There are many types of volumetric displays, but they are most often portrayed as a large transparent sphere with imagery seeming to hover inside. Designers are required to grasp complex 3-D scenes--for example, they may need to know if any of several hundred components of an engine are mistakenly overlapping. While physical prototypes provide insight, they are costly and take time to produce.

A 3-D display, on the other hand, can be used as a "virtual prototyping station," in which multiple designs can be inspected, enlarged, rotated and simulated in a natural viewing environment. The promise of volumetric displays is that they will encourage collaboration, enhance the design process and make complex data much easier to understand.

Volumetric displays create images that occupy a true volume. In the still nascent vocabulary of 3-D display technology, it could also be argued that reimaging displays, such as those using concave mirror optics, are volumetric, since they form a real image of an object or display system placed inside.

Unlike many stereoscopic displays and old "3-D" movies, most volumetric displays are autostereoscopic; that is, they produce imagery that appears three-dimensional without the use of additional eyewear. The 3-D imagery appears to hover inside of a viewing zone, such as a transparent dome, which could contain any of a long list of potential display elements. Some common architectures use a rotating projection screen, a gas or a series of liquid crystal panels.

Another benefit of volumetric displays is that they often have large fields of view, such as 360 degrees around the display, viewable simultaneously by an almost unlimited number of people.

But if they're so great, why aren't volumetric displays available at every corner store? One reason is that they are difficult to design, requiring the interplay of several disciplines. Also, bleeding-edge optoelectronics are required for high resolution imagery. For example, the core component of many 3-D displays is the reflective microdisplay, a two-dimensional spatial light modulator (SLM) that must project a sequence of images at over 4,000 frames per second.

In a volumetric display, fast 3-D rasterization algorithms are needed to create fluid animation. Once the algorithms "rasterize" the scene by converting 3-D data into voxels, the data slices are piped into graphics memory and then into a projection engine. In this case, a three-chip XGA MEMS-based projector is used with a color-mixing prism. The projector illuminates a spinning projection screen that rotates at 730 rpm. The screen is encased in a rotating dome. A high-resolution volumetric image, in this case a fighter jet flying over synthetic terrain, can be used as a ''virtual prototyping station,'' where multiple designs can be inspected, rotated and simulated.

A typical high-end volumetric display, as bounded by the performance of SLM technology, might have a resolution of roughly 100 million voxels. (Voxels, short for "volume elements," are the 3-D analog to pixels.) Because it's difficult to set up 100 million emitters in a volume, many 3-D displays project a swift series of 2-D imagery onto a moving projection screen. Persistence of vision blends the stack of 2-D images into a sharp 3-D picture.

For example, a swept-screen volumetric display creates imagery in a series of steps:

1. A three-dimensional data set is converted into a series of "slices," similar to thin slices of an apple around its core. For example, a high-end 3-D display creates imagery composed of 198 slices, where each slice is roughly 768 x 768 pixels. The conversion method is proprietary, as the efficiency of the algorithms relates to the display's interactive performance.

2. Once processed by a voxel-processing engine, the slices are stored in a memory bank, such as 6 Gbits of double rata rate synchronous DRAM (DDR SDRAM).

3. A high-speed digital projector illuminates a rotating screen with the 198 slices of voxel data at 24 Hz. This is equivalent to almost 5,000 single-bit-depth frames per second!

4. Your persistence of vision fuses these slices into a sharp 3-D image. For the display described here, the imagery contains roughly 100 million voxels ( = 768 x 768 x 198) that is refreshed at 24 Hz--or 8.4 Gbits/second of "optical bandwidth" for a 3-bit display.

In a volumetric display product, three-dimensional data is captured in real time from a host PC, which is sent over SCSI to the display's rasterization system. Proprietary algorithms "rasterize" the scene; that is, they convert the 3-D data into voxels.

