You are on page 1of 32


D4.1 Hardware platform for bio-electrical neuroimaging

and stimulation.


TYPE OF DOCUMENT Deliverable D4.1.
DATE 14/10/2015
ABSTRACT This deliverable describes the hardware platforms for
EEG, EIT measurements, data processing and
stimulation. Two platforms are presented: (1) one
platform based on active electrodes capable of
measuring EEG, EIT providing electrical stimulation, (2)
a low-cost, quick-setup platform for EEG only.
AUTHOR, COMPANY M. Guermandi, M. Chiesi, E. Franchi Scarselli, R.
Guerrieri (UNIBO), C. Hintermueller (GTEC)
VALIDATOR, COMPANY C. Windischberger (MUW)
NATURE P: Prototype
CONFIDENTIALITY LEVEL CO: Confidential thus for consortium members only

Release Date Reason of change Status Distribution

R0.1 04/09/2015 Contribution from Sent to GTEC for Mailing list
UNIBO review and
R0.2 11/09/2015 Contribution from Sent to UNIBO for Mailing list
GTEC review
R0.3 14/09/2015 Review by UNIBO Sent to GTEC for Mailing list
review and
R0.4 14/09/2015 Internal Review Finalized and sent Mailing list
to consortium for
internal review
R1.0 14.10.2015 Revised version Incorporated Restricted area on
remarks from web-page
internal Review
R1.0 14.10.2015 Final version Formal review Sent to EU
Table of Contents

1. INTRODUCTION ................................................................................................................................. 1
2. EXECUTIVE SUMMARY (ALWAYS PUBLIC)............................................................................. 3
STIMULATION ............................................................................................................................................ 3
3.1.1 Active Electrodes for EEG/EIT and electrical stimulation. .........................................4
3.1.2 Backend hardware .......................................................................................................................8
3.1.3 Control hardware for programming the electrode behaviour..................................9
3.1.4 Electrode programming software ...................................................................................... 13
3.1.5 Acquisition software................................................................................................................. 15
3.1.6 Measurements ............................................................................................................................. 17
3.2. COST-EFFECTIVE, QUICK-SETUP EEG ACQUISITION SYSTEM ..................................................... 21
3.2.1 Active Electrodes........................................................................................................................ 23
3.2.2 Backend hardware .................................................................................................................... 23
3.2.3 Acquisition software................................................................................................................. 24
3.2.4 Measurements ............................................................................................................................. 25
4. CONCLUDING SECTION ................................................................................................................. 28
4.1.1 Conclusion ..................................................................................................................................... 28
4.2. BIBLIOGRAPHY ............................................................................................................................... 28
4.3. GLOSSARY ........................................................................................................................................ 29
ICT 612022 - CREAM 14/10/2015

1. Introduction
Neuroimaging is a discipline which deals with the ability to image the structure and/or
functionality of the central nervous system (CNS). Among the techniques currently used
to provide functional maps of CNS activity in clinical practice or at research-level, we may
mention Positron Emission Tomography (PET), Functional Magnetic Resonance Imaging
(fMRI) [1], Electro- and Magneto-Encephalography (EEG and MEG) [2] and Diffuse Optical
Tomography (DOT, also known as NIRS). The first two techniques provide the highest
level of spatial accuracy at the price of rather low temporal resolution (in the order of
seconds). Moreover, the expensive and voluminous hardware limits use of them to
hospital settings, and patient discomfort hampers several possible applications, for
example in sleep research or epileptic focus localization which require long-term
monitoring. By contrast, the other techniques trade spatial accuracy for compactness,
movability and ease of use. In particular, the localization of EEG signal sources is a fairly
well established method of providing functional maps of brain activity and one which
guarantees good levels of portability and comfort for the patient [2]. The EEG source
localization method suffers from a lower spatial resolution than fMRI and PET; however,
images can be acquired at a much higher sampling rate, allowing analysis of phenomena
occurring at time intervals of fractions of seconds. Systems based on EEG source
localization can therefore be ideal for settings such as doctors surgeries, the home or
ambulances, provided an improvement in spatial accuracy can be achieved. Since EEG
source localization relies on the ability to reconstruct current dipole locations inside the
patients head by measuring voltages generated on its surface, a high-density system is
required so as to optimize localization accuracy. EEG source localization algorithms rely
heavily on the availability of an accurate, patient-specific head model to describe the
electrical behaviour of the domain. To this end, if one could obtain structural information,
e.g. from a MRI scan, reconstruction quality would be significantly increased. However,
this would still not provide direct information about the conductivity of the various
tissues or regions which are in practice commonly approximated by simply assigning a
single, time-invariant conductivity value to each of the few tissues types into which the
head volume is usually segmented. To overcome these limitations, it has been proposed
that local, time-dependent conductivity information be provided by Electrical Impedance
Tomography (EIT). EIT is a less established imaging technique which allows to estimate
the conductivity distribution inside a body by injecting small AC currents at frequencies
ranging from a few kHz to a few MHz on the surface and measuring the resulting voltage
on the same surface [3] [4].
Since EIT already requires the injection of low-intensity and high-frequency currents on
the surface of the head, the next step is to exploit the existing EIT system and extend it by
the capability to inject currents for transcranial stimulation.
Having a unique system for neuroimaging from EEG/EIT and stimulation would greatly
improve the coupling of the three techniques, reducing and simplifying setup times and
ultimately moving closer to applications where the system setup might be performed by
non-clinical personnel, such as those envisioned for the outcomes of the CREAM project.
The recording of EEG as well as the results of its processing (for example source
localization and functional connectivity analysis) can greatly improve the understanding

