Howland current source for high impedance load applications

K. F. Morcelles, V. G. Sirtoli, P. Bertemes-Filho, and V. C. Vincence

Citation: Review of Scientific Instruments 88, 114705 (2017); doi: 10.1063/1.5005330

View online: https://doi.org/10.1063/1.5005330
View Table of Contents: http://aip.scitation.org/toc/rsi/88/11
Published by the American Institute of Physics

Articles you may be interested in

The differential Howland current source with high signal to noise ratio for bioimpedance measurement system
Review of Scientific Instruments 85, 055111 (2014); 10.1063/1.4878255

Erratum: “Howland current source for high impedance load applications” [Rev. Sci. Instrum. 88, 114705
(2017)]
Review of Scientific Instruments 88, 129901 (2017); 10.1063/1.5016966

Biased four-point probe resistance

Review of Scientific Instruments 88, 114701 (2017); 10.1063/1.4995389

Noise spectra in balanced optical detectors based on transimpedance amplifiers

Review of Scientific Instruments 88, 113109 (2017); 10.1063/1.5004561

Note: Temperature effects in the modified Howland current source for electrical bioimpedance spectroscopy
Review of Scientific Instruments 88, 076103 (2017); 10.1063/1.4991829

Comparison of cryogenic low-pass filters

Review of Scientific Instruments 88, 114703 (2017); 10.1063/1.4995076
 REVIEW OF SCIENTIFIC INSTRUMENTS 88, 114705 (2017)

Howland current source for high impedance load applications

K. F. Morcelles,a) V. G. Sirtoli, P. Bertemes-Filho, and V. C. Vincence
Department of Electrical Engineering, Santa Catarina State University, Joinville 89.219-710, Brazil
(Received 18 September 2017; accepted 4 November 2017; published online 21 November 2017)
For Electrical Impedance Spectroscopy (EIS) applications, the Enhanced Howland Current Source
(EHCS) is a popular choice as an excitation circuit due to its simplicity, reliability, and safety. However,
its output impedance degradation at high frequency leads to errors that are unacceptable for high load
impedance applications, such as the ones which use dry or microelectrodes. Based on a proposed
mathematical model, this work designed an EHCS circuit which includes an output current buffer
and frequency compensation. PSpice simulations were performed as proof of concept, and then the
measured data were collected for comparison. For the proposed circuit, called here Load-in-the-Loop
Compensated Enhanced Howland Source (LLC-EHCS), the results showed that the output current
errors are lower than 1% up to 3.7 MHz over the load range of 560–2200 Ω and 1.2 MHz with 5.6 kΩ.
On the other hand, for the case of the standard EHCS circuit, these frequencies are 170 and 80 kHz,
respectively. Also, the output linear swing was found to be 3 times higher than the EHCS. It can be
concluded that the proposed LLC-EHCS may be widely used as an excitation circuit for high load and
wide bandwidth EIS applications. Published by AIP Publishing. https://doi.org/10.1063/1.5005330

I. INTRODUCTION use of current sources simplifies the measurement and allows

direct methods of identification.7 On the other hand, a volt-
Electrical Impedance Spectroscopy (EIS) is a non-
age source has better frequency performance than the current
invasive technique based on the analysis of electrical properties
source,5,8,9 which has its output impedance degraded at higher
to identify the composition of biological material under study.1
frequencies10 (>1 MHz). When high frequency EIS or EIT
The method consists of the injection of an electrical signal
systems are needed, designers usually choose a voltage source
and measurement of the material electrical response. Simi-
for their equipment, sometimes having an alternative current
larly, Electrical Impedance Tomography (EIT) uses the same
mode option.9,11–14 They may also use controlled current pat-
technique as EIS (see the definitions in the Nomenclature).
terns with voltage sources6 despite the practical advantages of
EIT is an imaging method which uses electrical impedance to
current source excitation.
assess the material.1 This technique may be used as an alter-
Most current sources used in bioimpedance applications
native to other imaging methods, such as X-ray, ultrasound,
are usually Voltage Controlled Current Sources (VCCSs).
and magnetic resonance imaging. X-ray suffers from radiation
They can be split into two groups: voltage mode and current
hazards and significant rate of inconclusive results.2 Further-
mode.15 In the voltage mode group, the current is generated by
more, magnetic resonance imaging is an expensive technique,2
a voltage in the active circuit,15 and the most popular topolo-
and ultrasound may not be an efficient method for detecting
gies are the load-in-the-loop, Howland, and Tietze sources.
microcalcifications.3 EIT can also detect early malignancy in
On the other hand, in the current mode group, the current
tissue, as changes in the electrical properties start to appear
is generated directly by an active device, such as transcon-
in malignant tissue before symptoms develop.1 However, the
ductance amplifiers (OTAs), current conveyors (CCs), and
evaluation of impedance values in EIT is a non-linear ill-posed
current mirrors (CMs). Some designers mix both modes by
problem,4 requiring highly precise instruments for both signal
using a CC or OTA device with a load-in-the-loop stage.16–18
generation and acquisition.
The voltage mode is the preferred choice by many design-
One of the main components in an impedance interro-
ers since it is easy to implement in a discrete design and it
gation system is the excitation source. This excitation can be
is only necessary to have one or two operational amplifiers
applied by either an electrical voltage or current, depending on
(OPAMPs) with few passive elements (resistors and capac-
the topology of the circuit. The main EIS systems use a current
itors). The major drawback of this mode is that the output
source. The current source can be adjusted to deliver the maxi-
impedance is highly affected by the resistor mismatch and
mum recommended current into tissue, without being affected
non-idealities of the OPAMP.19–25 The current mode is almost
by the skin-electrode impedance. This ensures a maximum
independent of external components but not very accessible
resolution within the medical safety limits, a load flexibility,
since CCs and OTAs are not widely available commercial
and optimization of image distinction,1,5,6 which are inherent
devices as OPAMPs, pushing the researchers to design and
advantages of this operation mode over voltage sources. Also,
manufacture the Integrated Circuit (IC).16–18,25,26
in time-variant impedance assess with harmonic spectra, the
The Enhanced Howland Current Source (EHCS) circuit is
one of the most used current sources in EIS measurements19,20
a)Authorto whom correspondence should be addressed: kauefelipems@ due to its versatility, low complexity, and single active device
gmail.com topology.27 This well-known circuit is suitable for most of the

