An FPGA-Based Vision Prosthesis Prototype: Implementing An Efficient Multiplexing Method For Addressing Electrodes

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Proceedings of the 2005 IEEE

Engineering in Medicine and Biology 27th Annual Conference
Shanghai, China, September 1­4, 2005

An FPGA-Based Vision Prosthesis Prototype: Implementing an Efficient


Multiplexing Method for Addressing Electrodes

Y.T. Wong1, G.J. Suaning1,2, S. Dokos1, P.J. Preston1, N. Dommel1, D. Grace1, N.H. Lovell1,3
1
Graduate School of Biomedical Engineering, University of New South Wales, Sydney Australia
2
School of Engineering, University of Newcastle, Newcastle Australia
3
National Information and Communications Technology Australia (NICTA)

Abstract — A prototype of an epi-retinal vision prosthesis different stimulation strategy from the conventional serial
based upon an efficient electrode addressing schema has been mode is needed if patterns are to be conveyed to a patient
developed. This system has the ability to stimulate multiple without flicker.
electrode regions simultaneously, hence greatly improving the This group [6][7] has been developing alternate schemes
maximum rate of stimulation compared to many currently for stimulation with the most promising results emerging
available neural stimulation devices based on serial stimulation from concurrent stimulation based schema, commonly
protocols. To minimize the problem of cross talk between
referred to as a parallel mode of stimulation. Parallel
stimulating electrodes, a hexagon layout of electrodes was
implemented. Basic tests were completed using a field stimulation offers a considerable increase in stimulation
programmable gate array logic system driving analogue throughput and is scaleable for implants of larger electrode
circuitry to inject current into physiological saline via quantities.
electrodes in hexagon arrangements and in a simple paired A drawback of this method is that there exists a paucity
arrangement. The hexagon layout of electrodes was shown to of data on the effects of simultaneous stimulations in
clearly reduce the interaction between multiple current sources excitable neural tissue or models thereof within the
and hence cross talk. literature. It is known however, that without careful
consideration of electrode design and stimulation
Keywords – vision prosthesis, FPGA, electrodes parameters, electrical cross talk between multiple current
sources and their respective return electrodes will occur.
I. INTRODUCTION This electrical cross talk will greatly reduce the
Most current neural stimulation devices multiplex a predictability of current flow in stimuli to localised regions.
single current source across a few electrodes, commonly In [6], an efficient method to multiplex 14 current
referred to as a serial mode of stimulation [1][2]. This sources between 98 electrodes is postulated. This paper
method is sufficient for applications where a low stimulation illustrates how a prototype of this system was developed and
rate is required such as in cardiac pacemakers but is unlikely tested using field programmable gate array (FPGA) logic
to suffice for an epi-retinal prosthesis owing to its high and associated analogue stimulation circuitry. Furthermore,
electrode quantities and rate requirements. this system was used to investigate the effects of delivering
The critical flicker fusion (CFF) frequency is the current through physiological saline with various electrode
frequency in which image fusion is perceived and a arrangements such as hexagonal mosaics and a simple
“flickering” light interacting with a person’s retina appears electrode pair configuration.
continuous. The CFF frequency of a normal sighted person
varies with luminance in accordance with the Ferry-Porter II. METHODOLOGY
Law and is approximately 20-40Hz at normal luminance. The prototype system was developed based on an
The CFF is not a constant frequency but changes with efficient multiplexing scheme for 14 current sources and 98
luminance and stimulus size and differs from person to electrodes arranged in a hexagonal layout. This system
person even under the same conditions [3][4]. In a person could be scaled for even larger electrode numbers at
with a diseased retina wherein vision is conveyed via neural minimal, and in some applications, no expense to data
stimulation of retinal neurons, this CFF frequency can be up requirements.
to two or three times that of a normal observer (equivalent The electrodes are grouped into “clusters” with a center
CFF - eCFF) [5]. electrode, referred to as the stimulating electrode, and six
Due to this eCFF, the required stimulation rates that an surrounding electrodes serving as guards (return electrodes).
epi-retinal prosthesis needs to provide will exceed the One current source is assigned to each of these 14 clusters
limitations of serial mode stimulations. Compounding this with each current source having the ability to be multiplexed
problem, is the large number of electrodes that will need to to seven different electrodes, corresponding to the seven
be addressed in such a prosthesis when compared to the methods of grouping the electrodes in 14 clusters. Further,
single electrode systems used in cardiac pacemakers and the to facilitate using any electrode within the array as a
tens of electrodes used in cochlear neural prostheses for the stimulating electrode, each electrode may be connected to as
deaf and hearing impaired. It is highly probable that a many as two additional current sources from neighboring
hexagons.

