Home Search Collections Journals About Contact us My IOPscience

Broadband discrete-level excitations for improved extraction of information in bioimpedance

measurements

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2014 Physiol. Meas. 35 997

(http://iopscience.iop.org/0967-3334/35/6/997)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 193.40.240.154
This content was downloaded on 06/06/2014 at 16:07

Please note that terms and conditions apply.

 Institute of Physics and Engineering in Medicine Physiological Measurement

Physiol. Meas. 35 (2014) 997–1010 doi:10.1088/0967-3334/35/6/997

Broadband discrete-level excitations for

improved extraction of information in
bioimpedance measurements
Mart Min and Toivo Paavle
Thomas J Seebeck Department of Electronics, Tallinn University of Technology,
Ehitajate tee 5, Tallinn 19086, Estonia

E-mail: min@elin.ttu.ee

Received 30 November 2013, revised 20 February 2014

Accepted for publication 3 April 2014
Published 20 May 2014

Abstract
The implementation of bioimpedance-based methods in implantable and
wearable medical devices requires simple, cheap and low energy consuming
measurement settings for enabling impedance spectroscopy at a wide range of
frequencies. In the present paper, such a wideband bioimpedance measurement
method is discussed, which embodies two-channel impedance measurement
for monitoring of the frequency-dependent phase shift between the channels
(phase spectrum). In addition, the improved resolution is achieved by
employing comparative measurements by introducing the predetermined
reference impedance into one of the measurement channels. The proposed and
analyzed measurement system uses a binary excitation signal that simplifies
signal generation and processing hardware and does not need sophisticated
software—low-complexity devices can be designed this way. It is shown
that in particular the binary chirp excitation has some essential advantages
compared with its counterparts—the maximum length sequence and binary
multifrequency excitations. The spectra of chirps of the binary chirp excitation,
including their discrete-level modifications, are continuous and flat at the same
time. Due to the independent scalability in time and frequency domains and
very high chirping rate, the chirps are especially suitable as excitation signals
for wideband spectroscopy of dynamic objects with changing impedances in
devices such as implantable heart monitors, pacemakers and high-throughput
microfluidic lab-on-chip-type devices for performing bioimpedance-based
monitoring of cells and droplets.

Keywords: bioimpedance, impedance spectroscopy, broadband excitation,

discrete signals, binary chirp, dynamic impedance
(Some figures may appear in colour only in the online journal)

0967-3334/14/060997+14$33.00 © 2014 Institute of Physics and Engineering in Medicine Printed in the UK 997
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

1. Introduction

The study of passive properties of a biological object is often carried out by the analysis of
its electrical equivalent EBI—electrical bioimpedance of the living tissue or cells. The EBI
represents a definite set of resistive and capacitive components connected into an equivalent
circuit, which can be described by a frequency-dependent complex impedance Z(jω) =
|Z(ω)|exp(jz(ω)), where ω = 2π f is the angular frequency in rad s−1. Usually it is expected
that the impedance Z(jω) is time-invariant, but when measuring the impedance of a beating
heart and/or breathing lungs (figure 1) the impedance is certainly time-variant, Z(jω,t), as
considered in Sanchez et al (2013). To remain adequate in such non-stationary situations,
we have to fulfil two conditions: (i) the frequency ω = 2π f must be high in comparison
with the speed of changes, and (ii) the measurement time must be so short that changes are
negligible during the measurement procedure. The measurement results can be presented in a
wide frequency range as impedance spectrograms Z(jω, ti) measured at different time moments
t1, t2, t3, . . . (Sanchez et al 2011, 2013).
The impedance in the complex plane Z(jω) describes the state and properties of biological
objects under study within the range of a few kHz to several MHz (β-region) depending on the
structural properties of the object (Grimnes and Martinsen 2008). In our study, we focused on
the frequency region below 100 kHz, which is characteristic of implantable medical devices
(Min et al 2004, Min and Parve 2007) used in cardiac control and monitoring (Howie et al
2001, Kink et al 2006, 2011, Min et al 2005a, 2005b). The examples in the following chapters
illustrate the spectral behavior of excitation signals and responses only in that frequency region.
To obtain information about the object in a wide frequency range, it is appropriate to
study the impedance Z in the frequency domain applying the Fourier transform F( · ) to both
the output response voltage, Vz(t), and the excitation (stimulus) current, Iexc(t):
Z(jω) = F (Vz (t ))/F (Iexc (t )). (1)
In principle, many different waveform patterns can be used for forming the excitation
signals (Min et al 2008); for example, the multisine waveform is preferred in several
applications (Sanchez et al 2011). Multisine excitation enables an exact but discrete
(the number of sine wave components is limited) coverage of the excitation bandwidth
Bexc coinciding with the frequency range of interest (Annus et al 2012b, Ojarand et al 2012).
Moreover, multisine excitation can also be used for wideband characterization of such time-
variant impedances (Sanchez et al 2013) as is inherent to a blood pumping heart. However,
here another problem is raised: the characteristic frequencies of the time-variant impedance
will also change in time. As a result, we cannot determine optimal frequencies as proposed
in Sanchez et al (2011, 2013). To overcome the problem, we should generate a continuous
spectrum excitation covering the frequency range without any interruptions. The chirp signal
with scalable frequency range and duration has such a spectrum (Paavle et al 2008).
However, the bandwidth coverage is not the only requirement for the excitation. Several
other requirements are important; for example, simplicity of generation, flatness or other
predetermined shapes of the amplitude spectrum, and flexibility of scaling. Moreover, ultra-
low energy consumption of battery energy becomes especially important (Yan et al 2013)
for implantable and wearable devices. Hereby, the discrete-level excitation signals, especially
binary ones, have clear advantages, reducing the complexity of generation and reducing the
power consumption to a minimum (Min et al 2006, Min and Parve 2007), converting the electric
charge of the battery into excitation energy with minimal losses not exceeding about 5%.
In this paper, we compare some often used discrete-level broadband signals: maximum
length sequences (MLS) (Gawad et al 2007), roughly digitized multisines or few-level
998
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

