Multilead measurement system for the time-domain analysis of bioimpedance magnitude.

Abstract

Bioimpedance measurement applications range from the characterization of organic matter to the monitoring of biological signals and physiological parameters. Occasionally, multiple bioimpedances measured in different locations are combined in order to solve complex problems or produce enhanced physiological measures. The present multilead bioimpedance measurement methods are mainly focused on electrical impedance tomography. Systems designed to suit other multilead applications are lacking. In this study, a novel multilead bioimpedance measurement system was designed. This was particularly aimed at the time-domain analysis of bioimpedance magnitude. Frequency division multiplexing was used to avoid overlapping between excitation signals; undersampling, to reduce the hardware requirements; and power isolated active current sources, to reduce the electrical interactions between leads. These theoretical concepts were implemented on a prototype device. The prototype was tested on equivalent circuits and a saline tank in order to assess excitation signal interferences and electrical interactions between leads. The results showed that the proposed techniques are functional and the system's validity was demonstrated on a real application, multilead impedance pneumography. Potential applications and further improvements were discussed. It was concluded that the novel approach potentially enables accurate and relatively low-power multilead bioimpedance measurements systems.

Full text

Page 1

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 59, NO. 8, AUGUST 20122273
Multilead Measurement System for the Time-Domain
Analysis of Bioimpedance Magnitude
Javier Gracia∗, Ville-Pekka Sepp¨ a, Jari Viik, and Jari Hyttinen
Abstract—Bioimpedance measurement applications range from
the characterization of organic matter to the monitoring of biolog-
ical signals and physiological parameters. Occasionally, multiple
bioimpedances measured in different locations are combined in
order to solve complex problems or produce enhanced physiolog-
ical measures. The present multilead bioimpedance measurement
methods are mainly focused on electrical impedance tomography.
Systems designed to suit other multilead applications are lacking.
In this study, a novel multilead bioimpedance measurement sys-
tem was designed. This was particularly aimed at the time-domain
analysis of bioimpedance magnitude. Frequency division multi-
plexing was used to avoid overlapping between excitation signals;
undersampling, to reduce the hardware requirements; and power
isolated active current sources, to reduce the electrical interac-
tions between leads. These theoretical concepts were implemented
on a prototype device. The prototype was tested on equivalent
circuits and a saline tank in order to assess excitation signal in-
terferences and electrical interactions between leads. The results
showed that the proposed techniques are functional and the sys-
tem’s validity was demonstrated on a real application, multilead
impedance pneumography. Potential applications and further im-
provements were discussed. It was concluded that the novel ap-
proach potentially enables accurate and relatively low-power mul-
tilead bioimpedance measurements systems.
Index Terms—Bioimpedance, frequency division multiplexing
(FDM), impedance measurement, interference channels.
I. INTRODUCTION
B
distribution of matter. It is traditionally determined by injecting
an alternating current excitation signal through the tissue and
recording the voltage response (see Fig. 1). Bioimpedance mea-
surement has a wide range of applications, which differ in the
excitation waveform and the analysis of the voltage response.
The most common analysis techniques are:
1) Frequency domain analysis ˙Z(w): The frequency re-
sponse of a single wide-spectrum excitation signal or sev-
eral individual frequency components provides detailed
IOIMPEDANCE represents a set of electrical properties
of living organisms directly related to the composition and
Manuscript received October 19, 2011; revised March 31, 2012; accepted
May 19, 2012. Date of publication June 5, 2012; date of current version July
18, 2012. Asterisk indicates corresponding author.
∗J. Gracia is with the Department of Biomedical Engineering, Tam-
pere University of Technology, Tampere FIN-33720, Finland (e-mail:
javier.gracia.tabuenca@gmail.com).
V.-P. Sepp¨ a, J. Viik and J. Hyttinen are with the Department of Biomedical
Engineering, Tampere University of Technology, Tampere FIN-33720, Finland
(e-mail: ville-pekka.seppa@tut.fi; jari.viik@tut.fi; jari.hyttinen@tut.fi).
Digital Object Identifier 10.1109/TBME.2012.2202318
information about the composition of organic matter. This
technique is utilized by different applications to estimate
the composition of the human body, tissues, and single
cells [1].
2) Time domain analysis ˙Z(t): Variations in composition
are often related to biological and physiological changes.
Tracking the variations in bioimpedance over time en-
ables the extraction of biological signals and dynamic
parameters. For example, impedance cardiography tracks
impedance changes caused by blood to produce haemo-
dynamicparameters[2].Typically,time-domainmeasure-
ments are based on the variations of a single-frequency
component excitation signal. Moreover, in many situa-
tions, variation in the magnitude |˙Z | (t) is sufficient.
This is the case in impedance plethysmography applica-
tions such as impedance pneumography (IP). IP estimates
changes in lung volume and pulmonary flow rate by mon-
itoring variations in the transthoracic bioimpedance [3].
In certain cases, the information provided by a single
bioimpedance measurement may not be enough to resolve com-
plex problems or reliably produce physiological parameters.
Often, this lack of information may be solved by combining
bioimpedances simultaneously recorded in multiple locations.
For example, the combining of multiple signals may allow the
removal of movement artifacts [4], as well as the separation
of overlapping frequency signals [5]. However, unwanted in-
teractions between single-lead bioimpedance measurement sys-
tems working on the same body may degrade the results. It is
necessary to use specific methods for multilead bioimpedance
measurement.
Most of the actual multilead methods for measuring
bioimpedances are aimed at electroimpedance tomogra-
phy (EIT). EIT is a noninvasive technique that combines
bioimpedances recorded between several electrodes to dynami-
callyimagebodyregions[6].EITsolutionsmaybeunsuitableor
excessivelycomplexforotherpurposessuchasthetime-domain
analysis of bioimpedance magnitude.
This article briefly reviews the existing approaches for mul-
tilead bioimpedance measurement and presents the design
and testing of a novel system specifically oriented to the
time-domain analysis of bioimpedance magnitude in multiple
locations.
II. MULTILEAD BIOIMPEDANCE MEASUREMENT BACKGROUND
If multiple single-lead bioimpedance measurement systems
are connected to the same organism, their excitation signals
may overlap, and the behavior of their electrical circuits may be
altered. These effects can cause corrupted results.
0018-9294/$31.00 © 2012 IEEE