0­7803­8740­6/05/$20.00 ©2005 IEEE. 5268
This prototype system is comprised of four main
components, a radio frequency (RF) receiving circuit to
capture incoming data signals, controlling circuitry
implemented in FPGAs, 14 current sources implemented
with digital to analogue converters (DACs) and CMOS
switching circuitry to allow the DACs and current return
paths to be connected to specific electrodes. In the present
study, two of the 14 hexagons were used to investigate the
electrical cross-talk properties in saline. Figure 2. Parallel stimulation protocol which includes a single packet for
the stimulus delivery times of all 14 clusters.
The prototype design is illustrated in Fig. 1 with the
addition of a camera/processor/transmitter unit designed to
capture, process and then transfer images across the eye
tissue using an RF link [7].

Figure 1. The full visual prosthesis prototype system starting at the left
with a camera for image acquisition, then a unit to process images into an
encoded RF signal for a transmitter to send to the implant. Both the Figure 3. Serial stimulation protocol which includes two separate packets
external and implanted systems have been designed, produced and tested. for the stimulus delivery times of each of the 14 clusters.

The FPGA unit decodes incoming RF packets that are The prototype has been designed with special modes to
received into signals indicating the stimulation and current allow control of as many stimulation parameters as possible.
return positions, and the duration and strengths of the current Two of the more significant modes are an electrode shorting
comprising the stimuli. The FPGA acts upon these signals mode and a DAC scaling mode. When stimulating in retinal
to close and open switches in the multiplexing circuit, tissue it is important that any charge that is introduced into
thereby allowing current to flow through the electrode array the system is subsequently removed. This is accomplished
and retinal tissue. Stimulation data is encoded using a with accurate current sources and a balanced bi-phasic
simple pulse counting method whereby a number n is current injection. Should residual charge remain on the
represented by 8n+8 packets of RF [7]. While this encoding electrodes, the shorting mode allows all electrodes to be
scheme has significant redundancies, it does offer an shorted to the implant’s ground between stimuli, thereby
increased error detection ability which is desirable for an dissipating excess charge.
implant destined for human use where conservative Inter-stimulus shorting was found to be the most
approaches are prudent. appropriate type of shorting for this prototype with shorting
Both a serial schema and parallel schema for stimulation occurring before and after the current injection as in Fig. 4.
have been implemented in this system. The protocol for the The prototype does not keep state between sequences for
parallel mode is described in [6] and offers a considerable safety reasons, so to increase the electrode shorting time the
increase in stimulation throughput over the serial protocol. shorting enable is programmed to reset two sequence
Figs 2 and 3 demonstrate the reduction in data that is made packets after a global reset occurs at the beginning of each
when using the parallel mode. sequence.
By referencing the 14 stimulus delivery times with one The DAC scaling mode allows the current injection to be
single packet and stimulating the 14 clusters in parallel, the changed from the normal operating range of 5 µA to 100 µA
prosthesis stimulation rate is effectively increased by a to a scaled mode delivering current from 100 µA to 2 mA
factor of 14 over that of serial stimulation. thus adding to the versatility of the system that is being used
As the electrodes are split up into 14 identical clusters, for both visual and functional applications.
electrode addressing overhead is also reduced as all 14
stimulating electrodes can be described with three-bits
indicating to which of the seven possible configurations the
clusters will be assigned. A further three bits specify the
return path configuration.

5269
measured using a Fluke 77 digital multimeter for the
differing current injection parameters.

Figure 4. Shorting mode signals for the inter-stimulus shorting protocol.


