IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL.

70, 2021 4500710

An FPGA-Based Multifrequency EIT System

With Reference Signal Measurement
Yanbin Xu , Member, IEEE, Zhicheng Yan , Bing Han , and Feng Dong , Senior Member, IEEE

Abstract— Multifrequency electrical impedance tomogra- through surface electrodes, and the induced voltage on the
phy (MFEIT) has been rapidly developed and has shown boundary was measured, and the image reconstruction algo-
great potential for biomedical analysis and imaging. This arti- rithm is used to reconstruct the impedance distribution. Con-
cle presents the design and evaluation of a wideband and
high-precision MFEIT system which incorporates novel features ventional EIT used a fixed single frequency as the excitation
to fit biomedical application. First, a high-accuracy multifre- frequency [5], [6]. Since the electrical properties of biological
quency current source with adjustable output frequency and tissues usually vary over a wide range of frequencies, the data
amplitude as well as high-output impedance is designed based on analysis and imaging results over the single frequency may
the combination of the classic current mirror and a differential not provide a wealth of information for bioprocess evaluation.
topology, which demonstrates stable performance in high fre-
quencies. Second, a reference measurement circuit is developed Therefore, compared with the traditional monofrequency EIT,
for the calibration of phase measurement data through which Multifrequency electrical impedance tomography (MFEIT) is
precise measurements of both amplitude data and phase data more suitable for biomedical detection and biological process
can be obtained. Furthermore, the low-pass filters utilized in research, which utilizes excitation with multiple frequency
the system have been designed based on the Butterworth filter components. MFEIT made rapid progress in recent years [7],
theory to realize the flattest frequency response in the passband.
The MFEIT system is systematically tested and evaluated in [8], [38], which has become a research hotspot in the field of
terms of filter performance, signal to noise ratio (SNR), channel electrical tomography.
consistency and measurement repeatability, flatness analysis and In 2007, Oh et al. [9] designed the KHU Mark1 system for
load dependence as well as imaging experiments. Experimental brain functional imaging with a frequency range of 10 Hz to
results showed that the maximum amplitude SNR and phase 500 kHz. In 2011, Oh et al. [10] developed the Mark2 system,
SNR of the boundary measurements are 71.55 and 64.12 dB,
respectively. The evaluation results indicate that the system has which had some features including flexible electrode config-
good performance over a wide frequency band (100 Hz–1 MHz) urations, multiple independent current sources and voltmeters
and achieves a relatively high SNR. for fully parallel operations, where the signal-to-noise ratio
Index Terms— Current source, electrical impedance tomogra- reached 84 dB. In the same year, the University of Sussex and
phy (EIT), multifrequency, reference signal, system design. the University of Brighton jointly developed a high-precision
multifrequency EIT system for breast cancer detection, which
operated at 10 Hz–20 MHz with low cost, high precision, and
I. I NTRODUCTION simultaneous multifrequency signal excitation [11]. In 2011,
Liu et al. [12] designed a microscopic EIT system for bioassay

E LECTRICAL impedance tomography (EIT) is an imag-

ing modality that estimates the electrical properties of
an object from measurements made on its surface [1]. It has
that included two pairs of current-injection electrodes and
360 voltage-sensing electrodes. In 2014, Wi et al. [13] pub-
lished the Mark 2.5 system with automatic self-calibration for
advantages such as fast, nonintrusive, nonradioactive, and low long-term monitoring. In the same year, Sun et al. [14] devel-
cost. It can make full use of the rich physiological and patho- oped a high-speed EIT system based on information filtering
logical information contained in the bioimpedance to detect and demodulation, which can be demodulated in one-third of
lesions according to the principle of different electrical prop- the signal period. In 2015, Guermandi et al. [15] designed the
erties in different biological tissues [2]–[4]. In general, a small EIT system with electrode-skin contact impedance monitoring,
safe excitation current was injected into the biological tissue which was specially optimized for human body applications
and used EEG electrodes for long-term stable monitoring of
Manuscript received July 26, 2020; revised September 25, 2020; accepted patients. In 2016, Santos et al. [16] published the first complete
September 27, 2020. Date of publication October 15, 2020; date of current
version November 24, 2020. This work was supported in part by the mixed-frequency EIT system paper, from circuit structure,
National Natural Science Foundation of China under Grant 61671322 and in which excitation measurement method, demodulation algo-
Grant 51976137 and in part by the Natural Science Foundation of Tianjin rithm, data normalization, human–machine interaction soft-
under Grant 19JCZDJC38900. The Associate Editor coordinating the review
process was Dr. Amitava Chatterjee. (Corresponding author: Feng Dong.) ware, and other aspects have been elaborated. At the same
The authors are with the Tianjin Key Laboratory of Process Mea- time, the model based on the amplitude-phase, real-imaginary
surement and Control, School of Electrical and Information Engineering, part and the measured results of human body were given.
Tianjin University, Tianjin 300072, China (e-mail: [email protected];
[email protected]; [email protected]; [email protected]). In 2017, Yang and Jia [17] published a MFEIT system using
Digital Object Identifier 10.1109/TIM.2020.3031158 semi-parallel structure for real-time 2-D and 3-D imaging.
1557-9662 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on April 15,2024 at 19:25:50 UTC from IEEE Xplore. Restrictions apply.
 4500710 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021

Fig. 1. Structure block diagram of the FPGA-based MFEIT system with reference signal measurement.

