An open-source optical bhangmeter — a fast photodetector that recognises the unique two-peaked light curve of a nuclear detonation and estimates its yield from the timing of that signature. This repository holds the complete build: firmware, host software, PCB, and enclosure.
A bhangmeter estimates yield from the optical waveform alone. No other natural or artificial event reproduces the double-pulse signature, which makes the waveform both the detection trigger and the measurement.
A tour of the build, front to back:
Carrier board, front — the analogue signal chain: AD8304 log amp, LT1711 comparator for edge timing, and AD7091R ADC for waveform capture.
Rear — the BHG-3000 optical sensor head and the RP2350 Pico 2 W (RP2W) that runs the firmware and handles WiFi upload.
The companion app — a PySide6 desktop console for driving the device, watching live waveforms, and reviewing captured events.
A high-yield fireball radiates ~35–45 % of its energy as light, but not smoothly. Three phases produce two distinct optical peaks:
- First pulse — radiative fireball. The fireball surface cools through ~10⁴ K and briefly shines bright in the visible — a ~10,000 K source, peaking in the UV (~290 nm) with a strong visible tail. Short (~1 ms) and nearly independent of yield — its timescale is set by air opacity, not energy.
- The minimum. A hydrodynamic shock detaches and races ahead, ionising the air into an opaque shell that hides the (still hotter) interior. Visible emission drops about one order of magnitude (ten-fold) — the minimum sits at roughly 0.1 of the second-peak power, only ≈250 mV below the peaks on the log amp's output. The light never goes dark between the pulses; the dip is shallow (NWA FAQ §5; G&D Fig. 7.84).
- Second pulse — shock breakaway. The shock cools below ~3000 K and turns transparent again, revealing a ~7500 K interior — near solar temperature, its blackbody peak ~380 nm with energy spread across the whole visible band. Lower peak, but it lasts hundreds of ms to seconds and carries ~99 % of the integrated thermal yield.
The gap between the two peaks scales with yield — and that gap is what the instrument measures. The firmware uses standard low-altitude Glasstone & Dolan scaling[^gd]:
t_inter ≈ 41.7·W^0.44 ms t_min ≈ 1.87·W^0.576 ms
The two times scale with different exponents, so t_min/t_inter is not a
fixed 1/10 — it runs from ≈0.06 at 10 kt to ≈0.115 at 1 Mt. Inverting gives two
yield estimates, but they are not co-equal: t_inter is the primary estimator
— a sharp feature, and simulation against the literature waveforms recovers yield
to roughly ±15 % across 10 kt – 1 Mt at 5–20 km — while t_min is a coarse
cross-check. The minimum is a shallow, flat-bottomed feature, so its measured
time carries tens-of-percent uncertainty that the W ∝ t^1.7 inversion amplifies.
Acceptance is therefore driven by t_inter, with t_min only required to land in
its (wide) window.
| Yield | T_min (peak→min) | T_inter (peak→peak) |
|---|---|---|
| 10 kt | 7 ms | 115 ms |
| 50 kt | 18 ms | 233 ms |
| 100 kt | 26 ms | 316 ms |
| 250 kt | 45 ms | 473 ms |
| 500 kt | 67 ms | 642 ms |
| 1 Mt | 100 ms | 871 ms |
The device reports a yield only when the timing lands in the valid window —
t_inter 90–1050 ms, t_min 5–130 ms — i.e. the 10 kt – 1 Mt design envelope
with slack. (High-altitude bursts lack the hydrodynamic minimum and fall outside
this instrument's scope.)
The comparator's job is onset timestamping only: it captures the sub-µs
leading edge at its 1.03 V threshold (over a ~0.9 V background) and triggers the
ADC capture. Both t_min and t_inter are then measured from the ADC waveform
using reversal-hysteresis peak/valley detection — because the shallow,
one-decade minimum never recrosses the comparator threshold mid-event for any
in-envelope signal, so the timing cannot be taken from the comparator alone.
A detection therefore requires all of: a fast leading edge, then a
peak → shallow-minimum → second-peak shape on the ADC trace, with t_inter
falling in the valid 10 kt – 1 Mt band. That combination rejects the plausible
impostors — none of which reproduce the full sequence:
- Lightning — a single peak, typically <50 ms, no slow second hump.
- Camera flash, welder strike, arc fault — one fast pulse, no second peak.
- Sun glint, aircraft strobe — slow rise, no millisecond-scale leading edge.
- Meteor / bolide — one broad envelope, not peak → minimum → peak.
The photodiode sees a huge range of light: at 5 km a 1 Mt first pulse drives milliamps of photocurrent, while a far, hazy 10 kt second pulse sits down in the nanoamps — about seven decades. Rather than switch gain mid-event, the AD8304 logarithmic transimpedance amplifier compresses that whole span onto one analog output:
VLOG = 246 mV × log₁₀(I_pd / 1 pA) ⇔ I_pd = 10^(VLOG / 0.246 − 12)
The 246 mV/decade slope is the front-end's measured/design figure. A fixed OD 2.0 neutral-density filter (1 %
transmission, effective aperture A_eff = 1.51 × 10⁻⁸ m²) pulls the bright
extreme back under the AD8304's 10 mA ceiling. First-pulse saturation is
accepted: the comparator has already timestamped the leading edge to sub-µs, and the
log output stays valid through the dimmer, slower second pulse that carries the timing.
