Annals of Biomedical Engineering, Vol. 21, pp.

135-146, 1993

0090-6964/93 $6.00 + .00

1993 PergamonPress Ltd.

Printed in the USA. All rights reserved.

Complex Bioelectric Impedance Measurement System

for the Frequency Range from 5 Hz to 1 MHz
JAMES J. ACKMANN
Department of Veterans Affairs Medical Center and Department of Neurosurgery,
Medical College of Wisconsin, Milwaukee, WI

Abstract-Analytic techniques that have been successfully employed in materials science, and to a lesser extent in the study
of biologic systems, have potential for improving the application of bioelectric impedance provided that both real and imaginary impedance components can be measured with sufficient
accuracy over a given frequency range. Since biologic tissue,
particularly animal tissue, is typically highly conductive, phase
angles are small, making accurate measurements difficult. A
practical four-terminal system employing commercial lock-in
amplifiers is described and error sources and corrective techniques are discussed.

Keywords--Electric conductivity, Electronic instrumentation.

INTRODUCTION
Electrical impedance measurements have been used
since the late 1800s to study the electrical properties of
biologic tissue and to measure physiologic events. The
methodology has been previously reviewed (1,4). While
some investigations of tissue and cell suspensions have
employed complex measurements, the majority of published studies have utilized only the absolute value or the
real component. In contrast, in materials science research,
numerous studies employing complex impedance measurements have been conducted and a significant methodology using complex-plane plotting techniques along
with analytic methods has been developed; an overview
of the methodology has been presented (7). The techniques can be extended to biologic systems provided that
both real and reactive impedance components can be measured with sufficient accuracy over the frequency range
of interest. Biologic tissue, particularly animal tissue, is
typically highly conductive, and dissipation factors (defined as the ratio of the real component R to the reactive
component X ) are high; corresponding phase angles

Acknowledgment-This work was supported in part by the Department of Veterans Affairs Medical Center research funds.
Address correspondenceto James J. Ackmann, Departmentof Neurosurgery, Medical College of Wisconsin, 9200 W. Wisconsin Avenue,
Milwaukee, WI 53226.
(Received 4/22/91; Revised 1/29/92)

(tan -1 X / R ) are therefore small and obtaining the necessary data is not a trivial matter.
Instrumentation techniques for complex impedance
measurements have been previously reviewed (1,7). The
most commonly used technique applied to biologic tissue
has been the two-terminal bridge; however, two m a j o r
limitations of the technique limit applicability. Interfacial electrode impedance can be significantly greater than
sample impedance and techniques permitting corrections,
such as a variable-length conductivity cell, must be applied. The technique is limited to experiments in which
a tissue sample can be introduced into the cell and is, in
general, not applicable to studies in intact systems. O f
equal importance is the inherent limitation of bridge methods in measuring impedances with high dissipation factors (7). While impedance analyzers are commercially
available, these instruments are intended primarily for
network and component analysis and are basically autobalance bridges, which are subject to the same limitations
noted above. Measurements are usually restricted to systems with dissipation factors of less than ten. In contrast,
dissipation factors in animal tissue are frequently greater
than 100 in the frequency range o f interest. In addition,
the measurement terminals of the instrument are configured as "four-terminal pairs," that is, the current and
voltage terminals are connected internally, and the measurements are essentially two-terminal.
To eliminate electrode impedances f r o m the measurement, a four-probe configuration in which a known
measurement current is applied to a sample with one electrode pair, and the attendant voltage is measured with a
second electrode pair, is frequently used. For complex
measurements, a system capable of measuring both the
amplitude and the phase of the measured voltage with
respect to the applied current is required. This can be
accomplished by a technique known as phase-sensitive
detection. Commercial instruments, termed lock-in amplifiers, employing the method were introduced in 1972
and are used in a number of disciplines. Phase-sensitive
detection is accomplished by sequential mixing and timeaveraging operations. In addition to the signal to be mea135