Page 2

2274IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 59, NO. 8, AUGUST 2012
Fig. 1.
rent I(t) is injected and the resulting voltage response V (t) is recorded. The
value of the complex impedance is obtained through˙Z =˙V /˙I (1). I+, I−,
V + and V − represent the four terminals of the tetrapolar lead.
Ideal tetrapolar bioimpedance measurement circuit. An excitation cur-
A. Excitation Signal Overlapping
If two or more excitation signals are simultaneously injected
into the same organism, the voltage response corresponds with
the sum of the voltage responses that each single excitation
signal would produce on its own. Special strategies should be
followed to avoid the overlapping of the information conveyed
in each individual voltage response signal.
1) Time division multiplexing (TDM): The classic solution
utilizes fast switching between multiple bioimpedance
leads, resulting in a pseudosimultaneous measurement.
Since leads are not connected at the same time, the excita-
tion signals do not overlap. However, an increment of the
switching rate may cause transient processes in the tissues
and the instrument, which will distort the results [7].
2) Frequency division multiplexing (FDM): Some studies
suggest injecting the excitation signals concurrently, and
preventing the overlapping using FDM methods [8]. If
nonoverlapping frequency signals are injected, their re-
sponses can be separated again through frequency separa-
tion techniques.
3) Synchronous sampling: A recent study claims that the
frequency separation methods utilized in FDM involve
high computational resources [9]. The researchers pro-
posesbasingthebioimpedanceanalysisexclusivelyonex-
citation sine waves. Sampling frequencies and sine waves
can be synchronized in order to record when antinodes of
a particular sine match with the nodes of the other signals.
This method simplifies the signal processing calculations.
B. Electrical Interactions
As Fig. 2 shows, in a real single-lead bioimpedance mea-
surement system, there are two nonideal impedances˙Zo,˙Ziin
parallel to the bioimpedance being measured˙Z [6]. The value
estimated by the measurement system is equivalent to the par-
allel circuit of the three elements˙ZT, see Fig. 2. Typically,
the two nonideal impedances are significantly higher than the
organism’s bioimpedance. Hence, the error they cause may be
considered negligible. However, if more leads are connected to
the same organism, the number of parallel nonideal impedances
increases, and the error may become considerable.
Fig. 2.
the output impedance of the CI,˙Zi the input impedance of the VM and˙Z
the bioimpedance. The real impedance measured by the whole system˙ZT is
calculated by ˙ ZT = ˙Zo?˙Z ?˙Zi(2).
Nonideal tetrapolar bioimpedance measurement circuit.˙Zorepresents
Fig. 3.
represent the boundary of an organism, and the different types of dashed lines
the paths for different currents inside the organism. (a) Simplified schema of
a typical simultaneous current injection EIT system. Current sources with a
common ground facilitate the spreading of currents in all areas of the body.
(b) Simplified schema of an independent bioimpedance measurement system.
Independent current sources allow the distribution of currents in the areas of
interest, regardless of the location of the other leads.
Strategies for simultaneous current feeding. The thick gray circles
Electrical interactions are not an issue for the TDM method,
asitisnotatrulysimultaneousapproach.However,fortheFDM
and synchronous sampling methods, the only solution is to use
circuits with a higher hardware performance.
The design of high output impedance current sources is a
common problem in bioimpedance measurement. In the case
of simultaneous current feeding, this may become additionally
complex. There are two strategies for feeding multiple currents
into the body that affect the design of the current sources.
1) Current sources with shared ground: Currents are simul-
taneously injected in different locations and collected
at a common point [see Fig. 3(a)]. This strategy sim-
plifies the system design when measuring contiguous
bioimpedances, such as in the case of EIT. Since all cur-
rentsourcessharethesameground,currentsourcecircuits
from single-lead systems can be directly used. Many ex-
isting active current source circuits can be employed for
this strategy.
2) Independent current sources: Currents are simultane-
ously fed between two independent points for each lead
[see Fig. 3(b)]. This solution facilitates the measurement

