Chapter Five

Data Acquisition

5.1 Introduction

Three-dimensional (3-d) eye and head rotation data were collected using the high
resolution, low noise magnetic search coil technique. The configuration of the magnetic
search coil system and the other systems used in the data acquisition process will be
described in this chapter. Details of the experimental procedures, stimulus paradigms, data
analysis, and mathematical modelling will be considered in the following chapters.

5.2 Magnetic Search Coil Technique

Robinson (1963) developed a low noise, high resolution system by which eye
rotations could be measured in 3-d space. The technique involves recording electrical
signals from a soft rubber search coil positioned on the anaesthetized eye of a subject
sitting in an oscillating magnetic field.

In earlier studies, a simplified version of the system was used to measure eye
position in two-dimensional space. However, practical improvements to the system have
made it possible to accurately measure eye position in 3-d space (Collewijn et al. 1985).
Although invasive, the method remains the gold standard for recording eye movements.

5.2a Search Coils

Dual search coils (manufactured by Skalar Instruments, Delft, The Netherlands)

were used to record the eye and head rotation data. Each dual search coil consists of an
annular soft rubber contact lens that has two fine copper induction coils – the “direction”
and “torsion” coils – implanted in it. The direction coil consists of several turns of copper

64
wire wound in the frontal plane, while the torsion coil consists of several turns of copper
wire wound effectively in the sagittal plane. Both direction and torsion induction coils are
connected to preamplifier leads by insulated copper leads, which are tightly twisted in
order to minimize magnetic artifacts. The preamplifier leads are connected to amplifiers,
from which the signals are passed to phase detectors (see sections 5.2e and 5.2f).

The search coils are designed so that they lie on the conjunctiva over the sclera
surrounding the cornea and not on the cornea itself (see figure 5.1). The search coil
therefore does not obscure the subject’s vision, and the risk of acquiring a vision-
threatening complication, such as a corneal abrasion, is minimized.

Figure 5.1 A sterilized search coil in place on the limbal conjunctiva (from Leigh and Zee 1999).

5.2b Principle of Operation of the Search Coil System

When a coil of wire is positioned in the centre of an oscillating magnetic field, an

electromotive force (EMF) is induced in that coil, in accordance with Faraday’s law.

65
Faraday’s law states that if the magnetic flux φ across a coil of n turns is changing with
respect to time, then the EMF induced in the coil will be given by
dφ
EMF ∝ − n
dt
where dφ/dt is the change in magnetic flux with respect to time. When the coil is
stationary, the EMF oscillates, since the magnetic field is oscillating regularly. However,
the amplitude of the EMF induced in the coil remains constant. When the coil rotates into
a new position, the amplitude of the EMF induced in the coil changes if the area of the coil
that is perpendicular to the magnetic field changes.

The search coil system works on the above principle. The amplitude of the EMF
induced in each induction coil is dependent on the area of the coil that is perpendicular to
the magnetic field and, hence, the position of the coil in space. The search coil system used
to gather data in these experiments had two earth-fixed magnetic fields, which were
orientated horizontally and vertically in space. Movement of the search coil therefore
resulted in two EMF signals being induced in each induction coil. The vertical magnetic
field oscillated at one-and-a-half times the frequency of the horizontal magnetic field,
thereby allowing the two signals from each of the induction coils to be extracted by a
phase detector using synchronous demodulation. The signals from the phase detectors
were precision rectified, and the square root of the mean of the square (the RMS) of each
of the signals was calculated. The RMS of each signal, a DC voltage, was low-pass filtered
and recorded by the PC. In total, four signals were recorded from each coil. The horizontal
and vertical components of the data were derived from the signals induced in the direction
coil, while the torsional component of the data was derived from the signals induced in
both the direction and torsion coils (see chapter 7).

During the experiment, the subject’s head was positioned so that it was in the
centre of the two oscillating magnetic fields. After sterilization, one search coil was placed
on the conjunctiva overlying the sclera of the subject’s left eye; the search coil was
positioned so that the exiting wire was positioned nasally (see figure 5.1). Another search
coil was secured to a plastic spectacle frame which was in turn secured to the subject’s
head (see section 6.4c). The eye and head rotation data were derived from the signals
recorded from these two search coils.

66
5.2c Search Coil Faults

Ideally, the direction and torsion induction coils should be orthogonal to one
another, since the oscillating magnetic fields they are placed in are orthogonal to one
another. In practice, many search coils were found to have non-orthogonal induction coils.
The effects of the non-orthogonality were most obvious during vertical in vitro
calibrations, where purely vertical movements of the search coil were observed to induce
voltages in both induction coils. Bruno and Van den Berg (1997a) developed an algorithm
to correct for non-orthogonality between the direction and torsion coils, which has been
used to correct data from previous experiments (Thurtell et al. 1999) and in the current set
of experiments (see section 7.3a).

Significant non-orthogonality between the induction coils was found to be present

in most new search coils. The effects of non-orthogonality were, however, exaggerated in
search coils that had been subject to excessive manipulation, so handling of the search coils
was therefore minimized before and between experiments.

