The article is a brief summary of the fundamental principles
of EEG system design. It is based on almost twenty-year experience
in the creation of systems of functional diagnostics and contains
strict and clear interpretation of the main technical terms, which
characterize the quality of electroencephalographic systems. Although
the article requires special knowledge from the reader, it is
of interest for professionals in the field of electroencephalography,
as well as for system engineers, who independently develop a system
of functional diagnostics. Taking into account that many terms
used in the article require explanations, the article is completed
with a simplified glossary. Harshness of some definitions of the
author is an inevitable reaction to the real state of the matter
with electroencephalographic equipment in Ukraine.
Recording an EEG in the routine version of examination is not
a technically difficult task. However, obtained results are far
from always of high diagnostic value. It is connected not only
with physiological limitations of the method, but often also with
the quality of recording, which not only shows no increase, but
in a number of cases evokes associations from the domain of obscene
The reason lies not only in the degradation of electroencephalographic
equipment in Ukraine (sic!), but also in the absence of the stuff
training and frequent complete dilettantism of the doctor.
Really, if twenty years ago the devices from respected
companies - NIHON_KOHDEN, ORION, etc. were used, completed with
proprietary electrodes and expendable materials, than now they use
computers, "home-made" by Ukrainian companies and certified
a la Ukraine. The masterpiece of squalor, if one may say so, is
their completion with homemade iron electrodes wrapped in rags to
be moistened with physiological solution. It is clear that, dumping
at tenders, these companies compete in reducing the cost of production
and as a consequence, in reducing the quality of EEG recording.
Naturally, producing more and more "full-featured" software
they try to play the fool with the doctor. They offer thereat either
nonexistent methods of analysis, or methods for removing the artifacts,
born by, let's say it flat out, a shitty electroencephalograph.
Most often they manage to draw the wool over the doctor's eyes,
because for almost 20 years there were no full value courses in
this specialty. Eventually the most eye-popping results are achieved
in commercial polyclinics, where the main requirement to the equipment
is the availability of a system of automatic diagnostics in it.
But, in spite of our nationally isolated, cheap
at tenders, domestic collective farm, the technical progress in
the world goes on, and the quality of electroencephalographs of
the world leaders grows up. And thereat not only doctors, but some
domestic developers of medical diagnostic equipment are convinced
that such pure recordings cannot exist, and bourgeoises have invented
some trick, by means of which they draw such nice electroencephalograms,
electrocardiograms, etc. Generally speaking, the ghost of second-rateness
and provinciality is tacitly, but menacingly present at any discussion
of the problems of electrophysiological signal recording in our
domestically developing consciousness. The best evidence of this
thesis will be the fact that nobody even thinks of using homebred
Ukrainian EEG junk for diagnosing the death of the brain (it is
not just telling users and patients horror fairy tales, "everything
you say can be used against you"). Or for operation with a
connected artificial lung1 (it's obvious that the device will produce
nothing but interference), and for any serious examination at all.
But, regardless the local (featherbed!) state of the matter, it
is necessary to strive to improve the quality of recording, even
disregarding the fact that not competition, not doctors and not
patients require this. The worse the users, the better the equipment
must be. And, if earlier only the Big League dealt with electroencephalography,
now there will be nobody except sleeping on the go dumb Dora Nakenavel
as the nurse and a prescriber of biological food additives as the
doctor. There are lucky exceptions (very few of them), but they
are still exceptions.
Theoretically the problem of EEG recording is
a problem of multichannel synchronous recording of microvolt infra-low
frequency2 signals with a typical internal resistance of the signal
source between several and hundreds of kiloohms. One must take thereat
into account the a priori presence of additive3 systematic noise
with the mains frequency with the signal-to-noise ratio worse than
a unit, additive white Gaussian4 noise (AWGN) of the electrodes,
zero-frequency component in the form of electrochemical polarization
potential, trends caused by changes in osmotic pressure5 if electrode
liquid, and multiplicative noise caused by changes in electrode
resistance. Perhaps, we must place systematic and radio-frequency
noise in a separate group, as well as physiological artifacts.
Historically, this problem was rather successfully
solved by means of differential amplifiers7 with a large common-mode
rejection ratio (CMRR8) and high input impedance. No other solution
was proposed up till now. And even now, in spite of all impressive
progress of electronics, the majority of devices are designed in
the same way. Beautycraft in the form of widened dynamic range9
or decimation10 after high-frequency digitization is presented as
a revolutionary breakthrough to the shining heights. Especially
good (particularly for a psychiatrist) are the statements on achieving
the CMRR of more than 140 decibel (160 dB11 - I saw it with my own
eyes in one advertising booklet of an Ukrainian manufacturer). Amazingly,
there is no such equipment either in Germany, or the USA, or Japan.