0034-6748/2017/88(11)/114705/8/$30.00 88, 114705-1 Published by AIP Publishing.

114705-2 Morcelles et al. Rev. Sci. Instrum. 88, 114705 (2017)

low-frequency applications. However, complex applications

may need improved EHCS circuits, which can be developed by
a variety of topologies, methods, and auxiliary circuits. These
may include circuits such as (i) mirrored current pumps28,29
and bridge configurations30–32 to reduce common-mode inter-
ference and increase compliance, (ii) buffers in the feed-
back branch to increase output impedance,30,33,34 (iii) lead-lag
compensations19,31 to get stability, (iv) General Impedance
Converters (GICs)35,36 and Negative Impedance Converters
(NICs)29,37 to reduce stray capacitance effects, and (v) How-
land followed by a load-in-the-loop current buffer stage38 to
increase compliance and output impedance.
The objective of this work is the design of a wide fre-
quency bandwidth enhanced Howland source, working beyond
1 MHz with a wide range of loads without the need of constant
calibration. The Howland circuit will need low output current
error and high voltage compliance. Also, it is proposed as a FIG. 1. Schematic of the EHCS circuit, where RL is the resistive load and
reliable mathematical model for designing the optimal circuit. Vin+ and Vin are the input voltages. In this work, Vin+ is the signal and Vin
is at 0 V.

II. METHODOLOGY
The components were selected according to a developed
The EHCS was divided into two parts: core and aux-
MATLAB algorithm to find the resistor values which give the
iliary circuit. The core represents the main topology; this
maximum output impedance. It also considers the transcon-
may be the basic Howland current source (BHCS) or the
ductance (G), desired bandwidth (BW), output swing, load
enhanced one (EHCS) with or without a buffer in the posi-
range, and amplifier parameters (gain and bandwidth). The
tive feedback loop [buffered EHCS (BEHCS)]. The auxiliary
algorithm conducts a sweep on parameters α (=R2A /R2B ) and
circuit represents the circuitry for increasing the performance
β (=R4 /R3 ) (these resistances are shown in Fig. 1), to find
of the core in a desired way, such as GIC, NIC, and lead-lag
the maximum output impedance. This algorithm is based on
compensators.
the equations presented in Sec. III. Care was taken in the
The decision procedure to choose the best combination of
layout of the circuit by making a good ground plane and min-
core and auxiliary circuit was based on the literature and theo-
imum of length traces, reducing stray capacitances, parasitic
retical evaluation. Also, by using the governing equations and
inductances, cross talk, and electromagnetic interferences.
simulations of the circuit, the optimal configuration was found
and tested experimentally. To compare the novel circuit to the
literature topologies, measurements were made by simulations
and experimental analysis of frequency response, phase, and III. THEORETICAL EVALUATION
output swing. A. The core
Simulations were conducted in PSPICE version 16.3,
and experimental measurements were collected by using a The EHCS topology was chosen over the BHCS since its
Tektronix oscilloscope (model MSO 4034) and a Tektronix performance is better.27 In order to decide among the EHCS
function generator (model AFG 3251). Both theoretical evalu- topology variations found in the literature and design the cir-
ations and experimental data analysis were processed in MAT- cuit with optimal performance, this work used a mathematical
LAB version 7.8.0. The PSPICE model of the AD825, pro- model of the EHCS based on the equations presented by
vided by Analog Devices, was used to simulate the OPAMPs Tucker et al.19 These equations describe the output impedance,
implemented in this work. transconductance, noise, and the maximum load for a linear
The output current was calculated by dividing the mea- response. It considers the input signal connected to the positive
sured voltage across the load to the measured resistance. input of the EHCS and a first order OPAMP model. Figure 1
Frequency response was tested, both in experiment and shows the EHCS circuit used here. To increase the accuracy
in simulation, using 4 different resistance values ranging in the calculations, some modifications in the equations were
from 560 to 5600 Ω. 25 discrete frequencies were analyzed made based on circuit analysis. It was considered the transcon-
ranging from 100 Hz to 10 MHz. The performance of the ductance bandwidth, which was neglected by Tucker.19
circuits was evaluated based on the output current error com- A simple circuit inspection showed that the transconductance
pared to the low frequency transconductance gain. The out- bandwidth is dictated by the amplifier gain-bandwidth product
put swing was measured by applying a low frequency input (f H ) and the negative feedback gain. The following equation
voltage with different load values while the input amplitude describes the transconductance G with its frequency dependent
was kept at 1.65 V. Then, the maximum load that main- term:
tains a linear operation could be evaluated. Average was R2A + R2B 1
G+ = . (1)
used in the measurements to reduce the effect of noise and R1 R2A 1 + R4 /R3 s + 1
disturbances. 2πfH
114705-3 Morcelles et al. Rev. Sci. Instrum. 88, 114705 (2017)