In the same year, Avery et al. [18] developed a highly ver- 1) The reference measurement circuit was constructed for
satile and reproducible multifrequency EIT system. In 2018, the calibration of phase data through which the phase
Singh et al. [19] introduced a novel wireless high-speed delays caused by the stray capacitance or the hardware
frequency division multiplexing multifrequency EIT system. system itself can be removed. Moreover, the amplitude
The current injection method adopted simultaneous injection results of demodulation can also be calibrated by this
to all the electrodes having different frequencies, which can reference structure and therefore both the amplitude and
reconstruct real-time 2-D images for continuously monitor- phase data can be measured very precisely.
ing applications where impedance changes rapidly. In 2019, 2) A wideband, high-precision current source with multiple
Wu et al. [20] proposed a highly integrated wearable multi- frequency components is realized by combination of the
frequency EIT system, which used 16 new active electrodes classic current mirror and a differential topology. Instead
and had an image frame rate of 122 frames/s with a wide of being released to the ground, the current after flowing
operating bandwidth of 1 MHz. through the load is connected to the differential topology
It is known from bioimpedance spectroscopy that by mea- by which the output impedance of the current source can
suring tissue impedance at different frequencies, the electrical be further improved.
properties of biological tissues and more deeply the physical 3) The low-pass filters (LPFs) in this system are care-
and pathological information can be explored [21]. MFEIT is fully designed based on the Butterworth filter theory to
able to take advantage of the frequency-dependence of the achieve the flattest frequency response in the passband
complex impedances of biological tissues by injecting currents and hence the variety of measured signals can be only
at multiple frequencies. Most MFEIT systems utilize the adjusted by the electrical properties of the biological
digital quadrature demodulation to deal with the measured data tissues in frequency domain.
after ADC in the voltage measuring unit by which the demod- Besides, the performance of the MFEIT system has been
ulated amplitude data and phase data can be obtained [9], [12]. systematically evaluated. The modules in the system have
Due to the nonignorable parasitic or stray capacitance of the been tested and the signal-to-noise ratio (SNR) analysis aimed
circuits in high frequencies and the system-based phase delays at both amplitude measurement and phase measurement has
such as the delays caused by filters, the precise measurement been given. Channel consistency and measurement repeatabil-
of phase data is much harder than that of amplitude data. ity, flatness, and load dependence of the system have been
Therefore, most EIT systems would ignore the phase data and also tested and discussed. Moreover, the imaging experiment
utilize the amplitude data only [22]. This strategy is quite easy has been carried out with objects as the biological tissues
to perform and has been widely used in traditional mono- and nonliving things. The results have been carefully ana-
frequency EIT and so extended to MFEIT systems. However, lyzed combining current knowledge of biological impedance
the phase data measured in MFEIT plays an important role to spectrum.
discover the electrical properties of biological tissues which
II. M ETHODS
cannot be ignored. Therefore, a reference phase measure-
ment method is proposed in the development of the MFEIT A. System Structure
system [23]. This article is an extension from its conference To achieve phase measurement calibration and improve
version. measurement accuracy, a MFEIT system based on FPGA is
In this article, a novel serial MFEIT system with designed. Fig. 1 shows the structure of the MFEIT system,
16 electrodes is proposed, which can be used for both which comprises the FPGA with 200 000 system gates and
mixed-frequency and sweep-frequency excitation from 100 Hz peripheral circuit, the multifrequency current source, the ref-
to 1 MHz. The novel features of the proposed MFEIT system erence signal measurement module, the switching circuit,
include the following. the acquisition and signal process module, and a PC for

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on April 15,2024 at 19:25:50 UTC from IEEE Xplore. Restrictions apply.
 XU et al.: FPGA-BASED MFEIT SYSTEM WITH REFERENCE SIGNAL MEASUREMENT 4500710

Fig. 3. Reference signal measurement circuit.

can be adjusted flexibly by the amplitude control module.

