Low-Noise Magnetic Coil System for Recording 3-Dimensional Eye Movements

Abstract

Objective:

Vestibular and oculomotor research often requires measurement of 3-dimensional (3D) eye orientation and movement with high spatial and temporal precision and accuracy. We describe the design, implementation, validation and use of a new magnetic coil system optimized for recording 3D eye movements using small scleral coils in animals.

Methods:

Like older systems, the system design uses off-the-shelf components to drive three mutually orthogonal alternating magnetic fields at different frequencies. The scleral coil voltage induced by those fields is decomposed into 3 signals, each related to the coil’s orientation relative to the axis of one field component. Unlike older systems based on analog demodulation and filtering, this system uses a field-programmable gate array (FPGA) to oversample each induced scleral coil voltage (at 25 Msamples/s), demodulate in the digital domain, and average over 25 ksamples per data point to generate 1 ksamples/s output in real time.

Results:

Noise floor is <0.036° peak-to-peak and linearity error is < 0.1° during 345° rotations in all three dimensions.

Conclusion and Significance:

This FPGA-based design, which is both reprogrammable and freely available upon request, delivers sufficient performance to record eye movements at high spatial and temporal precision and accuracy using coils small enough for use with small animals.

I. Introduction

PRECISE, accurate and high-speed three dimensional (3D) oculography (measurement of eye movement) is often required for studying vestibular and oculomotor reflexes, which in turn are key model systems for biomedical engineering and control systems approaches to the study of sensorimotor integration and neuronal mechanisms underlying learning in the central nervous system [1], [2]. Oculography can also provide clues to diagnose pathologic conditions of the inner ear, brain, and oculomotor system, including dysfunction of the inner ear’s vestibular labyrinth [3].

Three methods are commonly used for measurement of eye movements: (1) electrical-oculography (EOG, also called electronystagmography, ENG), (2) video-oculography (VOG, also called video-nystagmography, VNG), and (3) the scleral search coil technique. Each method has strengths and weaknesses for use in clinical and research settings.

EOG uses surface electrodes placed on the face above, below and beside each eye to measure changes in electric potential generated by the corneo-retinal dipole [4]. Although this method is simple and noninvasive, it has a low accuracy for measurement of small eye movements due to high noise levels, drift, and electromyographic artifacts from blinks and facial muscle activity. Eye rotation about the line of sight does not cause a detectable EOG signal, so EOG can only measure eye movements in 2D (horizontal and vertical) [4]–[6].

Technological advancements in video image acquisition equipment have increased the use of VOG over the past couple decades [5]–[16]. The majority of VOG systems track movement of the cornea, pupil, or landmarks in the iris coloration [6], [10], [15], corneal reflections and Purkinje images [6], [9], or an added eye marker or fluorescent dye [8], [14]. In a clinical setting, VOG without an eye marker is common due to its noninvasive nature and relies on instruction to the patient to keep their eyes open. In addition, VOG recordings often contain artifacts due to blinking and head-mounted camera slippage during quick head movements. For research labs, VOG can be used if the animal is always alert and awake with eyes open; however, this can be difficult to do with an animal kept in darkness for vestibular testing, where any eye closure causes loss of data collection. Additionally, VOG recording in 3D can be difficult in animals that do not have readily visible iris striations. Markers placed temporarily on the eye’s surface or permanent markers implanted beneath the conjunctiva can overcome this problem (e.g., [8], [14], [17]); however, closure of the lid obscures the camera’s view of the marker, and artifacts from a marker hitting the lid or falling off can limit the duration of experiments and repeatability of experiments.

The “gold standard” oculographic technique is the scleral search coil method. First described by Robinson in 1963, it uses voltages magnetically induced in a coil of wire on the surface of the eye to measure eye movements with high spatial and temporal resolution [1]. Robinson used two orthogonal alternating magnetic fields sinusoidally modulating 90° out of phase with each other to induce a scleral coil voltage that he demodulated with a phase detector to determine the scleral coil’s orientation with respect to field-generating frame coils. A single coil cannot detect eye rotation about that coil’s axis, but addition of a second scleral coil enabled measurement of 3D eye movements (horizontal, vertical, and torsional) [1].

