| Publications Page | HITL Home |
The Virtual Retinal Display (VRD) is a visual
display that scans modulated laser light on the retina of the
viewer's eye to create an image. Maximum Permissible Exposures
(MPE) have been calculated for the VRD in both normal viewing
and possible failure modes. The MPE power levels are compared
to the measured power that enters the eye while viewing images
with the VRD. The power levels indicate that the VRD is safe in
normal operating mode and in failure modes.
Keywords: display, laser, scanning, safety analysis
The Human Interface Technology Laboratory
(HITL) at the University of Washington has developed a display
technology called the Virtual Retinal Display (VRD)tm,
a device that scans an image, one pixel at a time, onto the retina
of the viewer's eye. The VRD is based on a novel scanning engine,
the Mechanical Resonance Scanner (MRS). The MRS is small enough
that when combined with a red laser diode, a portable display
can be created. When blue and green laser diodes are perfected,
portable, full color systems will be possible. The VRD approach
for displaying images has several advantages compared to cathode
ray tube or flat panel displays for head mounted and other portable
display applications. The contrast, brightness and dynamic color
range of the VRD are extremely good. These features are especially
important in multiplexed applications (where the display takes
up a portion of the field-of-view) and augmented applications
(where the display is superimposed on the outside environment).
In such applications, portable displays will often be used in
high ambient brightness environments. The conventional displays
do not offer the desired combination of high resolution, low power,
and sufficient brightness to operate in these situations.
The process of scanning laser light on the
retina was used by Webb in the development of the scanning
laser ophthalmoscope (SLO). In the SLO a raster pattern is scanned
on that part of the retina the operator wishes to view. A portion
of the optical beam is reflected off the retina and passes back
through the lens and cornea of the eye where it is detected.
The intensity of the reflected light is used to modulate a synchronized
video signal, allowing an image of the retina to be displayed.
Webb noted that if the input optical beam was modulated by a video
source, the user would see an image. SLO mechanisms have been
used in this mode for evoked potential stimuli[2, 3] and as test
stimuli for low vision subjects. Kleinbeil performed an analysis
of the safety aspects of the SLO. He observed that the power levels
required to illuminate the retina sufficiently to measure back
scattered light for scanning purposes were lower than the calculated
Maximal Permissible Exposures (MPE). He performed an extensive
analysis of the heating effects on the retina of the laser illumination
source. His analysis serves as an important basis for considering
safety aspects of lasers scanned on the retina.
The Virtual Retinal Display
The VRD is comprised of five basic parts:
a light source, a modulation mechanism, horizontal and vertical
scanners, delivery optics and controlling electronics. A block
diagram of the VRD is shown in Figure 1. The light source in
the VRD is typically a laser. Methods have been developed and
demonstrated to use an LED as the light source in a monochrome
display. A single laser is used to create a monochrome display
and three lasers are used for the creation of a color display.
Each laser must be individually modulated such that its intensity
matches that of the image pixel being drawn. For the modulation
of the light source, current to the laser diode itself can be
varied. For systems such as the green and blue gas lasers, which
cannot be directly modulated at video rates, the modulation is
performed by controlling an external device, such as an acousto-optic
modulator. In the case of the laser diode, optics to equalize
the horizontal and vertical divergence of the beam and to minimize
the astigmatism are used. If a multi-color system is being built
the light from all lasers is combined into a single, full color
FIGURE 1: System block diagram of the virtual retinal display.
The resulting modulated beam is then scanned
to place each image pixel at the proper position on the retina.
Our controlling electronics use the raster method of scanning
an image which allows the VRD to be driven by standard video sources.
To draw the raster, a horizontal scanner moves the beam to draw
a row of pixels. A vertical scanner then moves the beam to the
next line where another row of pixels is drawn. The Mechanical
Resonance Scanner (MRS) (figure 2) was developed at HITL for the
VRD. The MRS operates at frequencies on the order of 15.1 kHz,
which allows the creation of high resolution images.
The scanned beam is passed through a lens
system which forms an exit pupil about which the scanned beam
pivots. The user places themselves such that their pupil is
positioned at the exit pupil of the system. This is called a
Maxwellian view optical system. The lens of the eye focuses the
light beam on the retina, forming a pixel image. As the beam
scans across the retina, an image is formed (figure 3).