Shorting of the electrodes occurs preceding and following the bi-phasic
current injection.

Table I lists the varying number of data packets that


need to be transmitted to the prosthesis for each of the
operating modes. Each instruction packet varies in time Figure 5. Electrode positions for the current injection experiment into saline
using two full hexagon clusters with 6 guard electrodes each. The solid red
from 3.2 µs to 25.6 µs and the stimulus delivery times will circles indicate that current is flowing into these electrodes in the first phase
vary from 5 µs to 2 ms. All modes use the same first 32 of the injected sine wave and the striped blue circles indicate that current is
packet protocol but it is evident that the series stimulation flowing out of these electrodes in the second phase.
mode is less efficient in terms of the amount of data needing
to be transferred to the implant compared to the parallel III. RESULTS
stimulation mode. As electrode numbers increase the The controlling logic for the prototype system has been
difference between the serial and parallel mode protocols tested using an Altera Hardware Description Language
will also increase. (AHDL) simulation and implemented on an Altera Flex
EPF81188. All valid inputs and boundary case invalid
TABLE I
STIMULATION MODE RF PACKETS inputs have been simulated returning logically correct
Mode Sync Stimulus Return responses in accordance with the design specification. To
Position Position confirm the logic, the FPGA was connected to a DAC and
Parallel 1 1 2 switching board. A balanced bi-phasic current has been
Series 1 1 2
Shorting 1 1 2
switched through a two hexagon electrode arrangement.
Scaling 1 1 2 Fig. 6 illustrates the voltage profile of saline resulting
Mode DAC Special Stimulus Total from a balanced 1 mA peak to peak 1 KHz sine wave current
Programming Mode Delivery injection with two hexagons and a full guard of electrodes.
Selection Times
Parallel 28 1 2 35
Series 28 2 28 62
Shorting 28 3 2 37
Scaling 28 4 2 38

To examine the interaction of the DACs when used in


parallel stimulations, currents were injected into an electrode
array comprising of 14 platinum ball electrodes with
diameters of 430 µm + 30 µm arranged in a two hexagon
patterns as shown in Fig. 5. The electrodes had a horizontal
separation of 1000 µm and a vertical separation of 866 µm
allowing each electrode center to be separated by 1000 µm.
The electrodes were immersed in a thin layer of
physiological saline.
A 1 kHz sine wave signal was injected into the center
hex locations and the current returned using a single
electrode or a full guard configuration. The peak to peak
voltage levels of the saline with respect to a distant ground
were sampled in 3600 locations in an equi-spaced grid of 60
x 60 using a silver-silver chloride measuring electrode fixed
to an XY indexer, with 1mA peak to peak being injected into Figure 6. Voltage profile for a current injection using two hexagon clusters
both stimulating hexagons [8]. Voltages were measured with 6 “guard” electrodes for each cluster.
using a National Instruments PCI-6254 M Series data
acquisition card with the measuring electrode moved by an Table II lists the return currents for a current injection
Autoscan 3000B Microscope stage with a position resolution using the two hexagons with a full guard of electrodes and
of 0.25 µm and a position repeatability error of +2.0 µm. also using a two electrode pair configuration. Results for a
The current returning through the electrodes was also balanced 1 mA peak to peak sinusoidal injection into the
right and left hexagons and two imbalanced injections are
5270
included. For the experiments using the full guard, the closer to providing restored vision of continuous patterns
currents from the six return electrodes are summed and free of flicker.
tabulated. This prototype system has been shown to operate well
and with further work by members of this group will be
TABLE II minimized into an Application Specific Integrated Circuit
RETURN CURRENTS FOR DIFFERING ELECTRODE CONFIGURATIONS
(ASIC) for bench and in vivo testing.
Current Injection(mA) Single Return(mA) All Guards (mA)
Left H Right H Left H Right H Left H Right H
1 1 1.10 0.90 0.99 1.01 ACKNOWLEDGMENT
1 0.50 0.83 0.68 0.90 0.60
1 0.10 0.63 0.47 0.70 0.39 This study was funded in part by the Australian Research
Council and the Universities of NSW and Newcastle,
IV. DISCUSSION Australia.

