|
SPIE - The International Society for Optical
Engineering White Paper and Technical Briefing
1.0 ABSTRACT
Groups of video projectors can be arrayed into
electronic displays that offer larger, brighter, and higher
resolution images. This paper provides a broad overview of
the design considerations and opportunities inherent in arraying
projectors in a group. A projector array provides:
- Increased image size
- Increased image brightness
- Increased image resolution
- Reduced projection distances
- Increased depth of focus
Although video walls are the most common example
of electronic image arrays, the limiting factor of traditional
video walls is the visual segmentation. Minimizing the segregation
between arrayed images is highly desirable with the goal being
to make the segregation indistinguishable.
Overlapping and seamlessly blending multiple
video projectors into a single composite image goes a long
way toward eliminating the segregation of projector elements
and opens the way to many new practical applications. This
is particularly significant in displaying computer graphics.
Computers have the ability to generate multi-channel
composite images at resolutions that far exceed traditional
electronic media and even the maximum resolution of any single
monitor or projector. These images can then only be displayed
using an arrayed system.
The challenge is to make the entire projection
array behave as a single imaging device. This can seem to
be a daunting task with large arrays, but one whose solutions
are already clear. Consider that a typical CRT projector is,
in fact, an array of three projectors (R, G, & B) engineered
into a single plastic housing and arrayed in a superimposed
and converged geometry.
Similarly, an array can be managed with integrating
electronics to create a "virtual" packaging around
multiple projectors, arrayed in adjacent and registered
geometries, with the entire package behaving as a single
cohesive imaging device.
2.1 Edge matched formats
Edge matched formats rely on very clean projection
geometries where the two projected images are immediately
adjacent to one another. Defining a perfect single pixel row
edge between projectors is improbable because small variations
in the vertical geometry cause small overlaps and brightness
aberrations between the images. It is typically better to
have a definitive mechanical edge, and it is recommended that
there be an actual "mullion" or black separation
between array elements in such applications.
2.2 Edge blended formats
Edge blended formats rely on an overlap region with redundant
picture information from each projector in the overlap. The
overlap region is, therefore, double the brightness of the
rest of the image and needs to be blended using an edge blending
processor capable of fading down each overlapping edge of
the projectors in such a way as to compensate for the gamma
(video signal reduction vs light output curve) of the phosphor,
light valve or LCD. The goal is a uniform brightness level
across the overlap region.
2.3 Wide-field arrays
Wide-field arrays use adjacent geometries.
These are typically laid out in horizontal patterns to create
wide scope images. 
In some instances, the projectors may be stacked vertically
creating very tall images.
It is also possible to vary this geometry format
by rotating the projectors 90 degrees, then, laying out the
format horizontally. This is particularly useful if each array
element has a 9:16 aspect ratio format. With a 20% overlap,
the resulting display has a very pleasing aspect ratio of
1:1.46. Using HDTV resolution for each array element, the
resulting display is 2690 X1920 pixels for a composite resolution
of over 5,000,000 pixels.
2.4 Matrixed arrays
Matrixed arrays organize the projectors in XY
grids. Video walls are an example of a matrixed array. Alternatively,
using a soft edge blending processor capable of affecting
all four sides of the projector, it is possible to construct
huge seamless projections. There are two challenges in this
geometry format. The first is the need for separate soft edge
blend control in the corner regions where all four projectors
overlap, The second is physical placement of the projectors
so as not to block the viewer. The obvious answer would be
to rear project the image, which is logical but nevertheless
challenging since rear projection materials are manufactured
in finite sizes, and thus can be the limiting factor.

3.1 Increased image size
It is obvious that by adding more array cells,
that the aggregate size of a composite image can me made larger
and larger.
3.2 Increase image brightness
Given a specific image size for each array element
or projector, the overall brightness of the display remains
constant as more and more array elements are added. This is
in contrast to a single projector whose brightness is diminished
as the size of the display is enlarged.
3.3 Increase resolution
By feeding each projector a portion of the image
at a specific resolution of video, the "overall"
resolution of the display is increased. Each projector provides
its own elemental percentage of the overall result providing
an "additive resolution" effect. This is especially
evident in edge blended applications, where the result is
a single composited image. For example, a triple VGA resolution
display with a 25% overlap would result in a 1600X480 or 768,000
pixel composite resolution after subtracting the redundant
overlapping pixels.

