per second
Total Packets per Second
Same Room
Adjacent Room
(b) Total packets received per second
Figure 15: Wireless reception rate for PowerBlade. Unique packets
per second as well as total packets per second are recorded for
each test. PowerBlade units are deployed singly, throughout a room,
and adjacent on a power strip. The lab environment includes many
other background sources of 2.4 GHz communications which may
interfere with PowerBlade broadcasts. When in the same room as or
adjacent room to a PowerBlade deployment, advertised power data
is received once per second with a high probability.
The primary safety requirement is preventing electric shock by the
unit. Figure 16 shows a PowerBlade unit after coating in a resistive
material that protects the user from live parts. The resistive material
does not noticeably affect wireless performance, and increases the
total volume to 0.074 in3 (12% increase). This currently requires
a labor intensive application process, but for future builds we have
identified a supplier capable of adding a heat-resistant overmolding
in a scalable manufacturing process.
In addition to applying the conformal coating, we evaluate the
system based on two sets of safety tests from UL 2735. The first is
single fault testing for the electrical system, to determine if a failure
in any one component could result in unsafe conditions. We identify
three components that, in the event of a fault, will overload other
components. In the power supply, ZIN failing closed (short circuit)
will expose the system to increased current, and DZ failing open
will no longer limit VCAP to 9 V. In the voltage measurement circuit,
RI1 failing closed (short circuit) will expose multiple components
to 120 V. Circuit analysis indicates that ZIN and RI1 failing open
will result in no connection (no testing required). DZ failing short
will simply prevent charging, although we do test this condition.
As expected, shorting either ZIN or RI1 results in secondary
faults in other components. Shorting ZIN results in open circuit
faults in both D1 and DZ (D1 is visibly damaged) but after 30
minutes there is no detectable change in temperature in the unit.
Shorting RI1 results in open circuit faults and visible damage in both
RF and the voltage measurement amplifier. This test also results in
a bright spark, audible pop, and momentary increase in temperature
(to 33 degrees Celsius), but the temperature quickly decreases, and
after 30 minutes there is no detectable ongoing change. Opening
DZ results in increasing VCAP to 23 V, but the 3.3 V regulator
continues to operate as normal and the system continues to function.
Shorting DZ results in no charging, but also no unsafe conditions.
Figure 16: Conformally coated PowerBlade. Shown are the unit
after conformal coating has been applied, as well as the conformally
coated unit attached to a load. Although this method requires a
manual application process that is not manufacturable in large scale,
it does protect the user from live parts. In the future an overmolding
process will provide similar protection in a manufacturable way.
The second set of tests from UL 2735 is mechanical, to determine
if the protective coating in Figure 16 can withstand real world conditions. Note that this conformal coating is a temporary solution;
we use it to determine its effectiveness. The future overmolding
technique will result in more durable protection. The conformally
coated PowerBlade passes the static force and drop tests in UL 2735
without damage, but the impact test does destroy the unit and expose
live parts. The live parts do not extend to the edge, however, and
therefore do not present a safety hazard when the unit is plugged in.
In this section, we discuss some limitations of the current design,
explore some possible workarounds, and propose some future directions for improvement. In particular, the accuracy could be further
improved, the wireless system could be better utilized, and interval
data could be collected with periodic timestamps.
Improving Accuracy
We identify two changes that could be taken in future designs
to improve accuracy. First, the MSP430 microcontroller software
algorithm that integrates ISEN SE to obtain the current waveform
could be changed. Although the present implementation provides
a usable signal, we believe a hardware implementation may better
address baseline drift. Second, the acquisition of the current and
voltage waveforms could be more tightly synchronized. For example,
if the samples are acquired with a separation of 500 µs, that delay
will introduce an measurement error of 6% on a 150 W resistive
load. PowerBlade has a 22.2 µs delay between current and voltage
sampling, but synchronous acquisition of the signals would be ideal.
Although error in real power is only 1.13% for the resistive loads
and 6.5% for household devices, PowerBlade’s error in power factor
is often higher. Figure 12 shows the accuracy in power factor (which
is actually accuracy in apparent power) decreases significantly at
lower wattages. A similar effect is observed in the household loads.
Real power is the metric used to assess utility charges, so we believe
this to be the more important value, but we have yet to determine
the source of the error in apparent power.
Wireless Communications
The advertisement-only communication model that PowerBlade
employs presents two areas for improvement. First, any BLE receiver
can eavesdrop on these broadcasts as the data are transmitted in
clear-text, so data security is an obvious area of improvement. We
have verified that a nearby smartphone can respond to a PowerBlade
advertisement to open a BLE connection, showing that changes from
the advertisement-only model are also possible.
Interval Metering
PowerBlade can store accumulated energy in non-volatile memory, allowing it to aggregate total load usage. However, this data is
a scalar value representing total load since boot or since inception.
PowerBlade could be much more useful if it could serve as a finegrained interval meter, providing energy usage data broken down by
periodic intervals—typically 1, 15, or 60 minutes in duration—that
are synchronized in time with other meters and “wall” time as this
allows better visibility into energy use over time.
To support interval metering, PowerBlade needs a reliable method
of obtaining and keeping the time, likely a combination of a wireless
time synchronization protocol and a real-time clock. The primary
difficulty in adding a real-time clock (RTC) is limited long-term
energy availability in this form factor. To that end, we have identified
supercapacitors capable of storing sufficient energy to operate an
RTC for extended periods [12]. We intend to explore the viability of
interval metering in this form factor in future work.
The state-of-the-art in plug-load metering fails to provide consumers and corporations the detailed knowledge they need to understand and adjust their energy consumption patterns at a size, cost,
power, and usability point that permits widespread adoption. While
plug loads represent one of the faster growing segments of electrical
loads, existing systems for measuring them remain too expensive,
draw too much idle power, lack a wireless interface, and are often
too large or too cumbersome to easily deploy.