In order to model the effects of resistor mismatch, finite achieve better results with the BEHCS circuit, but at the cost
gain, and bandwidth of the OPAMP, the output impedance of transconductance bandwidth, according to Eq. (1). There-
(Z 0 ) can be represented in terms of two resistances (R01 and fore, it was considered the simple EHCS topology for the core
R02 ) and a capacitance C 0 , according to Tucker.19 However, circuit.
it does not consider the effects of common mode gain, which
are significant in high gain amplifiers. Therefore, this effect B. The auxiliary circuit
was modeled by adding this term, as shown in the follow-
ing equation, where AOL is the open-loop DC gain of the It has been found in the literature that the EHCS design
amplifier: has its output impedance degraded by parasitic capacitances
and non-idealities,19,20 but this may be reduced using auxiliary
R
1 + AOL R +3 R circuits. Generalized impedance converters35,36 and negative
3 4
RO2 = (R2B //(R2A + R1 )) . (2) impedance converters29,37 have been used to reduce capaci-
AOL R3
1 + CMRR R + R tance effects, but the improvement is very limited in band, not
3 4
being suitable for broadband applications, for example. One
Equation (3) represents the maximum acceptable load to
topology which increases the output impedance and swing,
get a linear operation for a given fixed output current. This
keeping its robustness in a wide bandwidth without calibra-
is limited by voltage compliance, and it was modified from
tions, is the one proposed by Zanganeh38 (see Fig. 2). This
the study of Tucker et al.19 by adding a denominator term,
is a load-in-the-loop current source at the output of an EHCS
representing the current that flows through the positive feed-
circuit, working as a current buffer output stage. The load-in-
back resistors, where V SAT is the saturation voltage of the
the-loop EHCS (LL-EHCS) circuit isolates the load from the
amplifier,
EHCS source, increasing both the output swing and output
VSAT
I − R2B impedance (which is 1 + AOL 2 times higher than Z 0 ). AOL 2 is
RL,max = OUT . (3) the frequency dependent open-loop gain of the second ampli-
1 + R R+2BR fier. The results obtained by Zanganeh38 were only based on
1 2A
The rest of the equations presented by Tucker 19 were kept simulations, so a deeper theoretical analysis and experimental
unchanged. All equations used here were previously proved data were needed.
by simulations with several OPAMPs. By using this model, it By using the proposed EHCS model and some analysis
was possible to choose the best configuration for application over the LL-EHCS circuit, it further investigated the theoret-
required and then to design an appropriate auxiliary circuit. ical improvements obtained from this topology. It was found
The circuit used by Constantin and Gheorghe30 introduces that the negative input capacitance of the second amplifier is
a buffer between the output of the Howland source and its pos- a critical parasitic component to this topology, reducing the
itive feedback loop, called here BEHCS. This buffer increases overall output impedance of the circuit and generating a peak
both the output resistance and swing of the EHCS circuit, as response at high frequency. The following equation describes
it forces the output current to flow only in the load and not the output impedance considering the negative input capac-
through the feedback loop. Adapting Eqs. (2) and (3) with the itance of the second amplifier (C in2 ), the finite gain (AOL 2 ),
model presented by Tucker 19 to the BEHCS case, one can find the bandwidth (f C2 ), and the output impedance of the EHCS
that both the output resistance and swing are increased by the stage (Z 0 ):
terms ∆Z 0 and ∆RL ,m áx , considering the ratio R4 /R3 = 1, as !
ZO AOL2 +
shown in the following equation: ZO,LL−EHCS = × *.1 + s /. (5)
sCin2 ZO + 1 +1
R2B 2πfC2
∆RL,máx = ∆ZO = 100%
, -
. (4)
R2B + 2R2A The output swing of the overall circuit is higher than the
The high output impedance of the EHCS circuit can only standard circuit, but the first term shows the degradation caused
be achieved if its resistor values follow the well-known rela- by C in2 . Equations (6) and (7) describe the maximum load
tionship, which is (R2A + R2B )/R1 = R4 /R3 .27 For the BEHCS supported by the first and the second stages, respectively. The
case, the relationship is R2A /R1 = R4 /R3 . Equation (4) sug- maximum load supported by the whole circuit is the minimum
gests that the increasing performance provided by the BEHCS between the values resulted from Eqs. (6) and (7). It can be
is proportional to the ratio R2B /R2A and reaches a maximum noticed that both equations are frequency dependent, which
of 100% when R2B  2R2A . However, increasing the ratio
R2B /R2A without precaution may lead to similar problems of
the BHCS topology.27 For the BEHCS circuit, in the typical
case of R2B = R2A , both the output resistance and swing are,
at maximum, 33.3% higher than the EHCS. This results in an
increase of 33.3% in the range of loads that can be used to a
fixed output current error. To the requirements of this work,
that improvement is not enough to justify the introduction of
another amplifier in the circuit, which increases cost and com-
plexity and may introduce new parasitic components that can FIG. 2. The EHCS circuit followed by a load-in-the-loop stage,38 where RL
compromise the design. The increase of the ratio R4 /R3 may is a resistive load.
114705-4 Morcelles et al. Rev. Sci. Instrum. 88, 114705 (2017)

