J. Sens. Sens. Syst.

, 5, 389400, 2016
www.j-sens-sens-syst.net/5/389/2016/
doi:10.5194/jsss-5-389-2016
Author(s) 2016. CC Attribution 3.0 License.

High-accuracy current measurement with low-cost

shunts by means of dynamic error correction
Patrick Wekamp and Joachim Melbert
Institute of Electronic Circuits, Ruhr-Universitt Bochum, Bochum, Germany
Correspondence to: Patrick Wekamp ([email protected])
Received: 18 August 2016 Accepted: 17 October 2016 Published: 8 November 2016

Abstract. Measurement of electrical current is often performed by using shunt resistors. Thermal effects due to
self-heating and ambient temperature variation limit the achievable accuracy, especially if low-cost shunt resis-
tors with increased temperature coefficients are utilized. In this work, a compensation method is presented which
takes static and dynamic temperature drift effects into account and provides a significant reduction of measure-
ment error. A thermal model of the shunt resistor setup is derived for this purpose and a suitable calibration
method is developed. The correction algorithm is based upon a digital filter bank and is optimized for micro-
controllers with low computational complexity. It is implemented in laboratory test equipment for long-term
studies on automotive lithium-ion cells. For a 600 A current pulse, it reduces the measurement error from 2 % to
less than 0.1 %. Measurements with a real-life testing profile show a reduction of remaining measurement error
by 60 %. Statistical results for 100 test systems and long-term drift measurements prove the reliability of the
method. The proposed dynamic error correction algorithm therefore allows high measurement accuracy despite
the use of low-cost shunt resistors.

1 Introduction significantly increases the total system cost (Ziegler et al.,

2009a; Ripka, 2010).
Measurement of electrical current is important for many sci- Shunt resistors therefore represent a low-cost alternative,
entific and industrial applications. One example is the grow- especially if electrical isolation is not required and isolation
ing number of hybrid and electrical vehicles, which require amplifiers can be avoided. This is the case for single-cell bat-
precise on-board measurement of current in the range of tery test equipment (Lohmann et al., 2015; Wekamp et al.,
several hundreds of ampere for tracking the state of charge 2016).
(SoC) of the battery. In addition, preliminary testing of the The value of the shunt resistor has to be chosen with regard
automotive battery cells is necessary to determine the elec- to signal dynamic range and power dissipation: the desired
trical behavior as a function of aging. voltage drop across the shunt for applications with high mea-
Suitable laboratory equipment for this task has even higher surement accuracy is typically in the range of several hun-
demands on the accuracy, bandwidth and noise of the current dred mV in order to reduce influence of amplifier noise, to
measurement. Frequencies of interest range from DC up to increase the effective resolution and to limit the thermoelec-
several tens of kilohertz. Sensor concepts for these applica- tric effect. On the downside, the high currents used for au-
tions include shunt resistors and current transducers based tomotive battery testing lead to significant power dissipation
on the Hall effect or on the fluxgate measurement princi- and self-heating of the shunt, resulting in a change in resis-
ple (Table 1). Although current transducers offer electrical tance described by the temperature coefficient TK . Without
isolation and negligible power losses, they are susceptible compensation, this leads to increased measurement errors.
to magnetic stray fields and temperature-dependent offsets, Straightforward solutions encompass application of resis-
require large operation currents, add additional noise from tances with very low temperature coefficients, better cool-
the control electronics and have limited measurement band- ing concepts or oversizing in terms of maximum power rat-
width, depending on the sensor type. Their complexity also ings, all of which might reduce the thermal effects below an

Published by Copernicus Publications on behalf of the AMA Association for Sensor Technology.
390 P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts

