NOVOVIEWTMLCV: BALANCING PERFORMANCE AND COST FOR A "LOW COST" VISUAL SYSTEM Dr. James L. ~ a v i s * Downloaded by UNIVERSITY OF NEW SOUTH WALES (UNSW) on October 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.1989-3321 Rediffusion Simulation, Ltd. Crawley, West Sussex, England ,4'W Abstract Aircrew training with flight simulators is accepted as being a valuable supplement to training in the actual aircraft. Enhancing a simulator with an out-of-cockpit visual simulation system further expands this training role, yielding improved training or comparable training at reduced cost. However, a problem exists in providing visual simulation to aircraft users who can't justify typical visual system expense. The chosen approach examined total system cost rather than component costs, and sought to strike a new balance between cost and performance. Selective capability with flexibility was found to be the key to good performance, while increased standardization was critical for reducing cost. One possible implementation of these findings is Novoviewm LCV, a complete visual system package comprised of computer image generator, generic data base, one of several standardized displays, a visual control console, installation & integration support, and overall product support. Training aircrew in flight simulators, even those without visual systems, is becoming more commonplace owing to well-established cost savings and trainee safe0 considerations. Despite simulator costs of about $10 million, choosing to use simulator training is relatively easy for major commercial airlines since regulatory agencies such as America's FAA and the UK's CAA allow simulator training to replace time otherwise spent in the actual aircraft. Similarly, it is an easy choice for the military in many cases, either because of the high costs incurred in operating an actual aircraft, or because peacetime flight rules do not permit the range of training needed for combat preparation. Justifying this level of expenditure is not difficult when dealing with aircraft having a purchase cost of * Manager, Low Cost Visual Products Member AIAA several tens of millions of dollars and an operating cost of several thousand dollars per hour. However, where does this leave the operators of smaller aircraft costing less than one-tenth as much? Certainly they stand to reap the same benefits from training on a visuallyequipped flight simulator. One scheme for satisfying the simulation needs of the smaller operators is to provide less expensive simulators possessing less expensive visual systems. However, where does one cut comers in providing the visual aspect of the simulation? The view from the flight deck of a Shorts 360 is not much different from that of a Boeing 747. And the pilot of the smaller aircraft is probably less experienced than the 747 pilot and thus needs proper training that much more. To appreciate what "low cost" means in terms of this paper, it's beneficial to first examine the spectrum of simulation devices currently available. Figure 1 graphically illustrates the range of entries in today's simulation marketplace. Plotted are approximate ranges of cost for both the simulator (without visual capability) and a typical visual system appropriate to the device type. These devices range from simple fixed-base procedure and instrument trainers, to motion-base procedure and instrument trainers, to FAA Phase I1 and Phase I11 certified flight simulators, to military air combat and weapons tactics simulators/trainers. As shown, the fixed-based trainers cost about $100,000 and can justify a visual system cost of about $50,000; anything much more or less would not offer the user a balanced capability. This caliber of visual is often satisfied by graphics workstation technology (or its equivalent) running customized flight simulation software. The motion-based trainers typically cost $1-2 million and require a visual system costing about $0.5-1 million. Some companies have sold into this market; however, it is not a thriving market niche. Historically, this can be attributed to an inability to get "good" visual value at this price level. Such is certainly not the case with the well-regulated commercial flight simulation. Simulators costing between $5- 10 million usually have visual systems costing several million dollars. This Copyright Q 1989 by James L. Davis. Published by the American Institute of Aemautiu and Astronautics with pamission. 416 Fixed-Base CPT's and 'Instrument Trainers Downloaded by UNIVERSITY OF NEW SOUTH WALES (UNSW) on October 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.1989-3321 Motion-Base CPT's and Instrument Trainers Phase I1 and Phase I11 Full Flight Simulators KEY Air Combat And Weapons Tactics Trainers/Simulators Q ::.. . VISUAL SIMULATOR . . . . . . . . 0.1 1.0 10 $ (MILLIONS) FIGURE 1, SIMULATOR AND VISUAL SYSTEM RELATIVE COSTS AS A FUNCTION OF MARKET SEGMENT trend holds as well in the military arena, where both simulator and visual system can each cost close to $10 million. The concern here is not the bottom end of the marketplace, as represented by the left side of Figure 1. Rather, the particular low-cost visual market segment to be addressed is those fixed-base and motion-base cockpit procedure and instrument trainers (costing $1-2 million) that would benefit from a visual package costing $0.5-1 million. In defining and developing Novoview LCV, Rediffusion Simulation had as its goal satisfying a new visual cost/performance equation aimed at certain users willing to purchase sophisticated flight trainers, but unable to justify a comparable or larger capital expenditure for a visual system. In the past, this problem was attacked by producing an IG of limited capability (e.g., flat-world, little or no surface texture, digital artifacts) and