Figure 1. A bioimpedance measurement set up for measurement of variations of the

respiratory impedance ZR of breathing lungs and the ventricular impedance ZV of
the beating heart—a microelectronic bioimpedance measurement unit is placed into
a cardiac pacemaker to estimate patients’ physical workload (variations of ZR) and
ventricular ability (variations of ZV).

multifrequency signals (Yan et al 2013, Ojarand et al 2012), and discrete-level or pseudo chirps
(Pollakowski and Ermert 1994). As the spectrum of discrete signals is strongly influenced by
higher harmonic components, certain extra means were applied in measurements to maintain
satisfactory results.
To improve the quality of monitoring, we suggest a two-channel measurement method and
the corresponding system. It incorporates measurement and reference channels together with
spectral analysis of both the response signal from the measurement channel and the reference
signal from the other channel.
Finally, analysis of the phase spectrum instead of the amplitude spectrum is proposed.
This method is highly insensitive to unevenness in the spectrum, which is characteristic
of discrete-level signals containing higher harmonic components. Usually, the presence of
higher harmonics causes excessive blur in the amplitude spectrum and the spectra of real and
imaginary parts of signals. As the phase spectrum is evaluated through the ratio of imaginary
and real spectra of signals, the blur is suppressed greatly.
Use of the phase spectrum to monitor changes in the bioimpedance has been employed
in earlier papers by several authors (Paavle et al 2008, Min et al 2009, Trebbels et al 2010).
However, it is not the phase spectrum itself but the monitoring of its changes that is the
most informative. For that reason, a pre-dimensioned impedance is included into the reference
channel of the measurement system. This way we can detect tiny changes in the object under
study.
Our particular aim is to open and emphasize the advantages of chirp signals. The chirp
excitation can be scaled easily in the frequency as well as the time domain. It means that we
can change the frequency range at the same duration of excitation and, vice versa, change
the duration of excitation leaving the frequency range unchanged. Besides, the chirps have an
almost flat or tunable amplitude spectrum over the whole generated frequency range, which
enables increased accuracy of measurement (Min et al 2008).
Of course, almost perfect results of measurements can be achieved by using the sine
wave chirp, but the generation of sine wave chirps is more complicated and much more energy
consuming. Fortunately, their binary equivalents can successfully serve as the excitation signal,
rivalling the MLS or digitized (discrete binary or multilevel) multifrequency excitations (Annus
et al 2012a, 2012b). In addition, electronic circuits are simpler and consume less energy (Min
et al 2006).
999
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