Table 1. Comparison of current measurement techniques suitable digital filter structure, a method to determine the correction
from DC up to several tens of kilohertz. parameters by an automated calibration sequence and statis-
tical results obtained during operation of 100 distinct mea-
Pro Contra surement systems.
Shunt Low complexity, Power dissipation, no
low cost, wide isolation, self-heating 2 Theory and principle of operation
bandwidth
Hall effect Isolation, no power Temperature-dependent 2.1 Current measurement for test equipment for
losses offset, limited accuracy
automotive lithium-ion cells
Fluxgate Isolation, no power Disturbance due to
losses, good accu- excitation frequency, For aging studies on automotive lithium-ion cells, special-
racy large operation current, ized test equipment is necessary to cycle the cells with re-
limited bandwidth alistic currents and to measure small variations in the elec-
w/o additional
trical behavior with high precision. Currents of up to 600 A
transformer, cost
with frequencies from DC up to the kilohertz range are typi-
cal (Wekamp et al., 2016). Due to the low voltage range of
a single battery cell, electrical isolation is not required and
acceptable threshold. In any case, these methods typically shunt resistors can be used for current measurement. Their
lead to increased material and space requirements as well as low cost compared to other solutions is advantageous for
higher cost. building a large number of test systems; additionally, their
Direct compensation methods are an alternative approach wide bandwidth and inherent accuracy for low currents are
and are known to improve the measurement accuracy: a tem- also favorable (see Table 1).
perature sensor measures the temperature of the shunt ma- The cell tester developed by our research group uses 12
terial and corrects the current measurement based on the 10 m thick-film shunt resistors (TO-247 package) in par-
known material characteristics (Ziegler et al., 2009b). allel connection, mounted on a heat sink with water cool-
However, this direct approach is limited if there is a ther- ing. Electrical connection is accomplished by copper bus-
mal resistance between the shunt material and the location bars. The overall resistance is approximately 1 m, taking
of the temperature sensor, for example, if the resistance is the resistance of the interconnection into account. A temper-
enclosed in a package and the shunt material is not directly ature sensor allows the measurement of the surface tempera-
accessible. This is the case for nearly all available shunt re- ture of the heat sink (Fig. 1).
sistors intended for heat-sink mounting. In this situation, the For this application, the measurement error of the electri-
inner shunt temperature cannot be measured directly and is cal current should be about 0.1 % for a wide current range
typically higher than the temperature at the heat-sink surface. up to 600 A in order to determine the capacity of lithium-ion
For the thermal steady state, the inner temperature can still be cells by integration of the current flowing into the cell.
calculated if the thermal resistance is known (Ziegler et al.,
2009b). 2.2 Shunt resistor: temperature coefficient
Unfortunately, these static methods do not take into ac-
count the thermal capacitances of the materials along the path For low currents this maximum error can easily be accom-
of the heat flow. As shown in the following sections, they plished by periodic linear calibration and measurement of the
lead to different dynamic behavior of shunt and sensor tem- actual shunt resistance, which eliminates any static errors due
perature and therefore induce time-dependent measurement to resistance tolerances and slow drift effects over time.
errors. However, a linear calibration technique cannot compen-
In this paper a method is presented to overcome these sate for the effect of the temperature coefficient of the shunt,
challenges and to provide a dynamic error correction pro- which is determined to TK = 422 ppm K1 from the mea-
cedure. With this method it is possible to calculate the in- surement in Fig. 2. For the maximum current of 600 A the
ner shunt temperature also during arbitrary thermal transients power dissipation is 25 W per resistor. Although this is suf-
and to dynamically compensate for the temperature drift of ficiently below the respective absolute maximum rating of
a shunt resistor. The proposed algorithm has been optimized 100 W, the thermal resistance of the shunt and the heat sink
for real-time applications with low-cost microcontrollers and lead to an inner temperature rise of 30 C, equivalent to a
can also be used to increase the accuracy of existing current resistance variation of 1.5 %. Without further compensation,
measurement systems. the relative current measurement error has the same value
This paper is an extended version of the previously pub- and is an order of magnitude higher than desired.
lished conference paper (Wekamp and Melbert, 2016a) and In the shown configuration, the copper busbars and other
includes an in-depth discussion of the practical implementa- interconnections also contribute a small part to the total shunt
tion in an embedded system, including the derivation of the resistance (< 10 %) and to the overall temperature coeffi-

J. Sens. Sens. Syst., 5, 389400, 2016 www.j-sens-sens-syst.net/5/389/2016/

P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts 391

Figure 1. The 1 m shunt resistor for lithium-ion cell test equipment, consisting of 12 10 m shunt resistances in a TO-247 package
mounted on a heat sink with water cooling. Left: photograph with annotations. Right: cross section.

10.35 Here, R0 depicts the resistance value at the (arbitrary) tem-

Measurement
Fit
perature T0 . Typically, R0 is measured during calibration and
10.3
T0 is set by the temperature during this procedure. Two ef-
Resistance [m]

10.25 Temperature fects can lead to a temperature variation during operation:

coefficient:
self-heating and ambient temperature change.
10.2 422 ppm/K

10.15 1T = 1Tselfheating + 1Tamb (2)

R = 1.5 %
10.1 The ambient temperature rise is given by the difference
T=30 C
between the present ambient temperature Tamb and the cali-
10.05
bration temperature T0 :
10
20 40 60 80 100 1Tamb = Tamb T0 . (3)
Temperature [C]
The self-heating is caused by the power dissipation Ploss
Figure 2. Measured temperature dependence of a single 10 m
in the resistance material due to the electrical current I . The
shunt resistor, approximated by a linear fit. The total resistance
associated temperature rise for the thermal steady state is cal-
change for a temperature increase of 1T = 30 C is highlighted
(Grundktter, 2016). culated using the total thermal resistance Rth,total between the
resistor and the environment:

cient. Because the temperature coefficient of copper is quite Ploss = R(T )I 2 (4)
high ( 3900 ppm K1 ), the additional resistance increases 1Tselfheating = Rth,total Ploss
the total temperature coefficient to TK 600 ppm K1 . = Rth,total R(T )I 2 ,
Temperature variation of the cooling water (or more gen-
erally of the ambient temperature) will also lead to a drift of Rth,total R0 I 2 . (5)
the shunt resistance value, and has to be taken into account.
The approximation R(T ) R0 made in the last step of
Eq. (5) simplifies the following calculation. It can be shown
2.3 Nonlinear calibration for thermal steady state that the overall calibration error associated with this assump-
The temperature dependence of the shunt resistance used in tion is below 0.04 % and therefore can be disregarded.
this work can be approximated with good concordance by a Substituting Eqs. (2)(5) in Eq. (1) yields
linear function, as demonstrated in Fig. 2, and is given by the  
equation R(I, 1Tamb ) = R0 1 + TK 1Tamb + TK Rth,total R0 I 2 . (6)

R(T ) = R0 (1 + TK (T T0 )) (1) This equation describes the resistance variation due to self-
= R0 (1 + TK 1T ) . heating and ambient temperature change.