The amplitude control module is constructed based on an
8-bit high-precision DAC which can provide reference voltage
for the 14-bit DAC.
Fig. 2. Structure block diagram of the developed multifrequency current The VCCS circuit is based on the combination of the
source.
classic current mirror and a differential topology. Compared
with the popular Howland topology which can be constructed
image reconstruction. The working process of this system is using a single operational amplifier and a handful of resis-
as follows: 1) the waveform synthesizer module of FPGA tors [27], [28], the current mirror is more complicated yet
generates the digital waveform with multiple frequencies and requires less for the accuracy of resistors [24]. In the proposed
the current source transforms this digital signal into stable multifrequency current source, the current mirror is built based
multiple harmonic excitation; 2) the switching circuit adjusts on the current feedback amplifier (Analog Devices, AD844).
the electrode status as excitation, grounding, measuring, or idle The dc feedback topology can remove the residual dc compo-
and the digital switch of the reference measurement module nent of the output and address the probable charging prob-
decides whether the reference resistance or the real load is to lem of the contact resistance with detected object such as
be measured; and 3) the measurement signal is processed and human bodies. In most current source design, the current is
acquired by the FPGA and then demodulated, the results of released to ground after flowing through the load [27]–[30],
which are transformed to PC through the USB interface. which leads to the high common-mode interference and degra-
A FPGA is used as the main controller for digital signal dation of output impedance. In the differential topology of this
processing and control tasks while the serial control logic of current source, the reverse amplifier circuit from one end of the
the system is arranged by an 8-bit embedded microcontroller load is connected to the other end; thus, ideally the voltages
unit (MCU). It is a kind of soft core generated and embedded on two sides of the load are opposite and the common-mode
in the FPGA by which the trigger of the multiple harmonic signal should be zero.
waveform, the arrangement of electrodes status, the setting of
the programmable gain amplifier (PGA), the implementation C. Reference Signal Measurement
of the multifrequency demodulation as well as the data trans-
mission can be controlled and handled. The boundary voltage measurement actually includes volt-
age amplitude and voltage phase, where the absolute value
of voltage phase information is generally small and the
B. Multifrequency Current Source measurement is susceptible to interference. The ADC, DAC,
In MFEIT, the current source is required to be stable over filter, stray capacitance, and measurement object will introduce
a wide frequency band with driving different loads. In the phase shift to the modulated signal; however, the phase shift
proposed multifrequency EIT system, the current source with introduced by the measurement object is the only effective data
high precision, and adjustable amplitude is adopted [24]. needed. Since the phase measurement data plays an important
Fig. 2 shows the diagram of the developed multifrequency role in MFEIT, the reference signal measurement circuit is
current source comprising a 14-bit DAC and amplitude control designed to deal with phase data as shown in Fig. 3. A high-
module, a differential amplifier circuit, a LPF, and voltage- precision resistance is put in series with the following loads,
controlled current source (VCCS) circuit. The waveform syn- specifically the switching circuit and the measured objects.
thesizer generates the digital multiple harmonic waveforms by The input ports of signal processing and acquisition module
summing several mono-frequency sinusoidal signals produced are sequentially connected to the reference resistance and
with intellectual property (IP) cores. The optimal phase offsets the load by which the induced voltages on both of them
of the sinusoidal waveforms are calculated to obtain a lowest can be collected and demodulated. The method maintains the
crest factor [25], [26] to realize further SNR improvement. consistency of signal acquisition circuit while effectively sim-
A 14-bit high-speed DAC with 125 million samples per second plifying the circuit. Therefore, the real phase data modulated
(MSPS) update rate is utilized to convert the digital multiple only by the load can be measured meanwhile the amplitude
harmonic waveform into analog voltage whose amplitude measurement can also be calibrated. The measured effective

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on April 15,2024 at 19:25:50 UTC from IEEE Xplore. Restrictions apply.
 4500710 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021

Fig. 4. Adjacent-excitation adjacent-measurement scheme. Fig. 5. PCBs of the system.

amplitude data A and phase data ϕ are calculated as

⎧
⎨ A = UZ · R
ref
Uref (1)
⎩ϕ = ϕ − ϕ
Z ref

where Rref denotes the resistance of the fixed-value resistor,

U Z and Uref represent the voltage across the load and the
reference resistor, respectively, ϕ Z and ϕref represent the phase
of the load and the reference resistor, respectively.
Fig. 6. Signal processing and acquisition module.
D. Switch Control
To obtain the induced voltage data according to excita- differential measurement signal and convey it to the follow-
tion and distribution of the measured object, certain exci- ing FPGA. The preprocess circuit contains a high-pass filter
tation and measurement scheme is to be implemented. (HPF), the differential amplification circuit, and a LPF, which
Fig. 4 shows the most commonly used adjacent-excitation can transform the collected difference signal into noiseless
adjacent-measurement scheme. Excitation current is injected amplified unipolar signal. The following PGA enlarges this
at the excitation electrode pair and the differential voltages signal with adjustable gain controlled by FPGA; therefore,
are acquired at all adjacent electrodes excluding the current the signal collected by the ADC can be put into the best range
injection; thus, totally 208 voltages can be obtained for one to achieve further SNR improve.
frame.
For the sake of flexible configuration of user-defined exci- F. Multifrequency Demodulation
tation and measurement scheme, the structure of the PCBs is
constructed as shown in Fig. 5, which contains one main board, The multifrequency digital quadrature demodulation is
one baseplate, and four channels boards. Because different designed and implemented to obtain the amplitude and phase
scanning patterns are processed through imaging algorithms, data from the digital signal converted by the ADC. The mul-
the shape arrangement of PCBs will not affect imaging results. tifrequency discrete digital signal collected by the ADC can
The main board implements the principal system functions. be expressed as
The function of the channels board of PCB is to switch the  m  
n
excitation and measurement channels among 16 electrodes. U (n) = A j sin 2π + ϕj (n = 1, 2, . . . , K ) (2)
j =1
Nj
The channels board connects to the electric cables and elec-
trodes, and preprocesses the collected voltages. Each channel where m denotes the amount of all the frequency components,
board has four interfaces for connecting electrodes, so there A j and ϕ j represent the amplitude and initial phase of the
are four identical PCB boards. The baseplate achieves the j th component, respectively, N j is the sampling number in one
signal communication between the main board and the chan- period of the j th frequency component. K represents the total
nels boards. The 16-bit electrode control signal can be used sampling number of the ADC and n is the index of sampling
by all the channels boards and the switching signal can point which ranges from 1 to K .
choose certain channels board to implement the electrode The principle of multifrequency digital quadrature demodu-
switch. Therefore, it is easy for the control to be realized lation is based on the orthogonal characteristic of trigonomet-
through program design. ric function, which can be formulated as
⎧ −1  
⎪
⎪ 
K
n 1
E. Signal Processing and Acquistion Module ⎪
⎪ R = U (n) sin 2π = K A j cos ϕ j
⎨ j N j 2
0
As shown in Fig. 6, the signal processing and acquisition
⎪ 
K −1   (3)
module comprises the preprocess circuit, the PGA circuit as ⎪
⎪ n 1
⎪
⎩ Ij = U (n) cos 2π = K A j sin ϕ j
well as the ADC circuit, which is able to deal with the 0
N j 2

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on April 15,2024 at 19:25:50 UTC from IEEE Xplore. Restrictions apply.
 XU et al.: FPGA-BASED MFEIT SYSTEM WITH REFERENCE SIGNAL MEASUREMENT 4500710

Fig. 7. Realization of multifrequency demodulation.