Page 3

GRACIA∗et al.: MULTILEAD MEASUREMENT SYSTEM FOR THE TIME-DOMAIN ANALYSIS OF BIOIMPEDANCE MAGNITUDE2275
Fig. 4.
frequencies. In subfigure (a), the undersampling is performed without an antialiasing filter, resulting in all three signals being preserved, and potentially overlapped.
In subfigure (b), an antialiasing filter is applied to select the frequency band of interest before undersampling.
Graphical representation of undersampling effects. The frequency spectrum is illustrated as a fan-fold paper containing three distinct signals at different
of separated bioimpedances. However, it can only be
achieved through independent current sources. At the
present, there is a lack of current source circuits aimed at
independent bioimpedance measurement. To our knowl-
edge, only differential resistive current sources have been
utilized for this purpose [9].
III. SYSTEM IMPLEMENTATION
The target of this research was to establish a system for the
time-domain analysis of multiple independent bioimpedance
magnitudes |˙Z | (t) with maximal measurement accuracy and
minimal energy consumption. The system’s design approach
and the implementation of a prototype device are presented in
this section.
A. Theoretical Approach
1) Electrical Interactions: The improved Howland circuit
is one of the most popular active current sources utilized in
bioimpedance measurement [10]–[12]. In order to use it as in-
dependent current source, it was fed with an independent power
supply. This solution represents the first attempt to apply active
circuits for the measurement of independent bioimpedances.
2) Excitation Signal Overlapping: Although synchronous
sampling seems to be accurate and computationally efficient for
frequency-domain analysis, use of such approach may result a
complex solution for tracking only the variations in magnitude.
Instead, a simple FDM approach was utilised. This consisted
of assigning a different frequency excitation sine signal to each
lead,andseparatingtheresponsesinthefrequencydomainusing
fast Fourier transform (FFT).
3) Undersampling: A satisfactory separation of close fre-
quencies can only be achieved through digital methods. Hence,
the response signals had to be digitized. The recommended
excitation frequencies are over 100 kHz for many applications.
Therefore,inaccordancewiththeNyquisttheorem,theresponse
signals had to be sampled at frequencies over 200 kHz. How-
ever,thefrequencyofphysiologicalchangesconveyedonthere-
sponse signal are two to three decades lower. In order to achieve
a more efficient use of the hardware resources, an undersam-
pling strategy was applied. Undersampling enables particular
kinds of signals to be digitized at below the Nyquist sampling
rate without information loss [13]. Consequently, the sampling
rate could be reduced to approach the output frequency while
maintaining the accuracy.
Fig. 4(a) shows a graphical representation of the effects of
undersampling. The whole frequency spectrum can be seen as a
fan-fold printer paper. This infinite paper should have the folds
in the vertical direction. Hence, in this study, the inward creases
correspond to the multiples of the sampling frequency, and the
outward creases with the odd multiples of half the sampling fre-
quency. The superimposition of all the sheets after collapsing
the paper corresponds to the frequency spectrum after under-
sampling. Fig. 4(b) shows how a high-frequency band-pass sig-
nal can be digitized at a low sampling rate without loss of
information. It also highlights the importance of an antialiasing
filter. Filtering the target signal before undersampling prevents
overlapping with other signals.
B. Hardware
Basedonthepreviousapproach,athree-leadprototypedevice
was implemented, as shown in Fig. 5. The independent current
sources were composed of a digital signal processor (DSP) to
produce the excitation sine signal, and an improved Howland
circuit [10]–[12] was used as a voltage-to-current converter.
Both were isolated from the main power supply through a dc–
dc power isolator. Moreover, a digital isolator was included to
enable communication with the DSP. The voltage meter (VM)
consisted of an instrumentation amplifier and a second-order
antialiasingfilterconnectingtheamplifiertoananalog-to-digital
converter.
C. Signal Processing
Although the prototype device only implements hardware for
three leads, the signalprocessing was designed for an eight-lead
system. As the three hardware leads are frequency configurable,

