More Pixels or More Eyes? SCD's Mini Blackbird 1280 vs the Boson 640

Why compare these two

The two cores sit at opposite ends of the counter-drone design space. The Boson 640 is a small, low-power uncooled long-wave core, widely used in staring cameras. SCD (SemiConductor Devices)'s Mini Blackbird 1280 is one of the most compact cooled, high-definition mid-wave cores. Every number comes from the manufacturer's own published material, listed at the end.

The two cores on paper

SCD publishes two variants: HOT XBn, and HFM ("Hot Full Mid-wave", 2024 datasheet) with a wider band and higher frame rate.

Two cautions. A Mini Blackbird is not a camera: SCD supplies the detector, dewar and cooler, and an integrator adds the lens, enclosure and video pipeline; a Boson ships as a complete core with a lens. And the NETD figures are measured at different f-numbers, F/4 for the Blackbird and f/1.0 for the Boson, so they are not one test.

What each band actually sees

A small quadcopter's heat peaks in the long-wave band: it comes from motors, speed controllers and the battery, and most of the airframe sits near ambient. Wien's law puts the emission peak of a 300 K (27 °C) object at about 9.7 µm, and of a 340 K (67 °C) motor housing at about 8.5 µm: inside the Boson's 8–14 µm band, well outside the Blackbird's 3.6–4.2 µm.

A 300 K surface emits about 147 times more radiance in 8–14 µm than in 3.6–4.2 µm, and about 405 times more photons (about 36 and 89 against the HFM's 3.6–4.9 µm). Mid-wave compensates twice: its radiance changes faster with temperature, about 4% per kelvin at 300 K against about 1.5% for 8–14 µm, and a cooled photon detector is far more sensitive per photon than a microbolometer. Teledyne FLIR's own guidance places uncooled cameras in the long-wave band because that is where ambient-temperature targets emit most of their energy.

In Teledyne FLIR's fog modelling, LWIR penetrated fog better than MWIR in every case studied, about four times the range in Category II fog; MWIR did best in hot summer or tropical conditions and was more affected by pollutants; in rain the two bands performed similarly. Fudala et al. (2019) measured a DJI Phantom 4 and a DJI Inspire in both bands against clear sky, cloud and land. Such results depend on the scene, so neither band wins outright.

Against a clear sky, a drone in LWIR is a warm object on a cold background; Teledyne FLIR notes that an infrared camera can often detect a moving drone there as a cluster as small as 2 × 2 pixels. Neither band removes warm cloud edges, birds or sun glints; those are an algorithm problem.

Pixels on target: same pixels, fewer cameras

Assumptions: a 0.35 m square target; pixels across the target = size ÷ (range × IFOV), where IFOV = pixel pitch ÷ focal length; Boson lenses are FLIR catalogue lenses. SCD publishes no lens, so every Blackbird focal length below is an assumption, chosen either to match the Boson's field of view or to represent a typical long cooled zoom.

*Computed from pitch, array width and focal length. FLIR publishes 32.0° for this lens, about 20 pixels per degree, which changes these counts by about 4%.

Same pixels on the drone, fewer cameras. Behind an 11.7 mm lens, each 10 µm Blackbird pixel subtends the same 0.857 mrad as a 12 µm Boson pixel behind 14 mm, so a 0.35 m drone covers the same pixels on both: about 1.6 at 250 m, 0.8 at 500 m and 0.4 at 1 km. But the Blackbird spreads them over 57.5° × 47.4° against 30.7° × 24.7°, about 3.6 times the sky per camera, so a 360° ring takes 7 (6.3 before rounding up, with no overlap) instead of 12. Figures are for the image centre.

At the same field of view, the Blackbird has twice the linear resolution, and in pure geometry a sub-pixel drone fills four times as much of each pixel. Its published 25 mK NETD, against 20–60 mK across the Boson family, also gives it the lower noise floor. Sensitivity is its first clear win.

There is a catch at wide fields. Diffraction blurs a point into a spot about 2.44 × wavelength × f-number ÷ focal length wide in angle: about 1.7 mrad (two pixels) for the Boson's 14 mm f/1.0 lens at 10 µm, but about 3.3 mrad (nearly four pixels) for a Blackbird at F/4 and 3.9 µm behind 11.7 mm, and 1.6 mrad behind 23 mm. So at equal pixel density the Blackbird's point image is twice as wide, and its lower noise floor must make up for the spread signal; at matched wide fields its extra pixels sample almost the same blur, and the four-fold fill gain shrinks. This simple model leaves out aberrations and turbulence.

The cooled core earns its keep at long focal lengths — its second and bigger win. Teledyne FLIR notes that cooled MWIR can use lenses as slow as f/5.5, allowing focal lengths beyond 1,000 mm at manageable cost, and puts the cost crossover from uncooled LWIR above about 250 mm. An uncooled core needs f/1-class optics, so its aperture grows with focal length. Behind a 300 mm lens the Blackbird resolves a 0.35 m drone at 2 km with about five pixels, while the 14 mm Boson sees a fifth of a pixel. To identify a small drone at long range, this is the right tool.

Independent commentary by GOTEAM. Not affiliated with or endorsed by SCD or Teledyne FLIR. Specifications are quoted from each manufacturer's published material and have not been independently tested by us. Pixel counts, diffraction spots, blackbody ratios and ring sizes are our own arithmetic, and every Blackbird focal length in them is an assumption, not an SCD product.