136

J.J. ACKMANN

number. Maximum input sensitivities are approximately

100/~V full scale.
When properly applied, a four-probe measurement
system including a lock-in amplifier overcomes the limitations of bridge techniques: the effects of electrode impedance are minimized, measurements can be made in
systems with high dissipation factors, and the frequency
selectivity and sensitivity o f the amplifier permit measurements in a noisy environment with small excitation currents. While conceptually simple, a number of factors
can cause serious measurement errors. This communication describes a system consisting of two lock-in amplifiers along with ancillary circuits which are useful in the
frequency range from approximately 5 Hz to 1 MHz in
biologic tissue with dissipation factors up to approximately 200; the inherent error sources and system limitations are delineated. A Princeton Applied Research (PAR)
Model 5204 lock-in amplifier is used in the range from
5 Hz-100 kHz and a PAR Model 5202 lock-in amplifier
in the range from 100 kHz to 1 MHz.

sured, a reference signal at the frequency of interest must

be applied to the instrument. The reference is first converted to a square-wave; the signal to be measured, after
suitable amplification, is then multiplied by the reference
square wave. It is readily shown (7) that the resultant signal contains a DC term proportional to the amplitude of
the input signal the cosine of the phase angle @ between
the reference and measured signal plus time-varying terms
which are even harmonics of the reference signal; the latter components are removed with a low-pass filter. By internally shifting the phase of the reference signal by a
known amount until a maximum DC output is obtained
(cos @ = 1), both the amplitude and phase of the input
signal with respect to the reference can be determined.
In some instruments, two square-wave reference signals,
one in-phase and one at 90 ~ with respect to the reference
input, are generated internally and the multiplexingfiltering operation is performed with both reference signals to simultaneously measure both the in-phase and
quadrature components of the input signal. By separate
amplification of the two components, signals with realto-reactive ratios of several hundred to one can be resolved. It is also readily shown (7) that the detection
technique is highly frequency-selective with the exception
of odd-order harmonics of the reference signal; these are
attenuated by a factor of 1/n where n is the harmonic

ERROR SOURCES

A simple four-probe system using a PAR Model 5204

two-phase lock-in amplifier along with test results obtained in a physiologic saline bath are illustrated in Fig. 1.

REFERENCE

> 8.2. kA

(IZ

> 1,2 kA

>
T

SALINE BATH

-5
-10

-20

L
-25

-30
10J

104

10 s

FREQUENCY (Hz)

FIGURE 1. (Top) Basic four-probe impedance measurement system using lock-in amplifier; (bottom) phase errors.

Complex Bioelectric Impedance System

A Wavetek Model 185 function generator, in series with
a resistor significantly greater than the sample impedance,
is used to deliver a sinusoidal constant current to the sample. The reference signal for the lock-in amplifier is derived from a voltage divider connected to the same
oscillator. A second electrode pair is used to measure the
resulting voltage. Assuming that the amplifier input impedance is infinite, the sample impedance between the
sensing electrodes is computed as the ratio of the measured voltage to the applied current. Since both the real
and quadrature components are measured, a complex
computation can be made. The electrodes consist of four
0.3 m m diameter by 1 cm long platinum needle electrodes
with 30 cm long leadwires mounted colinearly in a plastic substrate at 0.5 cm intervals and are submerged to a
depth of 0.5 cm. Connection to the amplifier is made
through 4-foot lengths of RG-59/U coaxial cable. Since
a saline solution is purely resistive over a wide frequency
range, measured phase should be zero. To apply impedance spectroscopy, maximum phase errors on the order
o f 0.2 ~ are required; thus, the measured phase errors are
clearly unacceptable. It should be noted that results are
highly dependent upon lead placement and that virtually
any set o f data including positive phase errors can be
obtained.
While errors may be analyzed by computing commonmode rejection ratios (10), error sources can be conveniently illustrated heuristically as in Fig. 2. In an ideal
four-probe system, the impedance of the voltage sensing
system is infinite; whereas, in a real situation, amplifier
input impedance, stray capacitance, and input cable capacitance combine to present a finite impedance at the
sensing electrodes. Due to imbalances in both capacitances
and sensing electrode impedances, the leakage currents
il and i 2 of Fig. 2 will not in general be equal, resulting
in a differential voltage at the amplifier input terminals.
In studies o f biologic systems which have high conductivity, the resulting phase errors can be particularly trou-