is because the load seen by the EHCS stage depends on the

second amplifier’s gain,
VSAT
I − R2B AOL2 +
Zlmax1,LL−EHCS = OUT × *.1 + s / , (6)
1 + R R+2BR 2πfC2
+1
1 2A , -
s +1
VSAT * 2πfC2
Zlmax2,LL−EHCS = × .1 + +/ . (7)
IOUT AOL2
, -
The peak response generated by the second amplifiers,
due to its input capacitance, was not modeled in this work, but
PSPICE simulations pointed out this peak, which increases
with increasing load. A compensation circuit was designed to
FIG. 4. Output impedance calculated from the equations (EQ) and simulated
reduce the error, consisting of a capacitor C and a resistor R (SIM) for the 3 different types of Howland circuits.
in series which, in turn, are in parallel to the load (see Fig. 3).
This introduces a pole and a zero in the output current equa-
tion, as shown in Eq. (8). It can be noticed that the pole is
A. Simulations
controlled by the compensation circuit (R and C) and the load
(RL ). This reduces the output current in the peak frequency. The compensation circuit was adjusted to achieve the low-
Also, the frequency and amount of reduction are proportional est current errors over the desired frequency range, resulting in
to RL , resulting in a compensation which is proportional to the a resistor R of 8.1 kΩ and a capacitor C of 2.0 pF. Simulations
peak to be removed. The use of a resistor R allows the compen- were performed for the 3 types of circuits (EHCS, LL-EHCS,
sation circuit to be precisely adjusted empirically by using a LLC-EHCS) in the frequency range of 100 Hz–100 MHz. The
variable resistor. The circuit shown in Fig. 3 is called the LLC- output impedance spectra provided by the presented equations
EHCS (stands for Load-in-the-Loop Compensated Enhanced and PSPICE simulations are shown in Fig. 4.
Howland Current Source), The dependency of the output current to different load val-
! ues (0.00, 0.56, 1.00, 2.20, and 5.60 kΩ) was investigated over
sCR + 1 the frequency range of 10 kHz and 100 MHz. The results are
IO,LLC−EHCS = IO,LL−EHCS × . (8)
sC(R + RL ) + 1 shown in Fig. 5. The specific values of the loads were chosen
to match the resistor values used in the posterior experimental
procedure.
IV. RESULTS
Table I shows the best results of the mainly important char-
acteristics of two Howland topologies. Z lmax is defined as the
The equations presented here were implemented in a
MATLAB algorithm in order to find the optimum circuit con-
figuration, once a given input specification is known. It used an
input voltage of 3.3 VPP , a maximum output current of 1 mAPP ,
and a symmetrical power supply of ±5 V. The source transcon-
ductance gain was set to 303.03 µS for a maximum BW of
10 MHz. The OPAMP AD825 was used due to its high com-
mon mode rejection (=80 dB), high input impedance (=500
MΩ in parallel to 6 pF), and sufficient gain-bandwidth prod-
uct (=40 MHz) to operate in the megahertz range. The results
of the implemented algorithm showed a best solution for the
circuit by using the resistors R2A = R2B = R1 = R4 = 6.6 kΩ
and a resistor R3 = 3.3 kΩ.

FIG. 3. The LLC-EHCS circuit with the compensation network, where the FIG. 5. Frequency response of (a) EHCS, (b) LL-EHCS, and (c) LLC-EHCS
values of R and C were empirically calibrated. circuits for the transconductance gain at five different loads.
114705-5 Morcelles et al. Rev. Sci. Instrum. 88, 114705 (2017)

TABLE I. Calculated and simulated circuit parameters of both EHCS and

LLC-EHCS topologies.

Parameter EHCS LLC-EHCS

G (µS) 302.9 302.9

BW (MHz) 13.1 13.2
Z lmax (equation) (kΩ) 2.3 10.0
Z lmax (simulation) (kΩ) 3.0 9.5
CV G (%) 7.7 5.8
CV BW (%) 8.9 4.4

maximum load which gives a linear current-to-voltage ratio.

This parameter was evaluated by measuring the voltage drop
on the load (for the EHCS) and the output voltage of the sec-
ond amplifier (for the LLC-EHCS). This is because, in the case
of the LLC-EHCS, the second stage saturates before the first
stage, changing the topology back to the EHCS (this obser-
vation can be made considering the second OPAMP, a fixed
voltage source after saturation), creating a non-linear operation
that results in distortion in high frequencies. The Monte Carlo FIG. 6. Transconductance gain frequency response of three Howland topolo-
analysis was also made, using 1000 random samples. Table I gies at different loads. (a) EHCS, (b) LL-EHCS, and (c) LLC-EHCS.
shows the ratio between the standard deviation and average
transconductance (coefficient of variation—CV G ) and 3 dB
bandwidth (CV BW ), considering resistors of 5% tolerance and
a load of 2.2 kΩ.