www.j-sens-sens-syst.net/5/389/2016/ J. Sens. Sens. Syst., 5, 389400, 2016

392 P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts

The shunt is used for current sensing by measuring the

voltage drop over the resistor: Cth, 1 Cth, 2 Cth, 3

Ushunt = R(I, 1Tamb )I Tshunt

3


= R0 (1 + TK 1Tamb ) I + R0 TK Rth,total R0 I
Rth, 0
= a1 I + a3 I 3 . (7) Ploss
Rth, 1 Rth, 2 Rth, 3
Tamb
As demonstrated in Eq. (7), the relationship between voltage
and current is expressed by a polynomial function of third T1 T2 T3
degree, with a clear separation between effects due to self- Cth, 4
heating ( I 3 ) and ambient temperature change ( I ).
During calibration, a reference current measurement is Thermal model for
Tsensor internal shunt
available and the parameters a1 and a3 are obtained by poly- temperature Tshunt
nomial curve fitting of the measured voltage Ushunt as a func-
tion of varying reference currents Iref in thermal steady state, Ploss
Rth, 4 Tamb (optional) Thermal
as explained in Sect. 3.2, and are used to calculate the model
model for sensor
parameters TK , R0 and Rth,total . T4 temperature Tsensor
For the final application, solving Eq. (7) for the current I
therefore allows nonlinear calibration of the current measure-
ment in thermal steady state. Figure 3. Thermal model used for calculation of the internal shunt
temperature and the variation of the sensor temperature due to self-
heating.
2.4 Heat flow and thermal modeling
The nonlinear calibration technique discussed in the previous
section is only valid for the thermal steady state, i.e., when lumped element approach, consisting of a small number of
the inner shunt temperature is constant and does not change thermal resistances and thermal capacitances.
anymore. Obviously, this also implies a constant electrical The corresponding thermal model is given in Fig. 3 and
current I and sufficient settling time after a current step, typ- was optimized empirically. The characteristics of the heat
ically in the range of several minutes. However, realistic test- conduction are represented by an equivalent electrical circuit
ing procedures for lithium-ion cells require much faster cur- diagram, substituting temperature with voltage and heat flow
rent transients. The static calibration approach therefore has with current (Mrz and Nance, 2000). Therefore, well-known
to be extended by a dynamic thermal model for this applica- methods for electrical circuit analysis can also be applied for
tion. thermal analysis.
For the thermal analysis, a simplified shunt resistor setup The upper half of Fig. 3 describes the inner shunt tem-
with the following assumptions is used. perature Tshunt . The total thermal resistance from Eq. (7) is
split into four separate terms, so that Rth,total = Rth,0 +Rth,1 +
1. The 12 distributed shunt resistors are combined into 1 Rth,2 + Rth,3 , and thermal capacitances are added. They rep-
equivalent lumped shunt resistor. This is feasible if the resent the dynamic behavior of the heat transfer through dif-
electrical and thermal conditions are similar for all indi- ferent materials between the inner shunt resistance and the
vidual resistors, which holds true for this application. heat sink.
2. The lumped shunt resistor is mounted on the surface of The lower half of Fig. 3 describes the influence of the
a metal heat sink with an internal water cooling pipe power dissipation Ploss on the optional temperature sensor.
(cf. Fig. 1). The temperature sensor is also mounted on Ideally, the sensor would directly measure the ambient tem-
the heat-sink surface with some clearance to the lumped perature Tamb , but usually some part of the heat generated
resistor. by the shunt resistor is also coupled to the sensor, increasing
Tsensor . This impacts the measurement of the true ambient
3. The cooling water is considered to be an ideal heat sink temperature and needs to be taken into account.
with the invariant temperature Tamb . The chosen model structure (Foster network) simplifies
the following calculations, but at the disadvantage of no di-
The simplified structure is a sufficiently accurate approx-
rect physical equivalent of the lumped circuit elements, as ex-
imation of the actual shunt resistor setup. Unfortunately, ex-
plained in Bagnoli et al. (1998). The analysis of the circuit is
act (three-dimensional) thermal modeling is still complicated
best done in the frequency domain, using the complex-valued
and requires the use of simulation and the finite element
signals, e.g., Tshunt (t) c s Tshunt (j ). For the upper circuit,
thermal impedance Zth (j ) and the Fourier transform of the
method. The resulting thermal model is of high complexity
and cannot be analyzed in real time on an embedded sys-
the resulting equation is
tem. More suitable for this application is a one-dimensional

J. Sens. Sens. Syst., 5, 389400, 2016 www.j-sens-sens-syst.net/5/389/2016/

P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts 393

Tshunt (j ) = Tamb (j ) + Ploss (j ) Rth,0 results of Eq. (15) to the derivation of the static calibration
3 approach (Eq. 17) yields
X 1
+ Ploss (j ) Rth,i . (8)
i=1
1 + j i Ushunt (t) = R0 (1 + TK 1Tamb (t)) I (t)
+ R0 TK Rth,total R0 (I 2 (t)h1 (t))I (t)


The dynamic effects of self-heating therefore are given by  
the sum of three independent low-pass filters with time con- = a1 I (t) + a3 I 2 (t)h1 (t) I (t). (16)
stants i = Rth,i Cth,i and a time-invariant part due to Rth,0 .
Likewise, the transfer function for the sensor temperature is Equation (16) is then solved for the unknown current I :
obtained as
Ushunt (t)
1 I (t) =
(17)
Tsensor (j ) = Tamb (j ) + Ploss (j )Rth,4 . (9) a1 + a3 I 2 (t)h1 (t)
1 + j 4
Ushunt (t)
=
. (18)
By using the approximation Ploss = R(T )I 2 R0 I 2 , R0 (1 + TK 1Tamb (t)) + a3 I 2 (t)h1 (t)
Eq. (8) is rewritten as
On the right-hand side of Eq. (18) there are still two un-
known quantities: the change in ambient temperature 1Tamb