2. Broadband discrete-level signals

This chapter presents a short survey of commonly used discrete-level excitation signals,
and discusses their generation, waveforms and spectral features as the result of fast Fourier
transform (FFT) processing. In addition to binary signals, we also introduce some multilevel
signals. In our studies, we have paid focused attention to the energy parameters—mostly to
the energy efficiency δ E of signals, which we have stated as the ratio of the energy within the
desired excitation bandwidth Bexc and the total generated energy.
In general, the energy efficiency δ E (or partial energy for an arbitrary frequency range)
can be obtained with satisfactory precision using the values of the amplitude spectrum from
FFT processing as

N2  N
max −1

δE = |Vs (φi )| 2
|Vs (φi )|2 , (2)
i=N1 i=0
where |Vs (φ i)| is the spectral amplitude at ith frequency bin φ i, and N1 and N2 are the numbers
of frequency bins, corresponding to the initial and final frequencies of the arbitrary frequency
range Bx of interest (Bx = Bexc, usually). Nmax is the total number of frequency bins used in
FFT processing, and the divisor in (2) corresponds to the total energy Etot (Min et al 2009).
For spectrally sparse multifrequency signals, it is practical to estimate the energy efficiency
δ E as a ratio of summarized energy of signal components from i to k in the excitation range
Bexc over the total energy of all generated signal components Nmax from i = 0 to i = (Nmax −1):


k  N
max −1

δE = |Vs ( fi )| 2
|Vs (φi )|2 , (3)
i=1 i=0
where k is the number of spectral components in a useful excitation signal, and |Vs(fi)| is the
level of the spectral component with ith frequency.

2.1. Discrete-level multifrequency signals

Different methods are used for discrete representation of a single sine wave signal in impedance
measurements (Yan et al 2013, Min and Parve 2007). The simplest method for discretization
of a multisine signal to a binary (two-level) multifrequency signal VBMF(t) is expressed
mathematically as the signum function of summarized k sinusoids of frequency fk, amplitude
Ak and initial phase ϕ 0k:
 k 

VBMF (t ) = sign Ak sin(2π fk t + ϕ0k ) . (4)
1
Thus, the waveform of such a composite (eight-frequency) binary multifrequency (BMF)
signal is a sequence of binary values. In figure 2(a), a fragment of it (total duration TBMF =
5 ms) is shown, where k = 8, Ak = A = 1, ϕ 0k = 0, f 1 = 800 Hz and (fk+1/fk) = 2. The usable
frequency range spreads from f 1 to fk and has spectral lines with peaks at all frequencies
fk. Unfortunately, although the amplitudes Ak of all the sine wave components are equal in
expression (4), the spectral peaks of the binary signal (amplitudes of spectral components)
are not (see figure 2(b)). The deviations from the mean value of peaks are about 20% or
even more, depending on the combination and parameters of signal components. If this raises
problems, then for the equalizing or shaping of spectral peaks, a sophisticated algorithm should
be applied to ascertain convenient values for the location, amplitudes and initial phases of the
components (Ojarand et al 2012).

1000
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

(a)

(b)

Figure 2. (a) Waveforms of eight-component sine wave signal V(t) and the corresponding
binary multifrequency signal VBMF(t); (b) PSD of the 1 V BMF signal VBMF(t) given in
(a) at the resistive 1 k load (k = 8, f 1 = 0.8 kHz, f 8 = 102.4 kHz).

There is another way to compose the multilevel multifrequency (MMF) signal:

 k 

VMMF (t ) = sign (Ak sin(2π fk t + ϕ0k )) k. (5)
1

In equation (5), the signum functions of each of 1 to k sine wave frequency components are
summarized as binary signals, and to maintain the amplitude of the MMF signal normalized
to ± 1, the sum must be divided by the number of components k. The resulting multilevel
discrete signal is more complicated, having k non-zero levels with equal resolution intervals
1/k (see figure 3(a)), and a zero level (altogether k +1 levels). Simulations have proved that
in this case the spectral peaks are practically even without any modifying of equation (5), as
shown in figure 3(b)—dissipation of the spectral peaks remains under 2%.
The total power of this multilevel signal and absolute values of spectral peaks are
significantly less than those of the binary version, but its energy efficiency is higher. For the
binary excitation (BMF waveform), the energy efficiency was measured to be approximately
62 to 67%, while that for the MMF waveform is near to 75%.
Multifrequency excitation enables measurement of the impedance only at predetermined
frequencies and, for that reason, we must know the required frequencies of interest in advance.
Although this is not always possible, the advantage of multifrequency excitation is the freedom
in the choice of frequency range and particular frequency values within this range. However,
one must consider that the duration of excitation should be equal to or exceed 1/ f 1, where
f 1 is the lowest frequency component in the excitation signal.
A disturbing problem experienced in the use of discrete time (sampled) multifrequency
excitation is aliasing through the comparatively high level spectrum components, which appear
outside the determined useful frequency range. From the results of simulation (figures 2 and 3),
the partial energy of high frequency ( f > fk) components is about 17% for the binary signal