combining it with a simple data base and collimated or non-collimated display. Often the supplier of the IG or data base was not the same as that for the display. Indeed, often neither was the integrator of the visual system with the simulator; hence, yet another party became involved. For better or for worse, Rediffusion approached this problem from the standpoint of already being the major supplier of visual systems for the Phase 11 and Phase I11 commercial flight simulation market. The standard visual product was and still is customized high performance "complete" visual systems using proven, state-of-the art technology having all the "bells and whistles". Restating the problem from this angle, it is necessary to selectively alter the visual system to reduce cost, but not degrade performance to a point where users feel they are not getting value for money. The previous section pointed out how the visual system is only one of several systems in a flight simulator; the solution to the cost/performance problem begins with a similar decomposition of the visual system. A visual system is itself comprised of several subsystems, including Image Generator (IG), data base, and display. Because the visual must properly interface with the rest of the flight simulator, mechanical effort is needed to integrate it with the simulator fuselage and motion system (if present) and software effort is needed to integrate it with the host computer and Instructor Operator Station (10s). Additionally, out-of-cockpit imagery must correlate and be compatible with serisor imagery, cockpit avionics (e.g., HUD), and other perceptual cues (motion, sound, etc.). Downloaded by UNIVERSITY OF NEW SOUTH WALES (UNSW) on October 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.1989-3321 A cursory analysis reveals two important keys to the low-cost puzzle. First, care must be exercised to avoid reducing individual component costs at the expense of overall system cost. For example, there is no point in reducing manufacturing cost by relaxing tolerances on the display system if it produces an even larger increase in labor cost owing to complications in assembly and installation. Secondly, standardization must be maximized in order to reduce nonrecumng costs, yet the resulting system must not be so inflexible as to unduly limit its general applicability. Getting beyond top-level results requires a detailed examination of those visual system attributes that are both performance- and cost-drivers. Such an analysis has been previously described1. A summary of this analysis follows. It's not unusual for the IG to represent more than 50% of the cost of a visual system. Hence, attributes of the IG play a significant role in any trade-off analysis, and some impact the choice of display system as well. Important atmbutes include: Resolution - More pixels imply more video memory, higher video bandwidth, and more high-speed processing for pixel-rendering; all at increased cost. However, decreasing resolution reduces the maximum range at which objects can be discriminated. Of View (FOV) - Larger FOV generally demands more IG channels and displays, brighter display devices, or servo-controlled Area Of Interest (AOI) technology. Reducing FOV saves on cost but limits out-of-cockpit viewing and hence the tasks to be trained. Scene- More surfaces and lights in a scene require more IG processing to yield perspectivelycorrect imagery. Less scene content makes for simplistic imagery that can limit training effectiveness. Often 2-D texture is incorporated to enhance apparent scene content, the principle being that an expenditure for texture has more benefit than a comparable expenditure for additional surfaces. Co- The IG's ability to handle greater scene complexity relates to supporting more scene occultation. Though most flight tasks don't make heavy demands in this area, periods when the eyepoint is close to a 3-D ground and 3-D objects can increase the need for high scene complexities, as can the use of transparency. Picture 0- Good value requires that full use is constantly made of expensive IG processing capability. Additionally, any generated imagery should be free of computational artifacts arising from spatial and temporal aliasing, and jitter. For example, not having anti-aliasing can save on IG cost, but only at the risk of distracting the trainee with unrealistic scene content. Moving Oblech - Moving objects unrelated to the eyepoint (e.g., other aircraft, ground objects, and munitions) permit greater complexity of the training environment. However, aside from impact on IG complexity, moving objects serve as a cost driver since either software or Instructor control must be provided to direct their movements. Meteorological And E n v i r o n m e n t a l Effects - Because pilots usually fly at all times of day in all types of weather, it is desirable to have a visual system capable of day/dusk/night ambient lighting, limited visibility, reduced ceiling, etc. As with moving objects, flexibility demands control. So aside from increasing the complexity of the IG, variable environmental conditions demand that they be provided for during data base construction and that their control is imparted to the Instructor. These effects also have an impact on display capability and hence cost. For example, daylight simulation (as opposed to duswnight) generally demands (i) a full color gamut, (ii) more display intensity to yield simulated daylight brightness, and (iii) higher refresh rate to guard against image flicker at the higher brightness. Uadate - Increasing update rate means that the displayed scene is recomputed more often for changes in pilot eyepoint location or line-of-sight. It is a costdriver since rate increases are obtained only from increases in amount or sophistication of computational