Level of confidentiality: CO 1
ICT 612022 - CREAM 14/10/2015

of the mechanism and effects of stimulation or even provide tools which allow to establish
a neurofeedback loop between EEG recording and neuro-stimulation.
Additionally, reconstruction of local conductivities from EIT can also be used to model the
physical behaviour of stimulation currents and improve the capability of targeting certain
regions [5].
In this deliverable we report on the characteristics and performances of the developed
platform, based on active electrodes capable of measuring EEG, EIT and provide electrical
stimulation, in particular concerning the neuroimaging capabilities, since the stimulation
sub-block will be completed in Task 4.4 and reported in the relevant deliverable (D4.4).
Until then it will be integrated with the visual and acoustic stimulation libraries developed
in Task 4.5 and reported in detail in the relevant Deliverable (D4.5)

One relevant objective of the CREAM project is to develop hardware which can be used
outside clinical environments bearing reduced costs and complexity compared to high-
end EEG recording systems.
For these reasons, we have built a prototype of a modular, very fast to setup and cost-
effective platform for EEG acquisition providing recording quality comparable to those
obtained by high-end devices such as the g.HIamp from g.tec medical engineering
described in section 3.1.2.
Electroencephalogram (EEG) recording offers the capability to collect information on the
activity of the brain with some significant advantages over other techniques (such as fMRI,
PET, MEG, ECoG), which can be summarized as being the least invasive, the easiest to
setup and the one with the lowest cost. In the past years new applications for EEG
recording systems have emerged. These include Brain-Computer Interfaces (BCI) and
consumer-oriented applications ranging from home care to neurofeedback and gaming
controllers. There is, therefore, a demand for high-quality acquisition systems whose
montage is fast and does not require trained personnel. Systems based on dry electrodes
are particularly attractive since they make for very simple, almost instantaneous setup.
Clearly, also applications which can be envisioned for instrumentation, software and
protocols developed in the CREAM project would largely benefit from this kind of systems.
The presented system is foreseen to be used in Task 4.3 to validate software libraries for
neuroimaging. Also, it can be directly used in the activities of WP5 which do not involve
stimulation, such as Task 5.1. Moreover, this second prototype could be integrated in the
comprehensive neuroimaging/stimulation platform in order to reduce the costs of the
overall system. Even if this is outside the scope of the CREAM project, where the use of
g.HIamp from g.tec is foreseen in order to speed up the prototype development, the
system can also be adapted after the end of the project to be complemented with
visual/auditory stimuli production subsystem and through ad hoc interfacing board, it
can be used in conjunction with active electrodes for EEG/EIT and electrical stimulation.
In this deliverable we also report on the characteristics and performances of this second

Level of confidentiality: CO 2
ICT 612022 - CREAM 14/10/2015

2. Executive summary (always public)

Neuroimaging is a discipline which deals with the ability to image the structure and/or
functionality of the central nervous system (CNS). Several different techniques are used
to provide functional maps of CNS activity in clinical practice and at research level,
amongst others Positron Emission Tomography (PET), Functional Magnetic Resonance
Imaging. In addition transcranial and sub cranial electrical stimulation methods are used
to investigate, simulate, alter and enhance brain functions
In this deliverable a novel hardware platform for simultaneously performing
ElectroEncephaloGraphy (EEG) and Electrical Impedance Tomography (EIT)
measurements at high temporal resolution while delivering electrical stimuli to the brain.
The advantage of the presented platform is that it allows to use the same electrodes for
measuring EEG, EIT signals and for delivering a broad range of stimulation pulses to
nearly any brain region. This seamless integration of EEG and EIT recording and electrical
stimulation allows to closely observe the immediate and long-term effects of the
stimulation and enables the development of new ways to study the function of the brain,
as it is not necessary to alter electrode positions or to manually interchange EEG/EIT
electrodes with stimulation electrodes.
The electrodes are programmed and controlled through a dedicated controller unit. It
switches the electrode behaviour between stimulation and acquisition mode, controls the
currents injected for EIT and electrical stimulation, and modulates the EEG and EIT
signals onto a single carrier signal which is amplified, digitized, and transmitted to the
host computer for further processing by the biosignal amplifier backend.
On the backend a Matlab/Simulink based rapid prototyping system demodulates and
separates the EEG and EIT signals and allows to control the generation and delivery of a
broad range of electrical stimuli.

3. Hardware platforms for bio-electrical neuroimaging and


In the following two platforms are presented, one based on active electrodes capable of
measuring EEG, EIT and providing electrical stimulation and a low-cost, quick-setup
platform for EEG only.

3.1. Hardware platform for bio-electrical neuroimaging and stimulation.

The developed platform has the following functionalities (taken from D2.3 System-level
design of the neuroimaging/stimulation platform):
Extraction of ElectroEncephaloGraphy (EEG) and Electrical Impedance
Tomography (EIT) signals.
Production of visual and auditory stimuli.
Generation of currents for transcranial electrical stimulation (tACS: transcranial
Alternate Current Stimulation and tDCS: transcranial Direct Current Stimulation).
Functional brain imaging from the acquired bioelectrical signals.

Level of confidentiality: CO 3
ICT 612022 - CREAM 14/10/2015

The final objective is to provide a platform which is able to seamlessly couple

neuroimaging to different kinds of stimulation. This means, as an example, that the target
platform should be able to extract EEG data and perform electrical stimulation without
the need for switching electrode montage and perform complex operations.

The system is divided into the following blocks:

Reconfigurable, easy to apply cap for placing measurement and stimulation
Active electrodes able to measure EEG and EIT voltages (64 EEG channels, 0.5-100
Hz bandpass) compatible with ElectroOculoGraphy (EOG) and
Electrocardiography (ECG)
Electrodes for injecting EIT currents and stimulation currents.
Control hardware for programming the electrode behaviour (e.g. inject currents
and characteristics of such currents).
Back-end hardware for analogue-to-digital conversion of active electrode signals
and data transfer to computer.
Hardware for visual/acoustic stimuli production.
PC/Computational Platform for post-processing of data and control of the system.

Figure 1 depicts the sub-blocks and their interconnections.


Measuring (EEG amplifier)

active PC /
electrodes Computational
Stimulating control and
electrodes power supply

Figure 1: Schematic representation of the platform, from D2.3.