FIGURE 2: A photograph of a prototype mechanical resonance scanner shown beside a quarter for size comparison.
FIGURE 3: Comparison of illumination of the retina by a pixel-based display versus the VRD. Inset figures show schematized light intensity over any given retinal area in the image. Typical pixel-based displays such as CRTs have persistence of light emission over the frame refresh cycle, whereaws the VRD illuminates in brief exposures.
To demonstrate and test the capabilities
of the VRD a number of prototype systems have been assembled.
These include a color bench system and a monochrome portable system.
In the bench mounted system, the MRS performing
the horizontal scan has been packaged with a galvanometer, performing
the vertical scan, to form a small scanning engine. This engine
is used to scan the optical beam in a full color, VGA resolution
(640 by 480 pixel) see through VRD. Color is generated using
three lasers, a red diode laser at 650 nanometers and green helium
neon gas laser at 543.5 nm and blue argon gas laser at 488 nm.
The red laser is modulated directly and the green and blue lasers
are modulated with external modulators. The three optical beams
are combined and passed to the scanning engine. An eyepiece magnifies
the image field of view for direct viewing. An alternative viewing
arrangement allows the image to be viewed superimposed on the
real world. The scanned beam passes through a beamsplitter and
reflects off a magnifying mirror before reflecting off the same
beamsplitter into the eye. At the same time, a view of the outside
world is passed through the beamsplitter and into the eye. The
system can be focused to yield pixels smaller than 2.5 arcminutes
and with a horizontal field of view varying from 10 to 60 degrees.
The portable VRD prototype is housed in
a briefcase allowing for system demonstrations at remote locations.
It displays a VGA resolution, monochrome image with 6 bits of
intensity variation using a single red laser diode as a light
source. The image horizontal field of view is 40 degrees and
the vertical field of view is 30 degrees. Mounted in the briefcase
are the system electronics and power supplies.
Our approach to this problem was to calculate
the MPE in a variety of methods and then choose the most conservative
value. We also assumed conservative parameters such as an 8 hour
continuous exposure that would be the extreme of regular use.
Calculations of MPE were first made by determining the MPE per
pulse considering a pulsed source, then as a continuous wave
source. We then considered the VRD as an extended source. The
MPE values are compared against measured laser power from the
VRD to determine if the VRD is within limits.
VGA scanned beam
The following analysis was done for the
color VGA system. In the 640 X 480 pixel configuration, the sweep
time for each pixel is approximately 40 nanoseconds. The current
system operates at a frame rate of 60 Hz, giving a duration of
16.67 milliseconds per frame. The active time per frame is 12.19
Following the method of the ANSI standard Z136.1 (1993), we performed a worst case analysis for laser exposure in the visible range, which is in the 400 to 550 nanometer wavelength region. For wavelengths from 550 to 700 nm, the MPE value calculated for the 400 to 550 nm wavelength region is multiplied by a correction factor CB which is greater than one. Though we are creating full color images, we analyzed the power limits for the wavelengths between 400 and 550 nm for the more conservative approach.
We assumed an entrance pupil to the eye
of 7 mm, giving an area of 0.385 cm2(ANSI
table 8, p44). An 8 hour exposure was assumed based on a working
day for a user who would be wearing and viewing the display continuously.
MPE for Pulsed Lasers.
In the first method of analysis, the exposure
effects for repetitively pulsed lasers dispersed over an incident
area is used to determine exposure limits (ANSI appendix B3.1.2,
We start with the Maximal Permissible Exposure
(MPE) of a 40 nsec pulse in the visible and near infrared which
is 0.5 X 10-6
table 5, p41).
Using a continuous exposure of 8 hours (3 X 104 seconds),
The total number of pulses, n, is 3 X 104
seconds X 60 pulses/sec = 1.8 X 106.
To correct for repeated pulses, we introduce
a correction factor in the calculation of MPE of n-1/4
= 0.0273. And finally, we assume the beam is dispersed over the
whole aperture of the eye, 0.385 cm2.