Rth,0
Tshunt (j ) = Tamb (j ) + I 2 (j )R0 Rth,total and the current I 2 (t) itself, used as the input to the dynamic
Rth,total
3
# thermal model. The latter circular reference is resolved by
X Rth,i 1 approximating the current with the last known value, i.e.,
+
R
i=1 th,total
1 + j i I 2 (t) I 2 (t 1t). The error introduced by this approxima-
tion is small, because the thermal model consists of several
= Tamb (j ) + I 2 (j )R0 Rth,total H1 (j ). (10)
low-pass filters whose time constants are significantly higher
The term inside the brackets describes the normalized dy- than the discrete time step (i  1t) and therefore smooth
namic behavior of the system and is denoted by H1 (j ), the behavior of I 2 (t) in any case. Additionally, an imple-
mentation using digital filter structures can be constructed to
Rth,0 3
X Rth,i 1 eliminate the need for the most recent value of the electrical
H1 (j ) = + , (11) current at all (see Sect. 2.6).
Rth,total i=1 Rth,total 1 + j i
The change in the ambient temperature is given by Eq. (3),

whereas the term I 2 R0 Rth,total = 1Tselfheating describes the 1Tamb = Tamb T0 ,

steady-state temperature rise due to self-heating and has al-
ready been discussed in the previous section. The dynamic where T0 denotes the temperature at which the calibration
increase in the shunt temperature due to self-heating is there- measurement was performed. Ideally, the ambient temper-
fore given by ature could be measured directly by a temperature sen-
sor yielding Tamb = Tsensor , but in our application the self-
1Tselfheating (j ) = I 2 (j )R0 Rth,total H1 (j ). (12) heating of the shunt resistor dynamically influences the sen-
sor temperature, as shown in Fig. 3 and Eq. (9).
Similarly, the sensor temperature is described by Applying the same methodology as before, substituting
Eq. (13) in Eq. (3) and transforming to the time domain
Tsensor (j ) = Tamb (j ) + I 2 (j )R0 Rth,4 H2 (j ) (13) yields
1
with H2 (j ) = . (14) 1Tamb (t) = Tsensor (t) R0 Rth,4 I 2 (t 1t)h2 (t) T0 , (19)
1 + j 4
where h2 (t) c s H2 (j ). In this formula, the measured
2.5 Dynamic error correction procedure
sensor temperature Tsensor is corrected by the predicted tem-
The static and dynamic knowledge of the system is used to perature rise due to the power dissipated in the shunt resistor
develop a dynamic error correction procedure for the current and is used as an estimation of the true ambient temperature.

main using the impulse response h1 (t) c s H1 (j ) of the

measurement. First, Eq. (12) is transformed to the time do- A graphical illustration of the overall algorithm is given in
Fig. 4.
dynamic system:
2.6 Digital filter implementation
1Tselfheating (t) = R0 Rth,total I 2 (t)h1 (t). (15)
The correction algorithm is implemented on a digital pro-
Here the output of the dynamic thermal system is mathemati- cessor for real-time application in an embedded system. The
cally described by the convolution I 2 (t)h1 (t). Applying the equations therefore are converted to discrete time steps with

www.j-sens-sens-syst.net/5/389/2016/ J. Sens. Sens. Syst., 5, 389400, 2016

394 P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts

mental transfer function

Ushunt x/y Ishunt
1t 1
R(t) Y (z) 1 i z
H (z) = = = . (24)
+ 2 z1
I (z) 1 + 1t i 1 (1 1ti )z
1

TKR0 R0 TKRth,totalR02 Rewriting this equation and transforming to the time domain
yields
Tamb h(t) * I 2(t)
- T0 1t 1 1t 1 2
+ Y (z) = (1 )z Y (z) + z I (z) (25)
- i i
1t 1t 2
yk = (1 )yk1 + I , (26)
i i k1
where the relationship yk1 c s z1 Y (z) was used. The
difference equation (Eq. 26) realizes the desired low-pass
Ambient
filter characteristic as a first-order IIR (infinite impulse re-
temperature Self-heating sponse) filter and can easily be implemented on a microcon-
troller. It is stable if 0 < 1t < i , i.e., if the discrete time step
Tsensor 2 is sufficiently small.
Ishunt (t-t)
The use of the forward difference (Eq. 21) instead of the
Figure 4. Overall dynamic correction algorithm, taking into ac- backward difference or the bilinear transform (Proakis and
count effects due to self-heating and ambient temperature change. Manolakis, 2007) reduces the computation cost and ensures
that at time k the value Ik2 is not needed in the filter calcula-
tion.
interval 1t, such that t = k 1t with k = 0, 1, 2, . . .. In the fol- The fundamental filter structure is implemented four times
lowing discussion, the short-form notation in parallel (one for each RC element) and the weighted re-
sults are summed according to Eqs. (22) and (23) to imple-
I (t)|t=k 1t = I (k 1t) = Ik (20) ment the dynamic thermal behavior.

will be used for all time-varying quantities.

2.7 Discrete-time dynamic error correction procedure
The dynamic system behavior given by the continuous-
time convolutions h1 (t)I 2 (t) and h2 (t)I 2 (t) is transformed The complete dynamic error correction procedure is defined
into a digital filter structure. This is accomplished by substi- by the algorithm in Table 2. It consists of five steps, which
tuting the corresponding differential equations of the system have to be executed for each time step k.
by discrete-time difference equations using the forward dif- In the first step, the new states of the four digital low-pass
ference (Rabiner and Gold, 1975): filters are determined. They are used in the second step to
2
calculate the dynamic self-heating effect Ifiltered,k . This value