Since Robinson’s original description, other groups have developed systems using similar concepts but differing in number of magnetic fields and the demodulation method employed [18]–[20]. One commercially available system by Primelec (Zurich, Switzerland) creates three magnetic fields alternating at different frequencies, samples signals from scleral coils, and then demodulates in the digital domain using a real-time Fast Fourier Transform [21]. The design for that system is proprietary. We sought to create a digitally demodulated, high-performance, system design that could be shared freely to other laboratories upon request.

In this report, we present a new scleral coil system implementation that yields high resolution, low-noise eye movement recordings. Based on Robinson’s foundation [1], the system we describe uses three magnetic fields alternating at three different frequencies. However, deviating from the design in [1], the induced voltages in the scleral coils are demodulated in the digital domain using a field programmable gate array (FPGA) at a much higher sampling rate. This provides synchronous digital control and processing to obtain low-noise, precise, and accurate means to obtain 3D eye movements from the small scleral coils required for experiments involving small research animals and offers flexibility to customize, extend, and expand the technology for future experiments. It should be declared that this paper reuses some content from thesis [22] with permission.

II. Methods and Design

The system comprises two major parts: magnetic field driver circuitry and signal demodulation circuitry, as shown in Fig. 1A–D. It records data from up to twelve scleral search coils. The system’s field coils refer to the coil frame itself, where each field coil pair, +X/−X (front/back), +Y/−Y (left/right), and +Z/−Z (up/down), are wired in pairs to make up the three orthogonal magnetic fields, X, Y and Z (following the right-hand rule). Each magnetic field oscillates at a unique frequency. Each field induces a voltage on each scleral coil. The three signals demodulated from each coil are proportional to the cosine of the angle between the scleral coil’s axis and the axes of each of the three field coils. With a set of two coils orthogonally attached to the sclera, the 3D VOR can be recorded and analyzed with 3D rotational kinematics equations described by Haslwanter and Migliaccio [23], [24].

Fig. 1.

Block diagram of the coil system architecture. FPGA outputs drive MOSFET switches in the magnetic field driver circuity at three different resonant frequencies for the X, Y, and Z magnetic fields. The induced eye coil current is the sum of three components at different frequencies, plus their harmonics, with relative magnitudes dependent on the coil’s orientation relative to each of the three fields. After amplification and analog-to-digital conversion, the FPGA demodulates the coil signal into the individual X, Y, and Z components required for 3D rotational kinematics analysis. The circles labeled A-D refer to test points shown in Fig. 3. E shows the scleral coil design schematic, with copper magnet wire comprising 20 turns with one final turn of stainless-steel wire before coming off as leads. F shows where scleral coil pair are sutured to each eyeball.

A. Scleral Coil Design

Each scleral coil was made using a base of 20 turns of 42 awg copper magnet wire coiled to form a ~2 mm inner and ~4 mm outer diameter circle, ~1mm tall annulus (Fig. 1E). Eyes move thousands of times per day, and copper wire cannot endure that without breaking. We therefore soldered ~30 cm of multi-stranded, PFA-coated, stress annealed 316 stainless-steel wire (AM Systems 793200) to each end of the magnet wire then wound each of those leads once around the outer edge of the coil before tightly twisting them to minimize magnetic flux through anywhere other than the coil itself. We potted the soldered joints and coil assembly in enamel and then silicone. A diagram of this construction is shown in Fig. 1E.

B. Hardware

1). Magnetic Field Frame:

The magnetic field frame is a cube, 30.5 cm on each edge, and made of single turn solid aluminum square rods (1×1 cm cross section). Each field coil pair, +X/−X (front/back), + Y/−Y (left/right), and +Z/−Z (up/down), is wired in series to make a two-turn inductor. Although previous designs [1], [18], [19] use multi-turn field coils to create a larger magnetic field, this single turn frame and the circuitry described below are more than sufficient for driving a large magnetic field.

2). Magnetic Field Driver Circuitry:

The magnetic field for each of the three field coil pairs is driven by a MOSFET switching circuit operating at the three field frequencies, X: 245 kHz, Y: 498 kHz, Z: 763 kHz. The circuit uses a high-speed MOSFET driver (Texas Instruments, TPS28225), driven by a pulse width modulated (PWM) digital input from a FPGA (Xilinx Virtex-4 XC4VFX12-ST363), as the basis to control the switching of two N-MOSFETS (Texas Instruments, CSD18503KCS) at the resonant frequency of each field’s capacitor/inductor pair. The single-turn aluminum frame coil design has low inductance compared to other coil system designs. We used high carrier frequencies to avoid needing very large, tunable, high Q, low loss capacitors. We chose field frequencies compatible with standard off-the-shelf fixed capacitor values. Polypropylene film capacitors (KEMET Corporation) set the resonance frequency with the field coil inductance to generate each magnetic field. High-speed optocouplers (Vishay SFH6702) isolate the FPGA and field driver to prevent noise from ground loops.