Page 4

2276IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 59, NO. 8, AUGUST 2012
Fig. 5.
Ii+, Ii−, Vi+ and Vi−, where i ∈ {1,2,3} and indicates to which of the three leads the terminals belong.
Block diagram of each of the three independent leads implemented in the prototype device. The four terminals of each tetrapolar lead are represented as
Fig. 6.
(a) illustrates the division of a frequency spectrum sampled at 10 kHz and the
location of the eight excitation frequencies before undersampling. Subfigure
(b) is the result of collapsing the upper diagram. It corresponds to the result of
sampling the spectrum at 10 kHz and shows where the excitation frequencies
are located along the undersampled 5-kHz bandwidth.
Fan-fold paper visualization applied to the present case. Subfigure
they were configured to emulate an eight-lead system through
multiple rounds of testing.
Given the available resources, the undersampling frequency
was set at 10 kHz, providing a 5-kHz output bandwidth. As
shown in Fig. 6(b), this 5-kHz wide band was divided into eight
frequency channels, one for each of the eight virtual leads. The
frequencies of the excitation sine signals after undersampling
for each channel are presented in the third column of Table I.
It is recommended that excitation frequencies should be above
100 kHz. Therefore, the frequencies of the excitation sine sig-
nals were calculated by adding 100 kHz to the corresponding
undersampled frequency [second column in Table I; see also
Fig. 6(a)], [see Fig. 6(a)].
Theantialiasingfilterswereconfiguredat100and105kHzfor
their lower and upper cutoff frequencies, respectively. In order
to separate the response signals after undersampling, FFT was
computed. The amplitude of a particular excitation signal was
calculated from the magnitude of FFT at the signal’s frequency.
TABLE I
EXCITATION FREQUENCIES BEFORE AND AFTER THE UNDERSAMPLING FOR
EACH FREQUENCY CHANNEL
Fig. 7.
channelinterferencetestandtheelectricalinterferencetest.Equivalentterminals
of each lead are connected to the same point on the circuit.
Electronic circuit and connection of the terminals for the frequency
IV. PERFORMANCE TESTING
A. Frequency Channel Interference Tests
Method: The three leads were connected to the same con-
stant resistive value RL= 500Ω ± 1% (see Fig. 7), and were
configured to work on different groups of frequency channels
(see Table I). For all groups, the outputs of the three leads were
recorded for a period of 10 s. The mean and variance were
calculated for each of the three outputs.
Two groups of triads of frequency channels were selected for
two different purposes.
1) Theresultsof[f4,f5,f6],[f3,f5,f7],and[f2,f5,f8]were
compared in order to analyze the effects that distance
between frequencies may have on the results.
2) The results for [f1,f2,f3], [f2,f3,f4], [f3,f4,f5],
[f4,f5,f6], [f5,f6,f7], and [f6,f7,f8] were compared in
order to verify the correct functioning of the processing
block for the eight established channels.
ResultsandDiscussion:Thethreeleadsshowedconstantout-
put values over the 10-s period. The mean values for the three

Page 5

GRACIA∗et al.: MULTILEAD MEASUREMENT SYSTEM FOR THE TIME-DOMAIN ANALYSIS OF BIOIMPEDANCE MAGNITUDE 2277
TABLE II
LEAD CONNECTION COMBINATIONS USED IN THE ELECTRICAL
INTERFERENCE TEST
outputs agreed with the load value (500 Ω), with errors lower
that the resistance tolerance (±1%). The variance for the three
outputs ranged from 0.0083 to 0.0086. Discrepancies in the
mean and the variance between triads were lower than 0.01%.
These results prove that the undersampling and frequency sep-
aration methods work as expected. For the following tests, the