df (t)  f ((k + 1)1t) f (k1t)
= is the discrete-time equivalent of the convolution h1 (t)I 2 (t).
dt t=k1t 1t In step three, the fourth low-pass filter (Rth,4 , 4 ) is used to
fk+1 fk eliminate the influence of heat generated by the shunt resis-
= . (21)
1t tor on the temperature sensor and yields the true change in
ambient temperature 1Tamb .
This is equivalent to the substitution j = (z 1)/1t in Using these values, the calibration factor K and the cal-
the frequency domain, replacing the Fourier transform with ibrated current measurement Ik are calculated from a mea-
the z transform. Equations (11) and (14) therefore become surement of the shunt-resistor voltage Ushunt,k . Steps four and
3 five have been separated, because the sampling frequency fs
Y1 (z) Rth,0 X Rth,i 1 for the current measurement (I , Ushunt ) may be chosen higher
H1 (z) = = + , (22)
I 2 (z) Rth,total i=1 Rth,total 1 + z1
1t i than the frequency for the thermal compensation algorithm
Y2 (z) 1 fc = 1/1t. In this scenario, it is sufficient to update the cor-
H2 (z) = = , (23) rection factor K with the lower frequency fc .
I 2 (z) 1 + z1
1t 4 The overall dynamic correction approach has been opti-
where yk c s Y (z) denotes the output of the respective dig-
mized for fast computation speed allowing real-time opera-
tion in a low-cost embedded system.
ital filter.
Due to linearity, the individual low-pass filters can be im- 1. An efficient first-order low-pass filter bank with IIR
plemented independently of each other. Therefore it is suf- (infinite impulse response) structure instead of higher-
ficient to synthesize a digital filter for the underlying funda- order FIR (finite impulse response) filters is used and

J. Sens. Sens. Syst., 5, 389400, 2016 www.j-sens-sens-syst.net/5/389/2016/

P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts 395

Table 2. Algorithm: discrete-time dynamic error correction proce- First, small-current amplitudes of up to 150 A are investi-
dure. gated with alternating directions of current. For these ampli-
tudes, the self-heating effect is negligibly small and thermal
Repeat for each time step k. steady state (with respect to the measurement resolution) is
1. Thermal reached almost instantaneously. Additionally, the use of al-
 filtercalculation
1t 1t 2 ternating current directions allows compensation for any off-
yi,k = 1 yi,k1 + I , for i = 1, 2, 3, 4
i i k1 set errors in the measurement of the shunt voltage.
2. Calculate the self-heating effect.
The nonlinear self-heating effect is evaluated by longer
3 current pulses with amplitudes in the range 250 A . . . 600 A.
2 Rth,0 2 X Rth,i
Ifiltered,k = Ik1 + yi,k The pulse width is chosen sufficiently high to allow for set-
Rth,total R
i=1 th,total tling of thermal effects. This investigation is repeated at two
3. Calculate the ambient temperature change. fixed ambient temperatures to determine the temperature co-
1Tamb,k = Tsensor,k R0 Rth,4 y4,k T0 efficient of the shunt.
The sequence in Fig. 5 is applied to every test system by
4. Calculate the calibration factor.
an automated calibration test bench and the voltage over the
K = 1/ R0 + TK R0 1Tamb,k
 shunt resistor is measured. A reference value for the current
+TK R02 Rth,total Ifiltered,k
2
is obtained using an external high-power precision shunt re-
sistor with a temperature coefficient of only 1 ppm K1 and
5. Calculate the shunt-resistor current.
a measurement uncertainty of 0.026 %, which is assumed not
Ik = K Ushunt,k
to be affected by thermal drift. The measured signals are eval-
uated using Matlab on a PC.

models the dynamic thermal behavior with sufficient ac-

curacy. 3.2 Static parameter estimation
2. An iterative approach to solve the nonlinear voltage The currentvoltage relationship of the shunt resistor in ther-
current relationship avoids complex and expensive math mal steady state is described by Eq. (7):
functions (like square roots), except a single division.
Ushunt = R0 (1 + TK 1Tamb ) I + R0 TK Rth,total R0 I 3


3. For each time step k, the algorithm requires only 11 ad-
ditions, 16 multiplications and 1 division, as well as = a1 I + a3 I 3 .
storage for 5 numeric values. By setting fs > fc , the
computational impact of the correction algorithm can For each calibration current pulse at Tamb = 20 C (Fig. 5),
be reduced further. in Fig. 6 the shunt voltage at the end of the pulse is plotted
in relation to the reference current. Using polynomial curve
3 Application and measurement results fitting tools on the resulting data set, the parameters a1 and
a3 at Tamb = 20 C are determined.
In this section, the derived algorithm will be applied to the The error between the polynomial fit and the measurement
laboratory test equipment for automotive lithium-ion cells is depicted in the lower part of the figure. It is less than 20 mA
(see Sect. 2.1). First, an automated calibration sequence is over the full current range. The hypothetical error, if the ef-
discussed for obtaining the static and dynamic calibration fects due to self-heating were not included and a purely lin-
parameters. The accuracy of the algorithm is evaluated for ear model was used, is also shown and exceeds 10 A. The
real-life testing profiles. A comparison of parameter variance same evaluation is also performed for Tamb = 30 C, yield-
between different test systems and a discussion of long-term ing a second set of the coefficients a1 and a3 .
parameter drift conclude the paper. In Table 3 the determined coefficients from the polyno-
mial curve fitting are given for the two ambient temperatures.
As predicted by the theoretical approach in Eq. (7), the val-
3.1 Automated calibration sequence
ues of a3 are nearly independent of the ambient temperature
The parameters required by the static part of the error cor- rise and describe only effects due to self-heating. From the
rection procedure are R0 , T0 , Rth,total and TK , whereas the varying values of a1 the resistance R0 = 0.8868 m (T0 =
dynamic behavior is described by Rth,i /Rth,total and i (i = 20.4 C) and the temperature coefficient TK = 594 ppm K1
1, 2, 3). For the ambient temperature correction, additionally, can be determined. The overall thermal resistance is obtained
Rth,4 and 4 are needed. from a3 and equals Rth,total = 0.10 K W1 .
The current profile given in Fig. 5 is used as the calibration A similar approach is used to determine the thermal re-
excitation. It consists of rectangular current pulses with vary- sistance Rth,4 for the temperature sensor: at the end of each
ing amplitudes at two different ambient temperatures Tamb . current pulse, thermal steady state is assumed and the rise in

www.j-sens-sens-syst.net/5/389/2016/ J. Sens. Sens. Syst., 5, 389400, 2016

396 P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts

Figure 5. Automated calibration sequence consisting of rectangular current pulses with varying amplitude at two ambient temperatures.