3). Scleral Coil Signal Processing:

Scleral coil voltages are filtered and amplified by ultra-low noise amplifiers (Analog Devices, AD8331) before digitization and demodulation. The AD8331 includes a single-ended pre-amplifier followed by a variable gain amplifier (VGA) and a selectable gain post-amplifier. A software command from a personal computer user interface sets VGA gain using a digital-to-analog converter (Texas Instruments, DAC7574). For the scleral coil design described above, the overall amplification is configured to provide ~100x gain of raw scleral coil signals. Bandpass filters are located at three stages throughout the amplification circuitry, with overall cutoff frequencies between 24 kHz and 1.5 MHz. After amplification, each coil’s amplified signal is digitized using a 12-bit, high-performance analog-to-digital converter (ADC, Texas Instruments, ADS5242). The FPGA then simultaneously samples up to 12 scleral coil signals at 25 Msamples/sec each and then demodulates them using a multiply accumulate (MAC) unit. This high sample rate improves noise performance via two variations on oversampling. First, the 25Msamples/S rate is about 4–8 times the Nyquist rate required for sampling the output of the 24kHz-1.5MHz bandpass preamplifiers. Oversampling by 2²ⁿ = 4 times the Nyquist rate reduces ADC quantization noise and effectively adds n≈1 bit to the 12-bit ADC’s effective resolution, depending on the spectral content of the signal and the degree to which the noise is uncorrelated with the signal. [25],[26] The MAC unit multiplies each signal by three pseudo-sinusoids (with the same frequency and phase as the field) to extract each component of scleral coil angular position. To report X, Y, and Z coil signal components at a final rate of 1kSample/s, the FPGA performs a MAC operation over N=245, 498, and 763 cycles, respectively, of the scleral coil signal multiplied by the corresponding field coil driving signal. To the extent that noise at this stage in processing is uncorrelated with signals, this averaging would yield a √N≈16–28-fold reduction in noise amplitude relative to signal amplitude In addition to reduced noise, this approach also achieves much higher bandwidth relative to [14], which handled noise by passing analog-demodulated signals through a bank of 200Hz 8-pole Butterworth analog low-pass filters.. User interface software written in C and running on a PC acquires the demodulated signal at 1 kHz via a UM232H-B-NC serial-to-USB interface (Future Technology Devices International Limited).

The FPGA also provides digital pulse train driving signals for the magnetic fields. The pulsatile driving signals operate at 20% duty cycle to reduce high-current load on the MOSFET circuitry while generating strong enough fields to obtain a scleral coil signal well above the system’s noise floor. This duty cycle is software programmable. Adjusting it can increase field intensity, increasing signal strength for smaller scleral coils.

C. Surgical and Experimental Methods

Animal experiments were performed in accordance with a protocol approved by the Johns Hopkins Animal Care and Use Committee, which is accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC) International and consistent with European Community Directive 86/609/EEC. Under general anesthesia (isoflurane, 1.5%−5%), an adult chinchilla was fit with a head post (to stabilize the head during experiments) and head cap (to enclose the scleral coil wires between experiments) using dental acrylic. Two scleral coils per eye (four total) were sutured to the sclera within subconjunctival pockets at approximately the temporal and nasal aspects of the eye (Fig. 1F), away from extraocular muscles. To ensure that the eyes are not tethered or otherwise restrained by the stainless-steel leads, we add an additional loop of wire within orbit fat to provide slack in the system. After suturing the coils in place on the sclera and before closing conjunctival incisions, we apply passive movements to each eye to confirm that the eye can move freely over a normal oculomotor range without palpable restraint by the stainless-steel wires. Conjunctiva was closed using fast gut suture. Stainless-steel twisted pair leads from the coils were routed subcutaneously from the orbit to the head cap. The animal recovered for 10–14 days before recording sessions. After the animal fully recovered from surgery, we check whether the eyes are mechanically restrained by testing the vestibulo-ocular reflex and comparing eye movement responses to normative data in [27] for the same species.