The declared specifications of their devices are tragically inferior
to the Ukrainian ones, and it is not clear, what insidious Japan
specialists do in the field of industrial espionage.
However, in order to understand the process of
development of equipment for EEG examinations, it is expedient to
recollect its history with unforgettable tube Alvars, the nostalgia
for which clings to the good old guard, going into oblivion.
So, let us take as an axiom the requirement of
high impedance and common-mode rejection ratio of the input amplifier.
Paradoxically, but tube circuits met this requirement in the best
way, especially in a nuvistor12 low-noise version. The reason is
simple (although it is ignored by contemporary developers) - these
stages preserved high CMRR up to the frequencies much higher than
the working ones and had very insignificant (almost unreachable
even up till now) input capacitance. So, as a consequence, the device
successfully filtered all HF components of external interference
and physiological artifacts. And an irremovable drawback was a parametric
trend, that is, these assemblies required fine adjustment and were
not a LIS13 system. At present it is difficult to imagine an electroencephalographist
adjusting the CMRR (!), he has very vague notion even about the
operation of the device's filters (the better cars, the worse the
The development of semiconductor technology has
predetermined the appearance of electroencephalographs with field-effect
transistors at the input and the input amplifier balanced at the
factory. In spite of the development of special-purpose chip assemblies
of type KPS 104, these devices were inferior to tube ones by all
parameters (except dimensions, mass and energy consumption). Higher
flicker14, the necessity to protect the inputs from static discharges
and impossibility of adjustment after repair and replacement of
components have made these devices non-competitive. As a consequence,
parametric input amplifiers (PIA15) began to appear, where the main
gain in current was provided by a stage based on varicaps16. High
characteristics in the working band were combined in them with low
efficiency of HF interference rejection and complexity of circuitry.
But the necessity to space apart the working frequencies of the
electroencephalograph amplifiers has become the main problem. That
is why such devices were manufactured only in versions with small
number of channels.
First attempts to build the amplifiers on integrated17
operational amplifiers have shown that everything is far from good
with internal noises and input impedance. And, in spite of special
methods for suppressing flicker (passivation18 of the crystal by
silicon nitride, etc.), only recently the attempts to achieve the
desired characteristics succeeded. Nevertheless, the problem of
drift of the main parameters as frequencies approach the boundary
ones still persists even in operational amplifiers manufactured
by AD, Burr-Brown and other. That is exactly why the designing of
a modern electroencephalograph somewhat differs as to complexity
from assembling a LEGO toy. Utilization of good components by no
means guarantees the good quality of the device (NB!).
Modem operational amplifiers19 in the integrated
circuit version have turned out to be suitable for utilization as
the electroencephalograph input stages. Disregarding rather modest
figures of the main parameters declared by the manufacturer (chip
140UD13), they ensured reliable and stable operation without the
drift of characteristics throughout the entire period of operation.
However, some peculiar features have floated up. For example, it
is not possible to use electrodes with polyurethane foam filler
- "bridges"20 - due to the device's response to periodical
changes of the active and reactive components of their total impedance.
The reason is simple - the modem operational amplifier pumps the
RF energy "back" through the spurious capacitance21 of
the modulator gates22. Unfortunately, this fact is ignored by some
Ukrainian manufacturers, who have obtained the device in "solid-drawn"
form. The results of polyurethane foam bridge application are horrible
for the user, but wonderful for the manufacturer - it's cheap, cool,
and nothing is clear.
Our experiments have shown that the attempt to
build modem amplifiers on small-scale integrated circuits in order
to increase the channel area of field-effect transistor and, as
a consequence, to reduce internal noise, have yielded excellent
results in separate specimens and demonstrated complete inadequacy
for mass production. We have obtained the peak-to-peak noise23 reduced
to the input less than 0.35 ?V with CMRR not less than 115 dB and
differential input impedance not less than 20 MOhm. But a multichannel
electroencephalograph built on this principle required enormously
large labor consumption for adjustment and tuning.
Out of the hardware exotics, we must note commutation
amplifiers24, which, according to the results of investigation in
our laboratory, remain promising, although problematic (internal
noise grows proportionally to the square root of the number of channels).
But one cannot exclude that application of optimum circuitry solutions
can help to solve this problem, that is to lower the noise to acceptable
Among the loss-making products of the Goodwin's
factory of magical articles (this factory was recently renamed in
one of Ukrainian companies), we can name circuits with "flying"
capacity25, which simulate a high CMRR by means of lowering the
differential input impedance of the stage (the energy of commutation
noise is compensated due to continuous recharging of capacitor).