(0.5 X 10-6
X (0.385 cm2)
X (0.0273) = 5.25 X 10-9
is equivalent to 0.13 watts, given no further correction factors.
This is 5 to 6 orders of magnitude above the typical power output
of the VRD: 100-300 nanowatts.
MPE for Continuous Wave Sources.
In the second method of analysis, the exposure
is calculated for a continuous wave laser source dispersed over
a given area. We then divide by the number of pulses and calculate
the MPE per pulse. As above, the estimates are most conservative
for wavelengths at 400 to 550 nm. For an eight hour exposure at
these wavelengths, the MPE for a continuous source is 10-6
(ANSI table 5).
/ 60 pulses/sec= 16.67 X 10-9
The overall MPE is the MPE per pulse divided
by the pulse duration and multiplied by the aperture:
MPE= (16.67 X 10-9
/ 40 X 10-9nsec
) X 0.385 cm2 = 0.16 watts (3)
This value is essentially the same as calculated
in method one.
MPE Calculations for Extended Sources.
For a further estimate of MPE, the VRD could
be considered as an extended source. An extended laser source
is considered to be a system with an intrabeam angular subtense
over 11 millirad which for source viewing is longer than 10 seconds.
As pointed out in Marshall almost all laser sources, including
collimated laser diodes that are pulsed, actually are less than
the 11 millirad restriction. This is consistent with the values
for the VRD, whose source standard angle is 1.1 millirad. However,
use of the correction factors for extended sources further reduces
the estimated MPE which causes a more severe restrictions for
the VRD power limits. Further because the scanned image is swept
over an angular extent of 40 by 30 degrees, one might consider
the VRD as an extended source. This is the assumption that is
used in analyses of scanning laser ophthalmoscopes. The
analysis considering scanned sources as an extended source is
not explicit in the ANSI standard. The solid angle covered by
the source () is approximately 0.36 steradians.
Extended Source Analysis for pulses
In ANSI Appendix B 3.2, MPE's for extended
sources greater than 0.1 rad and less than 0.7 seconds in duration
are expressed as follows:
MPE:Lp = 8.5 X 103
X MPE (ANSI table 5) Jcm-2
using the MPE for 40 nsec pulses from equation 1 above which includes factors for pulse repetition, and aperture area
MPE:Lp = 8.5 X 103 X 5.25 X 10-9 Jsr-1
= 4.46 X 10-5 Jsr-1
The time per raster sweep (i.e. the time to illuminate the extended source once) is 12.19 msec so
IntensityMPE:Lp = (4.46 X 10-5 Jsr-1) / 12.19 msec
= .00366 Wsr-1
Thus for a 0.36 steradian display,
= .0013 W
In Klingbeil , a correction factor of 0.8 is used to correct for the short term temperature effects of scanning. Using the correction factor per Klingbeil of 0.8:
Pmaxk = 1.05 x 10-3 W
This result is still 4 orders of magnitude
greater than the typical VRD power output.
Extended Source Analysis by Video Frame
In the following calculations, we will consider the source as covering the whole video frame (at 60 Hz) and then we will apply the maximum correction factor for extended sources: CE (ANSI table 6). CE is the correction factor used for extended sources. It depends on the source angle. For an extended source with pulse duration of less than 0.7 sec the most conservative correction factor CE assumes an extended source of over 100 millirad. For an extended source with exposure greater than 10 seconds ANSI Section B3.2 shows:
MPE (extended source) = CE X (MPE ANSI table 5) Wcm-2 sr-1
MPE (extended source) =1.15 X 103 X 10-6 Wcm-2 sr-1
= 0.00115 Wcm-2 sr-1. (5)
For a 0.385 cm2 pupil (the limiting aperture),
MPE= 4.43 X 10-4 W/sr
Each video frame occurs 60 times per second,
and each frame actually lasts only 12.19 msec per frame so we
can determine the power per frame.
= 6.05 X 10-4
For a display area of 0.36 sr, and applying the 80% correction factor as per Klingbiel,
MPE = 1.74 X 10-4 W, or about 150 microwatts.