During experiments, each animal was head-fixed in a plastic enclosure, with its head centered in the field coil frame (Fig. 2). The animal and surrounding coil frame are moved together using a 6DOF Motion Platform (Moog 6DOF200E) either translating or rotating sinusoidally along or about the three cardinal axes at 1–2Hz with 3m/s² or 30deg/s amplitude respectively. Gains were collected at the beginning of each experiment by facing a ‘dummy coil’ of the same design as the implanted scleral coils into each of the three magnetic fields to get the relative maximum magnitudes of the three fields. Offsets were determined from recordings with the animal in the coil frame and the implanted coils connected to the VGA. The first recorded was with each scleral coil connected and the second was with each scleral coil connector flipped. Offsets arising from capacitive effects of the coils’ presence in vivo or caused by other signal pick-up other than through induction in the scleral coil itself were calculated by taking the difference in the recorded positions after connectors were flipped to get the full range of demodulated output voltage. Considering the safety of our animals, the maximum magnetic field for our experimental setup is 84.8 μT in the center of the frame, well below the strength of a clinical 1.5–3 Tesla MRI machine and on the same order of magnitude of Earth’s magnetic field.

Fig. 2.

(A) a photo and (B) a sketch of the experimental setup. All components are fixed to the 6DOF Moog Motion Platform that can translate or rotate along/about any 3D axis from center (so that the animal head never moves with respect to the coil frame). The animal is head-fixed and secured to a chair that is positioned so that the head is in the center of the 3-field coil frame (cube). Four twisted-pair stainless steel leads each come from a scleral coil implanted on the sclera of the animal eye (two coils on each eye as shown in Fig. 1EF) and leave the coil frame to connect to amplifiers, ADC, and demodulation circuitry as depicted in Fig. 1CD.

D. Data Analysis Methods

Demodulated X, Y and Z components from scleral coils were analyzed using software written in Matlab (Mathworks, Natick, Massachusetts) using 3D rotational kinematics to convert the six components from each pair of scleral coils to rotation vectors representing 3D eye orientation. The two 3-vectors of signals for scleral coils on one eye were orthogonalized prior to further analysis. [23],[24],[28],[29].

III. Results

Table I summarizes bench testing results, which unless otherwise specified were collected using scleral coils of the design detailed above. While the system has the capability to record from up to 12 coils, all testing was done using only four coils, as is our normal experimental setup (Fig. 2).

TABLE I.

System Characteristics

Field Coils	X Field	Y Field	Z Field
Size of Frame (cm)	30.5	30.5	30.5
Inductance @100 kHz, μH	2.00	2.23	2.02
Capacitor, nF	222.4	47.2	22.3
Resonant frequency, kHz	245.3	497.8	762.9
Q-Factor (@ 100 kHz)	20	24	25
Typical Scleral Coil Size	21 turns, 3–4 mm diameter
Scleral Coils (up to 121)
Coil/Lead Resistance2	Typically 100 – 200 Ohm
Sampling Frequency	1 kHz on each of 12 Channels
Power Consumption
Driver Board3, W	4.35	2.65	2.1
Optoisolator Board, W	0.825 for all channels
Amplifier+Demodulator4, W	7.15 for all channels
Demodulated Scleral Coil Signals
Maximum Coil Voltage, mV	20.0	37.2	51.6
Input Referred Noise (peak-peak, with amplifier input shorted)
μv	5.0	4.7	5.0
equivalent coil angle	0.0143°	0.0072°	0.0055°
Facing Field Noise5 (peak-peak, scleral coil aligned with field coil)
μV	12.7	12.9	26.0
equivalent coil angle	0.036°	0.020°	0.029°
Perpendicular Field Noise6 (peak-peak, scleral coil 90° vs field coil)
μv	9.8 μV	5.8 μV	6.4 μV
equivalent coil angle	0.028°	0.0089°	0.0071°
Worst Case Linearity	Roll:	Pitch:	Yaw:
Error	0.099°	0.099°	0.088°
Crosstalk Between Adjacent	0.0085%	0.012%	0.0099%
Amplifier Channels
Crosstalk From Other	Y:0.60%	X: 0.10%	X:0.070%
Fields After Demodulation	Z: 0.12%	Z:0.058%	Y:0.001%

¹Values reported here are with only 4 coils connected (our normal setup)

²DC resistance is mostly due to the stainless-steel leads from the coil.

³Using 5V power supply

⁴Using 12V power supply

⁵With scleral coil parallel to field, yielding maximum signal.

⁶With scleral coil perpendicular to field, yielding minimum signal.