Once the 3-D scene is rasterized, the slices are piped into graphics memory and then into a projection engine. For example. the projector may be a three-chip XGA MEMS-based projector with a color-mixing prism consisting of three SLMs, for red, green and blue. The projector illuminates a spinning projection screen that rotates at 730 rpm. The screen is encased in a rotating dome and a stationary second dome surrounding the rotating one.

The result of this conceptually exasperating process is a high-resolution volumetric image. To appreciate the design challenges involved, let's walk through a few multidisciplinary innovations that were required to bring this particular volumetric 3-D display to fruition.

Remember rasterization? In that process, specialized algorithms convert mathematical descriptions of a 3-D scene into the set of voxels that best visually represent that scene. These algorithms, which do the dirty work of mapping lines onto grids, are well-known in the world of graphics cards and 2-D CRTs (for example, the Bresenham line-drawing algorithm).

However, these algorithms don't directly apply to many volumetric displays, especially if the device's coordinate system is cylindrical. Therefore, you have to invent fast 3-D rasterization algorithms in order to create fluid animation.

Other aspects of the system require the mathematical equivalent of lobbing a water balloon from a moving roller coaster onto a target below. One consequence of illuminating a rotating screen and mirror assembly with a fixed projector is that the projector's image rotates in the plane of the projection screen as the projection screen rotates. It's sort of like a tumbling cylindrical coordinate system. Although it is possible to use standard optics to freeze the effect, a more cost-effective solution is to devise algorithms that eliminate the twists and turns through 3-D space.

Although every 3-D display is born different, there are usually some similarities in the graphics-processing system.

Typically, the volumetric or geometric 3-D customer data is loaded into a "raster engine" where it is converted into slices. Computations are performed on a Texas Instruments TMS320C6201 32-bit fixed-point digital signal processor (DSP). Executable code, computational tables and FPGA code are stored in synchronous burst flash ROMs permitting field upgrades over the SCSI interface. Most of the executable code is copied into faster SDRAM which is also used as scratch space and to buffer the active 3-D scene. Processed graphics data is stored in voxel form in a graphics memory module.

An example of a graphics memory module maybe 6 Gbits of DDR SDRAM in the graphics memory modules that are divided into three colors and two buffers. Data is routed among the DSP, graphics memory and digital projectors by three "voxel routers." Furthermore, an automatic read-modify-write capability has been designed into the voxel routers. The raster engine can initiate each read-modify-write operation with only one write instruction, performing operations such as AND, OR and XOR.

In addition, volumetric displays must be designed with an eye toward bandwidth constraints. The link from the host PC to the volumetric display may be 20 to 40 Mbytes/s, which is bound by the type of connection (SCSI, FireWire)--and utilization of the bandwidth depends on the complexity of the 3-D scene being rendered. Within the display, the bandwidth increases. For example, the capacity of the interface between the raster engine and the graphics memory module is approximately 133 Mbytes/s. To maintain a volume refresh rate of 24 Hz, data must be scanned out of graphics memory at approximately 1.4 Gbytes/s (three colors x 198 slices/color x 768,432 voxels/slice x 24 Hz = 1.4 Gbytes/s).

As one may imagine, volumetric 3-D displays require a careful synchronization of the screen's angle to the associated slice being projected. Furthermore, the following conditions must be met:

• The screen's location must be known to within a fraction of a slice (typically less than one degree).

• The slices must be projected at high brightness, high contrast and in focus.

• Since the 3-D volume is swept twice for every 360 degree rotation of the screen, care must be taken to properly register these two scans.

Specific approaches to each condition is beyond the scope of this article. However, it can be appreciated that volumetric 3-D displays require precise engineering of optomechanical and projection systems, as well as repeatable optoelectronic sensing. Furthermore, the optical path of the projection system must be folded in order to reduce the footprint of the display.

Fortunately, many existing third-party applications are based on one of a number of graphics standards. Therefore, it is often possible to hook in to the real-time graphics pipeline and send 3-D data out to the customer's volumetric display. This allows the user to continue working with 3-D software that he or she is already comfortable with, minimizing any change to the workflow.