3.1.1 Active Electrodes for EEG/EIT and electrical stimulation.

The active electrodes are based upon a previous design by University of Bologna [6]. The
current design improves the electrode performance for EEG and EIT measurements and
adds electrical stimulation capabilities.
One of the main issues arising when designing an active electrode which is capable of both
acquiring EEG/EIT signals and performing current stimulation is related to the high

Level of confidentiality: CO 4
ICT 612022 - CREAM 14/10/2015

contact impedance which arises at DC (for tDCS) or the low frequencies typical for tACS.
Since the contact impedance can be in the order of hundreds of k, even a current of 0.5
mA can lead to voltage drops on the contact impedance of several Volts. In order to
accommodate for this, a straightforward approach would require to design ICs able to
withstand these high-voltages. This solution has been discarded since it would lead to an
increase in fabrication costs and design complexity.
The approach which has been adopted is to modify the architecture at system level. In
particular, stimulating electrodes are powered by separate power supplies, whose ground
level can be adjusted in order to follow oscillations of the electrodes voltage, keeping all
the pins of the IC well within the power supply even when injecting currents. Further
details on this approach are presented in section 3.1.3.

Specifications for EEG signal quality are taken from Deliverable D2.3 and are basically
derived from the International Federation of Clinical Neurophysiology (IFCN)s standards
[7] which are commonly adopted in clinical practice. Preamplifier input impedances
should be higher than 100 M, the Common Mode Rejection Ratio (CMRR) of the system
must be at least 110 dB and additional noise introduced in the recording should be less
than 0.5 V root-mean-square at any frequency from 0.5 to 100 Hz including 5060 Hz.

Unlike EEG, EIT is currently only used in research settings and not included in standard
clinical practice. Also, no widely accepted clinical standard exists to define the minimum
requirements in terms of signal quality, noise value etc. Concerning maximum noise value,
our choice was to adopt the same level as used for EEG signals. This level is comparable
to noise introduced at the interface of the electrode in case no skin preparation is
performed. The maximum instantaneous value of EIT injected current is limited to 512
A, which is the same value as for transcranial stimulation. The Common mode rejection
of an EIT system mainly prevents the measurement from being affected by the common
mode signal introduced by the voltage drop which may be caused by unmatched EIT
currents flowing to ground. In general, values in the order of 60 to 70 dB are commonly
found in systems reported in the literature and are therefore used as the specification.
Common injection frequencies for EIT range from a few kHz to a few MHz. In this system,
the maximum injection frequency is limited to 256 kHz, since working at higher frequency
generally requires extensive and continuous calibration to compensate for different
behaviour among the full set of electrodes. Such calibration is unrealistic in a system
which is designed to be inexpensive, easy to use and set up.
Specifications are summarized in the following table.

Specification Value Comments

Number of Electrodes 64
Operating Voltage 3.3 V
Die Size <9 mm2
Number of External Components < 10
Electrode PCB size <1 cm2
Total Input Referred Noise (EEG) 0.5 Vrms 0.5 to 100 Hz System Level

Level of confidentiality: CO 5
ICT 612022 - CREAM 14/10/2015

Input Referred Noise (EIT) 50 nV/Hz

Total Input Referred Noise 10 rms
EEG Signal Bandwidth 0.5-100 Hz
CMRR 100 dB System Level
Output Dynamic Range 250 mV
Input Impedance >100 M (EEG and electrode-tissue only)
Load Drive Capability 100 pF
Number of Connection Lines to 64 + 1 GND 64 analog EEG/EIT/impedance
back-end channels + 1 GND channel;
Number of electrode configuration 2 per electrode

Concerning stimulation, due to the limited size of the interface between electrode and
skin, we chose to limit the maximum current injectable by a single active electrode to be
0.5 mA. Higher current levels can be achieved by using multiple electrodes to provide
higher current levels.
However, one (non-active) electrode is foreseen which can draw or provide the current
needed to balance that of the active electrodes. So, for example, this electrode can be
surrounded by 4 active electrodes, each injecting 0.5 mA, and the result will be a central
electrode drawing 2 mA altogether.
Current waveform is user-programmable and is stored in an onboard memory in 512
words of 8 bits which contains the instantaneous value of the current. The memory is read
in cycles during normal operation (except for ramp-up and down).
Specifications (derived from those in D2.3) are therefore the following ones.

Specification Value Comments

Number of locations 16
Max tDCS current per electrode 0.5 mA For active electrodes
Max tACS current amplitude per 0.5 mA
Stimulus frequency DC 100 Hz
Stimulus duration User-
Stimulus shape User- Direct synthesis from on-board
programmable memory

In order to support high-density system integration, the electrode PCB occupies less than
2 cm2 and is covered with epoxy resin in order to seal the electronics and allow safe
handling and cleaning and avoid damage to the electrode. The PCB has a 5 mm diameter
hole whose walls are gold-plated (see Figure 4); this is used for direct contact with the
patients skin, achieved by placing the PCB on the scalp and filling the hole with conductive
gel. The number of signals needed to connect each electrode to the back-end is minimized

Level of confidentiality: CO 6
ICT 612022 - CREAM 14/10/2015

to 4, namely the output signal (OUT), ground and two digital programming signals (CK,
DATA). The electrode is connected to the back-end by means of two 1 m two-core cables
terminating in touch-proof connectors, one for the analogue and one for the digital signals.
A dedicated power line is not foreseen and is shared with the output signals. In fact the
output stage of the IC is biased by an external programmable current source and the
voltage is fixed by the IC itself. The output line is shared among EEG, EIT signals by placing
them at different frequency bands.
It should be noted that none of the signals is transferred at baseband in order to minimize
noise and maximize the CMRR performances.

Figure 2: Schematic of the active electrode with the main components integrated in the IC

Figure 3: Frequency planning for the output signal in readout mode.