B. Experiments
To reduce the effects of resistor mismatch, R1 was replaced
by a variable resistance and then was empirically calibrated
to achieve the maximum output impedance. This allows the
output resistance to be affected only by the OPAMP non-
idealities, hence predicting it using only Eq. (2). A capacitor
was placed in series to R2B to avoid DC positive feedback.
In order to find the maximum output impedance operation
point, a low frequency voltage signal, in series with a resis-
tor of 62 kΩ, was applied at the output of the EHCS stage. It FIG. 7. Phase response of the LLC-EHCS topology with five different
resistive loads.
was assumed that the maximum output impedance gives the
minimum voltage drop on the 62 kΩ resistor, so by measur-
ing this drop it was possible to find the maximum Z 0 without
knowing its value. The compensation circuit was empirically Figure 8 shows the positive voltage swing and negative voltage
adjusted to get the lowest current errors over the frequency swing of both the EHCS and LLC-EHCS as a function of
range. The practical result was a resistor R of 1.33 kΩ and a loading.
capacitor C of 7 pF. Figure 6 shows the measured transcon-
ductance of the three Howland topologies using 4 different
resistive loads (0.56, 1.0, 2.2, and 5.6 kΩ). The base value
for transconductance (G) was measured using a 50 kHz sig- TABLE II. Experimental circuit parameters for both EHCS and LLC-EHCS
topologies.
nal with a load of 560 Ω, which is in the band of operation.
Figure 7 shows the measured phase response of the proposed Parameter EHCS LLC-EHCS
LLC-EHCS circuit. In order to investigate for abnormal peaks
between 2.2 and 5.6 kΩ loading, measurements using a 3.3 kΩ G 304.5 µS 304.2 µS
load was also collected for both LL-EHCS and LLC-EHCS BW (3 dB) 5.0 MHz 7.0 MHz
f 1% (@560 Ω) 490.0 kHz 3.7 MHz
circuits.
f 1% (@1 kΩ) 210.0 kHz 3.9 MHz
Measured transconductance (G), bandwidth (BW ), and
f 1% (@2.2 kΩ) 170.0 kHz 3.9 MHz
maximum load impedance (Z lmax ) are shown in Table II. Also, f 1% (@5.6 kΩ) 80.0 kHz 1.2 MHz
it shows the maximum frequency (f 1% ) that yielded output Z lmax 2.0 kΩ 7.5 kΩ
current error (relative to G) below 1% for the different loads.
114705-6 Morcelles et al. Rev. Sci. Instrum. 88, 114705 (2017)

whereas 10 times bigger at 1 MHz than the ones found in the

EHCS circuit (see Fig. 4). These results agree with the simula-
tion results reported by Zanganeh.38 However, the EHCS out-
put impedance was higher than the LL-EHCS above 20 MHz.
This effect was not modeled in Eq. (2), for simplification. How-
ever, this may be explained by the fact that close to the unity
gain frequency f H (e.g., 40 MHz for AD825) the influence of
the amplifier on the output impedance is very small. By cir-
cuit analysis, one can find that the EHCS output impedance
tends to R2B //(R2A + R1 ) as the frequency exceeds f H , while
the LL-EHCS output impedance is further reduced due to the
input capacitance of the second amplifier. Although it seems
to be a disadvantage for the proposed circuit, it only occurs
beyond the operation frequency range of the transconduc-
tance. It must be emphasized that the practical EHCS output
FIG. 8. Output voltage swing as a function of loading for both EHCS and impedance is easily degraded by parasitic capacitances, so the
LLC-EHCS circuits. The ideal curve represents a fully linear operation.
real output impedance will not remain in a constant value.
Furthermore, it can be seen in Fig. 4 that the calculated out-
V. DISCUSSIONS put impedance agreed with the simulated one for EHCS and
LL-EHCS.
Both simulation and experimental analyses showed As pointed out by Zanganeh,38 the LL-EHCS circuit per-
expected values of transconductance gain (G) with respect to mits the use of OPAMPs with much lower f H than in the EHCS
the ones from the MATLAB algorithm. Although one could case, facilitating the choice of the amplifier. However, Eq. (5)
expect an error in the experimental G value, as it was mea- shows that the input capacitance of the second amplifier is one
sured using 560 Ω load at 50 kHz, simulations showed that the of the main practical limitations of this topology. This capac-
error in the G value is negligible in these conditions. Accord- itance effect controls the output impedance of the first circuit
ing to Tables I and II, the simulated bandwidth is significantly at higher frequencies. This means that the choice for a good
bigger than the experimental one. The simulated and exper- amplifier can be still a challenge in this design. Furthermore,
imentally implemented compensation circuit components the OPAMP input capacitance also generates a transconduc-
(R and C) were also significantly different. Besides model tance peak at higher frequencies, as shown Fig. 5(b). This
errors and layout effects, it was verified that the capacitance effect was not modeled in this work, but it is well-known that
of the probe is added to the capacitance of the second ampli- the input capacitance reduces the phase margin of an OPAMP-
fier input. This effect increases the peak [see Fig. 6(b)], which based amplifier, generating ringing or instability.19 Since this
consequently demands more compensation than the simulated is a simulation result, the peak effect cannot be related to layout
circuit. In order to confirm this hypothesis, the LL-EHCS cir- effects or ground loops.
cuit was simulated by including the circuit probe model at the The proposed compensation circuit is just a simple way
measured points. The results from these simulations showed an of correcting the error introduced by the input capacitance of
increasing peak of the transconductance as observed in the real the second amplifier. The compensation is proportional to the
circuit. Figure 9 shows the simulated transconductance gain load [see Figs. 5(b) and 5(c)]. The drawback of this technique
with and without the oscilloscope probe effect, using a load of is the reduction of output impedance (see Fig. 4), as R and
2.2 kΩ and probe impedance according to the manufacturer’s C are placed in parallel to the load. Although one can argue
guide. that this nulls the objective of LL-EHCS topology, the rea-
The calculated and simulated output impedance of the LL- son why designers try to achieve high output impedance is
EHCS was approximately 5600 times bigger at low frequency, to reduce the output current error with increasing load. Given
that the compensation can be empirically adjusted by simply
varying the resistor R, the LLC-EHCS offers a low current
error even without having the very high output impedance
provided by the LL-EHCS. Other different compensation tech-
niques will be implemented in future studies, such as lead-lag
compensation,19 NIC circuit for cancelling out the OPAMP
input capacitance, or negative feedback loops for controlling
the critical poles of the system.
Results showed that the EHCS bandwidth is very depen-
dent on the load value [see Fig. 5(a)], resulting in a significantly
high frequency error. It is important to know that the less
the current at higher frequencies, the more the degradation of
the signal-to-noise ratio (SNR) is. In other words, the flatter the
FIG. 9. Simulated transconductance gain of the LL-EHCS circuit as a output current frequency response, the more predictable it is
function of frequency with and without the oscilloscope probes. [see Fig. 5(c)]. Figure 5(c) shows that the LLC-EHCS yielded
114705-7 Morcelles et al. Rev. Sci. Instrum. 88, 114705 (2017)