137

blesome. Also in an ideal system, the output impedance

of the current source is infinite but in a real situation is
finite. Conceptually, the current source output divides between the sample and the equivalent output impedance;
with increased impedance in the loop containing the sample, a larger portion of current is diverted through the output impedance. Depending on the output capacitance,
sample impedance, and electrode impedances, and how
these change with frequency, significant phase errors can
result.
Since surprisingly small stray capacitances coupled with
electrode impedances can cause large phase errors, the capacitances and sensing electrode impedances Z3 and Z4
depicted in Fig. 2 were manipulated both experimentally
and by means of an equivalent circuit model to provide
a more quantitative assessment. Using the same experimental configuration as that for Fig. 1, electrode leadwires and input cables were arranged to obtain a minimum
phase error. To simulate increased stray capacitance, an
additional 4-foot length of RG-59/U cable was inserted
into either the positive or negative amplifier input terminal, and to simulate electrode imbalances, a 100 l) resistor
was inserted in series with either the positive or negative
voltage-sensing electrode leadwire. Results are illustrated
in Fig. 3. Depending on experimental conditions, both
positive and negative phase errors may be encountered
and unacceptable errors can occur at frequencies as low as
5 kHz.
To quantify the magnitude of errors that can be expected for given parameters, an equivalent circuit model
of the Fig. 1 test configuration was constructed using the
linear circuit analysis program LINCAD (2) implemented
on a VAX computer. Sample impedance was modeled
with a 12 fl resistor. Amplifier input impedance was modeled with a 100 Ml) resister shunted by a 25 pF capacitor
according to manufacturer's specifications. Combined input cable and stray capacitances were represented with
150 pF capacitors; these were alternately increased to

el

e2
A
~r

_T_m
_

FIGURE 2. Error sources in basic system.

A (el-e2)/8

138

J.J. ACKMANN
I0

10

R IN - LEAD J

i o,

S-

i~
v

-S-

-S-

-10-

-10-

,<

<

It IN +L[AO

-t5-

(A) EQUALLEAD LENGTHS

........
i

-20

(A)

lOO00

tODD

-is-20 leee

fOs

FREQUENCY(HZJ

EQUAL INPUT CAPACITANCES

FNE~[NCT (NZ)

tO*

20.
20

15.

15

o
w
Q

lO-

I0.
a
v

5,
it

It IN +

/
LEAD

-S.

N IN 4.

(D) LONGERN(GATIV[ LEAD

.... i

-IO

........

IOOOO

FREQUENCY(HZ)

lOS

t
LEAD

(|) LARGE| CAPAClT~C( AT - TEmI~L

........
i
9
9 9
I0000
IOOO
rttQU(aCV (HZ)

I0 =

10-

s
i

/!

IN i LEAD

4-I0-

o10-I|-

-zo

-1|-

icl t0Mta eosmv( LUO

IOO0

tOOOO

FREQUENCY(HZ)

IOe

-20
Ioeo

10001

FREQUENCY(NZI

tO=

FIGURE 3.

E x p e r i m e n t a l l y d e t e r m i n e d p h a s e errors as f u n c t i o n of
imbalanced cable capacitances and electrode impedances.

270 pF to simulate an increased input cable length. Electrode impedances were modeled by a parallel combination of a 1.25/zf and 25 kfl resistor in series with a
100 fl resistor; these values approximate experimental data
previously obtained for the electrodes used. The 100 fl
value was increased to 200 fl to simulate an electrode imbalance; this value is also consistent with previous data
obtained during a study of electrode variability. The
model was exercised according to the conditions previously described for Fig. 3 and resulting phase errors are
shown in Fig. 4.
There is general agreement between experimental and
model results, and no attempt was made to adjust model

FIGURE 4.

R e s u l t s o f circuit m o d e l f o r d a t a o f

Fig. 3.

parameters for closer correspondence. With balanced electrodes and leakage capacitances (Fig. 4a, center trace),
phase error is - 1.1 o at 100 kHz and results primarily from
the combined effects of sensing electrode impedance and
amplifier input capacitance. Although not illustrated,
error increases to - 5 . 7 ~ at 500 kHz and to - 1 1 . 1 ~ at
1 MHz. Thus, even under balanced conditions, an amplifier input capacitance as low as 25 pF can result in intolerable phase shifts. Imbalances in electrode impedances
cause either additional positive or negative phase errors,
depending upon which lead contains the higher impedance electrode. With imbalanced electrode impedances
(Fig. 4b and 4c), errors are exacerbated; however, an in-