triad [f1,f2,f3] was used.
B. Electrical Interference Test
Method: This experiment aimed to quantify the error caused
by the addition of leads. The output error for one lead was
calculated while the current injectors (CI) and VM from the
other leads were connected to the same resistive load RL(see
Fig. 7) in different combinations (see Table II). Ten different
resistors’ between 100 and 1000 Ω were used for RL. In order
to remove the resistors tolerance error from the measure, the
measured values Rtest:n(RL) were normalized with respect to
the result of lead-1 connected on its own (Combination No.1,
Rtest:1(RL)) as
εtest:n(RL) =Rtest:n(RL) − Rtest:1(RL)
Rtest:1(RL)
· 100.
(3)
Eachresultwaspresentedasthemeanvalueof50repetitions.
ResultsandDiscussion:ThechartinFig.9illustratestheerror
for the set of ten resistors for every combination of connections
defined in Table I. Although no error exceeded ±0.8%, addition
of more leads may increase the error considerably.
The connection of elements involves the addition of an
impedance parallel to the load RL(see Section II-B). As these
impedances are much higher than the load RL, the error these
produce on the measure (εtest:n) should be close to 0 for low
loads, and increasingly noticeable for higher load values. The
growing rate of the error should differ between experiments and
be related to the total impedance value in parallel to the load.
This phenomenon was not noticed in the results. This led us
to believe that the current sources had a high output impedance
(over 50MΩ). Nevertheless, an unexpected offset error was ob-
Fig.8.
electrical model test. The circuit represents the electrical model of an organic
tissue, where RV I is the resistance between the electrodes (chosen as 50 Ω),
and R1, R2, and C have the values shown in Table III. Equivalent terminals of
each lead are connected to the same point.
Electroniccircuitandconnectionoftheterminalsforthebioimpedance
Fig. 9.
the load RL. The ordinate shows the error relative to the results of Combination
No. 1 (see Table II) as defined by (3). Each line represents the result for one
of the combinations defined in Table II. Since Combination No. 1 is used as
reference to calculate the error, the result is 0 for all load values.
Electrical interference test results. The abscissa shows the value for
served. Discrepancies between the offset errors for the connec-
tion of VMs (Combinations No. 5, 6, 7) were less than 0.2%.
On the other hand, discrepancies between the offset errors for
the connection of CIs (Combinations No. 8, 9, 10) were greater
than 1%. It is believed that despite the power isolation, the con-
nection of improved Holland circuits to the same load produces
interactions that cause the offset error.
C. Bioimpedance Electrical Model Test
Method: The previous experiments have been based on re-
sistive loads. However, bioimpedances also involves capacitive
elements. To assess the accuracy of the device for capacitive
loads, the three leads were simultaneously connected to the
same bioimpedance electrical model (see Fig. 8) [1]. Eight con-
figurationswithdifferentcomponentvaluesweremeasured(see
Table III). The error for each configuration was calculated with
respect to the theoretical values as the mean of 50 repetitions.
Results and Discussion: Table III shows the value of the
componentsforthedifferentconfigurationsandtherelativeerror
for each lead. The errors did not exceed ±1% and the variation
between the three leads was always less than 0.1%. Thus, it is
assumed that the errors were mainly caused by the tolerances of
the components and the electrical interactions explained in the
previous section.