1001
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

(a)

(b)

Figure 3. (a) Waveform of the eight-component MMF signal against its binary
counterpart (BMF), and (b) PSD of the MMF signal in (a) (k = 8, f 1 = 0.8 kHz,
f 8 = 102.4 kHz).

(BMF waveform in figure 2) and only near to 7% for the nine-level discrete signal (MMF
waveform in figure 3).

2.2. MLS signals

The MLS signal is a pseudorandom binary sequence, which can be easily generated using a
hardware-based shift register fed back through an XOR logic gate. In fact, the MLS signal
is a periodic one with the period of TMLS; however, the sequence of binary values within
every period behaves like a random (or pseudorandom) sequence. The basics of MLS and its
application as an excitation for the bioimpedance measurement is discussed by several authors,
for example in Sun et al (2007).
The MLS circuit of nth order (n is a number of stages in the shift register) generates a
binary sequence of (2n−1) pulses with a duration of TMLS = (2n−1)/ f clk, where f clk is the
clock frequency (figure 4(a)). The main (first lobe) bandwidth of the MLS signal spectrum is
B1 = f clk (B1 = 100 kHz in figure 4(b)) and its energy efficiency is about 90%.
Unfortunately, we cannot use all the bandwidth B1 for excitation, because the envelope of
its sinc function type (|sin(π f )/(π f )|) spectrum reduces to zero at every f = kfclk (k = 1, 2,
3, etc). It means that the spectrum is not flat and its power spectral density (PSD) function of
MLS signals reduces to half of its initial level, roughly at Bexc = f ≈ 0.44 f clk (Bexc = 44 kHz
in figure 4(b)), which is expected to be usable in real measurements (Min et al 2009). These
properties should be considered as the drawback of MLS excitation.
Figure 4(a) depicts a fragment of the ninth order MLS signal in the time domain. In
figure 4(b), the respective PSD is shown.

2.3. Discrete-level chirps

Chirp-type signals can be classified by the rule of changing their instantaneous frequency.
The basic class includes so-called power-law chirps, the frequency f (t) of which changes by

1002
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

(a)

(b)

Figure 4. The ninth order MLS signal (n = 9) with f clk = 100 kHz: (a) a fragment of
the waveform; (b) PSD of the MLS signal given in (a) at the resistive load Rload = 1 k .

a power function tn. A sine wave power-law chirp of the power of n (n = 1 for the simplest
linear chirp) and amplitude A can be described mathematically as follows:
Vch (t ) = A sin θ (t ) = A sin(2π ( f0t + βt n+1 /(n + 1))), (6)
in which β = Bexc/Tch characterizes the rate of frequency variation (chirping rate), Bexc =
f fin– f 0 is the excitation bandwidth between the final f fin and initial f 0 frequencies, and Tch is
the duration of chirp pulse. The quantity θ (t) stands for the current phase as an integral of
instantaneous frequency θ (t) = f (t) dt.
The simplest discrete-level version of chirps is the two-level binary chirp (BC). However,
in some circumstances three-level ternary chirps should be preferred due to their better spectral
properties and higher energy efficiency (Paavle and Min 2012).