a flat transconductance frequency response up to higher fre-

quencies than the EHCS, for a load range from 0 to 2.2 kΩ.
Also, in this load range, the transconductance LLC-EHCS
curves barely changed, which is not the case for the EHCS,
showing that the circuit is less dependent on the load value and,
therefore, more predictable. For the 5.6 kΩ load, the improve-
ment in the frequency response is almost one decade higher
than the EHCS circuit. Although Figs. 5 and 6 showed results
only above 100 kHz, measurements were made from 100 Hz,
but no additional information was founded below 100 kHz.
Experimental results confirmed the improvement obtained
by simulations, as shown in Fig. 6. A load of 3.3 kΩ was added
in the analysis for both LL-EHCS and LLC-EHCS circuits
to confirm the transconductance peak reduction at higher fre-
quencies, according to simulations. It must be emphasized that
the output impedance was not directly measured due to instru-
ment limitations. According to the results shown in Table II,
FIG. 10. Time response to (a) square (1 MHz) and (b) sinusoidal (5 MHz)
the proposed circuit reached 3.7 MHz with errors below 1% input waves.
over a load range from 560 to 2.2 kΩ, whereas 170 kHz for the
EHCS topology in the same conditions. When a load of 5.6 kΩ
was used, f 1% increased by more than one decade in the fre- signal up to 1 MHz whereas 5 MHz for a sine wave signal
quency bandwidth, confirming the superiority of the proposed (see Fig. 10). The square excitation showed only a small over-
circuit over the whole load range. The negative phase shift shoot, which was partially caused by the voltage generation
result was expected, which is due to the amplifier behavior on itself. This result suggests the possibility of using the LLC-
high frequencies. However, the LLC-EHCS phase shift was EHCS with DIBS signals up to 1 MHz. The approach used
significantly affected by the load (see Fig. 7), which was also by the LLC-EHCS is more suited to this kind of signal than
observed in the simulation results. The phase effects at lower the GIC or NIC configuration proposed in other studies29,35–37
frequencies are due to the DC decoupling capacitor. because while the latter improved performance on a very lim-
Output voltage swing was analyzed in terms of having a ited frequency band, the LLC-EHCS reduced output error in a
maximum load for a linear circuit operation. This parameter broadband width, which in this work covered frequencies from
is very important when designing this type of circuit, espe- 10 to 3700 kHz. No reducing noise techniques were applied
cially when both the input voltage and supply voltage cannot whatsoever in the signals seen in Fig. 10 (different from the
be changed in the design. In simulations, the AD825 model sat- previously showed measurements), keeping the raw data to
urated near 5 V. In these conditions, the LLC-EHCS maximum show the noise condition of the measured signal.
load was 3 times bigger than the EHCS (Table I), which agreed Most biological tissue measurements need to remove the
with the mathematical model results. Improvement was con- offset current to keep heat out of data, to avoid electrode polar-
firmed experimentally, as shown in Table II and Fig. 8. The ization, and for safety issues. This can easily be done putting
proposed topology showed better swing improvement com- a capacitor in series with tissue or using a DC feedback loop.
pared to the symmetrical EHCS28,29 and bridge EHCS,30,32 The drawback of this solution in the LLC-EHCS is the loss
which also uses two or more OPAMPs. When one can keep of polarization by the second amplifier (see Fig. 3). In order
Z lmax 1,LL-EHCS higher than Z lmax 2,LL-EHCS [see Eqs. (6) and (7)], to close the DC negative loop, a high value resistor Rf needs
then the LLC-EHCS maximum load will not depend on the to be put in parallel to the RC circuit of the second amplifier.
EHCS resistors’ values, increasing the degree of flexibility of However, this will reduce the low frequency output impedance
the design. of the LLC-EHCS circuit but will not interfere at high frequen-
The coefficients of variation for transconductance and cies if an optimum value is chosen to minimize the degradation.
bandwidth, shown in Table I, suggest that the LLC-EHCS is One may also use a DC feedback loop instead, such as the off-
also less susceptible to component variations, as the transcon- set compensation proposed by Pliquett,40 but the appropriate
ductance variations are smaller than the ones for the EHCS feedback circuit to be used with the LLC-EHCS is out of the
circuit. scope of this work.
Excitation methods for broadband signal applications may The electrical characteristics of the proposed topology
depend on an optimized waveform, such as Discrete-Interval suggest that this may be applied in applications using high
Binary Sequence (DIBS)39 and multisine.39 In these cases, impedance loads. Usually, the load impedance is dominated
it is very important to check for signal quality in terms of by the interface of the electrode and tissue, especially for
noise, distortion, and step response. The time response of the applications using small dry electrodes. These characteristics
LLC-EHCS to sinusoidal and square waves, with 5 MHz and are found to be true in EIT,41 cell culture monitoring,42 and
1 MHz, respectively, is shown in Fig. 10, to analyze the qual- microelectromechanical systems (MEMSs).43 However, for
ity of the signal in typical implementations. Both input signals applications using switching systems like EIT, the topology
had an amplitude of 1.65 V and a load of 1 kΩ. The pro- must be previously tested to accommodate the high leakage
posed circuit showed a predictable response for an input square capacitance of the multiplexing network.
114705-8 Morcelles et al. Rev. Sci. Instrum. 88, 114705 (2017)