Complex Bioelectric Impedance System

creased electrode impedance in the lead opposite to that
containing the increased capacitance reduces the error.
In summary, direct application of a commercially available instrument for measuring complex bioelectric impedance will in general produce unsatisfactory results. Phase
errors result from the loading effects of amplifier input
capacitance and imbalances in capacitive leakage currents
to ground. In addition, as subsequently discussed more
quantitatively, load-dependent phase shifts can occur in
a current source due to output capacitance.
The above problems may be minimized by the inclusion
of additional circuity: loading effects due to amplifier input capacitance can be greatly reduced by incorporating
low input capacitance buffer amplifiers, loading effects
due to voltage-sensing cable capacitance can be significantly lowered by the use of driven shields, leakage currents to ground can be largely eliminated by isolating the
current source, and current-source output capacitance can
be decreased by careful design.
CURRENT SOURCE

In a four-terminal system employing a lock-in amplifier, a common reference must be applied both to the current source and to the amplifier; however, to eliminate
a return path for capacitive leakage currents (see Fig. 2),
the current source must be isolated from ground. This may
be accomplished in several ways.
In a previous design (6), the drive to the current source
was isolated optically. While the circuit operated satisfactorily up to 100 kHz, phase shifts within the isolation circuitry were unacceptable at higher frequencies.
A scheme employing two current sources in series was
therefore used in subsequent designs. With reference to
Fig. 5, if the currents from two sources in series, one operating at 0 ~ and the second at 180 ~ are precisely balanced, the currents in the return lead cancel, effectively

139

isolating the overall source from ground. The feasibility

o f obtaining the necessary tracking to effect isolation was
demonstrated in a previous project in our laboratories (9).
An individual current source may be assembled using
a single operational amplifier in several different configurations. As in any operational-amplifier circuit, performance parameters are frequency dependent. Modest
frequency-dependent phase shifts can be accounted for
by calibration; however, corrections cannot be made for
phase shifts secondary to changes in load impedance
which occur because of finite source output capacitance.
In an earlier project (6), several conventional single
operation-amplifier circuits (8) all proved unsatisfactory
and a discrete-component circuit was required. While improved performance can be obtained in an operational
amplifier circuit with additional components (5), a convenient alternative is a single-chip operational transconductance amplifier. This component has the generic
characteristics of an operational voltage amplifier except
that the forward transfer characteristic is best described
as a transconductance rather than voltage gain, and the
output is current, not voltage. The magnitude of the output current is equal to the product of the transconductance and input voltage. A bias current is used to program
the transconductance. While intended as a gain-controlled
building bIock for various applications, a balanced dual
current source can be simply constructed.
With reference to Fig. 6, a single chip containing a dual
amplifier (CA3280) is used. Bias currents of approximately 1 mA are applied at pins 3 and 6 which establish
transconductances of approximately 25 mS. The balance
potentiometer permits matching of the two transconductances and thus output currents. Following attenuation,
the input voltage is applied to the positive input of one
amplifier and the negative input of the other, thus establishing output currents which are at 0 ~ and 180 ~ with
respect to the input signal. For the component values

REFERENCE

~
Y

I1/B
I

wI2=e

C---

I
t

v/ge

PREAPPt.]T
IER

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

FIGURE 5. Practical four-probe system.

LOCK-~
j

~t.IFIER

140

J.J. ACKMANN

- 111'r

3!11cll

t .,F

.tp
4"
<
8.~11
u

o z-...

t2
S,lkR

~ loft

lily

I.q -

>

13
OZ..

14
r
_- .I I !
UI: gA~2S0

+lSV
FIGURE 6.

-$5V

Current-source s c h e m a t i c diagram.

indicated, an output current of 30/~A RMS is produced

with an input voltage of approximately 1.0 volt RMS. Separate DC offset adjustments are provided for each channel. The circuit was packaged on a printed circuit board
contained in a 6.5 cm x 7.5 cm 4.2 cm die-cast aluminum box. Miniature pin-jacks are provided for electrode
connections.
Phase transfer characteristics as a function of load were
determined experimentally and further evaluated by an
equivalent circuit model using LINCAD (2). The output
source was modeled by a voltage-dependent transconductance given by:

gin(V)

U
=

0.025

1+

Jf

(1)

8.75(106 )