hardware. Lowering update rate saves money, but the penalty is increased temporal aliasing (manifested as image-stepping or double-imaging). Alternatively, training can be limited to slower eyepoint movement or to reduced FOV's, either tending to reduce the angular rates of objects in the visual scene and thereby making the scene more forgiving of low update rates. Data - Often a simulator user will want training to occur in a world closely resembling that in which the actual aircraft flies. For example, a pilot spending most of his time in the Boston area would probably obtain additional benefit from having a simulator data base depicting Eastern Massachusetts and Logan Airport. The cost of this "custom" data base will be related to (i) quantity of data involved (size of data base) and the difficulty of acquiring source data, (ii) the required fidelity of the data base, (iii) tools available for constructing the data base, and hence (iv) the time (labor) required to build the data base. Efforts exceeding one man-year are not unusual. An alternative to a custom data base is a "generic" data base. If care is taken initially to give it training flexibility, then often it can be provided more cheaply since many users can take advantage of it. The downside, however, is a lack of geospecific or airport-specific training. The display is comprised of (i) a display input device (monitors or projectors), (ii)optical elements (e.g., Downloaded by UNIVERSITY OF NEW SOUTH WALES (UNSW) on October 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.1989-3321 beamsplitters, mirrors, or screens), and (iii)a support structure. After the IG, the display sub-system shares with installationlintegration the distinction of being the next costliest portion of the visual system. The role of the display cannot be overstated; the quality of even the most pristine IG imagery will rise and fall with the presentation made by the display system to the trainee. Display attributes affecting both performance and cost are listed below. better contrast relative to the surfaces against which they appear, (iii) finer positioning, and (iv) improved dynamic behavior. However, calligraphy makes severe demands on a display device in terms of power required for linear deflection. In addition, care must be exercised to avoid hysterisis artifacts and improper alignment of rasterdrawn surfaces with calligraphically-drawn lights. Collimation - There is good reason why collimation has historically found a niche in aircraft flight simulation. First of all, it enhances eye relief (distance from eyepoint to first optical surface) for a FOV and thus often simplifies positioning of the display on the simulator fuselage. Secondly, image size doesn't change with fore and aft head movement, just as in the real world for distant objects). Thirdly, an object's angular position does not change with sideways head movement or, more importantly, with laterally-displaced eyepoints as found in multi-crew cockpits. Finally, the eyes accommodate (focus) and converge (tilt inwards or outwards) as if viewing distant objects, which is comparable to what occurs most of the time from an actual cockpit. Because additional optical components are needed to perform collimation, display costs are higher. Additionally, the additional weight of the optics can have a cost-impact on other parts of the simulator, such as the motion system. Though many users of computer image generation equipment simply need a graphics engine and a display for viewing, the user of a flight simulator is not so fortunate. The IG and display must be successfully integrated with the flight simulator in both a hardware and software sense. And, since the simulator is either a training or engineering device, integration must also be accomplished with an InstructorfOperator Station (10s) to permit control of visual system parameters such as ambient lighting, visibility, and location of moving objects. Some aspects of integration are discussed in more detail below. Increased FOV Via -ioning -A common technique for extending FOV is to mosaic several display channels (which in turn requires several IG channels). To obtain a continuous FOV, each display channel is often blended both in geometry and intensity to ensure continuity between adjacent channels. This blending increases cost owing to the added sophistication needed by the displays. An alternative is simply to abut adjacent channels, leaving a small gap over which no imagery is presented. The argument against abutment is that it's unrealistic and small objects can disappear into the gaps. However, abutment is less expensive both in initial display cost and routine alignment. Also, one can argue that small objects never disappear into gaps for very long owing to the dynamic nature of aircraft imagery. Field/Frame- The best way to handle a sudden excess in IG scene capacity is to simply extend the period during which that scene can be processed. The impact, as far as the display is concerned, is a longer field or frame time. Most commercial displays offer only a fmed field and frame rate; having an "extend" capability is more costly. However, the alternative to providing for sudden overloads is to ensure that they never occur. This is only accomplished by building sparser data bases and not utilizing the IG to its fullest capacity. r vs. - The calligraphic portrayal of lights yields enhanced realism compared with raster-drawn lights owing to (i) higher brightness, (ii) er Of mart - Increasing the number of available airports implies that a means of selection must be provided at the IOS. Additionally, runways need to be aligned with Radio Aids residing in the host computer to ensure correlation of visual with navigation instruments. .. i t i o u - Every environmental condition must be