A. Coil System Characteristics

Voltages in Fig. 3 were measured at test points throughout the system to show the general progression of the signal that generates the magnetic field. Each of these system test points are shown in Fig. 1. Each column was recorded with only one axis of the magnetic field on at a time. The PWM signal leaving the FPGA (Fig. 3A–C) drives the MOSFET switch at the resonant frequencies of 245.3 kHz, 497.8 kHz, and 762.9 kHz for the X, Y, and Z fields respectively. At these relative resonant frequencies, the voltage at each field’s capacitor rings up to 30.8 V, 42.4 V, and 48.8 V (Fig. 3D–F). The quality factor (Q) measured at 100 kHz is 20, 24, and 25 for the X, Y, and Z fields respectively. Fig. 3G–I shows raw, unfiltered voltages induced on the scleral coil when placed in the center of the magnetic field frame facing directly into the X-, Y-, and Z-field. At this stage, high frequency switching noise from the MOSFET switch is large; however, it is significantly reduced after filtering (Fig. 3J–L), digital averaging and synchronous demodulation.

Fig. 3.

Signals recorded at four stages in the system with only one field on at a time. Signals in each row are from test points labeled in Fig. 1. (ABC) The FPGA produces a digital pulse to control the frequency of each directional magnetic field (test point A in Fig. 1). (DEF) The resonant circuit creates a magnetic field that rings up to create a quasi-sinusoidal magnetic field (test point B in Fig. 1). (GHI) Raw, unfiltered signal of a scleral coil facing directly into the X, Y, and Z fields (test point C in Fig. 1). High frequency noise caused by MOSFET driver switching is evident; however, after the raw signal is amplified and filtered (JKL; test point D in Fig.1) high frequency noise is attenuated before digitization and demodulation.

To determine the noise contribution due to the field drivers, we measured the input-referred noise of the amplifiers with the coil frame’s magnetic field drivers off and the input to the amplifier shorted. The input-referred noise of ≤5 μV peak-to-peak is equivalent to ≤0.014° peak-to-peak at the output of the system. With the fields turned on and a scleral coil connected to the amplifier input, each demodulated component’s peak-to-peak input-referred noise with the coil’s axis aligned with the X, Y and Z fields, respectively, increases to 12.7 μV, Y: 12.9 μV and 26.0 μV, equivalent to peak-to-peak angular position noise at the system output of 0.036°, 0.02° and 0.029°. Fig. 4 compares system outputs for a stationary scleral coil aligned with the Z field of the new system in comparison to the old analog coil system. [14] With the scleral coil’s axis perpendicular to each field, to eliminate induced voltage from that field, the noise floor of that component decreases. The peak-to-peak maximum induced voltage on the scleral coil when facing into the X, Y, and Z fields is 20.0 mV, 37.2 mV, and 51.6 mV respectively, producing a signal to noise ratio (SNR_X: 2.48e6, SNR_Y: 8.32e6, SNR_Z: 3.94e6) to allow recording of small eye movements.

Fig. 4.

Z-component of a demodulated scleral coil signal, for the analog coil system described in [14] (blue) and the new system described in this report (red). In (A), signals were measured using the same scleral test coil and leads, made per the design in section IIA with the coil axis facing the Z field. Noise floor is ~0.3° peak-to-peak for the analog system and ten-fold smaller for the new system, at <0.036°. In vivo noise comparisons are shown while the animal and its eyes were stationary in (B) and while the animal (and therefore its ocular reflex) is sinusoidally rotating about the Z axis with a 2Hz sinusoidal motion stimulus of velocity amplitude 30 deg/s in (C). The analog system measurements are from 80-turn glue-on coils as described in [14] to provide a system-wide comparison.

The step response of two different coil systems is shown in Fig. 5. For both panels, each data point represents a sample from the demodulated signal with the scleral coil positioned to face approximately equally into each of the magnetic fields. In the Fig. 5A’s coil system [14], which followed the design of Remmel [19], a monkey sized scleral coil (5 turns, 13 mm diameter) was used and the induced scleral coil voltage passes through an eight-pole Butterworth anti-aliasing filter with a 100 Hz corner frequency, detailed in [14]. Similar to the step response shown for the original Robinson design [1], there is ringing before the signal settles. In our design, Fig. 5B, based on the high A/D input sampling rate and subsequent multiply and accumulate operation of the FPGA firmware (integrating and therefore effectively averaging over 25,000 ADC samples to produce each output sample), there is only one demodulated output data point during the transition from shorted to scleral coil input, thus eliminating any ringing of the signal after the step input. In the new system, each data sample is independent of its neighbors – there is no inter-sample filtering performed by the coil system.

