Introduction

The Staff at the Fels Planetarium at the Franklin Institute Science Museum were looking to create a low cost star projector. This paper is a feasibility study to determine the types of technology available, and the cost of deploying them.

Design Parameters

The Fels planetarium currently uses a Model I digistar from Evans and Sutherland. The DigiStar was installed in 1986 as a replacement for a Ziess optical projector that could no longer be repaired. There are several problems with the system:
  1. The resolution is somewhat poor, as compared to an optical projector.
  2. It is monochrome green
  3. It regularly blows out rather expensive components due to operator error.
  4. There are very small pool of programmers for the Vax that powers it
Whatever system that replaces the digistar will have to:
  1. Be monochrome white, or better yet, full color.
  2. Deliver an acceptable resolution to the viewer
  3. Use easily replaceable parts

Video Fidelity

The most straightforward problem to solve was determing the acceptable resolution. I used a 12" diagonal television at a distance of 0.5 meters as my bechmark for acceptable video fidelity for the average american.

A 0.5 meters a 12" screen covers ~30 degrees of the viewers field of vision. TV broadcasts are analog, but on a computer display they are 600 pixels across.
600 pixels / 30 degrees = 20 pixels/degree
A planetarium dome is 360 degress around, thus:
360 degress * 20 pixels/degree = 7200 pixels (horizontal)
90 degress * 20 pixels/degree = 1800 pixels (vertical)
This is a simplification of the problem, given the hemisphere shape of the dome. These numbers work for around the bottom of the dome. We can cut corners on the upper parts of the dome.

Techologies

Video Controllers, Graphics Generation

When the Digistar was created, video cards on computers were reserved for high-end cad stations. It made sense a that time to develop custom controller cards. In its day the VAX was quite the numerical cruncher. Today however, the laptop I am composing this report on could beat it up and steal its lunch money.

In todays day an age, computer graphics technology is well developed, cheap, and easy to work for. Whatever projection techology we use must interface with a standard vga graphics card in personal computer. Depending on what is easier to deploy, or cheaper to build, we should go with a cluster of workstations with a several video cards driving multiple projectors. Given the cost considerations, and the need to develop low level routines to map the images on the dome and handle overlap, I recommend Linux. You just can't pick a better platform for writing custom display drivers, or clustering for that matter.

Laser Projectors

Rasterized Image New technology in laser projectors allows them to render rasterized images. On closer examination I found that laser projectors require rather sophisticated (not to mention expensive) proprietary driver cards. (This is in addition to the already expensive laser.)

Laser projectors do not deliver a fixed size. They can change the shape of the frame at will. The pixel count delivered by one of these systems is about 130,000 dots in total. Our application requires (7200*1800=) 12.9 million pixels.

We would need 100 laser and controllers to cover the dome. Cost aside, these are not small components. Cost not aside, they are not cheap by any stretch of the word. A bottom end card with raster ability is ~$1200.

Even if we did get it working, the screen quality is blocky and the color pallete is lacking. The illustration to the right is a photograph of an image rendered by one of these rasterized systems.

The technology will no doubt improve over time. But for now, it is too expensive and looks terrible for this application.

Video Projectors

Video projectors are standard fare in offices, classrooms and theaters. The Fels Planetarium has used video and multimedia projectors to suppliment slide and star effects. Given the direct translation from the video signal from a PC to the projector this seems like a logical fit

A decent projector, delivers a resolution of 1024x748.  

7200 dots / 1024 (dots/projector) = ~7.03 Projectors
1800 dots / 768 (dots/projector) = ~2.3 Projectors
8*3=24

A higher end projector delivers 1280x1024

7200 dots / 1280 (dots/projector) = ~5.6 Projectors

1800 dots / 1024 (dots/projector) = ~1.8 Projectors
6*2=12

I ran across a paper that described putting an omnidirectional motion picture with 5 1200x1200 projectors. In our case we need to hit a larger area, and our stuff tends to focus attention on a single point at a time. We would need the extra resolution.

Video Projector Technologies

LCD display side by side a DLP display There are 2 major technologies used in video projectors: LCD and DLP. In an LCD projector, light is filtered through a screen, and then reflected.  DLP uses a high density matrix of mirrors to control what pixels are on and off.

LCD's tend to be a little cheaper and deliver better color matching, but they don't deliver very good contrast. DLP deliver very high contrast ratios. Black is black. White is white. There is a lot of room in between. DLP also delivers a better looking pixel. (See the illustration to the right.) This ensures that the person in the front row has a harder time seeing the border between pixels.

Projectors of the size and shape we would need run about $4000. (I recommend a DLP projector for the contrast and longevity. We aren't too concerned about color matching in this application.)

Estimated Costs

My design assumes we will be using 24 DLP projectors connected to Linux workstations. Each workstation could probably accommodate 4 graphics cards to drive the projectors, with one additional card for a standalone screen, and a network card. We would need a total of 6 workstations in this arrangement. The workstations could be built with commodity components at $~1200. Each projector is estimated to cost $4000. Assume another $1000 in networking equipment to interconnect the workstations at high speed.

There will need to be some sort of mounting arrangement for the system, with provisions to realign them periodically. Also the labor time required to write the display software and show production software.

We also need to build into the costs the time for a software engineer and a few line-coders to hack out the software required to make all of these pieces talk together.

The total hardware cost for the project:
Item
Count
Cost/unit
Subtotal
1024x768 DLP Projector
24
$4000
$96,000
Controller Workstations
6
$1200
$7,200
100Mb Network Switch
1
$1200
$1,200
Mounting Hardware
24
$200
$4,800
Hardware Total


$109,200
In-House Engineering Hours
1000
$25
$25,000
Line-Coder Labor Hours
2000
$15
$30,000
System Installation
500
$10
$5,000
Labor Total


$60,000
Grand Total


$169,200

Now the labor numbers are fuzzy since we have in-house people that could fill out some of these roles, as well as volunteers. All of the development work is pretty straightforward, merely applying present knowledge to a concrete problem. However, as I have come to respect, development time is still development time.

Also note that the labor hours are a pure guestimate as to how long it would take a team of people to put it together, added up, based on a sketchy design concept and on my own personal experience with software development (which may or may not apply.)