3.4 Extreme Resolution
With image generators or graphics computers
capable of multi-pipe high resolution output, it is possible
to generate composite images that are three or more channels
of 1600X1200 pixels each. With a 25% overlap, this would result
in a 4000 X 1200 or 4.8 million pixel resolution image after
subtracting the redundant overlapping pixels. There are currently
no commercially available single display devices capable of
operating at these extreme resolutions. The practical solution
is to use a projector array where each projector is working
at its own maximum resolution, with the aggregate or composite
resolution being the multiple of each element minus the overlap
region.
3.5 Projection on complex surfaces
In an example using a coved cylindrical or spherical
screen design, a single projector trying to display on the
entire screen surface would need to be substantially further
back from the screen than would an array. This can be problematic
from an architectural point of view.
Perhaps more important, the "chord"
described by the curved screen would be deeper than the projector's
depth of focus. This, then requires custom optics to maintain
edge to edge focus. Custom optics are generally impractical
for a three gun CRT type projector but more practical with
single lens system. Alternatively, by using a projection array,
the "chord" of each projector is dramatically shortened,
now falling into the depth of focus operating range of the
projector. The combined resolution, bright- ness and depth
of focus benefits makes a projector array a good choice for
curved screen applications.
4.0 TYPICAL SYSTEM ARCHITECTURE
AND COMPONENTS
The example we will use to examine the components of a typical
arrayed display uses 5 projectors with an adjacent, edge blended
geometry on a flat screen. The components of the projector
array include:

Each components represent another variable consideration
in the overall display design.
4.1 Sources
4.1.1 Prepared video
For a number of entertainment applications such as theme
parks and museums, it is desirable to have an electronic display
system, rather than a film display system, for operational,
maintenance and/or architectural reasons. This display nevertheless,
needs to be wide-field in scope, or otherwise 'spectacular',
bright and high resolution. In these cases, a custom media
production is shot on large format film and sectioned into
overlapping array elements in post production processes. Each
array element is then mastered onto a reliable and synchronizable
playback medium such as laser disc, and re-composited into
a single image for playback. This provides the size, resolution,
brightness, architectural, and operational advantages desirable
in such applications.
4.1.2 Image generators
Simulation image generators such as the Lockheed/Martin or
Evans and Sutherland systems as well as graphics oriented
computers such as the Silicon Graphics systems and even the
Macintosh can each be configured to provide multiple simultaneous
image channels at a variety of resolutions (depending on the
system). This can result in extreme resolution displays useful
for simulation, visualization, group VR, group multimedia,
as well as full immersive environments. A projected array
is an ideal output peripheral for electronic computer displays
-- with a myriad of unexplored applications.
4.1.3 Source processors:
When maximum resolution is not required, a 'single-source-input
to multiple-output' processor can be used to break out the
arrays elements. One example would be HDVS input to multiple
NTSC output 'video wall' processors. Another example are systems
where a single 1280X1024 computer graphics source is broken
out into a four channel, 2 by 2, XY array at VGA resolution,
such as with RGB Spectrum's Compuwall system. Neither example
takes advantage of the additive resolution capability of arrayed
projection. There is another category of processor that will
provide for additive resolution. These are typically graphics
subsystems developed for command and control displays running
under X-Windows such as M3i or Jupiter Systems. Unfortunately,
most of these system have not yet allowed the user to break
out the image with an overlap region to allow for soft edge
blending.
4.2 Array Processor
The geometrically correct images come from the sources and
are fed through an array processor designed to reintegrate
the array elements into a single composite image. Two of these
processing parameters are edge blending and projector matching.

The patented edge blending technology1 which Panoram Technologies,
Inc. (PTI) has developed consists of: 1.) a memory lookup
table that holds information on luminance, 2.) a method of
synchronizing that table to the video coming through, and
3.) a converter for affecting the projected brightness based
on the lookup table.
Using a one dimensional lookup table and re-addressing it
with each video line, it is possible to provide very detailed
adjustment of the image brightness in adjacent overlapping
areas. These "smoothing parameters" can be calibrated
to the gamma behavior of individual phosphor tubes or light
valves as well as a number of other projector aberrations,
providing a smooth transfer across the image region.
Having found that the behavior of each R,G & B channel
differs, PTI has implemented separately adjustable lookup
tables for each gun in the array.
By using a one dimensional lookup table vertically, it is
possible to provide edge blending functions for tops and bottoms
of the display. A separate horizontal and vertical lookup
table provides some measure of control over all four sides,
but does not address "4 corner regions" (see Par.