Fig. 5.

Comparison of the step response of an analog coil system [14] (A) and the step response of the new digitally demodulated coil system described in this report (B). For each example, the demodulated value recorded from a scleral coil is shown after being normalized to the maximum magnetic field strength for each axis of the magnetic field. The scleral coil was connected to the amplifier/demodulation circuitry facing approximately equally into each of the three fields. The recording started with the input shorted and then released to the coil input to show the step response dynamics of the system. The response of the analog system shown in the top panel illustrates ringing due to filtering of the analog demodulated signals (also shown in [1]). In the new system, due to the sampling rate of the coil signal (25Msamples/sec) and digital domain averaging over 25kSamples to produce each independent 1 kHz output sample, further filtering of the demodulated signals is unnecessary, and the step response rises to the final value in just 2 samples with no overshoot or ringing.

Crosstalk between demodulated field components for a scleral coil signal was measured with one field on at a time and is reported as a percentage of the intended demodulated signal. For example, with only the X-field on, a scleral coil was pointed directly into the +X and −X direction. The ratio of undesired Y and Z demodulation is presented as percentage of the total magnitude of the X demodulated value. Measurements repeated with each of the three fields yielded crosstalk < 0.6% (Table I).

Channel-to-channel crosstalk between the four scleral coil amplifier channels was measured with three of the channel inputs shorted, while a scleral coil was connected to the fourth channel. The scleral coil was moved to face into each of the three magnetic fields and any measurable crosstalk on the shorted channels was presented as a percentage of the maximum value. Crosstalk between amplifier channels was below the noise floor. The maximum measurable crosstalk between scleral coil channels was 0.012%.

The variation in field strength as a coil is moved along the center of each field is shown in Fig. 6. The voltage induced in a large coil (11 turns, 2 cm diameter) was measured at every 2.54 cm along each field’s center axis. A large coil was used here to obtain high induced voltage for simplicity in mapping the field since the general homogeneity is characteristic of the magnetic field and not the specific scleral coil used.

Fig. 6.

Magnetic field strength measured at different points within the field using a large test coil (11 turn, 2 cm diam.) and shown as the peak-to-peak voltage difference from the value measured at the center of the frame (X=236mV, Y=480mV, Z=512mV). With only one field on at a time, the field strength was measured as from one frame face to the other along the center of the field. +X is out the nose, +Y is out the left ear, and +Z is up. For example, the X trace shows values recorded from −X frame (−6, 0, 0) to +X frame (6, 0, 0), where (0, 0, 0) is the frame center. Field strength is highest at the edges, where the scleral coil is coplanar with one face of the frame coil pair creating the field. Asymmetry seen between−Z and +Z frames is due to the metal platform atop which the frame is mounted.

To test the non-linearity of the system, a rotary motion stage (Zaber LMR39, resolution of 0.000234 deg) rotated two roughly orthogonal scleral coils in 1.5° steps for a total 345° rotation (motor limitations prevented the full 360°) around three axes: roll (X), pitch (Y), and yaw (Z). The motor itself was located outside of the field so the metal would not distort the field. The demodulated X, Y, and Z components on a single coil rotating through roll, pitch, and yaw are shown in Fig. 7A–C. This normalized data from the single coil shows a representation of the proportionality of each magnetic field induced during a pure rotation in the indicated direction. The zoomed in portion within each window of Fig. 6 shows the detail of the 1.5° steps of the motor. Using the second coil that was held orthogonal to the first during the rotation, the coil system’s measurement of the 3D rotational movement of the motor was found using rotational kinematics (detailed in [23], [24],[28]), shown in Fig. 7D–F. The residuals, Fig. 7G–I, were calculated by the measured step size in the middle row versus the ideal 1.5° step, indicating the error in linearity. Non-linearity remained <0.1° for all three − − Z.rotations for the entire 345° range rotated. This compares favorably to non-linearity of 0.3° in the Primelec system [21].

Fig. 7.