8
600 Measurement

T Sensor [C]
Measurement Quadratic fit
Fit 6
500

4
Ushunt [mV]

400

300 2

200 0
0 100 200 300 400 500 600
100 Current I [A]
0 Figure 7. Rise of the heat-sink temperature 1Tsensor due to the
0 100 200 300 400 500 600
power dissipated in the shunt resistor and fit to the quadratic model.
Reference current [A]
12
a 1 I + a 3 I 3 (full model)
10
Calibration errorI [A]

a1 I (linear fit only) the sensor temperature is derived from Eq. (13) as
8

6
1Tsensor = Tsensor Tamb = I 2 R0 Rth,4 . (27)
4

2 The measured temperature differences and a quadratic fit as

0 a function of the current I are presented in Fig. 7. From the
fit, the thermal resistance equals Rth,4 = 0.021 K W1 .
-2
0 100 200 300 400 500 600
Reference current [A]

Figure 6. Calibration measurement and resulting error for Tamb = 3.3 Dynamic parameter estimation
20 C and thermal steady state. The error of the proposed nonlinear
model is below 20 mA, whereas the error for the linear fit (without As discussed in Sect. 2.4, the dynamic thermal behavior is
self-heating effects) exceeds 10 A. fully characterized by the functions H1 (j ) and H2 (j ).
Both functions are a linear superposition of the fundamental
first-order low-pass filter H (j ). Its step response hstep (t) in
Table 3. Extracted parameters a1 and a3 using polynomial curve the time domain is given by
fitting for different ambient temperatures.

a1 (V A1 ) a3 (pV A3 ) hstep (t) = 1 et/ (28)

1
Tamb = 20.4 C 886.77 46.683 with H (j ) = . (29)
Tamb = 28.1 C 890.85 46.685 1 + j
Physical Resistance and ambient Self-heating
representation temperature change After a current step 1I (starting from I = 0), the shunt
resistor voltage response in the time domain due to thermal

J. Sens. Sens. Syst., 5, 389400, 2016 www.j-sens-sens-syst.net/5/389/2016/

P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts 397

Current [A]
shunt 1
610
600
Normalized U

590
0.5
0 50 100 150 200

measurement error [%]

Measurement 3
Fit

Relative current
2
0
0 50 100 150 200 250 300 1
Time [s]
0
Figure 8. Normalized change in shunt voltage due to self-heating -1 No compensation
after a current step. Steady-state compensation
-2 New dynamic compensation
-3
Table 4. Extracted parameters for the dynamic thermal model due 0 50 100 150 200
to self-heating by nonlinear least-square curve fitting. Time [s]

Figure 9. Relative current measurement error after a current step

i=0 i=1 i=2 i=3
at t = 0 s from 0 to 600 A for different temperature compensation
Rth,i /Rth,total 0.10 0.52 0.21 0.15 schemes.
i [s] 0.67 16.82 107.8
Table 5. Arithmetic mean of the absolute value of the relative cur-
rent measurement error calculated for the complete testing profile
effects is therefore given by given in Fig. 10.

1Ushunt (t) = (30) Compensation algorithm Mean measurement error

!
3 No compensation 0.211 %
Rth,0 X Rth,i
a3 1I 3 + (1 et/i ) , Steady-state-only 0.139 %
Rth,total i=1 Rth,total
New dynamic method 0.085 %
if the current as well as the ambient temperature are held con-
stant during the pulse. This is the case for the current pulses
in the calibration sequence in Fig. 5. The resulting voltage namic compensation scheme reduces the relative measure-
during one of the pulses is given in Fig. 8, where the voltage ment error from 2 % (no temperature compensation) to less
change due to the shunt resistance itself (= R0 1I ) is already than 0.1 % after 2 s settling time. For comparison, the steady-
subtracted and the values are normalized to the range 0. . .1. state compensation without dynamic effects requires 180 s to
The unknown parameters Rth,i /Rth,total and i are deter- reach this band of error.
mined by nonlinear least-square curve fitting with the func- In order to demonstrate the real-life improvement factor
tion given in Eq. (30). Note that 1Ushunt (t ) = a3 1I 3 . of the algorithm, a realistic testing profile for automotive
Because the factor a3 is already known from the static cali- lithium-ion cells is evaluated in Fig. 10. The resulting rela-
bration, it is sufficient to analyze the normalized behavior of tive current measurement error is depicted in the lower graph
1Ushunt , which is independent of the magnitude or starting for an excerpt from the profile. For low currents, the ther-
point of the current step 1I . mal effects are negligible and there is no difference between
The fit presented in Fig. 8 accurately models the dynamic the compensation schemes. For higher currents the error in-
effects. The resulting model parameters are given in Table 4, creases significantly if no compensation or only steady-state
with time constants in the range of 0.5 to 120 s. compensation is used. In comparison, the proposed transient
For the thermal model of the temperature sensor, a similar correction algorithm is able to keep the measurement error
technique is used for parameter identification with the equa- within a small band of error.
tion The arithmetic mean of the relative current measurement
error given in Table 5 for the complete testing profile is a
1Tsensor = Rth,4 R0 1I 2 1 et/4 ,