VI. CONCLUSION 11 T. R. Qureshi, B. Mehboob, C. R. Chatwin, and W. Wang, in 38th Annual

Northeast Bioengineering Conference (NEBEC 2012) (IEEE, 2012),
This work proposed a modification in the Enhanced How- p. 353.
12 S. Khan, P. Manwaring, A. Borsic, and R. Halter, IEEE Trans. Med. Imaging
land Current Source (EHCS) by reducing the high frequency
output current error dependent on the load and increasing the 34, 888 (2015).
13 R. J. Halter, A. Hartov, and K. D. Paulsen, IEEE Trans. Biomed. Eng. 55,
voltage output swing of the current source. Those improve- 650 (2008).
ments may facilitate the use of discrete current sources in 14 G. J. Saulnier, N. Liu, C. Tamma, H. Xia, T.-J. Kao, J. C. Newell,

bioimpedance systems using the high impedance dry elec- and D. Isaacson, in Conference Proceedings of the IEEE Engineering in
Medicine and Biology Society (IEEE, 2007), p. 4154.
trodes or microelectrodes. Also, it may apply for applications 15 D. Bouchaala, O. Kanoun, and N. Derbel, Measurement 79, 339 (2016).
where there is a poor or failure contact between the electrode 16 F. Seoane, R. Bragós, and K. Lindecrantz, in Annual International Con-

and the tissue sample, especially for skin measurements where ference of the IEEE Engineering in Medicine and Biology Society (IEEE,
the stratum cornea impedance is very high. 2006), p. 5121.
17 P. Annus, A. Krivoshei, M. Min, and T. Parve, in I2MTC 2008-IEEE

Instrumentation and Measurement Technology Conference, 2008.

18 R. Macı́as, F. Seoane, and R. Bragós, J. Phys.: Conf. Ser. 224, 12008 (2010).
ACKNOWLEDGMENTS
19 A. S. Tucker, R. M. Fox, and R. J. Sadleir, IEEE Trans. Biomed. Circuits
We thank the State University of Santa Catarina (UDESC) Syst. 7, 63 (2013).
20 T. R. Qureshi, C. R. Chatwin, N. Huber, A. Zarafshani, B. Tunstall, and
for the institutional support and the Foundation Research of
W. Wang, J. Phys.: Conf. Ser. 224, 12167 (2010).
Santa Catarina (FAPESC) and Fulbright Brazil for the financial 21 H. Hong, M. Rahal, A. Demosthenous, and R. H. Bayford, in European Con-
support. ference on Circuit Theory and Design 2007, ECCTD 2007 (IEEE, 2008),
p. 208.
22 M. Rafiei-Naeini and H. McCann, Physiol. Meas. 29, S173 (2008).
NOMENCLATURE 23 D. Bouchaala, Q. Shi, X. Chen, O. Kanoun, and N. Derbel, in International

EHCS Enhanced Howland Current Source Multi-Conference on Systems, Signals and Devices, SSD 2012-Summary
Proceedings, 2012.
BHCS Basic Howland Current Source 24 X. Zhao, S. Kaufmann, and M. Ryschka, in Proceedings of the 5th
BEHCS Buffered EHCS International Workshop on Impedance Spectroscopy, S. 1-2.
25 H. Hong, M. Rahal, A. Demosthenous, and R. H. Bayford, Physiol. Meas.
LL-EHCS Load-in-the-Loop EHCS
30, 999 (2009).
LLC-EHCS Load-in-the-Loop Compensated EHCS 26 A. A. Silverio and A. A. Silverio, Int. J. Eng. Appl. Sci. 1, 68 (2012), ISSN
GIC General Impedance Converter 2305-8269.
NIC Negative Impedance Converter 27 R. A. Pease, A Comprehensive Study of the Howland Current Pump AN-

OPAMP Operational Amplifier 1515, 2008.