Namely for this purpose the specifications of interest to a psychiatrist
were proclaimed. According to the user's response such devices never
At present the development of the EEG circuitry
has halted on operational amplifiers. Virtually all modern electroencephalographs
use precision26 operational amplifiers as the main scaling27 amplifier.
And taking into account the choice of their types, we can say for
sure that an amplified EEG is present at the output terminals of
the operational amplifiers. In this connection the question arises,
why the majority of homemade electroencephalographs work so bad
and how they manage to create such problems at routine recording?
The explanation can be the only one - the device
has not been designed for operation under the conditions of actual
noise. In this connection let us examine the situation with external
interference in more detail.
Undoubtedly, the most important and maximal as
to level is the noise with the mains frequency - 50 Hz. It is caused
by pickups from power supply wires, including current loops, in
which the device and patient can find themselves. Very simply said,
the value of the noise can be considered proportional to the capacitance
of the system patient-device in respect to the power supply wires.
One must thereat keep in mind that when a device with internal power
supply is grounded, the noise increases proportionally to the increase
in capacitance (the ground/mains capacitance is always larger than
the patient/mains capacitance). Of course, we speak not only about
the common-mode component. Its antiphase (to be specific - out-of-phase)
value is determined by the asymmetry of the system of electrode
placement. Naturally, the more symmetrical is the system of electrode
placement, the lower is the out-of-phase component of the noise,
which arrives at the input of the electroencephalograph. According
to the data, obtained in our laboratory, in order to obtain the
amplifier's CMRR of 80 dB, it is necessary to balance the system
of electrode placement up to the value not less than 40 dB. Otherwise
the operation outside a shielded chamber will become problematic.
However, this requirement does not exist for an
infinitive dynamic band of the input amplifiers and ideal elimination
filter28. Surely, nothing is perfect in the world, and an electroencephalograph
with infinitive CMRR and asymmetrical system of electrode placement
will not work. Even at very small external noise. Generally speaking,
one must take into account that asymmetrical systems were designed
at ancient times (but sound and professional) - only for the conditions
of operation in "Faraday cage". As it is impossible to
use it under unscreened conditions, we can with good reason attribute
it to the realm of the designer's idiocy. Conclusion for the buyer:
being offered a device with whatever-you-want hardware calculators
of the system of electrode placement - know: some of them will never
work in a not screened room, even disregarding the outstanding specifications
of the amplifiers.
Well, what will happen to an electroencephalograph
when the noise exceeds the value of the margin, determined by the
dynamic range? The answer is quite simple - "dirty" electroencephalogram
will periodically appear and then disappear, and the "sections
of silence" will be slightly shifted with respect to each other
in various channels (Fig.1).
Fig.1. Appearance of "sections
of silence" on EEG
Namely, the 50 Hz noise is not seen, it has been
shamefacedly cut off by the rejector. But if we turn it off, the
beauty of the designer's ideas will manifest itself in full volume
Fig.2.EEG with filters turned
But it is possible to solve the 50 Hz problem in
he absolute version. With full symmetry of the system of electrode
placement and equal patient/ground and patient/mains capacitances
it (i.e. noise) tends to zero in its out-of-phase version. Naturally,
the provision of such conditions require the infinitive input impedance
of the device, no grounding and complete symmetry of the inputs.
No problem is to eliminate grounding by means of digital telemetry.
But the other conditions are both unreachable and mutually exclusive.
It is impossible to balance the input under the conditions of very
large impedances of the input resistive array, and it is impossible
to obtain infinitive values of resistance for a precision array
due to physical reasons. Do not thereat forget that the noise EMF30
grows as the input resistance increases. Nevertheless, an approximate
(and particular) solution exists in the form of the transfer of
calculations into a strong-current or digital domain, naturally,
at a sufficient dynamic range. As a result, the noise will be not
filtered, but eliminated primordially, that is the condition of
external noise rejection will be met. Developing this theme, we
can offer also other solutions within the scope of the "External
Noise Rejection" (ENR39) principle. As an illustration, we
can show the electroencephalogram of the same patient, recorded
in the same premises and under the same recording conditions (Fig.3).