The main blocks composing the active-electrodes IC are shown in Figure 2. The input stage
of the readout section is based on two passive mixers. In normal operation, the mixers run
at fEIT - 1 kHz and 2 kHz, respectively, for down converting the EIT signal around 1 kHz
and up-converting the EEG signal (as shown in Figure 3) The two signals are then

Level of confidentiality: CO 7
ICT 612022 - CREAM 14/10/2015

subtracted by a differential amplifier based on a differential difference amplifier (DDA). If

only the EEG signal is required, the mixers are driven in counter-phase by 1 KHz square
wave. Since the EEG signal already drops below system noise level at frequencies higher
than a few hundred Hz and EIT is a slowly modulated signal related to either the head
structure or the metabolism, there is no risk of the two signals overlapping in normal

The digital portion of the IC is mainly comprised of an input interface which acquires
programming data from the back-end, some registers for storing configuration bits and a
512x10 bit Sequential Access Memory (SAM), which stores the samples for direct
synthesis of the EIT/tACS current waveform (8 bit digital word) plus digital signals M1
and M2 which control the mixers for input signal down- and up-conversion.

Electrode-Skin contact impedance (ESI) monitoring is performed at the same time as the
EEG/EIT readout and simultaneously on all electrodes. The monitoring is based on
injecting a very small AC test current between adjacent electrodes and measuring the
differential voltage at the injection frequency across the two electrodes.

Figure 4: Photograph of the IC with the main blocks highlighted and of the full active electrode

3.1.2 Backend hardware

The EEG and EIT signals are recorded using a g.HIamp based biosignal amplifier. Figure 5
shows the device used within CREAM. It offers 144 general purpose input channels and
sends the digitized data via USB to the processing unit.

Level of confidentiality: CO 8
ICT 612022 - CREAM 14/10/2015

Figure 5: g.HIGHamp

The g.HIamp based device is FDA cleared and CE approved and offers ADC converters with
24 Bit for perfect signal resolution. It offers an input range of +/- 250 mV with a resolution
of <60nV, which is required for precise EEG and EIT measurements, preventing saturation
of the amplifier input. All channels are DC coupled. Internally, signal processing is
performed with using a fast floating point DSP and a Linux based firmware. The amplifier
relies on a very high oversampling to reduce the noise as much as possible by averaging
The amplifier is integrated into the Simulink-based rapid prototyping and processing
platform through the g.tec Highspeed Processing blocks. It is powered by a medical mains
power supply or by a battery pack, and the system is connected to the computer via USB.
In addition, 16 digital trigger channels can be used to synchronize the EEG and EIT data
with external events and triggers.
Data is sampled internally at 614.4 kHz, which is much higher than the sampling
frequency required by EEG, EIT and source localization applications. The floating point
DSP then internally performs the oversampling and averages several consecutive samples
to increase the signal-to-noise ratio. For example, if the amplifier operates at 256 Hz, then
2400 samples are averaged per data point, which suppresses the noise by a factor of 49.
In addition, the floating point DSP also performs real-time bandpass filtering and notch
filtering of the data. Several hundred different bandpass filters are predefined. Also,
bipolar derivations can be calculated by the DSP to work with a very high CMRR. As a side
effect of the combination of DSP-based filtering and oversampling is a dramatic reduction
of aliasing artefacts.
The amplifier uses 144 ADC for the 144 channels, and hence all signals are sampled
exactly at the same time point to avoid any time delay between channels. This is especially
important for precise EEG, EIT and source localization measurements.
3.1.3 Control hardware for programming the electrode behaviour

Interfacing hardware for the different electrodes is gathered in groups of 64-channels on

the same printed circuit board (see Figure 6 for a schematic drawing, Figure 7 for a picture
of the board). Each board is then connected to the backend (BE) through a high density
65-poles cable carrying the 64 electrode output signals plus the reference signal. The

Level of confidentiality: CO 9
ICT 612022 - CREAM 14/10/2015

reference signal is generated on the PCB and it is a constant voltage equal to the DC level
of the electrodes outputs.

Figure 6: Simplified scheme for interfacing hardware

Figure 7: Board for interfacing 64 active electrodes to the BE.

Electrode programming is performed without having to stop the acquisition of data from
the electrode. In order to provide synchronization of the timings at which new
programming words are sent to different electrodes, local memories (RAM) are used, in
which the streams of serial bits required for re-programming each electrode are stored.

Level of confidentiality: CO 10
ICT 612022 - CREAM 14/10/2015

An update signal is then sent to the board so that the new programming words are
simultaneously uploaded to each electrode.
Depending on the electrode behaviour, the power supply current can be programmed to
different values (i.e. higher currents for injecting electrodes) through ad-hoc Digital-to-
analog converters (DACs in Figure 6).

The PCB provides isolated power supply to every circuit and the electrodes. Isolation is
achieved by means of an off-the-shelf standard 80W medical approved external power
supply, developed for laptops. Programming of the electrode behaviour is controlled from
the PC, through a USB interface. Digital signals are isolated through high-speed opto-


Figure 8: Scheme for stimulating electrode interface (B)

with respect to straightforward solution (A).

As anticipated in the previous section, we considered two possible solutions to allow

active electrodes to provide stimulation at DC or low-frequency level. At these low
frequencies, the contact impedance between electrode and skin can increase to several
tens of k, leading to high voltage drops (higher than 10 V) when current is injected.
Standard systems for tDCS and tACS do not use active electrodes and provide high power
supply voltages to the current source circuit in order to allow current injection. With an
active electrode system, this would require for the active electrode IC to withstand such
high voltages, requiring non-standard fabrication processes with increased cost (Figure 8
A). Additionally, the use of such process can imply constraint/limitations on the analog
portions of the IC, possibly leading to reduced performances, increased area and power
consumption. Finally, the need for providing high differential voltages on the cables to the
electrodes increases the size, weight and cost of the cables.

In order to avoid these limitations, we used a different approach which is based on

modifying the interfacing hardware for those electrodes which can be programmed to
provide stimulation.