Fig. 8. Filter performance oscilloscope observation.

where R j and I j are the real part and imaginary part for the
j th frequency component, respectively. Therefore, the ampli-
tude and phase measurements of each frequency component
can be further obtained by
⎧ 2
⎪
⎨ Aj = R2 + I 2
K j j (4)
⎪ I
⎩ ϕ j = arctan j .
Rj
This process of calculation for multifrequency demodula-
tion with two frequency components is implemented in the
FPGA using IP cores, as shown in Fig. 7. The multiply
accumulator (MAC) and Coordinate Rotational Digital Com-
puter (CORDIC) IP cores are generated to realize the function Fig. 9. Frequency response curve of the filter from design and measurement.
for (3) and (4), respectively.
still good in the entire frequency band, and the attenuation of
the effective signal is below 5%.
G. Low-Pass Filter
LPFs have been used in this system to help removing high- III. R ESULTS
frequency noises or work as the antialiasing filters. As is
A. System Description
known, the principle of MFEIT measurement is based on
the dependence of electrical properties of biological tissues The main board is designed to implement the main function
on excitation frequency. The main information needed to be of the system according to principle of module-based structure,
collected comes from the modulation function of biological as shown in Fig. 10(a). Fig. 10(b) illustrates the total hardware
objects toward excitation signal. Ideally, the measured voltages platform of the system, which comprises the ac/dc power
should only be affected by the detected object. In this work, supply, the PCBs that contain four channels boards and one
the LPFs were carefully designed based on the Butterworth main board as well as one baseplate, the detection sensor, and
filter theory [31] to achieve the flattest frequency response in a PC. The ac/dc power supply is able to provide multiple
the passband. isolated power tracks. The main board of the system is
The performance of filter was verified by the oscilloscope connected to the PC through an isolated USB 2.0 interface.
as shown in Fig. 8. Fig. 8(a) was a part of circuit in the signal The channels boards are connected to the sensor through
acquisition and processing module. The observation points electric cables based on SMB interface. Moreover, the system
were selected at point m before the filter and point “n” after the can be extended to 32 channels without any revision of the
filter. The system uses the reference signal measurement mode, main board if required.
the excitation current amplitude is 3 mA, and the frequency
is 100 kHz. After the 100-kHz acquisition signal passed B. SNR
through the filter, a large amount of high-frequency noise The SNR of the proposed MFEIT system is tested at five
was filtered out. The signal purity was improved, as shown frequencies, from 50 kHz to 1 MHz. The calculation formula
in Fig. 8(b). of the SNR can be calculated as
The ideal and measured frequency response curves of the  
σ
designed Butterworth LPFs are shown in Fig. 9. In the SNR = −20 log (5)
operational band of the system (below 1 MHz), the amplitude μ
gain of the measured filter keeps over 0.99 below 550 kHz, where μ = (1/n) ni=1 x i is the average value of the mea-
and reaches around 0.95 at 1 MHz. The frequency response surement data while σ = ((1/n) ni=1 (x i − μ)2 )1/2 represents
curve of the measured filter is deviated downward from the the standard deviation. n is the number of the collected
ideal curve at 500 kHz or more. However, the performance is measurements.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on April 15,2024 at 19:25:50 UTC from IEEE Xplore. Restrictions apply.
 4500710 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021

Fig. 10. Developed main board and MFEIT system.

Fig. 12. SNR of the reference measurement. (a) Ampltitude SNR.

(b) Phase SNR.

Fig. 13. EIT boundary voltage measurement data (single set).

Fig. 11. SNR of the sensor measurement. (a) Ampltitude SNR.
(b) Phase SNR. Moreover, to remove the effect caused by the detected
sensor, the SNR results when the system is working at
Specific test conditions are as follows: excitation current reference measurement pattern are also calculated, as shown
is 3 mA; the test container is a plexiglass container with in Fig. 12. The highest SNR of amplitude measurement is
the inner diameter of 125 mm, surrounded by 16 stainless 69.91 dB at 50 kHz and the average SNR is 62.03 dB; the
steel electrodes equidistantly arranged; reference resistance is highest SNR of phase measurement is 64.12 dB at 50 kHz
a 100- fixed-value resistance with an error accuracy of 1%; and the average SNR is 55.34 dB. Compared with the sensor
the test solution is 0.9% NaCl solution; and the number of measurement results, there are some improvements to certain
data frames n is 1024. The test container is filled with the extent, especially in the amplitude results.
test solution as the reference field, and the boundary voltage
is measured at the same time. C. Channel Consistency and Measurement Repeatability
Fig. 11 shows the SNR of 208 measurements when adjacent- MFEIT system with 16 electrodes evenly distributed is
excitation adjacent-measurement scheme is utilized under a multichannel impedance measurement system, so channel
50 kHz, 100 kHz, 250 kHz, 500 kHz, and 1 MHz, respectively. consistency is an important indicator of system performance,
Both the amplitude and phase results are given. The SNR which reflects the consistency of each channel response
curves have a similar trend with the measured voltage data, under the same input. Factors that affect channel consis-
that is a higher SNR occurs at a higher voltage measurement. tency may include electrode geometric parameter and position,
The highest SNR of amplitude measurement is 71.55 dB at impedance of each channel of the analog switch circuit, etc.
50 kHz and the average SNR is 55.71 dB; the highest SNR The test container is filled with 0.9% NaCl solution and the
of phase measurement is 73.29 dB at 50 kHz and the average excitation current was 3 mA with a frequency of 100 kHz. The
SNR is 54.38 dB. In general, the SNR performance at low adjacent-excitation adjacent-measurement scheme is used, and
frequencies is better than that at high frequencies. Overall 208 voltage measurement data are obtained.
stable measurements can be acquired for image reconstruction The results in Fig. 13 show that the measured voltage
in the frequency band of interest in MFEIT. in a single channel exhibits a good “U-shaped” distribution