Fig.3.Gain 50 чV /10mm, sweep
speed 30 mm/s, HFF-OFF, LFF 70 Hz. (EEG is scaled, 50 чV, 1 s are
Pay attention that the elimination filter is off,
and the bandwidth is limited to 70 Hz. That is, noise is not filtered,
it is eliminated primordially. This is very important also because
high values of the noise signal are not multiplied inside the amplifier
with the useful signal (such multiplication is caused by nonideality
of the voltage-current characteristic of any physically feasible
amplifier). Besides, the absence of a band-rejection filter ensures
high fidelity of signal transmission, as an IPC31-eliminator always
"clinks" at pulse interference, and an FPC32-system has
in the time domain a preceding and succeeding response, caused by
the shape of the data window (obviously, specialists do not need
We must also keep in mind that, apart from the
50 Hz noise, its harmonics exist, especially when fluorescent lamps,
thyristor power controllers and other similar devices are used.
Therefore the electroencephalograph's immunity to other systematic
noise must be ensured as well.
Most generally speaking, we can say that an ideal
device must not register any physical artifacts at all.
When analyzing a group of artifacts it makes sense
not to analyze each of them separately in the time domain, but investigate
the spectrum of all possible interferences and compare it to the
spectrum of electroencephalogram. This work was carried out in 70s
in the New York Medical College and it has immediately become as
near as dammit a classical one. It has been shown that the EEG signal
in many cases "dives under the noise" at frequencies below
one Hertz and above 70 Hertz. And one can say nothing about the
electroencephalogram below and above the indicated frequency boundaries
when analyzing it visually. At the same time, if the signal and
the noise are not spaced apart in time and the spectra of the noise
and the signal overlap, its complete separation is impossible. This
is one of the most general and fundamental postulates of the theory
of informatics and attempt to bypass it is akin to the desire to
cheat the force of gravity. However, such exercises still go on
and are caused not so much by dilettantism, as by efforts to invent
(a perpetuum mobile?).
Taking into account two facts - the required bandpass
of the amplifier and requirement to minimize noise within the band,
it is obvious, that a properly designed amplifier limits the bandpass,
and the nearer to the input amplification states, the better, and
tends to limit the spatial filtering at the input for the noise
within the bandwidth. Otherwise there is no way to cut them off
from the signal. That is, the ENR principle quite logically follows
from the main task of designing.
It must be acknowledged, that the majority of
devices with to the utmost open bandwidth, redundant digitization
at the dynamic range determined by the gain of the first (upstream
of the ADC) stages, are designed improperly. Herewith the increase
of the ADC digit capacity can be used for advertising purposes only.
The actual tolerance of the device to noise will the smaller, the
larger its bandwidth, and the dynamic range will be "jammed"
the earlier, the more open is the band from the lower side. But
the most substantial will be the contribution of the noise passed
through an asymmetrical system of electrode placement.
Besides, the problem of adequate discretization
of input signals still has no rigorous solution and is rather a
technical compromise, completely subordinated to posed problems.
Even if we orient exceptionally at sigma-delta converters33, we
must acknowledge, that the frequency response of the channel (Fig.
4) is far from the one intended to cope with high-amplitude interference
in a wide frequency band (Fig. 5).
Obviously, the simplest addition in the form of
implementation of Hamming polynomials makes it possible to obtain
a much better protected digitizer. At the same time, discretization
by the sigma-delta method has no advantages, except a lower intrinsic
noise, against a trivial "voltage-frequency" converter
(VFC34), which uses the temporal aperture of the discretization
period, and, taking into account the possibility to use in a VFC
the temporal data window, is inferior to it. And even primitive
solutions in the form of the hardware Bartlett window35 ensures
the victory of the VFC over the sigma-delta digitizer (it is hardly
its fault - it has been designed not for these purposes). Nevertheless,
both the designers and advertisers of the device always demonstrate
the euphoria on the occasion of total interference immunity of their
device, which implements all gimmicks of the modern component base.
If we actually subjugate the digitizer technical
requirements to specific features of the hardware, which has to
operate under the conditions of a priori unknown interference, we
must first of all ensure the maximum out-of-band attenuation at
a linear frequency response in the band and realization of necessary
band-rejection filters, which are combined with the digitization
procedure and do not use infinite pulse characteristics. Please
pay attention that in this case the requirements to the dynamic
band become substantially lower (it becomes realistic) and the possibility
appears to raise the gain of first stages. In its turn, this, as
a consequence, lowers their intrinsic noise (NB!). In this case
there is no necessity to increase the number of ADC digits, it can
be subordinated to the requirements of a specified digitization
error within the scope of the entire project (but not competitive
Fig.6. Death of the brain.
Infant 1 year old, the seventh day of artificial lung operation.
Throughout the entire recording no phenomena with amplitude above
1.5 чV were registered. Ag/AgCl gel electrodes. Telemetric electroencephalograph.
Gain 25 чV/cm in the main image and 6.25 чV/cm in the enhanced one.
The image is scaled, +-1,5 чV is indicated with arrows in the enhanced