Page 6

2278IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 59, NO. 8, AUGUST 2012
TABLE III
COMPONENT VALUES AND RESULTING RELATIVE ERROR FOR THE
BIOIMPEDANCE ELECTRICAL MODEL TEST
Fig. 10.
constructed, the location of the 12 electrodes within the tank and the terminal
connected to each electrode.
Saline tank test diagram. This shows the dimensions of the tank
D. Saline Tank Test
Method:Inthepreviousexperiments,theleadsalwaysshared
the same connections to the same load. However, in a real appli-
cation, leads will most likely be connected to different locations
of the organism. Therefore, there will be some impedance be-
tween leads and theload values willdiffer foreach lead. A more
realistic test was carried out using a saline water tank.
An elliptic tank 280 cm in length and 160 cm in width was
filled in with 2 cm of a 1000 ppm NaCl dissolution. Twelve
Ag/AgCl electrodes were placed in the inner wall of the tank.
Fig. 10 shows the location of the electrodes and the terminals
assigned to them.
The theoretical resistive value for each individual lead was
calculatedwiththesimulationtoolCOMSOLMultiphysics.The
output of the three leads were recorded for different combina-
tions of lead connections (see Table IV).
Results and Discussion: Table IV shows the results for each
lead for the different experiments. Discrepancies between the
simulated and measured values were found to be less than 3%.
This may be due to environmental factors such as temperature.
TABLE IV
RESULTS FOR THE SALINE TANK TEST
Nevertheless, the simulation merely provided a guiding value.
More relevant was the error caused by the addition of leads,
which less than ±0.4% in all cases. As expected, the addition of
leads to the water tank caused an error inferior to the connection
of the leads to the same load (±0.8%).
E. Human Testing
Method:Inordertodemonstratethecorrectfunctioningofthe
system,itwasusedinaparticularapplication,specificallythesi-
multaneousmeasurementofthreeIPsignals.Thefourterminals
of the three leads were connected to a subject through 12 elec-
trodes placed around the thorax at the height of the xiphoid pro-
cess, as shown in Fig. 11. Variations in transthoracic impedance
were recorded while the patient was breathing spontaneously.
The lung volume changes were simultaneously recorded with
a pneumotachograph (A. Fleisch No. 3, Lausanne, Switzerland
with Biopac SS40L pressure transducer, Goleta, CA, USA) by
integrating the measured flow rate signal.
ResultsandDiscussion:ThechartsinFig.11showtheoutput
for the three leads and the spirometer over 45 s. A visual inspec-
tion showed that the results are in line with previous findings by
Sepp¨ a et al. [3].
V. DISCUSSION
This study presented the design, implementation, and test-
ing of a novel multilead bioimpedance measurement system.
New hardware and new signal processing approaches were in-
troduced. They were proven to be functional and highly accu-
rate for the time-domain analysis of the magnitude of multiple
bioimpedances.
A. Hardware
The system showed excellent performance when leads were
connected individually to the measurand. The small error
(< ±0.8%) seen when multiple leads were connected simul-
taneously is likely to have been caused by the current sources.
The isolation of active current sources has proven to be a
promising approach to ensure independence between fed cur-
rents. However, the improved Howland circuit may not be the