This result is our most conservative estimate
of MPE, giving values of about 3 orders of magnitude above typical
In the event that both the horizontal and
vertical beam controllers failed, one spot on the retina would
be exposed to the whole output of the laser system. The calculations
that follow assume the worst case, that the laser output is continuous
(continuous wave) rather than pulsed.
The first MPE limit will assume that an aversion response to bright visible light will move the eye within 0.25 seconds (ANSI Table 5). The MPE for wavelengths from 400 to 700 nm for a duration of t=0.25 seconds is:
= 0.636 X 10-3
or 2.55 X 10-3
Multiplying by the aperture of 0.385 cm2, we get: MPE= 0.98 X 10-3 Watts.
This value of approximately 1 milliwatt
again is quite high relative to the the actual output of the
VRD for typical images.
Scanner failure without Aversion Response.
If we assume a scanner case where the continuous wave was viewed for more than 104 seconds (2.78 hours) and there were no aversion response, the MPE depends on the aperture and a correction factor CB depending on the wavelength :
MPE =aperture area X CB
(ANSI Table 5.)
The following table results:
TABLE 1: The calculated MPE values as a function of wavelength
|Wavelength (nm)||MPE (microwatts)|
In the shorter wavelengths, this output
value approaches the power output value of the lasers in use in
the color VRD. To reiterate though, it would require 2.78 hours
of continuous viewing of a single bright laser spot to reach this
limit. A subject would have to first suppress the reflex drive
from a bright image and second, then perfectly stabilize the small
Multiple Laser Sources
Li points out that there needs to be
a method of determining MPE for scanned images with multiple wavelengths.
They indicate that retinal hazard comes from both the thermal
effects of laser spots focused on the retina and from photochemical
effects of the light interacting with photopigments[4, 8, 9].
The assumption used by Li for combining different wavelengths
is that the hazards add linearly. They found that the MPE of combining
an infrared laser with visible lasers for scanning the retina
while displaying an image was safe for durations on the order
of 10 minutes even for power levels 10 times greater than they
typically used. It is important to note that the heating effects
of the infrared laser were limiting in their analysis. The VRD
does not use an infrared laser. The individual MPE's for each
laser are determined as above, but include the correction factor
that varies with the wavelength. As described earlier, the effect
of the correction factor is to make the MPE higher for wavelengths
above 550 nm. In our color VRD, the blue and green sources are
below 550 and so have the same MPE. We can thus simply add the
power levels for blue and green. If we directly add in the power
from the red without correction, we end up with a power safety
estimate that is easy and is more conservative than the method
by Li. Essentially the MPE calculated earlier will be distributed
across all visible light sources. Correction factors as Li suggests
can be used if a more precise value is needed.
Measured VRD brightness
Preliminary tests and calculations of VRD
images demonstrated that the system's power output with typical
images is below the maximum permissible exposure (MPE) limits
established above .We designed a test to determine the range
of power levels required for images in the VRD that have the same
apparent brightness as images from a conventional cathode ray
tube (CRT) display.
A modified Minimally Distinct Border (MDB)
approach was used to perform the brightness match for this study.
In order to perform these tests, we used the color VRD unit in
an augmented vision mode as described above. Using a half silvered
mirror in the optical setup, we arranged the VRD image to be seen
simultaneously with an image from a CRT. The image areas were
1 by 2 degrees. A mask, flat black in color, surrounded the
dual image areas. The image field from the VRD was aligned above
that of the CRT image. A program was developed in JAVA for adjusting
the position of the VRD image. This allowed the subject to move
the VRD image so that it immediately abutted the image from the
CRT. The program also allowed the VRD color stimuli to be adjusted
over a range of 255 intensity steps. The intensity of the VRD
stimulus was adjusted by having the subject move a mouse.
Eight solid color CRT test stimuli were
used: two each of red, green, blue, and white. The CRT image field
was generated using a software color palette with 65,000 colors.
One of each pair of intensities for each color was chosen to be
at the upper limit of the brightness range. The CIE color coordinates
and photometric power levels from the CRT images are given in
table 1 . The CRT for this experiment was a NEC MultiSync 5FGp
monitor. The CRT and VRD color images were both at 640 x 480 pixel
resolution. Measurements for the CRT intensities were taken with
a Minolta TV-Color Analyzer II (TV-2150). Repeated measures over
several days showed these levels stayed quite consistent.