2.3.1. Binary chirps. A BC with two values of ± A, expressed mathematically as VBC(t) =

A · sign(Vch(t)), is illustrated in figure 5(a) together with the waveform of its sine wave
counterpart. For simplicity, the amplitudes are frequently normalized to A = 1 as in figure 4(a).
The PSD of perfect sine wave chirps covers the whole generated bandwidth Bexc almost
uniformly; see figure 5(b). A drawback of BC excitation is the strongly expressed roughness
of its spectrum. Fortunately, the drawback is recouped by several benefits. For example, the
average PSDav value of BCs exceeds the PSD of respective sinusoidal chirps by (4/π )2 ≈
1.62 times.
The BCs are able to ensure the utmost power and energy. Like the MLS, √ the crest factor
of BCs is 1 (let us remember that the crest factor of sine wave chirps is 2 ≈ 1.414). The
energy efficiency of BC s is, in fact, somewhat but not meaningfully less than that of MLS
signals, and remains sufficiently high in any case; theoretically δ E ≈ 0.853 (Min et al 2009).
To form a BC, it is not necessary to generate the sine wave chirp first. Instead, we have
to detect the change of current phase θ (t); see expression (6). For simplicity, let us consider

1003
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

(a)

(b)

(c)

Figure 5. (a) Waveforms of linear sine wave and BCs, (b) their PSD functions (Rload =
1 k ), and (c) generation of the linear BC with duration Tch = 0.1 ms and final frequency
f fin = 100 kHz at the sampling frequency f samp = 106 samp s−1.

the linear (n = 1) BC pulse with initial frequency f 0 = 0. In this case β = f fin/Tch and θ (t) =
π βt2. The polarity of the signal alternates if the current phase has the value θ (t) = kπ , where
1
k = 0,1,2, . . . , etc. Consequently, the polarity changes at t = (k/β) 2 for every integer k. If the
sampling frequency is f samp, then the order number of samples ik, at which the polarity has to
be changed, is

ik = round( fsamp k/β ). (7)
An example of the generation of a BC is illustrated in figure 4(c). For this linear chirp
(n = 1), f 0 = 0, f fin = 100 kHz, Tch = 1 ms (β = 108 Hz s−1) and f samp = 106 samp s−1.

2.3.2. Ternary chirps. The waveform of the ternary signal has three levels, typically
VTC∈{−A; 0; +A}. For simplicity, the levels are frequently normalized to +1, 0 and −1
(figure 6). The spectrum of ternary signals depends strongly on the duration of the zero-value
levels (shortening of the duty cycle by the phase angle α) and the specific harmonic components
are absent in the spectrum in some particular cases. For example, at shortening by α = π /6
per a quarter of one cycle of the signal, the 3rd, 9th, 15th, etc, harmonics are absent (Min et al
2004).
We can describe ternary chirps with shortening by α as

VTC = 0, if kπ − α  θ (t ) < kπ + α
(8)
VTC = sign(sin(θ (t ))), otherwise.
1004
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

Figure 6. Waveform of ternary linear chirp with shortening of cycles by α = π /6.

Similarly to BCs, the generation of ternary chirps can be accomplished by detecting

the current phase θ (t) considering expression (8). As a rule, the energy efficiency of ternary
chirps is higher than that of BCs, exceeding 90% and reaching a maximum value of 93% at
α ≈ π /8 (Paavle and Min 2012). Practical generation of chirp signals is described in the patent
specification (Min et al 2013).

3. Structure of the measurement system

The inherent roughness of the amplitude spectrum of discrete-level signals (see figure 5(b), for
example) complicates the interpretation of measurement results, especially if the amplitude
spectrum is the only source of information. Therefore, it is reasonable to take information from
the phase spectrum, which is insensitive to the amplitude irregularities within the excitation
bandwidth Bexc. However, we need a reference signal to evaluate the phase shift. That is why
we must implement a two-channel measurement system, which enables us to compare the
vectors of the excitation signal and response signals assuming the unity transfer factor (Zref =
1) in the reference channel.
Furthermore, the actual value of the impedance Z is often not of great importance. We
are more interested in observing Z variations caused by the changes and events inside the
object. Therefore, we consider the two-channel measurement system, wherein one of the
channels includes predefined reference impedance Zref, which is approximately equal to Z and
is composed on the basis of a priori information about it. The structure behaves as a matched
filter, showing a zero-value phase spectrum if Z = Zref and enabling us to detect tiny deviations
in the phase spectrum. Thereby, in the reference channel we can use as an equivalent analog
the RC circuit of Z, as well as a set of numerical values (a digital model) obtained from
previously stored results of measurements of the response Vz.
This task can be accomplished by a simple structure with two FFT blocks for transforming
the response and reference signals Vz and Vref separately (structure A, figure 7(a)). The
complex numbers from both blocks are divided to obtain the spectra of relative amplitudes
|Z( f )|/|Zref( f )| and phase differences zr( f ) = z( f ) −ref( f ). The advantage of this
structure is a fast evaluation process, while the quality of the results is less independent of the
type of excitation signal.
Another solution in figure 7(b) implements the cross-correlation procedure (structure B).
By the generalized Wiener–Khinchin theorem, the Fourier transform of the cross-correlation
function (CCF) with the product rzr is defined as the cross-spectrum of signals Vz and Vref;