The output source was shunted by a 3.3 Mfl resistor and

3.3 pf in accordance with measured values. Since the nominal output capacitance of the CA3280 per manufacturer's
specifications is 7.5 pF nominal, and two sources are in

series, the capacitance value is reasonable. The input section was modeled by a 1012, 8.2k12 voltage divider per
Fig. 6 with the amplifier input shunted by a 7 pf capacitor in accordance with specifications. The transconductance and corner frequency of the model were adjusted
to provide a best fit to the actual data, but they also agree
closely with nominal specified values. Both experimental
and model results are illustrated in Fig. 7. While the frequency dependence of phase of the model agrees with actual data to only 800 kHz due to the simplified single
pole representation, the phase change with load is well
represented and clearly demonstrates the *necessity o f considering the effects of current source output capacitance.
Output saturation occurs approximately at the power
supply voltages; thus, at current of 30/zA RMS (85/~A
p-p), a total load of 350 kfl can be driven.
BUFFER AMPLIFIERS

As previously discussed, loading effects o f cable and

amplifier input capacitance at the voltage-sensing termi-

Complex Bioelectric Impedance System

141

El/

-6
a.

-8

-10
10 4.

10 !

10 e

irrequenoy (Hz)
FIGURE 7. Current-source phase as function of frequency and load.

nals in a four-probe system can cause unacceptable phase

errors. As indicated in Fig. 5, the use of unity-gain buffer
amplifiers and shield drivers can minimize the problem.
The buffer amplifiers provide a high impedance, Low capacitance input. By maintaining the cable shield at the
same voltage as the input, current flow between the shield
and center conductor, except for distributed effects,, is significantly reduced, thereby reducing the effective cable
capacitance. The concept is usually applied by driving
shields at the common-mode voltage of a differential amplifier (5). In the design used here, the usual unity gain
operational-amplifier configuration is generally unsatisfactory because of instability, and at least formerly, discrete component circuits employing field-effect transistors
were required (6). In recent years, however, integratedcircuit buffer-amplifiers with satisfactory characteristics
have become available.
The circuit illustrated in Fig. 8 employs LH0033 devices. R-C networks in the power supply connections
provide current limiting and are configured according
to manufacturer's recommendations. 220 ft resistors in
series with the input and with the connection to the shield
are included to prevent oscillation and were determined
empirically. The PAR 5204 lock-in amplifier used in the
frequency range from 5 Hz-100 KHz has provisions for
differential input and the buffer outputs could be directly
connected; however, the PAR 5202 amplifier used at
higher frequencies has a 50 t] single-ended input. To
provide an input for this instrument, a conventional
differential-to-single ended convertor using an LH0033
high-speed operational amplifier is included. A gain of

5.0 was chosen as a compromise between stability and

bandwidth. Precision resistors are used to maintain common-mode rejection without additional adjustment. The
circuit was packaged on a printed circuit board contained
in a 6.5 cm x 7.5 cm x 4.2 cm die-cast aluminum box.
Input leads consist of 50 cm lengths of RG-174 coaxial
cable terminated with miniature alligator clips. Amplifier
characteristics were measured using the PAR 5204 and
PAR 5202 lock-in amplifiers. Gain is essentially constant
at 4.94 (buffer amp gain = 0.99) from 5 Hz-2Mhz. Phase
characteristics are illustrated in Fig. 9. Input capacitance
is estimated at 6 pf and was determined by changing
source impedance, measuring the additional phase shift,
and computing the value using an equivalent circuit model.
Common-mode rejection ratios were estimated with a
common-mode voltage of 10.0 volts peak-to-peak using
an oscilloscope to measure output; computed values were
-68 db, - 7 4 db, - 8 0 db, -74 db, -61 db, and -54 db
at frequencies of 100, 1 kHz, 10 kHz, 100 kHz, 500 kHz,
and 1 Mhz, respectively. Noise referred to the input is estimated as 75 # VRMS using an oscilloscope measurement.
OVERALL SYSTEM CHARACTERISTICS
AND LIMITATIONS
The individual sources of error inherent in the lock-in
amplifiers, current source, and preamplifier combine to
form an overall system error which is a function of frequency, sample impedance, and electrode impedance. Absolute magnitude and phase errors which depend only on
frequency and not on load (sample impedance combined

142

J.J. ACKMANN
1OO11
~tSvO

,1

Ly ~

,tpF
0 +%5V

+ , . ~ < : I Z Z Z Z Z ~
-""~

T --vvv
_J__
1~fl

"T

I
"--I--. 4 ~F

I k n ,n

9 I#
O

V,~t

0 -$15V

.,.vo

=,+r

_____%

Z T
2;~Ofl

,,T2

mm m
It

~ A

ULU~

- 16V

U~

FIGURE

8.