Page 7

GRACIA∗et al.: MULTILEAD MEASUREMENT SYSTEM FOR THE TIME-DOMAIN ANALYSIS OF BIOIMPEDANCE MAGNITUDE 2279
Fig. 11.
the thorax and the terminal connected to each electrode. The upper chart shows
the output signals for the three leads and the lower chart for the spirometer
during 45 s of tidal breathing.
Upper pictures show the locations of the 12 electrodes placed around
best current source circuit for this solution due to electrical
interactions.
The lack of solutions for simultaneous independent current
feeding calls for further research. We believe that a comparative
studyofdifferentdifferentialandpowerisolatedcurrentsources
shouldbeconducted.Thetestspresentedinthisarticlerepresent
a solid reference for this study.
B. Signal Processing
ThecombinationofFDMandundersamplinghasbeenshown
to separate frequencies accurately while reducing hardware re-
quirements.
Although energy efficiency was sacrificed in favor of con-
figurability in the implementation of the prototype device, the
processing approach potentially enables the reduction of energy
consumption. On the one hand, as FDM does not require syn-
chronization between leads, the DSPs can be replaced with ana-
log oscillator circuits. Vuorela et al. [14] have recently demon-
strated that this change may provide a decrement of 6 mA per
lead. On the other hand, the reduction in hardware requirements
enabled by undersampling involves a decrement in energy con-
sumption.
C. Applications
In addition to IP, the developed system can be used
for any other application based on time-domain analysis of
bioimpedance magnitude. The simultaneous access to multiple
bioimpedances is promising for:
1) Signal separation:
For
bioimpedance is affected by cardiac and respiratory activ-
ities. However, both contributions have overlapping fre-
quency bands that cannot be satisfactorily separated using
linear filters. Bioimpedance signals recorded in multiple
locations could be combined to separate cardiac and res-
piratory signals [5].
2) Motion artifact rejection: Motion artifacts are a common
probleminimpedanceplethysmographic.Thecomparison
of signals recorded in different locations for the removal
of movement effects has been already proven for elec-
trocardiograms [4]. Similar methods could be applied to
bioimpedance measurement.
3) Multiple systems coexistence: Unlike the TDM or syn-
chronous undersampling, the FDM method does not re-
quire any kind of communication between leads. This
advantage simplifies the coexistence of independent
bioimpedance measurement systems on the same body,
providing flexibility to body sensor networks such as the
DexterNet [15].
example,atransthoracic
VI. CONCLUSION
A simultaneous access to multiple bioimpedance signals has
the potential to improve the accuracy of actual applications, as
well as to develop new ones. The design of new approaches out
of the main trend focusing on EIT enables the simplification
of the hardware and signal processing. It is most promising for
increasing precision in measurement and reduce energy con-
sumption.
ACKNOWLEDGMENT
The authors would like to thank the Renewable Energies In-
vestigation Group at the CIRCE Foundation for selflessly al-
lowing them to use their facilities.
REFERENCES
[1] S. Grimnes and Ø. G. Martinsen, Bioimpedance & Bioelectricity Basics.
London, U.K.: Academic, 2000.
[2] R. Patterson, “Fundamentals of impedance cardiography,” IEEE Eng.
Med. Biol. Mag., vol. 8, no. 1, pp. 35–38, Mar. 1989.
[3] V.-P.Sepp¨ a,J.Viik,andJ.Hyttinen,“Assessmentofpulmonaryflowusing
impedance pneumography,” IEEE Trans. Biomed. Eng., vol. 57, no. 9,
pp. 2277–2285, Sep. 2010.
[4] M. Chawla, “PCA and ICA processing methods for removal of artifacts
and noise in electrocardiograms: A survey and comparison,” Appl. Soft
Comput., vol. 11, no. 2, pp. 2216–2226, 2011.
[5] H. Nakesch, H. Pf˜A1
4tzner, C. Ruhsam, P. Nopp, and K. Futschik, “Five-
electrode field plethysmography technique for separation of respiration
and cardiac signals,” Med. Biol. Eng. Comput., vol. 32, p. 65–70, 1994.
[6] D. S. Holder, Electrical Impedance Tomography, Methods, History and
Applications.New York: Taylor & Francis, 2004.
[7] M. Min, R. Land, O. Martens, T. Parve, and A. Ronk, “A sampling mul-
tichannel bioimpedance analyzer for tissue monitoring,” in Proc. 26th
Annu. Int. Conf. IEEE Eng. Med. Biol. Soc., 2005, vol. 1, pp. 902–905.

