In this chapter the data-acquisition system, which has been specially developed, is discussed in detail. Two methods to reduce environmental noise in the measurement signal in combination with a magnetically shielded room are discussed, i.e. electronic balancing and active shielding by means of a feedback loop.
A data-acquisition system basically samples and stores the data. The demands for the system at the Biomagnetic Centre were:
The main problem when sampling continuously over long periods of time is the
necessity to store data immediately on disk. In a PC this is done through
a DMA transfer process. However, most Analog/Digital converters,
especially the more advanced types, also use DMA to move the data from the A/D
card to the PCs memory. This usually leads to busconflicts, which force
the A/D converter to stop sampling during transfer to disk. Only
interrupted measurement is possible in this way over longer periods of time.
An easy way around this problem is to choose an A/D converter which uses
interrupts to signal the presence of new data, and to have the computer move the
data. Since interrupts are served during DMA transfers, no conflict will
occur. The time spent by the computer in the various routines is depicted
in figure 11.1. It also shows the nesting of interrupt routines during disk
transfer.
The
A/D converter used has four 12 bit AID converter chips, together with four 16 to
1 multiplexers, which gives 64 single-ended input channels. The A/D card,
made at the Technical University of Delft, has a built-in timer chip and
memory to store one sample of each channel. This memory is mapped into the
memory of the main computer.
The basic scheme of the AID card is shown in
Figure 11.2. Originally, the addresses of this memory were located in a part of
the memory used by VGA cards in graphics mode. Since the program which
takes in the data runs in graphics mode, it could not read the A/D card. A
small modification of the printed circuit board of the AID card shifted the
addresses used by the A/D card outside the area used by the VGA display.
The timer can be programmed, and on its command all channels are swept, four channels at a time. No sample and hold circuits are used. The time lag between the first and the last sampled channel is less than 0.5 ms. After all channels have been read, an interrupt signal is generated to signal the computer to move the data. The computer stores the data in a double buffer in the main memory. If one of the buffers is full, a disk write process is started. The PC used is equipped an Intel 80486 processor running at 25 MHz, and has a 340 Mb SCSI harddisk, for the measured data. The programs are stored on a separate 20 Mb harddisk. Further, the PC has a super-VGA display with a resolution of 1024x768 pixels and an ethernet card to connect to the network and allow data to be transferred to the Vaxstation 3520. The computer-connections are shown in Figure 11.3.
The
software written for this PC is able to display twelve channels on the screen
during measurement. This can also be a running average of the evoked
responses, together with an estimation of the noise obtained by computing the
plus-minus average of the signal (Schimmel, 1967). The system samples
standard at 1000 Hz, and if 64 channels are being measured, the disk is able to
store 40 minutes of continuous measurement. Since the number of channels
is usually less than 64, the largest possible measurement time increases
proportionally.
In order to be able to keep track of events, for instance the presentation of a stimulus of a certain type, an Event Data Multiplexer (EDM) can be connected to the printer port of the PC. The EDM was developed by the University of Groningen, The Netherlands, and generates an interrupt together with an event code on this printer port whenever it receives a signal. This signal can come from a second PC which generates stimuli, or from the subject who takes some action in response to a stimulus. The event code is a number from 0-15 which can be used to discriminate between various types of events. When the computer receives the interrupt, it reads the code and stores it in memory, together with the current sample number. After the measurement is completed, a separate file is created which contains all the events and the time at which they occurred. In this way, each event can be correlated to the sampling time at which it happened. The time resolution of the registration of the events is the same as the sampling interval used for the measurement.
If so desired, the sampling program can be configured to start sampling on an event which acts as a trigger. After the specified number of channels has been sampled the system waits for the next trigger event to start a new cycle. This ran be repeated automatically as often as required. However, most evoked response measurements are carried out by sampling continuously. All events are stored, and averaging is done afterwards. In this way pre-trigger data can also be averaged.