. i ~F

Buffer-amplifier

with electrode impedance) can be accounted for and removed by calibrating the system with a known simulated
sample. In practice, this accomplished with a low-capacitance precision resistor. The measuring current is applied
to the resistor and the complex voltage (real and reactive
components) measured at each of the desired measurement frequencies. Dividing the voltages by the calibration resistance results in an effective applied complex
current. Experimentally determined voltages may then be
divided by the effective current to yield the complex impedance. This procedure is effective to the extent that the
calibration current is repeatable and that phase corrections are small with respect to experimentally determined
values. Additional errors which result from load changes
as well as system noise then determine the overall absolute accuracy.
Errors within the lock-in amplifier include both absolute phase error as well as phase noise. Absolute phase
errors for the two instruments are illustrated in Fig. 10.
The data were obtained using a 1.0 volt RMS sine-wave
reference signal and a 15.0 m Vinput signal derived from
a voltage divider connected to the same source; equivalent source impedance was approximately 50 ft. The model
5204 phase was set to zero at 1 kHz and the model 5202
phase was set to zero at 200 kHz. As indicated, phase er-

schematic

I~OOS~?J

L.qOO'J2CO

0.01~

diagram.

ror for the Model 5204 is less than 1 o and error for the
Model 5202 is less than 3 ~ over the useable frequency
range of the system. After a 30-minute warmup period,
readings are repeatable within the phase noise of the instruments. For the Model 5202, the manufacturer's specified noise is 0.035 ~ peak-to-peak maximum over the
entire frequency range of the instrument. For the Model
5204, specified noise is 0.005 ~ peak-to-peak maximum at
frequencies greater than 50 Hz. While not specified, noise
at lower frequencies is substantially greater; measured
values in the range from 5 Hz-100 Hz are illustrated in
Fig. 11. The data were obtained by sampling the lock-in
amplifier outputs at a rate of 2 samples/s with an analog-to-digital convertor. It should be noted that the noise
frequency is quite low; with reference to Fig. 10, the data
for frequencies below 50 Hz were obtained by recording
the lock-in amplifier outputs on a storage oscilloscope
over a period of approximately two minutes and visually
determining average values.
In addition to phase noise, settling time of the lock-in
amplifiers must also be considered. For the Model 5204,
the time required for reference lock with abrupt changes
in reference frequency is less than two seconds at frequencies greater than 50 Hz; at lower frequencies, the required
time depends on the initial frequency and the difference

C o m p l e x Bioelectric I m p e d a n c e S y s t e m

143

2.5

0.0

. . . . . . .

.......

. . . . . . . . . . . . . . . . . . . . . . . . .

-2.5
r
L,.

\
9

-o
v
q)
(,,9
o
e,"

-5.0

-7.5
V o l t a g e Gain = 4 . 9 4

-10.0

.....................................................................................

-12.5

'

'

'

'

'

'

'

10 4

'

'

'

'

'

'

'

10 5

'

10 6

Frequency ( H z )
FIGURE 9. Buffer-amplifier phase vs. frequency.

between the initial and final frequencies. For example,

the time required for reference lock for a change from
30 Hz-40 Hz is approximately 45 seconds and the time
for a frequency change from 40 Hz-50 Hz is approximately 25 seconds. For the Model 5202 lock-in amplifier,

the settling time is less than 1 second for any changes over
the entire operating range of the instrument.
An overall system calibration curve recorded with a
197.6 ohm resistive load is illustrated in Fig. 12. Current
magnitude is 15 # A R M S and is essentially constant with

1.0

0.8

0.6
Q

2E

0.4

0.2

0.0
a_
o~ -0.2
O4
U3

O.