(31) measure for the real-life improvement factor. For the shown
profile, the dynamic algorithm reduces the mean error by
yielding the value for 4 = 48.6 s.
60 % compared to the case if no temperature compensation
is used, and by 40 % compared to the steady-state-only cali-
3.4 Measurement results bration.
The performance of the proposed algorithm is demonstrated
in Fig. 9 for a 600 A constant-current pulse. The new dy-

www.j-sens-sens-syst.net/5/389/2016/ J. Sens. Sens. Syst., 5, 389400, 2016

398 P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts

600 Table 6. Arithmetic mean and variance of the calibration parame-

Current |I| [A]
Testing profile for Li-Ion cell ters for 100 test systems. The last column lists the maximum current
400 error if the mean value instead of the actual value of the respective
parameter was used for error correction.
200
Parameter Mean 3 Max. current
error if mean
0
0 200 400 600 800 1000 was used
Time t [s] R0 = a1 884.1V A1 1.60 % 1.341 %
600 a3 48.1 pV A3 12.83 % 0.211 %
Current |I| [A]

Excerpt from profile

500 TK 623 ppm K1 8.7 % 0.055 %
400
1 0.655 s 20.7 % 0.163 %
300 2 16.8 s 23.3 % (all dynamic
200 3 114 s 37.1 % parameters
100 Rth,1 /Rth,total 0.434 20.0 % are jointly
Rth,2 /Rth,total 0.247 20.1 % set to
113 114 115 116 117 118 119 120
Rth,3 /Rth,total 0.219 39.5 % mean values)
Time t [s]
measurement error [%]

1.5
Relative current

1
0.5 the sequence shown in Fig. 5; calibration parameters have
0 been obtained individually and are used for optimal results.
-0.5 No compensation The maximum current deviation after calibration in thermal
-1
Steady-state compensation steady state for all test systems is shown in Fig. 11. It is well
New dynamic compensation
-1.5
below 0.1 %, proving the accuracy of the proposed method.
113 114 115 116 117 118 119 120 The statistical distribution of the calibration parameters
Time t [s] themselves is also evaluated and given in Fig. 12. Each pa-
rameter is approximated by a normal distribution, whose
Figure 10. Realistic testing profile for lithium-ion cells (top: full
mean and variance are given in Table 6.
profile; middle: excerpt) and associated relative current measure-
On the one hand, the nearly Gaussian distribution proves
ment errors (bottom).
that there are no outliers due to errors during manufacturing,
that all test systems behave similarly, and that the parameter
35 estimation procedure works robustly. On the other hand, it
30 allows one to determine whether a calibration on an individ-
No. of test systems

ual basis is actually necessary, or whether it is sufficient to

25 only measure a small number of shunt resistors and use the
20 arithmetic mean of the parameters for error correction in all
test systems. The maximum resulting current error for this
15 case is given in the last column of Table 6. These values have
10 been obtained through simulation with a rectangular current
pulse of 600 A at an ambient temperature of 30 C, using the
5 respective arithmetic mean of the parameter instead of the
0
true value.
-0.08 -0.06 -0.04 -0.02 0 0.02 0.04 The variation of the resistance R0 has a large effect on the
Max. static current error [%] calibration accuracy and implies that (for this application) the
resistance value has to be measured individually for each sys-
Figure 11. Maximum current deviation after calibration in thermal
tem, but for the other parameters the effect on the resulting
steady state for 100 test systems.
current error is smaller. If an error of approximately 0.5 %
was acceptable, the thermal calibration does not need to be
3.5 Statistical analysis of the calibration parameters and
executed for every test system, eliminating the long current
long-term drift
pulses in Fig. 5 and reducing the calibration time from 85 to
14 min.
The dynamic error correction procedure discussed in this pa- Other important aspects are the long-term parameter drift
per is currently in use for 100 test systems in a test center and the resulting current measurement error. The drift is
for aging analysis on automotive lithium-ion cells (Wekamp caused by the aging of the shunt resistors and the thermal
et al., 2016). Each of the test systems has been calibrated with stress during normal operation and is corrected by repeated

J. Sens. Sens. Syst., 5, 389400, 2016 www.j-sens-sens-syst.net/5/389/2016/

P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts 399

30 30 60 30
No. of test systems

No. of test systems

No. of test systems

No. of test systems

20 20 40 20

10 10 20 10

0 0 0 0
870 880 890 900 40 45 50 55 570 600 630 660 12 14 16 18 20 22
a1 [V/A] 3 TK [ppm/K] 2 [s]
a3 [pV/A ]

Figure 12. Statistical distribution of selected calibration parameters for 100 distinct test systems.