28 P. Bertemes-Filho, L. H. Negri, A. Felipe, and V. C. Vincence, J. Phys.:
EIS Electrical Impedance Spectroscopy Conf. Ser. 407, 12030 (2012).
EIT Electrical Impedance Tomography 29 P. Bertemes-Filho, “Tissue characterisation using an impedance spec-

CC Current Conveyor troscopy probe,” PhD thesis, University of Sheffield, 2002.

30 A. V. Constantin and G. I. Gheorghe, in 2016 International Semiconductor
OTA Operational Transconductance Amplifier
Conference (IEEE, 2016), pp. 183–186.
CM Current Mirror 31 J. Liu, X. Qiao, M. Wang, W. Zhang, G. Li, and L. Lin, Rev. Sci. Instrum.

85, 055111 (2014).

1 S. Grimnes and Ø. G. Martinsen, Bioimpedance and Bioelectricity Basics 32 C. J. Poletto and C. L. Van Doren, IEEE Trans. Biomed. Eng. 46, 929 (1999).

(Academic, 2008). 33 D. Bouchaala, O. Kanoun, and N. Derbel, in 3rd IMEKO TC13 Symposium
2 E. C. Fear, S. C. Hagness, P. M. Meaney, M. Okoniewski, and M. A. Stuchly, on Measurements in Biology and Medicine, 2014.
IEEE Microwave Mag. 3, 48 (2002). 34 D. Bouchaala, Q. Shi, X. Chen, O. Kanoun, and N. Derbel, in 10th Inter-
3 E. Y. K. Ng, S. V. Sree, K. H. Ng, and G. Kaw, Technol. Cancer Res. Treat. national Multi-Conference on Systems, Signals & Devices, SSD 2013,
7, 295 (2008). 2013.
4 X. Zhang, W. Wang, G. Sze, D. Barber, and C. Chatwin, IEEE Trans. Med. 35 T. R. Qureshi, C. Chatwin, and W. Wang, APCBEE Procedia 7, 42 (2013).

Imaging 33, 2223 (2014). 36 A. S. Ross, G. J. Saulnier, J. C. Newell, and D. Isaacson, Physiol. Meas. 24,
5 Y. Mohamadou, T. I. Oh, H. Wi, H. Sohal, A. Farooq, E. J. Woo, and A. L. 509 (2003).
McEwan, Meas. Sci. Technol. 23, 105703 (2012). 37 P. Bertemes-Filho, R. Lima, and M. Amato, in Proceedings of the 12th
6 M. H. Choi, D. Isaacson, G. J. Saulnier, and J. C. Newell, in Proceedings of International Electrical Bioimpedance and 5th Conference on Electrical
25th Annual International Conference of the IEEE Engineering in Medicine Impedance Tomography (Gdansk, Poland 2004), p. 645.
and Biology Society (IEEE, 2003), p. 3114, Cat. No. 03CH37439. 38 M. Zanganeh, Int. J. Comput. Network Technol. 1, 205 (2013).
7 B. Sanchez, E. Louarroudi, R. Bragos, and R. Pintelon, Physiol. Meas. 34, 39 B. Sanchez, G. Vandersteen, R. Bragos, and J. Schoukens, Meas. Sci.

1217 (2013). Technol. 23, 105501 (2012).

8 R. J. Halter, A. Hartov, S. P. Poplack, R. Diflorio-Alexander, W. A. Wells, 40 U. Pliquett, M. Schonfeldt, A. Barthel, D. Frense, T. Nacke, and D. Beck-

K. M. Rosenkranz, R. J. Barth, P. A. Kaufman, and K. D. Paulsen, IEEE mann, Physiol. Meas. 32, 927 (2011).
Trans. Med. Imaging 34, 38 (2015). 41 S. Hong, K. Lee, U. Ha, H. Kim, Y. Lee, Y. Kim, and H. J. Yoo, IEEE J.
9 A. Hartov, R. A. Mazzarese, F. R. Reiss, T. E. Kerner, K. S. Osterman, D. B. Solid-State Circuits 50, 245 (2015).
Williams, and K. D. Paulsen, IEEE Trans. Biomed. Eng. 47, 49 (2000). 42 A. R. Abdur Rahman, D. T. Price, and S. Bhansali, Sens. Actuators, B 127,
10 B. Geeraerts, W. Van Petegem, W. Dehaene, M. Steyaert, and W. Sansen, 89 (2007).
in 1992 14th Annual International Conference of the IEEE Engineering in 43 H. J. Pandya, H. T. Kim, R. Roy, W. Chen, L. Cong, H. Zhong, D. J. Foran,

Medicine and Biology Society (IEEE, 1992), Vol. 5, p. 1703. and J. P. Desai, Sens. Actuators, B 199, 259 (2014).