controllable from the 10s; this control ranges from a simple onfoff to selection from a range of parameters. M o v i n ~0- As previously stated, each moving object needs a provision for activation and subsequent control. ht Above Terrain & C o l l l. s .l p n - Both of these attributes often require that the IG pass information back to the controlling host computer, thereby complicating an interface that would be unidirectional otherwise. WeaDonry - It's straightforward graphically to depict weapons firing and those effects associated with a hit or miss. However, accurate simulation in a training environment requires computation of trajectory and target correlation with other onboard avionics (e.g.. radar, gunsight, or HUD). The implication on integration is added cost. DisDlav - Ease of fitting affects installation cost. For example, the rake of a canopy or the method by which it opens will influence the choice of display. The amount and type of light-tighting can also be a cost-driver. The presence of a motion system implies that greater rigidity with lower weight and inertia are desirable. Downloaded by UNIVERSITY OF NEW SOUTH WALES (UNSW) on October 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.1989-3321 Host -C - The IG demands of the host computer certain information at the commencement of a training exercise, and additional information at regular intervals thereafter. Initially, the host sends information regarding (i) choice of data base, (ii) environmental conditions, and (iii) status of moving objects. Afterwards, the host must communicate to the IG data concerning (i) positional and attitudinal updates of moving objects and (ii) any changes in environmental conditions. The integration required to fulfill this mission must include software to collect and transmit the data residing in the host, and it must include a hardware link capable of communicating the required information at the needed rate. Povoview LCVTM: A Solution The remainder of this paper addresses one possible approach to the problem of providing a relatively lowcost visual system having good value. Termed Novoview LCV, it is comprised of an IG, data base, display, a visual control console, installation & integration support, and product support (both before and after delivery).Each area will be described in more detail. The IG selected for Novoview LCV is the ESIG100 from Evans & Sutherland. Derived from the SP-X family of image generation equipment, it represents proven (over 100 sold to date) state-of-the-art technology with good reliability and low technical risk. In terms of performance, it offers dayldusklnight simulation for up to four channels, both intensity and color-blended texture on surfaces of any orientation, anti-aliasing, up to three simultaneous moving objects, and control of meteorological effects including visibility, RVR, cloud height and thickness, scud, and rain/snow/ice effects. Because of its importance, special mention will be made of scene management. By way of reminder, scene management refers to efforts taken to ensure that the IG (i) operates close to capacity most of the time, (ii) utilizes this capacity in a manner most beneficial to the pilot (i.e., concentrating detail in the foreground where the pilot can see it), and (iii) in the event of system overload, degrades the image in a graceful rather than abrupt and distracting manner. The ESIG-100 uses the following techniques to guarantee the optimum level of performance: Perspective culling - Small objects subtending an angle of less than a given threshold are eliminated from further processing by the IG, thus freeing processing capacity for more significant objects. Level-of-detail management - Objects are modelled in several different ways. Crude models having a small number of polygons are used for representation at medium to large distances; a detailed model having a higher polygon count is normally used for representation when the object is close to the eyepoint. Transitioning between models occurs at a range sufficiently large to minimize observance of a discontinuity. The net effect of this capability is that scene detail is concentrated close to the pilot where hisher acuity can resolve it, and not wasted in the distance where it's not readily discernible anyway. In the event that the IG detects an overload situation arising gradually, the transition distances used for switching among crude and detailed models are decreased. This results in the cruder models being used more often and for longer times than designed for originally, but it does reduce IG processing requirements until the overload condition is past. FieldIFrame extend - In the event of a sudden overload (as might occur if a complex moving object enters the FOV of an already complex scene) in which the IG does not have time to execute level-of-detail changes, field/frame extend provides a mechanism for avoiding scene collapse. The time available for processing the image is increased, and the net effect is that the drawing of the new image is delayed and the updatelfield rate momentarily decreases. The ESIG-100 differs from the more sophisticated SP-X and ESIG products in several respects. One of these is resolution. Whereas some products have over 700,000 pixels, the ESIG-100 has a bit over 300,000 (yielding roughly 4 arcmin resolution with most display systems). In those applications demanding detection or recognition of small objects at long ranges, this level of resolution will prove inadequate. However, many tasks such as low-level navigation can be satisfactorily trained with this resolution. Furthermore, the anti-aliasing found in the ESIG-100 tends to enhance discernibility of individual small objects! Update rate is another important differentiator between the ESIG-100 and other systems. Phase I1 and