Two caveats on the optical model. Spectral: the Glasstone & Dolan pulse-shape data is total thermal power, but the unfiltered silicon photodiode weights ~400–1100 nm, so the in-band peak-to-minimum contrast may differ modestly from the broadband ten-fold figure. Burst height: per G&D §7.85 a contact surface burst behaves like an air burst of roughly twice the yield, lengthening the timings ~36 % — one reason the accept window carries headroom at the top.
Across 10 kt – 1 Mt and 5–20 km the photocurrent stays inside the AD8304's 100 pA – 10 mA window and VLOG stays under the external ADC's 2.5 V reference — one fixed-gain front-end, no switching. Bandwidth is modest: the first pulse rises in ~100 µs and is a few ms wide, the second pulse is sub-kHz, and the comparator (not the ADC) carries the fast edge. The whole double pulse lands in one clean trace:
The literature waveforms make the detection strategy explicit. Mapped through the repo's amplitude calibration, the VLOG output clears the 1.05 V comparator threshold (red) well above the 0.9 V background (dashed) on the leading edge, but the inter-pulse minimum stays ≥ ~1.3 V at every preset — so the comparator never recrosses its threshold between the pulses, which is exactly why the second-peak timing is recovered from the ADC trace rather than a second edge:
| Folder | Contents |
|---|---|
| code/ | All software. The firmware (RP2350 / Pico 2 W) and the companion app (PySide6 desktop operator console). See code/README.md. |
| PCB/ | Hardware design — two boards: the carrier (bhangmeterv3) and the BHG-3000 sensor head (BHG-3000). Schematics, assembly drawings + BOMs, and Gerbers. See PCB/README.md. |
| mech/ | Mechanical / enclosure design — 3D-printable models (e.g. STL/STEP) for the housing and optical mount. |
| Block | Part | Why |
|---|---|---|
| Photodiode | VEMD2704 (Si PIN, ~400–1100 nm) | Both flashes are broadband thermal emission — a near-blackbody continuum across UV, visible, and IR, with the energy-dominant second pulse resembling sunlight (~6000–7000 K) (Glasstone & Dolan §7.23, §7.76). Left deliberately unfiltered so the detector gathers as much of that broadband light as possible. |
| Logarithmic amp | AD8304 | Log-compresses the photodiode current so one fixed-gain front-end spans the dim baseline up to a saturating first pulse (several decades) — no gain switching mid-event. |
| Edge timing | LT1711 comparator | Sub-µs timestamp of the leading edge regardless of amplitude (first-pulse saturation is accepted). |
| Waveform capture | AD7091R ADC (SPI) | Records the analog shape so peak/minimum/second-peak can be analysed. |
| Compute / comms | RP2350 (Pico 2 W) | Dual-core: detection + control on core 0, WiFi/NTP/HTTPS upload on core 1. |
| Blast cross-check | Honeywell HSC pressure sensor | (planned) overpressure reference to back up the optical yield estimate. |
VEMD2704 — 1.51 mm² active area, ~0.40 A/W broadband responsivity, 17.6 pF, 70 ns rise, ±67° FOV.
- Build the hardware from PCB/ and mech/.
- Flash the firmware — a prebuilt
.uf2is included, or build from source. See code/optical_bhang/README.md. - Configure on first boot over USB serial (WiFi + a GitHub repo to publish to). No credentials live in the firmware — they're entered at runtime and stored in flash.
- Drive and view the device with the companion app, or read the published JSON from your GitHub repo.
Work in progress. Yield estimation from timing is intrinsically coarse — a 10 %
error in t_inter becomes a ~25 % yield error (W ∝ T^2.27), which is why the
sharp t_inter feature is the primary estimate, t_min only a coarse
cross-check, and a blast cross-reference is planned.
The built-in self-test (BIST) emits a square double-flash whose gap is fully dark, unlike the real signature's shallow one-decade dip; it validates the signal chain end-to-end but is not a shape-fidelity test of the discriminator (the corrected analysis accepts both).
The firmware's TLS upload path does not yet verify the server certificate; use a dedicated, fine-grained GitHub token (see the firmware README's security notes).
- Glasstone, S. & Dolan, P. J. The Effects of Nuclear Weapons, 3rd ed., 1977 (§7.78–7.88).
- Bridgman, C. J. Introduction to the Physics of Nuclear Weapons Effects, DTRA, 2001.
- Brode, H. L. "Fireball Phenomenology," RAND, 1964.
- U.S. AEC, Capabilities of Nuclear Weapons (DNA EM-1), declassified excerpts.