To measure system linearity, two mutually orthogonal scleral coils were centered in the middle of the frame. A Zaber LMR39 motor with 0.000234° resolution was programmed to rotate in 1.5° steps, holding each position for 3 sec, over 345° (limited by the motor, not the recording system) about the X (roll), Y (pitch) and Z (yaw) axes, starting with one scleral coil’s axis aligned with the Z, X and X axes, respectively. (ABC) Demodulated components of the normalized raw signal collected from one of the two scleral coils during each rotation. Insets show the change in the recorded coil signal as the motor rotated through 5 steps. (DEF) Post-processing of the demodulated data recorded from both coils gives the 3D rotation. (GHI) Error residuals for rotation of scleral coil as measured by the coil system compared to the motor’s 1.5° step rotation. Worst-case linearity error is < 0.1°.

B. Animal Data

Using approximately orthogonal dual scleral coils implanted binocularly in a chinchilla, the vestibulo-ocular reflex was recorded with the new coil system. Fig. 8 shows the 3D eye angular position and velocity traces recorded from an animal during Yaw rotation (Fig. 8ABEFIJ) and a left/right lateral translation (Fig. 8CDGHKL). With the improvements in this coil system, we can precisely measure the very small eye movements that occur during translational movements. In comparison, Fig. 4 details what a stationary and Yaw position trace looked like on the older analog system [14], showing noise on the same order of magnitude of some of the positional changes in response to translation (Fig. 8GH).

Fig. 8.

Example recordings of angular vestibulo-ocular reflex (ABEFIJ) and translational vestibulo-ocular reflex (CDGHKL) recorded from a chinchilla with dual scleral coils implanted binocularly. The motion sensor data (A-D, shows the velocity during a 1 Hz yaw rotation of peak 30 degrees/sec (dps) and the 1 Hz sinusoidal lateral translation (left/right) with peak acceleration of 3 m/s². E-H shows the unfiltered rotational eye position recorded by the coil system. The primary elicited eye movement during yaw rotation is in the about the yaw axis with a gain of ~0.7 which is typical for chinchillas. The eye movement elicited during translation is much smaller, but still captured by this scleral coil system. Proving capable of measuring small eye movements during translations. I-L shows angular velocity calculated from positional signals in E-H.

IV. Discussion

In laboratory based vestibular research, it is important to be able to record precise eye movements from small animal models. Each method of eye movement recording, from EOG, VOG, and scleral search coil has its strengths and weaknesses for use in an animal model. With the “gold standard” of the scleral coil system [1] and ability to record repeatable eye movements from the same animal over time, robust small coils were developed to work with such a system. To keep these coils small while also using stainless steel wire to prevent breaking after implantation, the number of turns and diameter were minimized to 21 turns at 3–4 mm diameter. These smaller scleral coils do not provide a sufficient signal to noise ratio for measuring small eye movements in our previous system which was typically used with larger diameter implanted coils in non-human primates or 80-turn, glue-on scleral coils for smaller animals. Therefore, this new system was developed with a simple magnetic field driver scheme and straightforward demodulation method to provide low noise floor, high linearity, and ability to customize many features for other applications.

The design of the magnetic field driver circuitry uses a simple digitally driven pulse to drive the entire resonant circuit that provides for a magnetic field ring up. The switching noise could be fully eliminated by changing the driving source to a sinusoidal input, but that proved unnecessary, because the high frequency switching noise is mostly eliminated by bandpass filters before amplification, as shown in Fig. 3.

The inhomogeneity of the field is as expected using the cube field design with higher field strength closer to the six faces of the cube. For experiments with a head-fixed animal, the inhomogeneity is not a problem because each eye rotates about a fixed point. To further mitigate effects of field inhomogeneity without changing the design of the magnetic fields as described in [30], a reference coil placed near the animal’s eyes can be used instead of comparing the measured coil signal to the pseudo-sinusoid in the FPGA. This would help reduce all influences of inhomogeneity within the field and allow for experiments with a freely moving animal inside the frame.

The use of a single-ended amplifier requires the measurement of coil offsets at the beginning of each experiment. This is due to capacitive influences coupled into the single-ended input. This offset must be subtracted from each demodulated value before completing rotational kinematics calculations. Future development will investigate using a differential amplifier to avoid the need to measure offsets of implanted scleral coils.

The new coil system presented here provides the ability to record 3D eye movements with a much lower noise floor than our previous system when using these new scleral coils (0.036° vs 0.2° respectively). The low noise floor was achieved by using a strong magnetic field and by oversampling the raw scleral coil signal at 25 Msamples/sec, then averaging over 25 kSamples to give 1 kHz scleral coil sample. With this approach, sample rate can be increased or decreased based on the needs of the experiment. We used a 1 kHz sample rate because that is sufficient for recording quick phases of eye movements while still providing a low noise floor. Signal to noise ratio could also be increased by using larger scleral coils.