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on April 15,2024 at 19:25:50 UTC from IEEE Xplore. Restrictions apply.
 XU et al.: FPGA-BASED MFEIT SYSTEM WITH REFERENCE SIGNAL MEASUREMENT 4500710

Fig. 16. Load dependence analysis of the MFEIT system.

Fig. 14. EIT boundary voltage measurement data (three sets).

Fig. 17. Experiment models for imaging.

Fig. 15. Flatness analysis of the MFEIT system.

characteristic with good symmetry. In addition, the relative

changes in the corresponding measurement points of 16
“U-shaped” curves are small, which show good channel
consistency. The system can achieve the ideal “U-shaped”
curve, which is largely due to the adjacent-excitation
adjacent-measurement approach. All channels share the same
VCCS circuit and voltage measurement circuit. This approach
reduces hardware cost and makes channel consistency easier
to be guaranteed.
The verification of measurement repeatability uses the same
parameters as those of the channel consistency is tested
multiple times in a short time. As shown in Fig. 14, the three
sets of “U-shaped” curves are randomly selected and placed
in the same coordinate system, and the deviation analysis
is performed for all test data. The results of measurement Fig. 18. Time-difference imaging results of glass phantoms. (a) Amplitude.
repeatability reflect the repeatability and stability of the sys- (b) Phase.
tem. As can be seen from Fig. 14, the three “U-shaped” curves
due to the interaction between the pole setting of filters and
almost coincide with each other. Ten sets of experimental data
the high contact impedance [32]. Thus, the flatness of the
were randomly selected for error analysis, and the coefficient
broadband response is a considerable indicator for determining
of variation is between 0.01% and 5.97% at the same data
the stability of the system at different frequencies; in other
point. This error has little effect on the imaging accuracy. The
words, the system is effective and reliable in a frequency range
conclusion that the system has good accuracy and stability can
where the response curve is flat.
be obtained through the above results.
In the experiment of Fig. 15, nine precision-valued resistors
which have the resistance value from 11.4 to 1469.5  are
D. System Flatness and Load Dependence used as loads, and response voltages are obtained at different
In general, it is desirable that the response curve of sys- frequencies. As shown in the figure, the curve is almost
tem frequency maintains as flat as possible over the entire horizontal over the frequency range of 40 Hz–1 MHz when the
broadband, especially when measuring purely resistive loads. resistance of loads is 159.1  or less. As the load increases,
However, the flatness of the response is usually corrupted the flatness of the curve gradually deteriorates. When the

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on April 15,2024 at 19:25:50 UTC from IEEE Xplore. Restrictions apply.
 4500710 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021

Fig. 19. Impedance spectroscopy. (a) Carrot. (b) Cucumber.

