A backlit cockpit annunciator panel passes photometric acceptance at 0.6 foot-lamberts with clean NVIS Green A chromaticity, then fails MIL-STD-3009 radiance testing by a factor of three. The LED is correct. The filter is the specified minus-blue NVIS filter. The leak is coming from a 0.3 mm gap between the filter carrier and the panel bezel, where unfiltered LED output at 730 nm escapes into the cockpit and straight into the goggle passband.
NVIS compatibility failures are usually not spectral design failures. The LED and filter combination is normally chosen correctly from the start. The failures come from light paths nobody modeled: edge leakage around filters, light piped through the panel substrate to an unfiltered exit, reflections off a canopy that return filtered light through a second unfiltered surface. These are geometry problems, and geometry problems are what non-sequential ray tracing exists to solve.
TracePro models the panel assembly with its real filter geometry, traces the spectral output through every path including the ones the designer did not intend, and reports both the photometric luminance a pilot sees with unaided eyes and the NVIS radiance the goggles see. Both numbers come out of the same trace. This article covers how to set that analysis up and where the requirements interact.
The Two Requirements That Fight Each Other
MIL-STD-3009 asks a cockpit lighting design to satisfy two conditions at once, and they pull in opposite directions. The first is readability. Displays, annunciators, and panel markings must be visible to the unaided eye across a wide dimming range, typically from full daylight readable levels down to below 0.1 foot-lambert for dark adaptation. This requires enough photopic output to see, and enough uniformity that legends read consistently across a panel. The second is goggle compatibility. Radiance in the band where image intensifier tubes are sensitive, roughly 625 to 930 nm for Generation III tubes behind a Class B minus-blue filter, must stay below specified NVIS radiance limits. MIL-STD-3009 defines NRa and NRb metrics by weighting measured spectral radiance against the respective goggle response curves and integrating.
The conflict is that the visible output and the near-infrared leakage often share a source. A white or amber LED has substantial emission past 700 nm. Getting the photopic luminance high enough for daylight readability while pushing the weighted near-infrared radiance below the NRb limit requires filtering with high out-of-band rejection, and that filtering must apply to every path light can take out of the assembly. A filter that rejects 4 orders of magnitude at 800 nm accomplishes nothing for the 2 % of flux that goes around it.
MIL-STD-3009 also constrains chromaticity, specifying CIE 1976 coordinates for NVIS Green A, NVIS Green B, NVIS Yellow, NVIS Red, and NVIS White. The filter that solves the radiance problem shifts the chromaticity, so the three requirements have to be closed simultaneously rather than in sequence.
Modeling the Panel Assembly
The value of a TracePro model here comes from including the parts of the assembly that are not optical components. The leak paths run through them.
Import the full mechanical stack
Import the panel from SOLIDWORKS or STEP with the bezel, the filter carrier, the light guide or diffuser, the PCB, and the legend plate all present. Model the filter as a solid with real thickness and real edges rather than as a surface property applied to the diffuser, because the escape paths of interest travel around the filter edge and through the filter side walls. A filter modeled as a zero-thickness coating has no edges and therefore no edge leakage, which removes the failure mode from the simulation entirely.
Spectral source definition
Define the LED with its full spectral power distribution out to at least 1000 nm, not just the visible portion. This is the step most often done wrong. Manufacturer datasheets frequently plot spectra only to 700 or 780 nm, which is exactly where the NVIS problem starts. Request the extended spectral data from the supplier or measure it. An LED whose published spectrum ends at 700 nm may carry 0.5 % of its radiant power between 700 and 900 nm, and that fraction alone can drive an NRb failure.
Sample the source spectrum finely enough to resolve filter transitions. A minus-blue NVIS filter has a steep cut with several orders of magnitude of change across 30 nm. Sample at 5 nm intervals or finer through the transition region, since coarse sampling straddles the edge and produces transmission errors of an order of magnitude.
Filter and coating properties
Apply measured transmission data across the full modeled band, including the deep rejection region. Vendor curves often stop plotting once transmission drops below 0.1 %, but the NVIS calculation is sensitive to what happens at 0.001 % because the goggle weighting is aggressive there. Extrapolating the rejection floor optimistically produces a model that passes while the hardware fails. Where measured data runs out, hold the last measured value flat rather than extrapolating downward.