Page 8

2280IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 59, NO. 8, AUGUST 2012
[8] Y. Granot, A. Ivorra, and B. Rubinsky, “Frequency-division multiplexing
for electrical impedance tomography in biomedical applications,” Int. J.
Biomed. Imag., vol. 2007, no. 2, art. no. 2, 2007.
[9] P. Annus, “Multichannel bioimpedance spectroscopy: Instrumentation
methods and design principles,” Ph.D. dissertation, Tallinn Univ. Tech-
nol., Tallinn, Estonia, 2009.
[10] K.BooneandD.Holder,“Currentapproachestoanalogueinstrumentation
design in electrical impedance tomography,” Physiol. Meas., vol. 17,
pp. 229–247, 1996.
[11] W. Wang, M. Brien, D. Gu, and J. Yang, “A comprehensive study on
current source circuits,” in Proc. 13th Int. Conf. Bioimpedance 8th Conf.
Electr. Impedance Tomography, 2007, vol. 17, pp. 213–216.
[12] A.Ar-Rawi,M.Moghavvemi,andW.Wan-Ibrahim,“Comparingbetween
three current source circuits for using in bio electrical impedance design,”
in Proc. Int. Conf. Tech. Postgraduates, 2009, pp. 1–4.
[13] O. M¨ artens and M. Min, “Multifrequency bio-impedance measurement:
undersampling approach,” in Proc. 6th Nordic Process. Symp.-NORSIG,
2004, pp. 145–148.
[14] T. Vuorela, V.-P. Sepp¨ a, J. Vanhala, and J. Hyttinen, “Design and imple-
mentation of a portable long-term physiological signal recorder,” IEEE
Trans. Inf. Technol. Biomed., vol. 14, no. 3, pp. 718–725, May 2010.
[15] P. Kuryloski, A. Giani, K. Gilani, R. Gravina, and V.-P. Sepp¨ a, E. S. et al.,
“Dexternet: An open platform for heterogeneous body sensor networks
and its applications,” in Proc. 6th Workshop Body Sens. Netw., 2009,
pp. 92–97.
Authors’ photographs and biographies not available at the time of publication.