O~
--

,--

-0.4
9 PAR 5204

9 PAR 5202
-4

-0.8
'

'

101

'

'

10z

i'
I

'

'

103

'

'

104

105

! FREQUENCY(Hz)

FIGURE 10. Lock-in atnplifier phase characteristics vs. frequency,

'

~5

'

106

o4

o
o4
--2 tn
a,,

-5

-0.6

-1.0

--1

144

J.J. ACKMANN
0,15

20 Hz Reference
niter time constont=5 sec, -8db/oct

0.10
P

0,05
0,00

hi
(n
"1(
3:

-0,05
-0.10
-0,15

40

80

120

160

200

240

TIME (sec)

0.3

Ob
*0
-14
v
IAI
0
Z

0.1
0,05
0.01

Ul
,<
3:
a.

0,003

0.001

10

30

1O0

FREQUENCY (Hz)
FIGURE 11. Phase noise at 20 Hz (top) and phase noise vs. frequency (bottom). Dashed line indicates specified maximum noise
at frequencies above 50 Hz.

frequency; therefore, only phase is shown. Input to the

to the preamplifier is approximately 3.0 m V and input
to the lock-in amplifier is approximately 15 mV. Calibration phase is simply the sum o f phase errors of the current source, preamplifier, and lock-in amplifiers. From
approximately 10 kHz-100 kHz, the slightly negative
phase shifts of the preamplifier and current source are
balanced by the slightly positive phase of the lock-in amplifier (refer to Fig. 7, 9, and 10) and corrections are less
than 1 o. At higher frequencies corrections are larger due
to increased negative phase of the current source and preamplifier. For a load range of 20 fl- 500 f~, the data of
Fig. 12 are reproducible within the phase noise limits
of the lock-in amplifier f r o m 5-50 Hz, within +_ 0.075 ~
between 50 Hz-100 kHz, and within _+ 0.15 ~ from 100
kHz-1 Mhz. The reproducibility at frequencies lower than
50 Hz m a y be improved to approximately _+ 0.15 ~ with
time-averaging.
As previously noted, electrode impedances m a y result
in further errors. To quantify these effects, resistors were
inserted into all four system leads and the values required
to cause an additional phase error o f ! ~ at frequencies
of 100 kHz, 200 kHz, 500 kHz, and 1Mz were determined;
values were 1.5 ki], 750, fl 250 fl, and 100 f~, respectively.
Electrode impedances decrease as a power function of frequency and impedances of electrodes typically employed
in our laboratories (0.3 m m dia by 10 m m long platinum
or stainless steel needles) are at least one order of magnitude less than the above values.
Measurements obtained on a two time-constant complex load, which simulates portions of data f r o m an

1,0

0,5

0.0
-0.5

- 20 1

-1.0

-4

-1,5

-6

-2.0

-8

-2,5

- 1 0 oc)

-3.o

-12 ~

-3.5

-14

,~
<.)

-4,0

-16
101

102

105

104

FREQUENCY (Hz)
FIGURE 12. Calibration phase vs. frequency.

10s

10 s

Complex Bioelectric Impedance System

1.439

.0147

49.7

97.9

145

49.8

-100

Ideoeured

9
A

-'/5

--Computed

--1

O0 kHz

-50

-2S

0
50

100

150

200

250

REAL (Ohm.)

-5
-10

-20-'St

-2s
-35

Meoeurecl

--Com,"~ . . . . . . . . . . . . . .
9 9

I0 t

9 ,

101

10 i

~ 7 - -

104

--

1 0 II

,
! 0| e

mmu(NcY(xx)
FIGURE 13. Impedance and phase vs. frequency plots for two time-constant circuits.

isolated lung preparation previously recorded in our laboratories (1), are illustrated in Fig. 13. Data were corrected
using a calibration curve generated with a 197.6 fl resistive load. To generate the computed curve, component
values were measured using a Fluke Model 87 multi-meter.

-1500

Omtne

-1250

Note that changes in the reactive component are well resolved in the frequency range up to 1 kHz where phase
angles are less than 5 ~.
Finally, use of the system on a biologic specimen is
illustrated in Fig. 14. The data are part of an investiga-

erTtau-oeTt.es- peeked eelll

100 id~

E
-1000
E
tv

-750

_>
bd
I-O

-500
-250

5kl~
0

FIGURE 14.