Table 7. Parameter drift of a single shunt after operation for 1 year. thermal steady state. It is supplemented by a dynamic model
The total charge throughput during this time was approximately describing transient thermal effects due to self-heating and
185 000 Ah. heat transfer over a heat sink. A temperature sensor is used
to measure the ambient temperature. The effect of the power
Parameter Value Drift (1 year) Current dissipation in the shunt resistor on the measured ambient
error
temperature is compensated for by a correction algorithm.
due to
drift effect The proposed algorithm is used for current measurement
(I = 600 A) in test equipment for automotive lithium-ion cells employ-
ing low-cost shunt resistors. It is able to reduce the current
R0 884.5V A1 +0.09 % +0.06 % measurement error from over 2 % to less than 0.1 % for a
Rth,total 0.094 K W1 +15.7 % +0.26 % 600 A current pulse. For a real-life, dynamic testing profile
TK 635 ppm K1 6.43 % 0.17 %
the mean current measurement error is reduced by 60 %.
1 0.655 s +2.44 % +0.04 % The presented dynamic error correction procedure there-
2 14.7 s +4.01 % (all dynamic fore enables precise current measurement with low-cost
3 101 s +5.29 % parameters shunt resistors. The implementation as a discrete-time filter
Rth,1 /Rth,total 0.482 3.60 % are jointly structure on a microcontroller with low demands on com-
Rth,2 /Rth,total 0.227 +1.25 % evaluated)
putation power is discussed in detail. An automated calibra-
Rth,3 /Rth,total 0.191 +7.57 %
tion sequence composed of rectangular current pulses is pre-
sented, by which all required parameters are determined in-
dividually for each shunt resistor.
calibration. The parameter drift and the resulting current A statistical analysis of the calibration parameters of
measurement error (at I = 600 A) for one tester between suc- 100 shunt resistors is performed, proving the robustness and
cessive calibrations are given in Table 7. Between the two accuracy of this approach. Drift effects over time have also
calibrations, the tester has been used for 1 year in an aging been evaluated; the current measurement error after 1 year is
study on lithium-ion cells with a total charge throughput of still less than 0.2 %.
185 000 Ah. Due to low computational requirements the proposed al-
The overall steady-state current measurement error after gorithm can also be used to increase the accuracy of existing
1 year is 0.165 %, which is acceptable for this application. systems. Particularly, this is the case if the ambient tempera-
In comparison, the drift of the shunt resistance R0 is about ture does not change or is already known, allowing the omis-
0.1 %, which is equal to the current measurement error for sion of the optional temperature sensor used in this applica-
low currents. At higher currents, the error is dominated by tion.
the drift of the thermal resistance Rth,total and the temperature
coefficient TK , although these effects cancel each other out
5 Data availability
in parts. In contrast, the drift of the dynamic behavior has a
negligible impact.
The dataset used in this article is available at
doi:10.5281/zenodo.164820 (Wekamp and Melbert,
4 Summary and conclusions 2016b).

In this paper, a dynamic error compensation procedure is pre- Edited by: K.-D. Sommer
sented, which is able to correct the current measurement er- Reviewed by: two anonymous referees
rors of shunt resistors due to self-heating and ambient tem-
perature change. The method is based on a theoretical anal-
ysis of the shunt resistance variation due to temperature for

www.j-sens-sens-syst.net/5/389/2016/ J. Sens. Sens. Syst., 5, 389400, 2016

400 P. Wekamp and J. Melbert: High-accuracy current measurement with low-cost shunts

References Rabiner, L. R. and Gold, B.: Theory and application of digital signal
processing, Prentice-Hall, Englewood Cliffs, NJ, 762 pp., ISBN-
Bagnoli, P. E., Casarosa, C., Ciampi, M., and Dallago, E.: Ther- 10: 0-13-914101-4, 1975.
mal resistance analysis by induced transient (TRAIT) method Ripka, P.: Electric current sensors: a review, Meas. Sci. Technol.,
for power electronic devices thermal characterization. I. Fun- 21, 112001, doi:10.1088/0957-0233/21/11/112001, 2010.
damentals and theory, IEEE T. Power Electr., 13, 12081219, Wekamp, P. and Melbert, J.: 3.4.4 High Performance Cur-
doi:10.1109/63.728348, 1998. rent Measurement with Low-Cost Shunts by means of Dynamic
Grundktter, E.: Untersuchung des transienten thermischen Ver- Error Correction, in: 18. GMA/ITG Fachtagung Sensoren und
haltens von Shuntwiderstnden zur Strommessung und Op- Messsysteme 2016, 224230, doi:10.5162/sensoren2016/3.4.4,
timierung von Fehlerkorrekturverfahren, Masterarbeit, Ruhr- 2016a.
Universitt Bochum, Bochum, 2016. Wekamp, P. and Melbert, J.: Supplementary dataset for Journal Ar-
Lohmann, N., Wekamp, P., Haumann, P., Melbert, J., and Musch, ticle High-accuracy current measurement with low-cost shunts
T.: Electrochemical impedance spectroscopy for lithium-ion by means of dynamic error correction [Data set], Zenodo, avail-
cells: Test equipment and procedures for aging and fast charac- able at: doi:10.5281/zenodo.164820, 2016b.
terization in time and frequency domain, J. Power Sources, 273, Wekamp, P., Haumann, P., and Melbert, J.: 600-A Test System
613623, doi:10.1016/j.jpowsour.2014.09.132, 2015. for Aging Analysis of Automotive Li-Ion Cells With High Reso-
Mrz, M. and Nance, P.: Thermal Modeling of Power- lution and Wide Bandwidth, IEEE T. Instrum. Meas., 65, 1651
electronic Systems, available at: http://www.iisb.fraunhofer.de/ 1660, doi:10.1109/TIM.2016.2534379, 2016.
content/dam/iisb2014/en/Documents/Research-Areas/Energy_ Ziegler, S., Woodward, R. C., Iu, H. H.-C., and Borle, L. J.: Cur-
Electronics/publications_patents_downloads/Publications/ rent Sensing Techniques: A Review, IEEE Sens. J., 9, 354376,
Therm_Modelling_2000_IISB.pdf (last access: 2 Novem- doi:10.1109/JSEN.2009.2013914, 2009a.
ber 2016), 2000. Ziegler, S., Woodward, R. C., Iu, H. H.-C., and Borle, L. J.: In-
Proakis, J. G. and Manolakis, D. G.: Digital signal processing, Pear- vestigation Into Static and Dynamic Performance of the Cop-
son Education, Upper Saddle River, NJ, 4th ed., 1084 pp., ISBN- per Trace Current Sense Method, IEEE Sens. J., 9, 782792,
10: 0-13-187374-1, 2007. doi:10.1109/JSEN.2009.2021803, 2009b.

J. Sens. Sens. Syst., 5, 389400, 2016 www.j-sens-sens-syst.net/5/389/2016/