resistance reaches 1469.5 , the frequency response can be two frequencies

stable over the range of 500 Hz to 40 k Hz. Therefore,
b = b f 1 − b f 0 . (8)
the frequency band in which the system operates effectively
narrows at high loads. On the other hand, the system still has To evaluate the performance of the MFEIT system, both the
good performance within a certain frequency band when the time-difference and the frequency-difference imaging experi-
load is high. ments were carried out.
Load dependence is also known as “transimpedance error,” The test frequencies vary from 50 kHz to 1 MHz. The
which is one of the important sources of error in MFEIT sys- excitation current is set as 3 mA. The measurements were
tem, especially due to the interaction between load and stray performed on a plexiglass test container with an inner diam-
capacitance at higher frequencies of MFEIT system [32], [33]. eter of 125 mm, surrounded by 16 stainless steel electrodes
It is obvious that the voltage measurement increases along equidistantly arranged. The total variation (TV) regularization
with resistance at different frequencies in Fig. 16. The label algorithm [37] is used to reconstruct the images based on both
on the right marks all the tested frequencies, and ideally the amplitude and phase measurement data, where the regu-
at different frequencies these lines should overlap. However, larization parameter is decided based on line search method.
as the frequency increases, the response deviates from the orig- Besides, all the imaging results will be normalized into the
inal trend due to load dependence. The voltage measurement interval 0 to 1.
begins to drop for the same load as the frequency increases to The models and corresponding geometric distribution of
100 kHz or higher. For the system, such a performance at low the experiments are shown in Fig. 17. Two glass rods with
frequencies (kHz) is not bad, but more work needs to be done diameters of 30 mm were chosen as the nonliving object
at higher frequencies (MHz). The most effective way to reduce while a carrot cylinder and a cucumber cylinder with diameters
this error is the combination of high output impedance [34] of 40 and 38 mm, respectively, were chosen as the biological
and low stray capacitance [35]. objects. About 0.9% NaCl solution is selected as the back-
ground solution.
E. Imaging Results and Analyses Fig. 18 illustrates the time-difference image reconstruction
results of glass phantoms at 50 kHz, 100 kHz, 250 kHz,
The mathematical model of EIT imaging can be linearized
500 kHz, and 1 MHz. The reference measurement bt0 was
approximately as
collected when the sensor is full of NaCl solution and the
b = Ax (6) measurement bt1 was obtained when the detected glasses were
added. In terms of the results based on amplitude measure-
where b is the discrete boundary measurements difference, ment, the results at low frequencies are better than that in
x is the unknown difference of electric parameters of all the high frequencies due to the performance degradation of the
pixels in the reconstructed field, and A represents the Jacobian system. As for the phase results, as the glass almost has
matrix which is usually calculated based on the Geselowitz’s no modulating effect on the phase of the excitation signal,
sensitivity theorem [36]. reasonable results cannot be obtained. Moreover, since the
From (6), it is known that the implementation of EIT conductivity and permittivity of the glass keep constant in the
imaging needs difference voltage measurements. In mono- frequency band under 1 MHz, the frequency-difference results
frequency EIT, the time-difference strategy is usually utilized, from both the amplitude and phase measurements were similar
where a measurement at time t0 is collected as the reference to that of Fig. 18(b) and not given anymore.
data and the measurement variety at another time t1 with Fig. 19 illustrates the electric impedance spectroscopy of
respect to the reference data is used the tested biological materials, carrot and cucumber, obtained
b = bt1 − bt0 . (7) by an impedance analyzer. The impedance spectrum trend of
carrot and cucumber is basically consistent, the impedance
While in MFEIT, the frequency-difference strategy is modulus decreases monotonously, and impedance phase angle
often utilized which makes use of measurement variety at first decreases and then increases. In the measured frequency

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on April 15,2024 at 19:25:50 UTC from IEEE Xplore. Restrictions apply.
 XU et al.: FPGA-BASED MFEIT SYSTEM WITH REFERENCE SIGNAL MEASUREMENT 4500710

Fig. 21 shows the frequency-difference imaging results,

in which 50 kHz was set as the reference frequency and
the difference between the measurements of all tested high
frequencies and that of the reference frequency was utilized
for image reconstruction. The frequency-difference imaging
results in Fig. 21 become better with the increase of frequency
despite the performance degradation in high frequencies. The
reason is that as the frequency increases, the difference of
impedance modulus and phase angle between the tested high
frequencies and the reference frequency becomes larger. So,
for the frequency-difference imaging method, better imaging
results with more evident inclusion images can be obtained.
IV. C ONCLUSION
This article presents a novel MFEIT system, which has
been thoroughly described in terms of hardware and system
verification. New features of this MFEIT system include the
design of multifrequency current source with adjustable output
frequency and amplitude as well as high output impedance,
the construction of reference signal measurement circuit for
the accurate calibration of phase measurement data, as well as
the filter design based on Butterworth theory. The FPGA-based
system programming is also completed with modular design.
Fig. 20. Time-difference imaging results of biological materials phantoms. The soft core of embedded microcontroller is used as the serial
(a) Amplitude. (b) Phase. controller of system to realize the complete excitation and data
acquisition functions of system.
Extensive experiments have been carried out to evaluate
the performance of both specific modules and the whole
system. The maximum amplitude SNR and phase SNR of the
boundary measurements are 71.55 and 64.12 dB, respectively,
and the SNR performance at low frequencies is significantly
better than that at high frequencies. In addition, the
experimental results of channel consistency and measurement
repeatability demonstrate the good performance and accuracy
of the MFEIT system. Response flatness and load dependence
analysis show the load capacity and stability of the system
over broadband. Finally, image reconstruction results toward
biological and nonliving materials are analyzed and deduced,
and conclusions are in accordance with the impedance
spectroscopy, which demonstrate the potential of the system
in application of real biomedical detection and biological
process analysis. Further work will concentrate on the further
performance boost and the human detection applications
Fig. 21. Frequency-difference imaging results of biological materials based on the developed MFEIT system.
phantoms. (a) Amplitude. (b) Phase.
R EFERENCES
band from 50 kHz to 1 MHz, both the impedance modulus [1] R. H. Bayford, “Bioimpedance tomography (electrical impedance
and phase angle of carrot and cucumber increase or decline tomography),” Annu. Rev. Biomed. Eng., vol. 8, no. 1, pp. 63–91,
Aug. 2006.
monotonously. [2] L. Borcea, “Electrical impedance tomography,” Inverse problems,
Fig. 20 shows the time-difference image reconstruc- vol. 18, no. 6, pp. 99–136, 2002.
tion results of carrot and cucumber phantoms at 50 kHz, [3] M. Cheney, D. Isaacson, and J. C. Newell, “Electrical impedance
tomography,” SIAM review, vol. 41, no. 1, pp. 85–101, 1999.
100 kHz, 250 kHz, 500 kHz, and 1 MHz, respectively. The [4] H. C. N. Jongschaap, R. Wytch, J. M. S. Hutchison, and V. Kulkarni,
time-difference imaging results in Fig. 20 illustrate that both “Electrical impedance tomography: A review of current literature,” Eur.
the amplitude and phase imaging performance degrade with J. Radiol., vol. 18, no. 3, pp. 165–174, Aug. 1994.
[5] L. Miao, Y. Ma, and J. Wang, “ROI-based image reconstruction of
the increased frequency. The results come from the perfor- electrical impedance tomography used to detect regional conductivity
mance degradation of the MFEIT system in high frequencies variation,” IEEE Trans. Instrum. Meas., vol. 63, no. 12, pp. 2903–2910,
as well as the fact that the variety of impedance modulus and Dec. 2014.
[6] J. Chen, L. Xu, Z. Cao, and H. Zhou, “Four-terminal imaging using
phase angle between the detected biological objects and the a two-terminal electrical impedance tomography system,” IEEE Trans.
background saline water tends to diminish. Instrum. Meas., vol. 63, no. 2, pp. 432–440, Feb. 2014.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on April 15,2024 at 19:25:50 UTC from IEEE Xplore. Restrictions apply.
 4500710 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 70, 2021