Phase I11 visual systems generally update imagery at field rate; i.e., 50 or 60 Hz. A rate of 25 Hz was chosen for Novoview LCV because (i) it allows scene capacities of 500 surfaceslchannel and 1000 lightslsystem and (ii) it compares favorably with motion picture film rates of 24 Hz. In addition, because lower cost visual systems tend to have narrower fields of view than more expensive ones, streaming effects in the peripheral field are less pronounced, thus making temporal aliasing less of an issue. Downloaded by UNIVERSITY OF NEW SOUTH WALES (UNSW) on October 27, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.1989-3321 The ESIG-100 is a pure raster device unlike other products in the SP-X range. Though not a significant cost driver in the IG (it was already designed into it), it would have had a significant impact on display cost. The net effect is to compromise light fidelity; this is especially noticeable at dusklnight. However, light attributes such as (i) straight and curved strings, (ii) random intensity, (iii) horizontal and vertical directionality, and (iv) flashing, blinking, strobing, etc. are still present. Novoview LCV offers either a generic civil or military data base comprising (i) an airportlairfield, (ii) surrounding 3-D terrain with cultural features out to a radius of 40-50 miles, and (iii) single textured polygon representations of earth and sky that are automatically placed in the visual scene if the trainee travels beyond the boundary of the normal generic data base. To help cater for individual training needs, the airportlairfield model is adjustable by the instructor in terms of runway length, airport lighting, approach lighting, and terminal location. Modem display systems run the gamut from single-channel direct-viewing to wide-angle collimation to area-of-interest projection slaved to headleye movement. A low cost visual needs to be more modest in outlook, yet cater to a wide variety of aircraft types. Potential users encompass operators of fixed and rotary wing aircraft having both single- and multi-crew flight decks. Hence, Novoview LCV offers a family of display options in configurations of one to four channels. Monitor-based displays with beamsplitterlmirror collimation are advocated for multi-crew cockpits, whereas front-projectiononto an 8' radius spherical screen for direct viewing is recommended for single-seat cockpits or those multi-seat situations where only one crew member is trained at a time. Use of commercial equipment in conjunction with standardized display structures helps to ensure low cost. A standardized Workshare and Interface Control Document (ICD) is part of Novoview LCV and defines the responsibilities of both Rediffusion and the user in terms of installation and integration of the visual system with the simulator. Essentially, Rediffusion installs the equipment and ensures that the host computer can communicate with it (currently via Ethernet); the user is expected to ensure (with Rediffusion's technical advice) that the simulator is mechanically compatible and that the other simulator sub-systems are compatible with the addition of the visual. This permits the use of more standardized display structures than would otherwise be the case. Often this approach represents a turnkey approach for the user. To simplify integration and thereby reduce system cost, Novoview LCV comes with an Instructor's Visual Control Unit that provides control of visual functions in those cases where no capability exists at the 10s. Novoview LCV is provided with both componentlevel and system-level documentation. After acceptance, the customer is provided with an on-site operations and maintenance course for up to three weeks. And, as long as the equipment is in service, a Customer Service Engineer will pay a one-day visit annually. Optional services are also available at added cost; these include (i) spares, (ii) tools and test equipment, (iii) on-site technical support, (iv) customized data bases, and (v) additional integration tasks, to name just a few. Striking a balance between visual system performance and cost is very much a function of the type of simulator for which it is intended. Market trends indicate that sophisticated trainers costing about $1-2 million need a visual system costing on the order of $0.5-1 million. Because the bulk of the training market avails itself of more expensive Phase 11- and Phase IIIcompatible visual systems, it was necessary to find a means of trading-off performance and cost to yield a system capable of satisfying this A $1-2 million trainer market. Looking at the system rather than individual components, it was realized that the key to reducing cost lay in (i) judiciously reducing performance to reduce cost, (ii) implementing standardization to reduce non-recurring costs, and (iii) maintaining system flexibility to ensure economies of scale through broad applicability to a a large user population. Putting these three concepts into practice then required a careful examination of visual system attributes affecting both performance and cost. One result of this analysis is a product from Rediffusion termed Novoview* LCV. It is a complete visual system package comprised of (i) an ESIG-100 dayldusldnight high-performance computer image generator, (ii) generic commercial or military data base with adjustable airport, (iii) either collimated or noncollimated display of up to four distinct channels, (iv) Instructor's visual control console, (v) installation and integration support, and (vi) overall product support. A system has already been sold and is due for delivery later this year. References 1. James L. Davis, "Managing Performance Trade-offs in Low Cost Visuals", Low Cost Visual Systems Conference sponsored by the Royal Aeronautical Society (23 November 1988) London.