5.2d Magnetic Field Driver and Coils

The magnetic field driver and magnetic field-generating coils were manufactured
by CNC Engineering (Seattle, USA). The driver consisted of power oscillators that drove
the two pairs of magnetic field-generating coils, each at a different frequency (see section
5.2b). The oscillators were locked to an internal temperature-compensated oscillator. The
magnetic field-generating coils were enclosed in a mahogany frame with dimensions 1.9m
× 1.9m × 1.9m, and were orientated at ninety degrees to one another, thereby producing a
horizontal magnetic field and a vertical magnetic field.

5.2e Signal Amplifiers

Signals from both of the induction coils of each search coil were amplified to
increase the signal-to-noise ratio, before being passed to phase detectors. The amplifiers
were positioned on the back of the test chair and were located within mu-metal shields, to
prevent the electronics within the amplifiers from being disrupted by the magnetic fields.

67
5.2f Phase Detectors

The phase detectors, which were also manufactured by CNC Engineering (Seattle,
USA), recovered amplitude-modulated signals from each of the induction coils by acting
as synchronous demodulators. The signal from the direction coil was split into horizontal
(H) and vertical (V) components, while that from the torsion coil was split into torsional
(T) and auxiliary torsional (T2) components. Following phase detection, the RMS of the
precision rectified signals was calculated and then passed through an anti-aliasing filter.
Electrical cross coupling between the horizontal, vertical, and torsional channels was
minimized, so that it was less than 0.01%.

5.2g Aliasing and the Anti-Aliasing Filter

Digitizing data acquisition systems can incorrectly interpret signal components with
frequencies greater than half the sampling frequency as being lower frequency signals, a
phenomenon known as aliasing (see figure 5.2). The custom made anti-aliasing filter
prevented aliasing by acting as a low-pass filter, attenuating signal components with high
frequencies prior to sampling. The 3dB cut-off frequency (the frequency at which signal
amplitude is ~70.71% of the total signal) for the filter cards was about 115Hz on average,
allowing the sampling rate to be adjusted to any frequency above 230Hz with minimal
signal component aliasing.

high frequency signal

(before sampling)

signal after sampling

Figure 5.2 When a high frequency signal is sampled at a low sampling frequency (1/∆t), the signal is
interpreted as having a frequency lower than its true frequency and is, consequently, confused with
the lower-frequency data, a phenomenon known as aliasing (from Marmarelis and Marmarelis 1978).

68
 The anti-aliasing filter also filtered out noise and any other high frequency
interference, such as electromagnetic noise, that could have corrupted the signal (Proakis
and Manolakis 1992). Thus, low-pass filtering the signal attenuated noise outside the
filter’s pass-band, reducing the additive noise spectrum to that which falls within the pass-
band of the filter.

5.2h Data Acquisition and Storage

The 16-bit analogue-to-digital converter of the PC (model AT-MIO-16X, National

Instruments, Austin, USA) digitized the signals once they had been filtered. The data
signals were sampled with a sampling frequency of 1kHz. Once digitized, they were
displayed on-line in real time, using a LabVIEW (National Instruments, Austin, USA)
program, while concurrently being stored on the PC hard disk. The data files were later
copied to the hard disk of a DECstation 5000/240 (operating under the Ultrix system) and
analysed using programs written for C and Splus (AT&T, New York, USA).

5.2i Noise Reduction

To minimize the amount of noise present in the signals from the search coils,
superfluous system hardware components were removed, and old system components
were updated. All unused input sockets were terminated using 75Ω resistors. The power
supply for all components was obtained from filtered power boards, and the use of
fluorescent lighting, observed to introduce ~50Hz noise into the signals, was minimized
during the experiments. Following these changes, peak-to-peak noise in calibrated data
was found to be equivalent to <1 minute of arc, with maximum velocity peaks at ±3.5°/s.
Consequently, no smoothing stages were required at any point in the data analysis.

5.3 Laser Projection Apparatus

During the experiments, the subject was required, at various times, to fixate on a
laser-generated target. The laser target was rear-projected on to a Perspex screen, which
was attached to the frame of the magnetic field-generating coils. The screen sat 94cm in

69
front of the cornea when the head was in the centre of the magnetic fields. The laser, a
solid-state red laser, was projected by two mirror galvanometers positioned behind the
screen. The LabVIEW program responsible for displaying and recording the data drove the
mirror galvanometers. The laser beam was passed through a neutral density filter to reduce
the intensity of the target and to minimize afterimage on the retina. The filtered beam was
then projected onto the screen, which had semi-translucent paper securely attached to it.

The offset of the laser target was adjusted to lie in the centre of the screen, at the
level of the subject’s eyes. The input voltages for the galvanometers were specified in the
LabVIEW program, so that the laser target was accurately positioned during each of the
experimental paradigms (see section 6.3).

5.4 Summary

The data were acquired using the magnetic search coil technique, with dual search
coils to record head and eye rotations in 3-d space. Signals from the search coils were
amplified and then split into their component signals by phase detectors. The component
signals were low-pass filtered and digitized with 16-bit resolution at a sampling rate of
1kHz, prior to being stored to the hard disk of the operating PC.

A solid-state red laser, which was rear-projected onto a tangent screen placed
directly in front of the subject, provided a visual fixation point. The operating PC drove
the laser. Following each test, the data were transferred from the hard disk of the
operating PC to the hard disk of a DECstation 5000/240 for off-line analysis.

70