Level of confidentiality: CO 11
ICT 612022 - CREAM 14/10/2015

In normal operating condition (EEG/EIT acquisition), the hardware operates as the one
described in the previous paragraph. When stimulation is required, power supply to the
electrode is provided by an additional control circuit which modifies the electrodes
ground to float according to the electrode signal. In this way, the electrode voltage will
always be fixed with respect to the ICs ground and the IC can still work with a 3V power
supply as in normal operating conditions.
Currently, test boards have been developed in order to verify functionality of the
electrodes and of the system in such an operating condition. Finalization of the design,
fabrication and integration is currently under development in Task 4.4 and will be
reported in the relevant deliverable (D4.4).


Figure 9: Driving Right Leg Circuit scheme

In order to boost the CMRR of the system from the 60-70 dB of active electrodes, we make
use of a recently introduced Driving Right Leg (DgRL) scheme (Figure 9) where the
common mode interference is detected by the reference electrode and directly drives the
isolated ground of the instrumentation. VG1 therefore follows the common mode
interference and, in this way, a single-ended stage inherently acts as differential and will
encounter very little common mode at its input, automatically rejecting most of the
interference and loosening requirements on the maximum value and accuracy of the
amplification stages gain. Moreover, since the input impedance of the amplifier will be
directed towards the reference signal, potential divider effects at input will be
significantly reduced. Thanks to this technique, the improvement in the CMRR is higher
than 70 dB, largely enough to improve the CMRR of the system up to the 100 dB required
by the specifications.

Level of confidentiality: CO 12
ICT 612022 - CREAM 14/10/2015

It should be pointed out that, with respect to more standard techniques such as Driven
Right Leg (DRL), DgRL does not introduce feedback paths which could lead to instability.

3.1.4 Electrode programming software

The electrode behaviour can be programmed through a graphical user interface (GUI).
The interface has been designed using Java which allows users to employ the software on
multiple platforms while keeping the same graphical outlay, level of visualization and
operation. The developed interface allows users to create the programming file for
configuring electrodes in a simple way. An example is reported in Figure 10. It allows to
Number of electrodes used.
Electrode behaviour (EEG, EIT, combined EEG/EIT measurement, current
sourcing (INJ_P) or sinking (INJ_N) )
Electrode operating frequency.
Amplitude of the injected current.

Figure 10:Create Configuration File window

Level of confidentiality: CO 13
ICT 612022 - CREAM 14/10/2015

Once the programming file has been created it can be uploaded on the RAM to configure
the electrodes via another window called Active Electrode Control Panel (see also Figure
11 ).

Figure 11: Active Electrode Control Panel window

Finally, the interface allows users to start the data processing by calling the desired
Matlab/Simulink method.

Level of confidentiality: CO 14
ICT 612022 - CREAM 14/10/2015

Figure 12: Data Processing window

3.1.5 Acquisition software

Data processing is performed in Simulink by extending the standard library for the
backend (g.tecs g.HIamp) in order to perform EEG, EIT and ESI signal extraction. Figure
13 shows the model for EEG and ESI signal extraction.

The main functions of the Simulink model are:

Data acquisition from backend
Carrier recovery and demodulation of the signals
Scaling and filtering

Sample rate is fixed at 4800 Hz, no filter is introduced in the backend.

Level of confidentiality: CO 15
ICT 612022 - CREAM 14/10/2015

In order to avoid the need for external trigger signals for synchronization and since the
signals are basically modulated in a double-sideband suppressed-carrier fashion, the
carrier is automatically recovered in software through a Costas loop.
EEG and ESI signal are then demodulated to baseband and processed for the required
bandpass filtering and signal scaling to compensate for gains introduced by the
acquisition chain.

Figure 13: Simulink model for extraction of EEG and ESI signals

Figure 14: Bipolar EEG signal acquired by the first 8 channels (reference on CH1) with the
electrodes shorted in saline solution.

Figure 14 shows the real-time plot of the acquired data on the first 8 channels with the
electrodes shorted in saline solution (noise measurement).

Level of confidentiality: CO 16
ICT 612022 - CREAM 14/10/2015

3.1.6 Measurements
The following table presents results for measurements of the active electrodes as
compared to specification requirements. All specs are substantially fulfilled.
Specification Name Spec Value System / Measurement
Number of Electrodes 64 64 (expandable to 128)
Operating Voltage 3.3 V 3V
Die Size <9 mm2 4 mm2
Number of External Components < 10 8
Electrode PCB size <1 cm2 0.5 cm2 without electrode
< 1.5 cm2 with electrode
Total Input Referred Noise (EEG) 0.5 Vrms 0.35 VRMS (EIT OFF)
Input Referred Noise (EIT) 50 nV/Hz 30 nV/Hz (EEG OFF)
60 nV/Hz (EEG ON)
Total Input Referred Noise 10 rms < 6 rms
EEG Signal Bandwidth 0.5-100 Hz 0.1-100 Hz
CMRR 100 dB From electrodes:
66 dB (EIT), 60 dB (EEG)
+ > 50 dB from DgRL
Output Dynamic Range 250 mV 250 mV
Input Impedance >100 M >100 M (EEG)
Load Drive Capability 100 pF 200 pF
Number of Connection Lines to 64 + 1 GND 64 analog EEG/EIT/impedance
back-end channels + 1 GND channel;
Number of electrode configuration 2 per electrode 2 per electrode

Concerning stimulation, tests have been performed to verify the electrode functionality,
which is able to perform current injections of the required shape, frequency and
amplitude. Integration and final testing will be performed in task 4.4 and reported in the
relevant deliverable (D4.4, due M27).

Figure 15 shows EEG acquired in the occipital region with reference on Fpz with the
subject in resting state and switching eyes condition between open and closed. Alpha
rhythm is visible in eyes-closed condition (from 0s to 15s and from 25 s to 40 s) while
when the eyes are open, eye-movement artefacts are visible.