TABLE 2: Color co-ordinates and intensities for the CRT test images for brightness comparison tests. Intensity values varies less than 1%
Fourteen subjects participated in the experiment, four as pre-test subjects and ten in the actual experiment. Each subject was first presented with two test stimuli for each color (except white) to familiarize them with the procedure. The subject would vary the intensity of the VRD until the border between the VRD and CRT fields were minimally distinct or until they judged both areas to be equal in intensity. The brightness control allowed complete variation, so subjects could change the VRD image to both dimmer and brighter than the CRT image. They were allowed as much time as necessary for the match. In the actual test, the starting intensity of the VRD image was randomly selected to be brighter or dimmer than the CRT image. For each stimulus, there were two trials. The subjects were tested in a darkened room.
Once the test subjects had made their adjustments
for intensity matches, then the power output of the system was
measured. As mentioned above, the test images for intensity comparisons
were 1 by 2 degrees in size. For the power measures, the entire
extent of the VRD image was turned on to the same intensity. The
vVRD power measurements were taken with a Newport Multifunction
Optical Meter (model 1835-C) and sensor (model 818ST).The sensor
surface is 7 mm by 8 mm and it was placed at the exit pupil of
the VRD system. The sensor size corresponds to the 7mm limiting
aperture set by the ANSI standard (ANSI table 9 p44). The actual
exit pupil is approximately 1.5 mm in diameter. The radiometric
measures were taken with the meter set for the sensitivity at
the wavelength of the particular red, green or blue source being
tested. For white level power measures, the red, green and blue
were measured individually and the results summed. The power measures
were made in darkness and were derived by measuring the power
with the display off and then subtracting that value from the
power with the system turned on.
The averaged power levels from all subjects
are shown in figure 4. As can be seen, the output varied from
60 to 250nW. Most subjects found that they could match the intensity
reasonably well for green and blue. This can be seen in the low
standard error of the mean for the values with these colors. However,
the red matches were more difficult to make. We determined with
subsequent testing of the system that there was a large non-linearity
in the intensity output of the red source. The most rapid change
in intensity per incremental change in control value occurred
close to the brightness levels being tested. The wider range of
standard error of the mean for the red tests reflects this effect.
As a consequence due to the non linearity in red brightness levels,
the white was also somewhat difficult to match. Several subjects
noted the white acquiring a "pinkish tinge" at the brighter
end of the control scale. However, despite these difficulties,
subjects were still able to make consistent matches.
FIGURE 4: Power level histograms. Measurements of power output in nanowatts at the VRD exit pupil after brightness comparison tests. Each bar indicates the mean power level. Error bars indicate one standard error of the mean. Stimuli were: B1= Blue 1, B@= Blue 2, G1= Green 1; G2= Green 2, R1= Red 1, R2= Red 2, W1= White 1, W2= White 2.
Comparison to theoretical predictions
Converting the photometric output of a CRT
to a radiometric measure of power would allow us to directly compare
the power levels. Using a standard conversion we calculated the
power output levels of the CRT, which are seen above in table
2. We note that these power levels are approximately the same
given for a typical CRT.
Figure 5 shows the MPE power limits as they
vary with wavelength. Figure 6 shows MPE levels as they vary with
exposure duration and indicates power levels of VRD ambient light
sources and the scanning laser ophthalmoscope.
Conclusions and discussion
In our Maximal Permissible Exposure calculations,
we have used a number of conservative assumptions for our
estimations. Any change in these assumptions should thus result
in greater differences between the calculated limits and the
actual output values. For example, the current limits were calculated
based on an 8 hour exposure time. If the time is less than one
hour, the limits go up. Further, the blue end of the visible
spectrum was considered, which is only a part of the full spectrum
light of typical images. Finally, the most conservative measure
was taken when considering the VRD as an extended source. The
correction factor for broad extended sources adds several orders
of magnitude to the MPE. If the whole image frame is considered
an extended source the assumption is reasonable. It should also
be noted that the calculations presented here generally apply
to non coherent sources that could be used in this display.