The data-acquisition system provides the possibility to examine channels of
sampled files and generate an FFT on screen. Also, the system can convert
the measured file to other file formats to enable the use of available software
from other research groups, written for the analysis of EEG signals.
However, on the VAXstation a special-purpose program has been developed to
examine measurement files in detail called the Data Analysis Program
(DAP). After the data has been transferred, all channels, or a selection
of them, can be viewed simultaneously on the screen. A digital
finite-impulse response filter can be applied, with selectable
corner-frequencies for the high- and low-pass filter. An average can be
made from the event-data file, and can then be displayed with an estimate of the
measurement noise based on the plus-minus average or the standard deviation.
Signal characteristics can be found by matching a template, using correlation
techniques with an adjustable threshold. As an alternative, a
neural-network simulator is built into the software which can be taught to
recognize specific signal features. The network has already been taught to
detect the spike and wave complex in EEG and MEG recordings of epileptic
patients.
The software can also be used for other applications like the analysis of the
magnetocardiogram (MCG) or electrocardiogram (ECG). For this purpose a
heart-peak detector is built in, which first removes the low-frequency
variations in the signal, and then searches for the QRS-complexes. After
these have been identified, the QRS complexes can be averaged. This part
of the program is used for foetal magnetocardiograrn recordings, in order to
average these signals to show significant details in shape, like the p-wave
(Peters and Dunajski, 1991).
Averaging of the evoked responses is one way to improve the signal-to-noise
ratio significantly. Shielding the sensors is preferable in many ways,
certainly if spontaneous brain activity measurements are to be performed, for
instance to record brain activity of epileptic patients or the EEG/MEG during
sleep. A magnetically shielded room is the most obvious way to reduce the
influence of environmental noise on the measurement signal. The
Biomagnetic Centre is equipped with such a magnetically shielded room, but is
also situated at a remote site on the campus. The walls of such a room
consist of a sandwich structure of a high conductivity material (usually an
aluminum alloy) and a high permeability material (a nickel-iron alloy, usually
called µ-metal). The former establishes an eddy current shielding, of
which the efficiency increases with the frequency. It starts to be
effective above about 0.2 Hz in standard magnetically shielded rooms. The
µ-metal, however, gives a frequency-independent shielding for biomagnetic
relevant frequencies (up to 100 Hz). Although this room greatly reduces
the contribution of environmental noise, the shielding for lower frequencies is
relatively poor. Furthermore, the room also distorts the environmental
noise fields. This distortion creates gradients inside the room, even when
the disturbing field is homogeneous. The homogeneous field changes are not
detected by ideal (first-order) gradiometers outside the room, but inside the
room they do register the distorted field changes. Various other noise
reduction techniques are known from literature (Regan, 1989), of which two are
further examined here.
11.3.1 Electronic balancing
The method of electronic balancing is an alternative to the mechanical balancing methods. The purpose of balancing is to reduce the sensitivity of the imperfect gradiometers to homogeneous magnetic fields. The principle is that the magnetic field is measured by a magnetometer and that this measured field is attenuated and subtracted from every gradiometer signal. The attenuation factor depends on the imbalance of a gradiometer, so it is different for each gradiometer.
In the article by Ter Brake et al. (1989) the balancing was done with an
electronic device, but the method can be easily implemented on a computer.
If the gradiometer signals as well as the reference channels are read into a
computer, the attenuation factor can be calculated and the disturbance
subtracted from the signals. Three orthogonal reference channels are
needed for a complete balancing. The problem with digital balancing is
that the A/D computer has to have a sufficient signal range and resolution to
sample the signal with the disturbance and still have a detailed signal after
subtraction. Therefore, subtraction is usually best carried out before
sampling.
DETERMINATION OF THE A17ENUATION FACTORS
The gradiometer-signal G can be imagined to be composed of an ideal gradiometer signal S and contributions of the homogeneous magnetic field in all three orthogonal directions due to the imbalance of the gradiometer:
G=
xBx+yBy+zBz
With
i: the unknown sensitivity of the
gradiometer channel for homogeneous fields in the i-direction (the imbalance).