While the coils used for measurement differ, Table II provides specification comparisons between the foundation of this work [1], a commercially available system [21], and the work presented here. Our system – with the use of very small scleral coils – allows 360 degrees of measurement in any rotational direction with precise linearity and has a similar level of noise to the commercial system [21]. However, the size and number of turns of the coils used to measure the specifications for [21] are unknown.

TABLE II.

System Comparisons

	Robinson 1963 [1]	Primelec [21]	System Described by this Report
Eye Coil(s) Used to Measure Specifications	10 turn/coil 12 mm diam Contact lens	Not specified	21 turn 3–4 mm diam
Dimensions Measured	3D	3D	3D
Technology	Phase Detector	Digital Frequency FFT	Digital Frequency Demodulation
Input Referred Noise (Amplifiers Shorted)	< 2 uV*	--	5 uV
Noise	-	< 0.024 deg p-p “Full Bandwidth” (frequency unspecified, coil not facing field)	0.036 deg p-p (facing field, NO inter-sample filtering)
Linearity	2% full scale	Error ±0.3 deg	Error < 0.1 deg (<0.028%)
Recording Range	--	360 deg HZ ±80 deg Vert	360 deg all directions
Step Response Rise Time	0.5 ms with <10% overshoot	--	2 ms 0 overshoot

^*Robinson defined resolution as the induced scleral coil signal/input referred noise

V. Conclusion

The low noise floor, linearity, and precision of our system improved our ability to record small eye movements with implanted scleral coils in chinchillas. The system design uses off-the-shelf components. Customization of amplification, resonant frequency and sampling frequency can be completed by changing programmable FPGA settings and capacitors. This coil system design gives a worst-case noise floor of 12.7 μV (equivalent to ~0.036° of eye movement) and non-linearity <0.1° using a robust, small scleral coils.

Biographies

Kristin N. Hageman received the B.S. degree in biomedical engineering (with a specialty in bioelectrical engineering) from Case Western Reserve University, Cleveland, OH, USA, in 2011. She received the Ph.D. degree in biomedical engineering from Johns Hopkins University, Baltimore, MD, in 2017.

Her interests include neural stimulation and medical device development. Currently, she is a Senior Biomedical Engineering at Medtronic, working on spinal cord stimulation systems within the neuromodulation team in Minneapolis, MN.

Margaret R. Chow received the B.S. degree in biomedical engineering from the Johns Hopkins University, Baltimore, MD in 2014. She is currently pursuing the Ph.D. degree in biomedical engineering at Johns Hopkins University, Baltimore, MD.

Her research interested include neural stimulation and medical device development, focusing on developing future iterations of a multichannel vestibular prosthesis.

Dale C. Roberts received the B.S. degree in computer science from the University of Maryland, Baltimore County, MD in 1989 and the M.S. degree in computer science from the Johns Hopkins University, Baltimore, MD in 1994.

He is currently a Senior Research Systems Engineer in the Johns Hopkins School of Medicine department of Neurology with research interests in neural stimulation and neural recording, specifically focused on vestibular and auditory neuromodulation and acts as a senior engineer in multiple Johns Hopkins labs including the Vestibular NeuroEngineering Lab, the Vestibular and Ocular Motor Research Lab, the Laboratory of Vestibular NeuroAdaption, and the Cullen Lab.

Charles C. Della Santina received the Ph.D. degree in bioengineering, in 1994, from the University of California, Berkeley, where his work focused on development of micromachined silicon devices for chronic multiunit interfacing to the auditory/vestibular nerve. He received the M.D. degree from the University of California, San Francisco, in 1997, and completed residency at the Johns Hopkins School of Medicine, Baltimore, MD, in 2002.

Since completing the M.D. degree and residency, he has been a clinician-scientist at Johns Hopkins, where he is Director of the Johns Hopkins Vestibular NeuroEngineering Lab and the Johns Hopkins Listening Center and a Professor of otolaryngology—head & neck surgery and biomedical engineering. His research focuses on vestibular neurophysiology and development of a vestibular prosthesis for restoration of labyrinthine sensation. He holds a founding interest in Labyrinth Devices LLC.