1005
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

(a)

(c)

(b)

Figure 7. Two structures of measurement system: (a) with two separated FFT channels
(structure A); (b) with two channels and cross-correlation calculus (structure B); (c)
electrical model of the bioimpedance used for calculations.

that is, F(rzr) = Pzr( f ). This relationship can be described as Vz · Vr⇔ F(Vz) · F(Vr)∗ , where
F( · )∗ denotes the complex conjugate (Vaseghi 2006). For structure B, only a single FFT
block is required for producing the cross-power spectrum |Pzr( f )| and the spectrum of phase
differences zr( f ).
This method is somewhat slower than the previous one due to the calculation of the CCF,
but enables us to obtain additional information on the declinations of Z from Zref throughout
the CCF in the time domain, and has a better noise immunity. In addition, it appears that this
structure is more sensitive regarding the excitation waveform: it works well only in the case
of using continuous spectrum signals.

4. Simulation results

The simulation results are obtained by using both structures, A and B, while a simple three-
element RC circuit with R0 = 1 k , R1 = 200 and C1 = 20 nF (see figure 7(c)) serves as an
‘unknown’ object Z in addition to the approximated reference impedance Zref. However, for
some examples the reference impedance was set to Zref = 1.
Figure 8(a) shows the initial fragments of response signals Vz(t) at three different
excitations: (i) multifrequency signal of eight components with f 1 = 0.8, f 2 = 1.6, . . . ,
f 8 = 102.4 kHz; (ii) the seventh order MLS sequence with f clk = 100 kHz; (iii) BC with f 0 =
0 and f fin = 100 kHz. In all cases, structure A was used and the duration of excitation was
5 ms.
Figure 8(b) shows the phase and amplitude spectra as the results of simulation by structure
A and Zref = 1. In this case, the phase spectrum describes a frequency-dependent relative delay
or negative phase shift of the response signal against the excitation waveform and the amplitude
spectrum presents a relative decrease of the magnitude.
Figure 8(c) illustrates variations of the CCF from structure B. The normalized CCF is
depicted in two extreme cases: Z = Zref and Zref = 1.
In figure 9, one can compare the response spectra in the case of the BMF and BC excitation
generated bandwidth, which cover an almost equal frequency range. The measurement
structure B in figure 7(b) with Zref = 1 was implemented for this example. The continuous and

1006
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

(a) (b)

(c)

Figure 8. (a) Fragments of response signals (Vrsp) on the background of respective

excitations—BMF, MLS and BC; (b) phase and amplitude spectra from structure A
using BC excitation; (c) normalized cross-correlation functions at Zref = 1 and Zref = Z
from structure B using BC excitation.
(a) (b)

Figure 9. Comparison of logarithmic spectra of responses for use of BC from 0 to

100 kHz and BMF with components at eight discrete frequencies of 0.8, 1.6, . . . , and
102.4 kHz: (a) cross-power spectra; (b) phase spectra (measurement structure B was
implemented).

smooth phase spectrum within the whole chirp bandwidth from 0 to 100 kHz can be noted as
an advantage of using chirp excitation.
From the results of simulation with Zref = 1 (figures 8(b) and 9) it is difficult to detect
the smaller changes of the impedance under study if the full scale of measurement covers the
whole variation range of spectra. Thus, to improve the sensitivity of estimation the reference
channel should include an impedance model with stable predetermined parameters near to
those of unknown impedance to be measured.
Moreover, figure 9 shows that it is problematic to apply the excitation with a sparse
spectrum (multifrequency signal with binary or multisine waveform). It has well-defined
values only at the predefined frequencies, but at other frequencies both the amplitude and
phase values remain uncertain.
1007
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

(a) (b)

Figure 10. Spectra of phase declinations at the +1% increments of Z components in

the case of the following excitations: (a) BC; (b) BC and pseudorandom MLS when
increasing C1 by 1% (see figure 2(a)).