Level of confidentiality: CO 17
ICT 612022 - CREAM 14/10/2015

Figure 16 shows simultaneous EEG and ESI acquisition in three different conditions (one
per column) with different qualities of contact between electrode and skin. ESI is
measured at 125 Hz, outside the acquired EEG band. In all the three cases, the subject is
asked to keep still in the first half of the recording and to repeatedly and rapidly nod in
the second half. In the first recording, the contact quality is very good and in both cases
the EEG signal can be acquired with good quality. When the contact impedance increases
(second column), the presence of the active electrode still allow good quality recording
with the subject is sitting still, however large movement artefacts appear due to the rapid
movement of the head. When the contact impedance is extremely large (third column),
the quality of the signal is always bad.

Figure 15: EEG acquired in O2 vs. Fpz. Alpha rhythm is visible in eyes-closed condition (from 0s to
15s and from 25s to 40 s).

Level of confidentiality: CO 18
ICT 612022 - CREAM 14/10/2015

Figure 16: Electrode-skin contact impedance acquired during EEG acquisition

Figure 17: Phantom model to evaluate the functionality of EIT acquisition

In order to verify and demonstrate the functionality of EIT acquisition, we built a simple
cylindrical phantom (22 cm diameter), filled with saline water (0.33 S/m conductivity, 4
cm water height) and surrounded by 8 equally spaced active electrodes (Figure 17).

Level of confidentiality: CO 19
ICT 612022 - CREAM 14/10/2015

Non-conductive objects of different sizes are inserted to verify the ability of the system to
recognize local conductivity variations in the domain under analysis. An ad-hoc algorithm,
developed in Task 4.2 and targeting localization of compact anomalies (such as in cases,
e.g., of stroke or localized blood flow increase due to brain activity), is used to reconstruct
the position of the anomaly.
A 127 A sine wave current at 16 kHz frequency is injected sequentially in two different
patterns between opposite electrodes. We have assessed how increasing the number of
patterns greatly increases the algorithm capabilities to localize the anomaly. This is
particularly relevant since the developed active electrode system enables each electrode
to be sequentially chosen to either inject EIT current or perform EEG/EIT acquisition.
Figure 18 shows two examples of reconstructions. On the right, a picture of the position of
the anomaly is shown. On the left, the localization of the target is shown with the black
circle identifying the anomaly and the colour gradients identifying the computed potential
distributions due to the two different current patterns.
Since the performances largely depend on the reconstruction algorithm, characterization
will be completed in Task 4.2/4.3.
However, preliminary results show how the HW/SW system can localize cylindrical
anomalies down to less than 5 cm diameter and 1.5 cm height, corresponding to
conductivity variations of approximately 30%.

Figure 18: Examples of target localization in the 2D phantom, through EIT.

Level of confidentiality: CO 20
ICT 612022 - CREAM 14/10/2015

3.2. Cost-effective, quick-setup EEG acquisition system

Figure 19: Simplified representation of a DgRL-based acquisition system. Vmains couples to the
subject through Cmains and Cbody resulting in a common mode interference VCM which is
detected by electrode REF

In recent times, the fields of EEG application have widened from those typical of the
clinical setting to include Brain-Computer Interfaces (BCI) and also consumer-oriented
applications ranging from home care to neurofeedback and gaming controllers. There is
therefore a demand for high-quality acquisition systems whose montage is fast and does
not require trained personnel. Systems based on dry electrodes are particularly attractive
since they make for very simple, almost instantaneous setup.

For these reasons, we wanted to validate the performance of a cost-effective, quick-setup

EEG acquisition system which can work with both wet electrodes with no skin
preparation and with dry-contact electrodes [9]. The presented system is foreseen to be
used in Task 4.3 to validate software libraries for neuroimaging and it can be directly used
in the activities of WP5 which do not involve stimulation, such as Task 5.1. Moreover, this
second prototype could be integrated in the complete neuroimaging/stimulation
platform in order to reduce the costs of the overall system. Even if this is outside the scope
of the CREAM project, where the use of g.HIamp from g.tec is foreseen in order to speed
up the prototype development, the system can also be adapted after the end of the project
to be complemented with visual/auditory stimuli production subsystem and through ad
hoc interfacing board, it can be used in conjunction with active electrodes for EEG/EIT
and electrical stimulation.

The fabrication costs for this prototype, we present in this section, is approximately
50 Euros for the first 8 channels (working system, including wet-contact active

Level of confidentiality: CO 21
ICT 612022 - CREAM 14/10/2015

electrodes) and 30 Euros per each additional 8 channels. These numbers are particularly
attractive for systems designed to be used outside clinical environments, such as several
possible applications of the outcomes of the CREAM project.

Similarly to the high-end system it makes use of the Driving Right Leg (DgRL) scheme
recently introduced by UNIBO [8] (see Figure 19) where the common mode interference
VCM is detected by the reference electrode (REF) and directly drives a potential VR, fixed
with respect to the isolated ground of the instrumentation (G1). VG1 therefore follows the
common mode interference. In this way, a single-ended stage will inherently act as
differential and will encounter very little common mode at its input, automatically
rejecting most of the interference. This allows to use low-cost passive components with
loose tolerances and to increase the gain of the amplification stage so as to relax noise
specifications for the following analog-to-digital conversion stage. The system described
is the first implementation of a DgRL circuit to active electrode systems and it shows how
the levels of CMR which can be obtained are high enough to allow the system presented
to acquire high-quality EEG signals essentially removing network interference without
the need for skin preparation for both wet and dry-contact electrodes.

The acquisition system is depicted in Figure 20. Analog-to-Digital conversion is performed

for groups of 8 electrodes by 8-channel ADCs. The system is modular and can host from 8
to 128 channels.

Figure 20: Simplified representation of the developed system. The overall distance between
electrodes and back-end is approximately 1.2 meters.