500

1000

Complex resistivity

1500
2000
REAL(ohm-em)

2500

plot of packed canine erythrocytes at

38.5~

3000

146

J.J. ACKMANN

tion of the complex resistivity of blood. Briefly, complex

measurements were obtained from canine erythrocytes
in plasma at various hematocrit using a 9 m m diameter
cylindrical conductivity cell. Cylindrical current electrodes
were spaced at 4 cm, and 0.5 m m dia voltage-sensing electrodes were embedded in the chamber wall at a separation o f 1 cm. The data illustrated were obtained from
packed cells. An analog circuit model of blood as a function of hematocrit can be constructed f r o m the data and
will be reported elsewhere.
DISCUSSION
Complex bioelectric impedance measurements may be
accomplished in a number of ways with each method having advantages for a particular application. The fourprobe system employing lock-in amplifiers described in
this communication largely eliminates the problems associated with electrode impedances and stray capacitances
at frequencies up to approximately 1 MHz. Data with sufficient accuracy to permit application of various materials-science analysis techniques can be obtained in systems
with relatively high conductivity and resultant small phase
angles. At frequencies above 50 Hz, data can be acquired
at a rate of approximately one measurement every three
seconds. At lower frequencies, acquisition time is substantially longer with settling times between frequencies of up
to one minute. In addition, phase noise is considerably
larger; if time-averaging is required to reduce the effects
o f phase noise, acquisition time could exceed two minutes per data point. The compliance of the current source
is sufficient to tolerate a combined current-electrode impedance of approximately 350 k~ at a measuring current
of 30 #A RMS, thus permitting measurements at relatively
low frequencies, depending on electrode type. With the
Model 5204 lock-in amplifier, measurements down to
0.5 Hz are possible; however, in our experience, settling
time and phase noise render measurements below approximately 5 Hz impractical. Several amplifiers having a low
frequency limit o f 5 Hz and a settling time o f approximately one-tenth that of the Model 5204 are available;
however, the phase noise of these instruments is substantially greater up to frequencies of 1 k H z and has not
proven practical in our experience.
It is emphasized that the impedance computation assumes that both the excitation current and measured volt-

ages are sinusoidal. Electrode impedances are a function

o f current density and this effect is frequency dependent
with impedance decreasing as a power function o f frequency (3). At low frequencies, electrode impedances can
be extremely high and can potentially cause operation of
the current source outside the linear range with resultant
waveform distortion. While the lock-in amplifier is narrowband (but does respond to odd harmonics of the reference)
and would partially mitigate the error, the measured impedance would nonetheless be incorrect. It is therefore important to visually monitor the measured signal with an
oscilloscope; in the PAR Model 5204, a signal monitor jack
is provided for this purpose. With the electrodes and current densities (<0.3 m A / c m 2) employed in our studies, no
problems have been encountered with frequencies as low
as 5 Hz.
REFERENCES

1. Ackmann, J.J.; Seitz, M.A. Methods of complex impedance measurements in biologic tissue. CRC Crit. Rev. Biomed. Eng. 11:281-311;1984.
2. Cornetet, W.H.; Batocletti, EE. Electronic circuits by system and computer analysis. New York: McGraw-Hill; 1975.
3. Geddes, L.A. Electrodes and the measurement of bioelectric events. New York: John Wiley and Sons; 1972.
4. Geddes, L.A.; Baker, L.E. Principles of applied biomedical instrumentation. New York: Wiley-lnterscience; 1975.
5. Graeme, J.G. Applications of operational amplifiers: Thirdgeneration techniques. New York: McGraw-Hill; 1973.
6. King, M.A. A four-terminal instrumentation system for the
measurement of complex bioelectric impedance. Milwaukee: Marquette University; 1978. MS Thesis.
7. Macdonald, J.R. Impedance spectroscopy. New York: John
Wiley and Sons; 1987.
8. Millman, J.; Halkias, C.C. Integrated electronics. New
York: McGraw,Hill; 1972.
9. Smith, G.A. Analysis of the current pathways associated
with multifrequency impedance measurements of the lung.
Milwaukee: Marquette University; 1982. MS Thesis.
10. Winter, B.B.; Webster, J.G. Reduction of interference due
to common mode voltage in biopotential amplifiers. IEEE
Trans. BME 30:58-62;1983.
NOMENCLATURE
= transconductance
V = voltage
f = frequency
gm

=4~i