TABLE 3: Intensity measures for CRT versus VRD light comparisons. The VRD measurements are the mean values from the test
FIGURE 5: Power versus wavelength. MPE values for wavelengths from 400 to 700 nm for the VRD illumination characteristics. Symbols indicate the average power output for the VRD for targets 1 and 2 in the brightness tests of blue, green and red.
FIGURE 6: Exposure duration and MPE. The solid line indicates how the MPE varies with exposure duration for VRD illumination characteristics. Note the MPE also varies with wavelength in exposure duration greater than 1 s. Shown are power levels for the VRD as well as other light sources as per Klingbeil 
As section 8.3 of the ANSI standard points
out, lower MPE values are needed for ophthalmic applications where
the laser spot is either stabilized on the retina, or the pupil
of the eye is dilated. In normal use, the VRD will not have either
of these conditions. However, if the VRD is to be incorporated
into an ophthalmic device, such as a scanning laser ophthalmoscope
or visual fields and the image from it is stabilized, or the ocular
pupil is dilated, the lower MPE limits will have to be used.
Measures of power output with typical images
indicates that the VRD generates power on the order of 200 nanowatts
during normal operation. This is below the Class 1 laser power
limit of 400 nanowatts. If failure were to occur, i.e. if scanning
were to stop in one or both dimensions, the power limits indicate
the mechanism is still safe. To use the VRD in brighter light
conditions, such as ambient daylight, higher power levels will
be needed. The MPE limits calculated here should be the guideline
Klingbeil's power analysis for scanning
laser ophthalmoscopes found similar values to ours. He noted
that the power output for SLO's was higher than the VRD and correspondingly
the SLO in image acquisition mode was closer to limits. He recommended
that there be a safety interlock for the SLO that would turn off
the laser source within one millisecond of scanner failure. This
was necessary because the time for an aversion response was too
slow for safety. In the VRD, the safety limits for failure incorporate
the aversion response. Even so, the design of the VRD driver circuit
is such that the laser source is immediately extinguished with
VRD and Virtual Retinal Display are registered trademarks of Microvision Inc.
1. Webb, R.H., Hughes, G.W., Pomerantzeff, O., Flying Spot TV Ophthalmoscope. Appl. Opt. 1980; 19:2991-7.
2. Sakaue, H., Katsumi, O., Mehta, M., Hirose, T., Simultaneous pattern reversal ERG and VEP recordings. Invest. Ophthal. Vis. Sci. 1990; 31:506-11.
3. Katsumi, O., Timberlake, G.T., Hirose, T., Van de Velde, M.J., Sakaue, H., Recording Pattern reversal visual evoked potential response with the scanning laser ophthalmoscope. Acta Ophth. 1989; 67:243-8.
4. Klingbeil, U., Safety Aspects of laser scanning ophthalmoscopes. Health Phys. 1986; 51:81-93.
5. Institute ANS. American National Standard on the Safe use of Lasers ANSI Z136.1-1993.: American National Standards Institute, 1993.
6. Marshall, W.J., Understanding laser hazard evaluation. J. Laser Apps. 1995; 7:99-105.
7. Li, L., Rosenshein, J.S., Safety considerations for simultaneous multiple wavelength exposure in scanning laser ophthalmoscopes. Health Phys. 1993; 64:170-7.
8. Ham, W.T.J., Mueller HA, Wolbarsht ML, Sliney DH. Evaluation of retinal exposures from repetitively pulsed and scanning lasers. Health Phys. 1984; 54:337-344.
9. Greiss, G.A., Blankenstein MF. Repair and additivity of actinic retinal lesions. Invest. Ophthal. Vis. Sci. 1981; 1981:803-6.
10. Tidwell, M., A Virtual Retinal Display for Augmenting Ambient Visual Environments. Seattle WA.: University of Washington, 1995.
11. Boff, K.R., Lincoln, J.E., Photometric Techniques for measuring spectral sensitivity. In: Boff KR, Lincoln JE, eds. Engineering Data Compendium: Human Perception and Performance. Wright-Patterson AFB, Ohio: AAMRL, 1988:24-27.