Level of confidentiality: CO 22
ICT 612022 - CREAM 14/10/2015

3.2.1 Active Electrodes

Signal amplification is performed on the electrode by a low-power, low-noise, rail-to-rail
Operational Amplifier (O.A.) connected in non-inverting configuration (see Local
Amplification section in Figure 20). Protection resistors are used in order to limit patient
auxiliary current in case of system faults. The O.A. is an OPA378 from Texas Instruments,
which has a quiescent current of 125 A and low current and voltage noise, with no
significant increase at low frequency. In order to minimize the number of wires, the
output signal of the amplification stage is used for providing power supply to the O.A. as
well. A forward-biased diode connects the output pin of the O.A. to the power supply. The
output is biased by the following stage which entails a 6.8 k resistor toward a 3.5 V
power supply. The output of the amplification stage can swing between 2 V, which is the
minimum power supply voltage of the O.A. and 2.4 V which is the maximum input voltage
of the A/D conversion stage. The amplification stage is configured for a DC gain of 1.12
and an AC gain in the EEG signal bandwidth (0.5 to 100 Hz) of approximately 45, which is
enough to reduce the input referred noise of the 16-bit A/D conversion stage to acceptable
levels. Considering a maximum input signal of 100 V, the input DC voltage range is 1.96
V 174 mV, which is enough to accommodate polarization and offset voltages of gold and
Ag/AgCl electrodes [10]. The PCB size is 15 mm x 10 mm, including either the gold-ring
for wet-contact or the snap connector for dry-contact. This size is compatible with
integration on high-density EEG systems (see Figure 21).

Figure 21: Dry-contact active electrodes. On the left, sealed PCB with electronics and snap
connector for the dry-contact (right).

3.2.2 Backend hardware

The Analog/Digital conversion is based on a low-power, 8-channel, 16-bit analog front-
end for biopotential measurements (ADS1198 from Texas Instruments). Standard low-
cost ribbon cables connect the active electrode to the ADC-section. Each electrode needs
two wires, one for ground connection and the other for the active electrode output (and
power supply, as previously discussed). The electrode signals are low-pass filtered by a
passive RC-filter with a 16 kHz bandwidth and sent to the non-inverting inputs of the ADC.
In order to avoid the need to generate an additional reference signal (and add its noise
contribution to that of the acquisition chain), inverting inputs are tied to ground and the
input range of the ADC is comprised between 0 and 2.4 V.

Level of confidentiality: CO 23
ICT 612022 - CREAM 14/10/2015

In the first version of the system (shown in Figure 22), the ADC board is mounted on the
top side of an Arduino UNO board. Arduino is capable of communicating with the ADCs
via SPI. Arduino is also used for providing power supply to the active electrodes and the
ADCs through optical isolators. The entire system is powered via USB.

The USB communication on Arduino UNO limits the maximum channel count to 32 for a
250 samples-per-second (SPS) data rate. A second version of the system is compatible
with Arduino DUE board and allows to extend the system to 128 channels with 1 KSPS
data rate. It has been tested and the final prototype is currently under fabrication.

Figure 22: 32 channel system. Stacked from the bottom, Arduino UNO board, board for the first 8
channels and 3 expansion boards for additional 24 channels

3.2.3 Acquisition software

In order to control and manage the communication between the EEG acquisition system
and a standard PC, a graphical user interface (GUI) has been developed. It can be used to
acquire, filter, process, classify, visualize and store EEG signals in real time. The interface
has been designed using Java which allows users to employ the software on multiple
platforms while keeping the same graphical outlay, level of visualization and operation.
Hence the EEG acquisition system can be connected to Microsoft Windows, Linux and Mac
OS X operating systems. The software architecture is very flexible and allows users to run-
time change the functionality (such as filter type, number of channels visualized, etc.).
The main functionalities of the software are:
Real-time data visualization (an example is shown in Figure 23);

Level of confidentiality: CO 24
ICT 612022 - CREAM 14/10/2015

Data storage in binary format (tools are available for easy importing in MATLAB
environment for further processing);
Referencing (towards one channel or the average of two or more channels);
Modify channel montage and naming;
Modify filter options (low-pass, high-pass, band-pass with different cut-off
Notch filter for network interference removal.

Figure 23: Graphical User Interface: real time data visualization

3.2.4 Measurements

Electrical Characterization

Noise performance is tested by shorting the inputs of the active electrodes. The system
shows a noise level of 0.98 VRMS, integrated in the band between 0.5 and 100 Hz, with no
residual 50 Hz interference. If one computes the difference between two channels in the
digital domain after analog-to-digital conversion, the integrated noise decreases to 0.61
VRMS, due to the removal of common mode noise components. This result, though slightly
higher than the 0.5 VRMS recommended by IFCN [7], are compatible with high-quality
EEG recording and in line with or lower than previously reported active electrode systems
[6] [11] [10].

Level of confidentiality: CO 25
ICT 612022 - CREAM 14/10/2015

In order to evaluate the CMR performances we use the standard setup prescribed by IEC-
60601-2-27 [12]. The standard prescribes that, for a 50/60 Hz sine wave excitation signal
with an amplitude of 20 VRMS, and considering a complete mismatch in the signal
differential path, the output signal should be below 1 mV peak-to-peak, corresponding to
a CMR of 95 dB with respect to the mains signal.






1 5 10 20 30 40 50 60 70 80 90 100
Figure 24: Results for CMR performances as prescribed by standard IEC-60601-2-27, as a function
of frequency.

Figure 24 shows CMR performances for the system, at frequencies between 1 and 100 Hz.
The maximum value of the residual common mode interference is approximately 150 V,
a value which is more than 6 times smaller than the upper limit defined by the standard.
Current consumption of the different sections is:

Local amplification 190 A per channel

Conversion stage 200 A per channel
DgRL < 1 mA

Functional Characterization

Figure25 shows results for EEG acquisition on electrode O2 with respect to Fp2, which
corresponds to the REF electrode of the DgRL circuit. The ground electrode GND is located
in Fpz. The electrodes are completely unshielded and acquisitions are conducted in a
typical electrical engineering laboratory, which can be considered a noisy environment
and worse than most home environments. No skin preparation is needed, except a quick
and gentle cleaning of the whole head with a dry cloth.