In figure 10, the spectra of phase deviation are the results of simulation by the structure
B, while Zref had predefined constant parameters and Z was varied.
Figure 10(a) demonstrates the main result of the proposed measurement method—the
possibility to distinguish very small variations of components of Z through the spectra
of zr( f ). The concurrent simulations proved that MLS excitation cannot ensure such a
high level of clearness of declinations in the phase spectrum as the BC excitation does; see
figure 10(b).

5. Summary and conclusion

The proposed method conjoins different means and aspects to improve the quality of
bioimpedance measurements in implantable and wearable devices using simple excitation
signals. The simulation results confirm the possibility of designing reduced complexity
but high quality impedance spectroscopy devices in which discrete-level excitations are
applied, especially those with only two-level BC waveforms. Such impedance measurement
micropower devices (Yan et al 2013) are very welcome in implantable medical instruments
such as cardiac pacemakers and monitors (Min et al 2005a, 2005b) and lab-on-a-chip -type
biodevices for cell detection and analysis (Min and Paavle 2013).
First, the use of binary waveform chirps for excitation simplifies the required hardware and
software essentially and assures the measurement in a wide frequency band at the same time.
The advantages of BCs are their flat frequency spectrum over the entire generated bandwidth
and independent scaling in both the time and frequency domain. From this point of view, the
BC overcomes other popular binary excitations—MLS and BMF signals.
Second, focusing on the phase spectrum enables us to avoid the disturbing impact caused
by the higher harmonic components inherent to rectangular-wave excitations including the
BC.
Third, the use of the two-channel measurement system with the reference impedance
Zref designed on the basis of a priori knowledge in one of the channels increases the accuracy
of measurement substantially and permits the detection of tiny changes in the structure and/or
parameters of the unknown bioimpedance Z to be measured.
The next step for future research is finding the simple but effective signal processing
methods specially developed for the processing of responses to binary excitations such as
the fast M-sequence transform (Sun et al 2007) and fast Hadamard transform (Gawad et al
2007) in the time domain and the Walsh transform (Yang et al 2011) in the frequency domain.

1008
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

Traditional digital methods for correlation and convolution analysis in the time domain and
FFT in the frequency domain are too complicated and too energy consuming for implantable
devices. More simple solutions for signal processing, for example, as described in Annus
et al (2013), will enable us to implement the findings in the paper in microelectronic cardiac
monitors (Yan et al 2013) and microfluidic devices (Min et al 2014).

Acknowledgments

This research was supported by the European Union through the European Regional
Development Fund in frames of the research center CEBE and the competence center ELIKO
and financed partly by the Estonian Research Agency, project IUT19-11.