Bi: The magnitude of the homogeneous magnetic field in the i-direction.
The conversion of the (gradient of the) magnetic field to the measured
voltage should be described by the transfer function. It is assumed here
that this transfer function can be described by a factor which is constant for
all frequencies in the signal. This multiplication factor is denoted by
the symbol . If it is assumed that
the imbalance other than in the direction along the axis of the gradiometer can
be neglected, only one reference has to be taken into account; the
Z-reference. Calculating the cross-correlation of the gradiometer signal
with the reference signal R., yields:
with rz being the amplification factor
for the z-reference channel.
The cross-correlation of the gradiometer and the Z-reference can be divided by
the auto-correlation signal of the Z-reference Azz to yield:
Since the amplification factor of the Z-reference can be determined, the attenuation of the homogeneous magnetic field component in the gradiometer can be determined, and with this, the amount with which the reference channel has to be subtracted from the gradiometer output signal to reduce the environmental noise. These formulas also hold if the homogeneous magnetic field components are uncorrelated to each other. However, since the reference channels do not only sense the homogeneous part of the magnetic field, this method only works properly if the disturbing sources are far enough away that their field is approximately homogeneous.
This method can be expanded for all three orthogonal directions, but then crosscorrelation terms between references have to be taken into account. A disturbing signal is probably visible in more than one reference channel and each should be used to subtract part of the disturbance, For this purpose, the cross-correlation of the gradiometer with each reference has to be determined. They are:
This set of formulas can be written in matrix form:
The factors i/ri
can be determined from this equation by inverting the matrix. All the
cross and auto correlations can be calculated from the measured signals
MEASUREMENTS WITH ELECTRONIC BALANCING
This method has been applied to a real measurement. For this purpose one gradiometer channel and three reference channels were sampled outside the shielded room to guarantee enough environmental noise. The result is shown in Figure 11.4. The amplification factor for the Z-reference is much higher than those for the X- and Y- directions. Since the imbalance of a gradiometer is largest along its axis and the axis of the gradiometer was almost parallel to the Z-direction, this is what could be expected. The disturbance has been reduced with a factor of 10 to 20.
However, if the experiment is repeated inside the shielded room, the result is less encouraging as can be seen in figure 11.5. The noise is reduced after balancing, but when the gradiometer signal undergoes a significant change, a peak is left in the balanced signal, with an amplitude as high as the noise in the signal. This can only be caused if the response of the reference channel to the field change is no longer similar to that of the gradiometer. This indicates that the imbalance of the gradiometer is no longer the main contribution to the response to external disturbances.
In order to test this effect, a Helmholtz-like coil was constructed around the shielded room, so that by applying a current through the coil, artificial disturbances of the magnetic field could be created. A disturbance in the form of a step function was applied, and the responses of both the reference channel and a gradiometer channel were recorded. The result is shown in Figure 11.6. The response of the reference channel is as expected; it follows the step function but with a certain timeconstant. This is because the high-frequency components of the field change are shielded by the room. However, the gradiometer channel does not follow the stelpfunction directly, but shows a clear distortion of the magnetic field gradients inside the room.
The effect of the field distortion can be understood by taking a closer look
at the biomagnetic sensing system. Usually, biomagnetic experiments are
performed with gradiometers that measure field gradients. They should, in
principle, be insensitive to uniform noise fields. However, due to limited
accuracy and errors in the construction of such a gradiometer a sensitivity to
uniform fields is always present. This imbalance of the gradiometer is
represented by the factor Cb. This imbalance factor Cb is equal to the
ratio of the effectively measured field to the applied field.
Gradiometers; usually have Cb values of 10-2 to 10-3 .