The setup time of a 32 channel system is less than one minute for both wet- and dry-
contact electrodes. The setup of neither system requires intervention from trained
professionals, with the dry-contact electrode system possibly being worn by the subject
himself without external intervention.

Level of confidentiality: CO 26
ICT 612022 - CREAM 14/10/2015

Figure 25: EEG signal acquisition in O2 with reference in Fp2 for wet (up) and dry-contact (down)
electrodes. From top to bottom: short time Fourier transform showing spectral content increase in
the time intervals where the subject has eyes closed ([0-30] and [60-90] seconds), time domain
representation and power spectral density (PSD) of the acquired signal.

For the upper part of the figure, the contact with the patient skin is obtained by means of
the gold-ring contact integrated on the electrode and conductive gel. On the bottom,
measurements with dry- contact electrodes are presented. The upper part of the graph
shows the Short-Time Fourier Transform of the acquired signal which clearly highlights
an alpha rhythm at approximately 12 Hz frequency when the subject keeps his eyes
closed. This is also visible in the time domain representation of the acquired signal (full
EEG band from 0.5 to 100 Hz, without any kind of filtering at 50 Hz), which is presented
in the central part of Figure 25. Eye movement artefacts locate the time instants (time
equal to 90 s) at which the subject closes his eyes. After this event, the amplitude of the
EEG signal decreases significantly and the alpha-rhythmic oscillation disappears. The
lower part of the graph shows the Power Spectral Density (PSD) of the signal. The
common mode interference from the mains (50 Hz) has been measured to be 2.5 mVRMS
for wet electrodes and 5.2 mVRMS for dry electrodes. Thanks to DgRL, the common mode
interference at the output of the acquisition system is reduced to less than 1 VRMS and 1.8
VRMS respectively (with a reduction in excess of 70 dB) and does not affect the quality of
the signal.

Level of confidentiality: CO 27
ICT 612022 - CREAM 14/10/2015

4. Concluding section
4.1.1 Conclusion
The presented hardware platform allows to simultaneously record EEG and EIT data at
high temporal resolution and to deliver electrical stimuli to the brain using the same
electrodes. This seamless integration of EEG and EIT recording and electrical stimulation
provides a novel way to study the different cognitive processes of the brain and the
immediate and long term effect of electrical brain stimulation thereon. New procedures
and protocols to study the function of the brain will may be used as it is not necessary any
more to alter electrode positions or to manually interchange EEG/EIT electrodes with
stimulation electrodes.
The easy-to-use interface allows to program and to control various stimuli while
electrodes operating modes are managed through a dedicated controller unit. A special
biosignal amplifier system is used to sample and transmit the recorded signals to the host
computer for further processing.

4.2. Bibliography

[1] C. J. Price, "A review and synthesis of the first 20 years of PET and fMRI,"
NeuroImage, 2012.
[2] C. M. Michel et al., "EEG source imaging," Clinical Neurophisiology, 2004.
[3] R. H. Bayford, "Bioimpedance Tomography (Electrical Impedance Tomography),"
Ann. Rev. of Biom. Eng., 2006.
[4] S. I. Goncalves et al., "In Vivo Measurement of the Brain and Skull Resistivities using
an EIT-Based Method and Realistic Models for the Head," vol. 50, no. 6, 2003.
[5] M. Dannhauer et al., "Modeling of the human skull in EEG source analysis," Human
brain Mapping, 2011.
[6] M. Guermandi, R. Cardu, E. Franchi Scarselli, R. Guerrieri, "Active Electrode IC for
EEG and Electrical Impedance Tomography with Continuous Monitoring of Contact
Impedance," IEEE Transactions on Biomedical Circuits and Systems, 2015.
[7] M.R. Nuwer et al., "IFCN Standards for Digital Recording of Clinical EEG," vol. 106,
[8] M. Guermandi, E. Franchi Scarselli, R. Guerrieri, "A Driving Right Leg Circuit (DgRL)
for Improved Common Mode Rejection in Bio-potential Acquisition Systems," IEEE
Transactions on Biomedical Circuits and Systems, 2015.

Level of confidentiality: CO 28
ICT 612022 - CREAM 14/10/2015

[9] M. Guermandi, A. Bigucci, E. Franchi Scarselli, R. Guerrieri, "EEG Acquisition System

based on Active Electrodes with Common-Mode Interference Suppression by
Driving Right Leg Circuit," EMBC 2015, 2015.
[10] Y. M. Chi, G. Cauwenberghs, "Micropower Non-contact EEG Electrode with Active
Common-Mode Noise Suppression and Input Capacitance Cancellation," in 31st
Annual International Conference of the IEEE EMBS, Minneapolis, Minnesota, USA,
[11] J. Xu, R. F. Yazicioglu, B. Grundlehner, P. Harpe, K. A. A. Makinwa, C. Van Hoof , "A
160W 8-Channel Active Electrode System for EEG Monitoring," IEEE Transactions
on Biomedical Circuits and Systems, vol. 5, no. 6, pp. 555-567, 2011.
[12] IEC 60601-2-27, "Particular requirements for the basic safety and essential
performance of electrocardiographic monitoring equipment," in Medical electrical
equipment, pp. Part 2-27.
[13] International Electrotechnical Commission, IEC 60601-1 Medical Electrical
Equipment - part 1: General requirements for basic safety and essential
performance, Geneva.
[14] R. Matthews et al, "Novel hybrid bioelectrodes for ambulatory zeroprep eeg
measurements using multi-channel wireless EEG system," in Int. Conf. Foundations
Augmented Cognition, Berlin, Germany, 2007.

4.3. Glossary

CMR Common Mode Rejection

CMRR Common Mode Rejection Ratio
EEG Electroencephalography
EIT Electrical Impedance Tomography
ESI Electrode-Skin Impedance
IC Integrated Circuit
OA Operational Amplifier
PSD Power Spectral Density
GUI Graphic User Interface

Level of confidentiality: CO 29