Computing NVIS Radiance From the Trace
TracePro produces spectral radiance at a detector, and the NVIS metrics follow from weighting that output. Place a detector surface at the position and orientation representing the goggle location, typically viewing the panel along the normal at the design eye position. Collect radiance data with the spectral dimension retained rather than integrating to a single value during the trace, since the weighting has to be applied wavelength by wavelength. For NRa, weight the resulting spectral radiance by the Class A goggle response and integrate across the band. For NRb, apply the Class B response instead. The output is in NVIS radiance units directly comparable to the MIL-STD-3009 limit for the lighting class in question. Run the photopic calculation from the same trace by weighting the identical spectral radiance data with the photopic luminous efficiency function. This gives panel luminance in foot-lamberts. Reporting both from one simulation is what makes the tradeoff visible: raising drive current to gain 0.2 foot-lamberts of readability raises NVIS radiance by the same proportional amount, and the model tells you immediately whether the margin exists.
The ratio of photopic luminance to NVIS radiance is the figure of merit worth tracking through design iterations. Improvements to filter rejection or leak sealing raise this ratio. Increases in drive current leave it unchanged. A design change that raises luminance without moving the ratio has not solved anything, it has only spent margin.
Finding the Leak Paths
The specific value of TracePro in NVIS work is path identification, because the failure is almost always a path rather than a spectrum. Use ray path sorting to separate rays reaching the detector by the sequence of surfaces they encountered. Rays that passed through the filter solid form one group. Rays that reached the detector without intersecting the filter form another, and that second group is the leak. Its magnitude relative to the total, weighted by the goggle response, tells you how much of the NVIS radiance budget the leak consumes.
Common paths worth checking explicitly:
Edge gap leakage between the filter carrier and the bezel. A 0.3 mm annular gap around a 40 mm panel represents a small fraction of the emitting area but carries entirely unfiltered light, and unfiltered near-infrared is roughly three orders of magnitude brighter in the goggle band than filtered light. A gap contributing 0.2 % of the visible flux can contribute more than half the NVIS radiance.
Substrate piping. Light injected into an acrylic legend plate travels by total internal reflection and exits at a scratch, a mounting hole, or a machined edge, bypassing a filter that only covers the front face. Model the legend plate as a solid with its real refractive index and check for exit points around the perimeter.
Second-surface reflection. Light that leaves the panel, reflects from the canopy or a glareshield, and returns can reach the goggle without passing back through the filter. Include the canopy geometry with measured reflectance when the panel sits where this path is plausible. Once a leak is identified and quantified, the fix is mechanical: extend the filter to overlap the bezel, add a gasket, blacken a substrate edge, or relocate a mounting hole. Re-trace to confirm the path is closed rather than merely reduced.
Dimming Range and Uniformity
NVIS panels operate across a wide dimming range and the requirements do not scale together. At full brightness the design is usually limited by NVIS radiance. At the lowest dimming setting it is usually limited by uniformity and by the discrete steps of the dimming control, since a panel that looks uniform at 5 foot-lamberts often shows visible LED hot spots at 0.05 foot-lamberts where the eye is fully dark adapted and far more sensitive to spatial variation. Model uniformity at the low end explicitly. Place a detector at the panel front face, set the bin size to roughly 1 mm, and examine the luminance map. Variation that reads as acceptable at 10 % contrast in daylight becomes obvious at night. Target luminance uniformity better than 2:1 across a legend area, and better than 1.3:1 within any single legend character.
The design levers are LED count, light guide extraction pattern, and diffuser strength. Increasing diffuser haze improves uniformity and costs on-axis luminance, which is a tradeoff TracePro quantifies directly by tracing the same geometry with different measured BSDF data on the diffuser. Adding LEDs improves uniformity and raises total near-infrared output proportionally, which spends NVIS margin. Refining the light guide extraction pattern improves uniformity without either penalty, which is why it is usually the right place to invest simulation effort. NVIS compatibility is decided by geometry as much as by spectrum. The LED and filter selection is normally right from the beginning, and the failures come from edge gaps, substrate piping, and reflection paths that route unfiltered near-infrared light to the goggle. Those paths are invisible in a spectral calculation and obvious in a non-sequential ray trace.
TracePro models the panel as built, traces the full spectral output to 1000 nm through every path in the assembly, and reports photopic luminance and NVIS radiance from the same simulation so the readability and compatibility requirements can be closed together. Ray path sorting names the specific leak and quantifies its contribution to the radiance budget, which turns a test failure into a bounded mechanical fix.
To request a TracePro demo covering NVIS radiance weighting and panel luminance