[7] X. Yue and C. McLeod, “FPGA design and implementation for EIT [29] M. Rafiei-Naeini and H. McCann, “Low-noise current excitation sub-
data acquisition,” Physiol. Meas., vol. 29, no. 10, pp. 1233–1246, system for medical EIT,” Physiol. Meas., vol. 29, no. 6, pp. S173–S184,
Sep. 2008. Jun. 2008.
[8] R. J. Yerworth, R. H. Bayford, G. Cusick, M. Conway, and D. S. Holder, [30] F. Seoane, R. Macías, R. Bragós, and K. Lindecrantz, “Simple voltage-
“Design and performance of the UCLH mark 1b 64 channel electrical controlled current source for wideband electrical bioimpedance spec-
impedance tomography (EIT) system, optimized for imaging brain troscopy: Circuit dependences and limitations,” Meas. Sci. Technol.,
function,” Physiol. Meas., vol. 23, no. 1, pp. 149–158, Jan. 2002. vol. 22, no. 11, Sep. 2011, Art. no. 115801.
[9] T. I. Oh, E. J. Woo, and D. Holder, “Multi-frequency EIT system with [31] I. W. Selesnick and C. S. Burrus, “Generalized digital butterworth filter
radially symmetric architecture: KHU mark1,” Physiol. Meas., vol. 28, design,” IEEE Trans. Signal Process., vol. 46, no. 6, pp. 1688–1694,
no. 7, pp. S183–S196, Jun. 2007. Jun. 1998.
[10] T. I. Oh, H. Wi, D. Y. Kim, P. J. Yoo, and E. J. Woo, “A fully parallel [32] A. McEwan, G. Cusick, and D. S. Holder, “A review of errors in multi-
multi-frequency EIT system with flexible electrode configuration: KHU frequency EIT instrumentation,” Physiological Meas., vol. 28, no. 7,
mark2,” Physiol. Meas., vol. 32, no. 7, pp. 835–849, Jun. 2011. pp. S197–S215, Jun. 2007.
[11] A. Zarafshani, N. Huber, and N. Béqo, “A flexible low-cost, high- [33] J. Schlappa, E. Annese, and H. Griffiths, “Systematic errors in multi-
precision, single interface electrical impedance tomography system for frequency EIT,” Physiol. Meas., vol. 21, no. 1, pp. 111–118, Feb. 2000.
breast cancer detection using FPGA,” in Proc. 14th Int. Conf. Electr. [34] R. Halter, A. Hartov, and K. D. Paulsen, “Design and implementation
Bioimpedance 11th Conf. Biomed. Appl. Electr. Impedance Tomogr., of a high frequency electrical impedance tomography system,” Physiol.
vol. 224, no. 1, pp. 169–172, 2010. Meas., vol. 25, p. 379, Feb. 2004.
[12] Q. Liu, T. I. Oh, H. Wi, E. J. Lee, J. K. Seo, and E. J. Woo, “Design [35] T. I. Oh, K. H. Lee, S. M. Kim, H. Koo, E. J. Woo, and D. Holder,
of a microscopic electrical impedance tomography system using two “Calibration methods for a multi-channel multi-frequency EIT system,”
current injections,” Physiol. Meas., vol. 32, no. 9, pp. 1505–1516, Physiol. Meas., vol. 28, no. 10, pp. 1175–1188, 2007.
Aug. 2011. [36] D. B. Geselowitz, “An application of electrocardiographic lead theory to
[13] H. Wi, H. Sohal, A. L. McEwan, E. Je Woo, and T. In Oh, “Multi- impedance plethysmography,” IEEE Trans. Biomed. Eng., vol. BME-18,
frequency electrical impedance tomography system with automatic self- no. 1, pp. 38–41, Jan. 1971.
calibration for long-term monitoring,” IEEE Trans. Biomed. Circuits [37] L. I. Rudin, S. Osher, and E. Fatemi, “Nonlinear total variation based
Syst., vol. 8, no. 1, pp. 119–128, Feb. 2014. noise removal algorithms,” Phys. D, Nonlinear Phenomena, vol. 60,
[14] S. Sun, L. Xu, Z. Cao, H. Zhou, and W. Yang, “A high-speed electrical nos. 1–4, pp. 259–268, Nov. 1992.
impedance measurement circuit based on information-filtering demodu- [38] R. Kusche, A. Malhotra, and M. Ryschka, “A FPGA-based broadband
lation,” Meas. Sci. Technol., vol. 25, no. 7, May 2014, Art. no. 075010. EIT system for complex bioimpedance measurements-design and per-
[15] M. Guermandi, R. Cardu, E. Franchi Scarselli, and R. Guerrieri, “Active formance estimation,” Electronics, vol. 4, no. 3, pp. 507–525, Jul. 2015.
electrode IC for EEG and electrical impedance tomography with contin-
uous monitoring of contact impedance,” IEEE Trans. Biomed. Circuits Yanbin Xu (Member, IEEE) received the B.S., M.S.,
Syst., vol. 9, no. 1, pp. 21–33, Feb. 2015. and Ph.D. degrees in control science and engineering
[16] S. A. Santos, A. Robens, and A. Boehm, “System description and first from Tianjin University, Tianjin, China, in 2001,
application of an FPGA-based simultaneous multi-frequency electri- 2004, and 2009, respectively.