References

Annus P, Land R, Min M and Ojarand J 2012a Simple signals for system identification Fourier
Transform–Signal Processing (Rijeka: Intech Open) pp 257–76
Annus P, Land R, Reidla M, Ojarand J, Mughal Y and Min M 2013 Simplified signal processing for
impedance spectroscopy with spectrally sparse sequences J. Phys.: Conf. Ser. 434 012031
Annus P, Min M, Ojarand J, Paavle T, Land R, Ellervee P and Parve T 2012b Multisine and
binary multifrequency waveforms in impedance spectrum measurement—a comparative study 5th
European Conference of the International Federation for Medical and Biological Engineering
(IFMBE Proceedings vol 37) (Berlin: Springer) pp 265–8
Gawad S, Sun T, Green N G and Morgan H 2007 Impedance spectroscopy using maximum length
sequences: application to single cell analysis Rev. Sci. Instrum. 78 054301
Grimnes S and Martinsen Ø G 2008 Bioimpedance and Bioelectricity Basics 2nd edn (Amsterdam:
Elsevier)
Howie M B, Dzwonczyk R and McSweeney T D 2001 An evaluation of a new two-electrode myocardial
electrical impedance monitor for detecting myocardial ischemia Anesth. Analg. 92 12–18
Kink A, Rätsep I, Min M and Parve T 2011 Device and method for monitoring cardiac pacing rate
US Patent Specification 7894900B2
Kink A, Salo R, Min M, Parve T and Rätsep I 2006 Intracardiac electrical bioimpedance as a basis for
controlling of pacing rate limits EMBEC’06: Proc. IEEE Engineering in Medicine and Biology
Conf. pp 6308–11
Min M, Kink A, Land R, Parve T and Rätsep I 2004 Modification of pulse wave signals in electrical
bioimpedance analyzers for implantable medical devices Proc. 26th Int. Conf. of IEEE Engineering
in Medicine and Biological Society pp 2263–6
Min M, Kink A and Parve T 2005a Rate adaptive pacemaker using impedance measurements and stroke
volume calculations US Patent Specification 6975903B1
Min M, Kink A, Parve T and Rätsep I 2005b Bioimpedance based control of rate limits in artificial
cardiac pacing BioMED’05: Proc. IASTED Int. Conf. on Biomedical Engineering pp 697–703
Min M and Paavle T 2013 Improved extraction of information in bioimpedance measurements J. Phys.:
Conf. Ser. 434 012029
Min M, Paavle T, Annus P and Land R 2009 Rectangular wave excitation in wideband bioimpedance
spectroscopy MeMeA’09: Proc. IEEE 4th Int. Workshop on Medical Measurements and Applications
pp 268–71
Min M, Paavle T, Land R, Annus P and Parve T 2013 Method and device for fast measurement of
frequency response with scalable short chirp signals European Patent EP 2314217B1
Min M and Parve T 2007 Improvement of lock-in electrical bio-impedance analyzer for implantable
medical devices IEEE Trans. Instrum. Meas. 56 968–74
Min M, Parve T and Pliquett U 2014 Impedance detection Springer Encyclopedia of Microfluidics and
Nanofluidics 2nd edn (Heidelberg: Springer)
Min M, Parve T and Salo R 2006 Biomedical electronics Wiley Encyclopedia of Biomedical Éngineering
(Hoboken, NJ: Wiley) pp 500–14
Min M, Pliquett U, Nacke T, Barthel A, Annus P and Land R 2008 Broadband excitation for short-time
impedance spectroscopy Phys. Meas. 29 S185–92

1009
Physiol. Meas. 35 (2014) 997 M Min and T Paavle

Ojarand J, Land R and Min M 2012 Comparison of spectrally sparse excitation signals for fast
bioimpedance spectroscopy in the context of cytometry MeMeA’12: Proc. IEEE 7th Int. Workshop
on Medical Measurements and Applications pp 1–5
Paavle T and Min M 2012 Rectangular-wave chirps for broadband measurement: spectra and energy
BEC’12: Proc. 13th Baltic Eletronic Conf. pp 195–8
Paavle T, Min M and Parve T 2008 Using of chirp excitation for bioimpedance estimation: theoretical
aspects and modeling BEC’08: Proc. 11th Baltic Eletronic Conf. pp 325–8
Pollakowski M and Ermert H 1994 Chirp signal matching and signal power optimization in pulse echo
mode ultrasonic nondestructive testing IEEE Trans. Ultrason. Ferroelectr. Freq. Control 41 655–9
Sanchez B, Louarroudi E, Jorge E, Cinca J, Bragos R and Pintelon R 2013 A new measuring
and identificationapproach for time-varying bioimpedance using multisine electrical impedance
spectroscopy Phys. Meas. 34 339–57
Sanchez B, Vandersteen G, Bragos R and Schoukens J 2011 Optimal multisine excitation design for
broadband electrical impedance spectroscopy Meas. Sci. Technol. 22 115601
Sun T, Gawad S, Bernabini C, Green N G and Morgan H 2007 Broadband single cell impedance
spectroscopy using maximum length sequencies: theoretical analysis and practical considerations
Meas. Sci. Technol. 18 2859–68
Trebbels D, Woelki D and Zengerle R 2010 High precision phase measurement technique for cell
impedance spectroscopy J. Phys.: Conf. Ser. 224 012159
Vaseghi S V 2006 Advanced Digital Signal Processing and Noise Reduction (Chichester: Wiley)
Yan L et al 2013 A 13μA analog signal processing IC for accurate recognition of multiple intra-cardiac
signals IEEE Trans. Biomed. Circuits Syst. 7 785–95
Yang Y, Kang M, Lu Y, Wang J, Yue J and Gao Z 2011 Design of a wideband excitation source for fast
bioimpedance spectroscopy Meas. Sci. Technol. 22 013001

1010