2.4) of matrixed arrays.
In a new implementation of the same principals, PTI
has provided a true two dimensional lookup table, able to
map "smoothing factors" for the entire raster of
each gun in the array. This leads to an additional series
of array processing capabilities including automated field
flattening, radial anti hot-spotting to compensate for optical
vignetting, automated colorimitry adjustment between projectors,
automated blending of overlap regions as well as managing
trapezoidal and complex blend regions for systems with complex
geometries such as spherical screens and domes.
4.3 Array Processor
Control Since the array processor and projectors are typically
located in a different places, some form of remote control
needs to be accessible at the eye point of the display. We
have found that a laptop computer makes the most flexible
user interface for the array control. Since most of the components
of the system, including sources, array processor and projectors
tend to be addressable via serial communication, it is easiest
to integrate the various control software programs under one
platform. Thus, it becomes possible to provide user setup
procedures that address the appropriate components for each
setup mode. As manufacturers standardize and publish control
protocols, it becomes ever easier to integrate component controls
into system controls under one user interface.
4.4 Projector Arrays
The actual projector arrays can consist of CRT projectors,
Light Valve or LCD projectors. Using CRT projectors provides
the most flexibility in terms of geometry control but limits
each array element to the maximum size and brightness of CRT
projection.
Using bright light valve projectors allows you to expand
the overall size of the array, and since several light valve
imaging schemes are driven by scanning CRTs, these models
still allow for good geometry control of the image.
LCD projectors have the potential of offering a low
maintenance and low cost solution for arrayed projection.
This is appealing when you start to think of using projectors
in larger and larger groups. LCD projectors however, have
virtually no geometric controls. However, since they tend
to be single lens devices (see Par. 3.5), this limitation
may be overcome in pre-integrated arrayed systems where fixed
custom optics can be applied.
There is another design consideration with both light valve
and LCD projectors.
Somewhere in the crossover region, each projected image must
be reduced to zero light output. CRT projectors can be tuned
to provide true blacks. Light valves and LCDs however, do
not provide a maximum density of true black, even when fed
pure black video. We will call this residual light R. In the
overlap region, the residual light is equal to 2R. In the
four corner region of an XY matrix array, the residual light
is equal to 4R. Since this residual light is mechanical and
cannot be controlled via the video signal, it imposes a limitation
on arrays using these technologies. This is balanced by the
ambient light in the environment (which we will call Am) as
it will raise the black pedestal of the entire display and
thus mask the residual light (R). For adjacent geometries:
2R must be Am and for XY matrix geometries: 4R must be Am.
4.5 Screens
There is much to be said about screens and screen design,
much of it beyond the overview scope of this paper. We will,
however, address a key issue: Screen gain.
In applications where the projectors are spread horizontally
and the viewing area is spread out, it is imperative to use
a screen with a gain of 1.0! This design error appears in
installation after installation promoted by the desire to
deliver brighter images and the natural tendency to think
that more is better. In this case, it is not. It is easy to
understand when you consider the nature of passive gain achieved
from a screen in context to a projector array.
Screen gain is achieved by optically bending the light hitting
the screen back toward center. This is often accomplished
by narrowing the viewing angle of the screen and redirecting
the light toward the viewer - so what's the problem? This
scheme only works well when your light source is a single
point.
In horizontal arrays, however, your light
source is spread horizontally across the entire screen surface.
This means there is an extremely complex reflectivity relationship
between all these sources and the viewers eye. It is possible
to adjust all the colorimitry and edge blend parameters of
the array to make a perfectly seamless and integrated image
- in ONE location - but the moment the eye point is shifted
left or right, all the reflectivity relationships change and
brightness related aberrations are experienced in colorimitry
and blend regions.
This problem can be minimized with curved screens if the
projection design allows for the clustering of the light sources.
Cross firing projectors such as in the diagram below are a
good example. In such cases, a screen gain of up to 1.3 is
acceptable. Higher gains are still not recommended.
In 360 degree applications with viewers in several locations
such as with flight tower or ship simulators where the projector
lenses are not clustered but arrayed in a circle, again a
1.0 gain screen is essential to creating a single composite
image result for all viewers.
Other than the above consideration, arrays can be applied
to flat screens, cylindrical curved screen, spherical curved
or toroidal screen, front projection or rear projection.
|