cal impedance tomography,” Sensors, vol. 16, no. 8, pp. 1158–1178, She is currently an Associate Professor with the
Jul. 2016. School of Electrical and Information Engineering,
[17] Y. Yang and J. Jia, “A multi-frequency electrical impedance tomography Tianjin University. Her research interests include
system for real-time 2D and 3D imaging,” Rev. Sci. Instrum., vol. 88, process tomography, biomedical imaging, inverse
no. 8, Aug. 2017, Art. no. 085110. problem, and hybrid imaging.
[18] J. Avery, T. Dowrick, M. Faulkner, N. Goren, and D. Holder, “A versa-
tile and reproducible multi-frequency electrical impedance tomography
system,” Sensors, vol. 17, no. 2, p. 280, Jan. 2017. Zhicheng Yan received the B.S. degree from the
[19] G. Singh, S. Anand, B. Lall, A. Srivastava, and V. Singh, “A low- School of Control Science and Engineering, Hebei
cost portable wireless multi-frequency electrical impedance tomography University of Technology, Tianjin, China, 2018.
system,” Arabian J. Sci. Eng., vol. 44, no. 3, pp. 2305–2320, Jul. 2018. He is currently pursuing the M.S. degree with the
[20] Y. Wu, D. Jiang, A. Bardill, R. Bayford, and A. Demosthenous, School of Electrical and Information Engineering,
“A 122 fps, 1 MHz bandwidth multi-frequency wearable EIT belt Tianjin University, Tianjin.
featuring novel active electrode architecture for neonatal thorax vital His current research interests include electrical
sign monitoring,” IEEE Trans. Biomed. Circuits Syst., vol. 13, no. 5, tomography, acoustoelectric tomography, and other
pp. 927–937, Oct. 2019. related hybrid imaging technique.
[21] S. Grimnes and O. G. Martinsen, Bioimpedance and Bioelectricity
Basics, 2nd ed. Amsterdam, The Netherlands: Elsevier, 2008.
[22] D. Djajaputra, “Electrical impedance tomography: Methods, history and Bing Han received the B.S. and M.S. degrees from
applications,” Med. Phys., vol. 32, no. 8, pp. 2731–2731, Aug. 2005. the School of Electrical and Information Engineer-
[23] Z. Yan, Y. Xu, B. Han, and F. Dong, “An FPGA-based multi-frequency ing, Tianjin University, Tianjin, China, in 2016 and
EIT system with reference signal measurement,” in Proc. IEEE Int. 2019, respectively.
Instrum. Meas. Technol. Conf. (I2MTC), May 2020, pp. 1–6. His research interests include electrical tomog-
[24] B. Han, Y. Xu, and F. Dong, “Design of current source for multi- raphy, electrical impedance tomography, and other
frequency simultaneous electrical impedance tomography,” Rev. Sci. related hybrid imaging technique.
Instrum., vol. 88, no. 9, Sep. 2017, Art. no. 094709.
[25] Y. Yang, F. Zhang, K. Tao, L. Wang, H. Wen, and Z. Teng, “Multi-
frequency simultaneous measurement of bioimpedance spectroscopy
based on a low crest factor multisine excitation,” Physiol. Meas., vol. 36,
no. 3, pp. 489–501, Feb. 2015. Feng Dong (Senior Member, IEEE) received the
[26] Y. Mohamadou et al., “Performance evaluation of wideband bio- B.S., M.S., and Ph.D. degrees in control science and
impedance spectroscopy using constant voltage source and constant engineering from Tianjin University, Tianjin, China,
current source,” Meas. Sci. Technol., vol. 23, no. 10, Sep. 2012, in 1988, 1996, and 2002, respectively.
Art. no. 105703. He is currently a Professor with the School of
[27] A. S. Ross, G. J. Saulnier, J. C. Newell, and D. Isaacson, “Current source Electrical and Information Engineering, Tianjin Uni-
design for electrical impedance tomography,” Physiol. Meas., vol. 24, versity, and the Director of the Tianjin Key Lab-
no. 2, pp. 509–516, Apr. 2003. oratory of Process Measurement and Control. His
[28] A. S. Tucker, R. M. Fox, and R. J. Sadleir, “Biocompatible, high research interests include process parameter detec-
precision, wideband, improved howland current source with lead-lag tion and control system, industrial process tomog-
compensation,” IEEE Trans. Biomed. Circuits Syst., vol. 7, no. 1, raphy, multiphase flow measurement, and intelligent
pp. 63–70, Feb. 2013. information processing.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on April 15,2024 at 19:25:50 UTC from IEEE Xplore. Restrictions apply.