Fig. 6 and Table II show that current injection with a full REFERENCES
guard electrode configuration allows for better control of
[1] Winter K., Hartmann R., and Klinke R., “A simulator with
localized current injections and minimizes cross talk
wireless power and signal transmission for implantation in
between electrodes compared to the single electrode return animal experiments and other applications,” J. Neurosc. Vol.
configuration. 79, pp. 79-85, 1998
Fig. 6 shows that the current injection is mainly kept to [2] Mayr W., Bijak M., Rafolt D., Sauermann S., Unger E., and
within the outer ring of guard electrodes when the full Lanmiller H., “Basic design and construction of the Vienna FES
hexagon electrode configuration is used. Table II indicates implants: existing solutions and prospects for new generations
that cross talk is much more prevalent when the single of implants,” Med. Eng. & Phys. Vol. 23, pp. 53-60, 2001
electrode return configuration is used. This point is [3] J. Rovamo, and A. Raninen, “Critical flicker frequency as a
illustrated when 1 mA of current is injected into both function of stimulus area and luminance at various eccentricities
in the human cone vision: a revision of Granit Harper and
electrode pairs with 1.10 mA returning to the left electrode
Ferry-Potter laws,” in Vis. Res. , Vol. 28, no. 7, pp 785-790,
pair and only 0.90 mA returning to the right electrode pair. 1988
When the full guard of electrodes is used for the same [4] C. W. Tyler and R. D. Hamer, “Analysis of Visual Modulation
current injection 0.99 mA returns to the left electrode Sensitivity. IV. Validity of the Ferry-Porter Law,” J. Opt. Soc.
hexagon and 1.01 mA returns to the right electrode hexagon. Am. A., Vol. 7(4) pp. 743-758, 1990
When differential currents are injected into saline, the [5] M. S. Humayun, E. de Juan Jr., J. D. Weiland, G. Dagnelie, S.
amount of cross talk between the electrode groups is Katona, R. J. Greenberg, and S. Suzuki, “Pattern electrical
increased in both the electrode pair configuration and stimulation of the human retina,” in Vis. Res. , Vol. 39, no.15,
hexagon guard configuration. The hexagon arrangement pp. 2569-2576, 1999
[6] G. J. Suaning, L. E. Hallum, P. J. Preston, and N. H. Lovell,
however, significantly reduces the amount of charge
“An efficient multiplexing method for addressing large numbers
exchange occurring between electrode groups compared to of electrodes in a visual neuroprosthesis,” presented at 26th
the electrode pair configuration allowing for a greater Annual International Conference of the IEEE-EMBS, San
control over localized current injection. Francisco, CA USA, 2004
The efficient addressing scheme used as a basis for this [7] G. J. Suaning, L. E. Hallum, S.E. Chen, P. J. Preston, and N. H.
prototype has been shown to allow for a significant increase Lovell, “Phospene Vision: Development of a portable visual
in the stimulation rate by way of parallel stimulation. This prosthesis system for the blind,” Proc. 25th Annual
scheme also allows for the system to be scaled for larger International Conference of the IEEE EMBS, Cancun, Mexico,
electrode numbers by simply adding in extra hexagon 2003
[8] D. Merrill, M. Bikson, and J. Jefferys, “Electrical stimulation of
clusters to the current organization with the only addition to
excitable tissue: design of efficacious and safe protocols,” J of
the existing data protocol being its DAC configuration - NeuroSc. Meth. Vol 141, pp. 171 – 198, 2005
common DAC settings will afford additional hexagons to be
added at no expense to data throughput. As described in [6]
the addition of even a large number of electrodes does not
change the addressing scheme with three bits always being
sufficient to describe the seven meta-cluster organizations
that can occur.

V. CONCLUSION
These results indicate that the hexagonal layout with the
use of guard electrodes can aid in maintaining injected
currents within a localized area and significantly contribute
towards minimizing electrical cross talk. This allows
parallel stimulation schemes to become a plausible
alternative to series stimulation, thus pushing the prosthesis

5271

You might also like