If such a gradiometer is placed inside a shielded room the situation is much
different. In this case, a uniform magnetic field is distorted by the
walls and gradients arise inside the room, due to the eddy currents, which are
detected by the gradiometer. This can be seen from Figure 11.6. The full
curve shows the current through the coil, the broken curve shows the output from
the magnetometer while the dotted curve shows the output from the first order
gradiometer. The distortion of uniform magnetic fields can be quantified
as an effective contribution to the imbalance. We measured in our shielded
room, for a magnetic field in the vertical direction, a Cb of 3.10-3 in the
centre and a Clo of 1.2.10-2 at 60 cm above the floor in the centre, employing a
gradiometer with a baseline of 5 cm. The standard double p-metal walled
shielded rooms have low-frequency shielding factors somewhat below 100.
This means that a reasonably balanced gradiometer is more sensitive to uniform
environmental magnetic noise inside the room than outside it!
Furthermore, one may conclude that it makes no sense at all to try to make
gradiorneters which are highly balanced to be used inside a magnetically
shielded room. Finally, it has to be concluded that the use of electronic
balancing in combination with a shielded room is not an effective way to reduce
disturbances.
11.3.2 Active shielding
To reduce environmental noise, the idea of an active shield has been used by Marzetta (1961), although not for Biomagnetic measurements. The principle of active shielding is based on a feedback loop. A magnetic field sensor picks up the variations in the magnetic field. This signal is fed into a controller which in its simplest form consists of an amplifier, that drives a set of coils which generates a magnetic field in the direction opposite to the disturbance. In this way the disturbing field is compensated and the resulting field should be less noisy. In Biomagnetic measurements this method has been used by Donnelly et al. (1988), but a problem occurs if the disturbing field pattern cannot be compensated by the field-generating coils. In this case, the disturbing and compensating field do not coincide, and a large part of the noise field remains. This problem becomes worse if one tries to compensate disturbances over a large area.
In order to avoid this problem, the method of active shielding can be combined with the passive shield viz, the magnetically shielded room. The p-metal walls of the room absorb the field lines, thereby reducing the possible field configurations. By applying coil-sets around the room which generate a counter-magnetic field of which the field lines are also absorbed by the p-metal, the actual compensation of the disturbances takes place in the walls of the magnetically shielded room. This way the compensation is much more effective than the methods without the shielded room. This method has been tried by Kelha et al. (1982). They used a flux-gate magnetometer outside their shielded room as a sensor. By feeding the signal through a PID controller they established an improvement in shielding of 35 dB at 0.1 Hz. and 20 dB at 1 Hz. The numbers depend on the position of the sensor. However, this sensor picks up more of the disturbance than is actually needed to be compensated, because part of it is also shielded by the room. Furthermore, the compensation of the feedback loop is best at the location in the area where the measurements are performed, i.e. near the pick-up coils. It is therefore logical to place the sensor inside the shielded room, and use a SQUID as the sensor. Most multichannel systems are already equipped with reference channels.
An important advantage is that frequencies which are already shielded properly by the room, do not reach the sensor. Especially the mains frequency, which can swamp the feedback amplifier, is reduced significantly.
The principle of the active shielding method is shown in Figure 11.7. Ter Brake et al. (1991) showed that a low frequency shielding of up to 40 dB can be reached with this method in magnetometer channels. The disturbances in the gradiometer signals are reduced by at least a factor 5. Obviously the improvement is less spectacular than it is for magnetometers. This is caused by the fact that the transfer function of the environmental field to the magnetometer is quite different from that of the environmental field to the gradionneter. Since this improvement is very useful, a user-friendly PID controller is under construction, which should yield an improvement over the use of only an amplifier to generate the compensating field. Also, the method will be expanded to all three orthogonal directions. This should further improve the performance of the active shield.
(c) MEG, EEG and the integration with Magnetic Resonance Images, H.J. Wieringa, 1993
[<<Chapter I][contents][home][Chapter III: From electric and magnetic signals to their sources >>]