To address the gap between emerging needs and existing solutions, we present PowerBlade, a new power/energy meter design
that enables a new paradigm in metering by making the sensor so
small and unobtrusive that it can be permanently attached to a plug
rather than an outlet. PowerBlade accurately meters the power of a
load in real-time and wirelessly transmits that data to nearby smartphones or gateways using a Bluetooth Low Energy radio. With a
thickness of a mere 1/16", PowerBlade is the first power meter that is
truly plug-through. Realizing this diminutive form factor, however,
requires revisiting all of the key design choices for a power meter.
In this paper, we introduce several new methods—voltage sensing,
current sensing, and power supply miniaturization, among others—
to realize a new metering design point. This design offers accuracy
within 1.13% of ground truth on a resistive load and 6.5% on a selection of non-unity power factor household devices. Furthermore, we
show that the system’s volume, cost, wireless capability, and safety
combine to make a usable and deployable system. With this new
design in hand, researchers, ratepayers, and regulators will be able
to inexpensively gain new insights into electricity usage patterns,
hopefully yielding smarter and more energy-efficient choices.
We thank our shepherd, Jie Liu, and the anonymous reviewers for
their helpful feedback. This work was supported in part by the TerraSwarm Research Center, one of six centers supported by the STARnet phase of the Focus Center Research Program (FCRP) a Semiconductor Research Corporation program sponsored by MARCO
and DARPA. This material is based upon work supported by the
National Science Foundation under grant CNS-1350967, by the NSF/Intel Partnership on Cyber-Physical System (CPS) Security and
Privacy under Award proposal title “Synergy: End-to-End Security
for the Internet of Things, NSF proposal No. 1505684.”, and by the
Graduate Research Fellowship Program under grant number DGE1256260. This work partially supported by generous gifts from Intel
and Texas Instruments.
[1] ADE7753 energy metering IC. http://www.analog.com/en/
[2] APS 3B series datasheet.
[3] Belkin conserve insight energy use monitor.
[4] Coilcraft flyback transformers Y8844-AL.
[5] Electronic product design Inc. PLM-1.
[6] Kill-a-watt electricity use monitor.
[7] LEM TT 100-SD current transformer.
[8] Linear Technology LTC5800 wireless mote-on-chip.
[9] LNK302 off-line switcher IC. http://www.mouser.com/ds/2/
[10] Mill-Max discrete spring loaded contacts.
[11] Nordic Semiconductor Bluetooth Smart SoC.
[12] Panasonic electric double layer capacitors.
[13] SR086 adjustable off-line inductorless switching regulator.
[14] SR10 capacitor-coupled, switched shunt, (CCSS) regulator.
[15] Texas Instruments CC2420 RF transceiver.
[16] Texas Instruments TPS62122 step-down converter.
[17] Watts Up? Pro plug load meter.
[18] Standard for safety in electric utility meters. Technical Report
UL2735, UL, May 2013.
[19] Electric power monthly with data for June 2015. Technical
report, U.S. Energy Information Administration, aug 2015.
[20] B. Campbell and P. Dutta. Gemini: A non-invasive,
energy-harvesting true power meter. In Real-Time Systems
Symposium (RTSS), 2014 IEEE, pages 324–333. IEEE,
[21] R. Condit. Transformerless power supplies: Resistive and
capacitive. http://ww1.microchip.com/downloads/en/
AppNotes/00954A.pdf, 2004.
[22] B. Coxworth. Flipit lets you charge devices from outlets that
are in use. http://www.gizmag.com/flipit-usb-charger/20024/,
[23] S. DeBruin, B. Campbell, and P. Dutta. Monjolo: An
energy-harvesting energy meter architecture. In Proceedings
of the 11th ACM Conference on Embedded Networked
Sensor Systems, page 18. ACM, 2013.
[24] S. Gupta, M. S. Reynolds, and S. N. Patel. ElectriSense:
single-point sensing using EMI for electrical event detection
and classification in the home. In Proceedings of the 12th
ACM international conference on Ubiquitous
computing, pages 139–148. ACM, 2010.
[25] C. J. Hansen. Internetworking with Bluetooth Low Energy.
GetMobile, 19(2), Apr 2015.
[26] X. Jiang, S. Dawson-Haggerty, P. Dutta, and D. Culler. Design
and implementation of a high-fidelity AC metering network.
In Information Processing in Sensor Networks, 2009.
IPSN 2009. International Conference on, pages 253–264.
IEEE, 2009.
[27] S. Kwatra, J. T. Amann, and H. M. Sachs. Miscellaneous
energy loads in buildings. American Council for an
Energy-Efficient Economy, 2013.
[28] T. Schmid, D. Culler, and P. Dutta. Meter any wire, anywhere
by virtualizing the voltage channel. In Proceedings of the
2nd ACM Workshop on Embedded Sensing Systems
for Energy-Efficiency in Building, pages 25–30. ACM,
[29] S. Tumanski. Induction coil sensors a review. Measurement
Science and Technology, 18(3):R31, 2007.
[30] T. Wu and M. Srivastava. Low-cost appliance state sensing for
energy disaggregation. In Proceedings of the Fourth
ACM Workshop on Embedded Sensing Systems for
Energy-Efficiency in Buildings, pages 53–55. ACM, 2012.
[31] L. Yerva, B. Campbell, A. Bansal, T. Schmid, and P. Dutta.
Grafting energy-harvesting leaves onto the sensornet tree. In
Proceedings of the 11th International Conference on
Information Processing in Sensor Networks, IPSN ’12,
pages 197–208. ACM, 2012.