האם אפשר לגלות את סיב האופטי הדק של כטב״מ מונחה-סיב?
Fiber-guided FPV drones trail an ultra-thin glass fiber — no radio, so RF detection is blind. This is an honest, cited survey of whether we can detect the line itself: the physics, the real prior art, the range limits, and the cheap experiments a maker can run this weekend.
מוקדש לזכרו של סמל נהוראי לייזר ז״ל — לזכרו — אור ולייזר, כדי לראות את הקו הסמוי.
לזכר סמל נהוראי לייזר ז״ל — דף ההנצחה הרשמי, יזכור ↗
Fiber-optic-guided FPV drones (FOG-D) — now the dominant short-range threat in Ukraine, and reportedly fielded by Hezbollah — carry their video and control as light traveling inside a hair-thin glass fiber, not as radio in the air. There is no RF emission, so jammers, RF sniffers and direction-finders are blind to them by design. The detection conversation has therefore moved to the drone's physical signatures — and to one question almost nobody has seriously attacked: can we detect the fiber itself?
If we can see, range, or sense the trailing line, we get something no other sensor offers: a vector that points back toward the operator, and an unambiguous “fiber-guided” classification. That is the breakthrough this page evaluates — and we report honestly, including where the physics fights us.
Detecting the fiber is physically possible but hard, and difficulty varies wildly per method.
The most promising near-term path is passive optical / event-camera detection of the taut, sunlit fiber against sky: the fiber is provably visible in real battlefield photos, mature wire-detection AI & LIDAR prior art transfers directly, and it is the cheapest to prototype. The highest-upside research bet is catching the infrared light leaking out of the fiber (macrobend / spool / connector leakage at 1310 / 1550 nm) with a SWIR camera — an almost-unspoofable signature whose only unknown is how many photons leak at range, a number nobody seems to have published.
No fielded system today detects the fiber itself — vendors fall back to radar on the airframe. That gap is the opportunity. Most avenues work only at tens to low-hundreds of meters and pair best with conventional airframe detection (radar / acoustic / thermal), not as a replacement. False alarms — spider silk, power lines, guy-wires — are the real adversary, beaten by a polarization discriminator and sensor fusion.
Two facts decide everything: the light is invisible infrared (1310/1550 nm — past the eye and past phone cameras), and the fiber is sub-millimeter and often dark-jacketed. Yet frontline photos show fields and trees draped in spent fiber — so it is optically resolvable at close range.
| Property | Typical value |
|---|---|
| Fiber outer diameter (coated) | ~0.25–0.9 mm (spec sheets cite ~0.35 mm) |
| Glass core / cladding | 9 µm core / 125 µm cladding (single-mode) |
| Jacket | often dark/black polymer — low-visibility by design |
| Operating wavelength | 1310 & 1550 nm infrared — not 850 nm, not visible |
| Tether / spool length | typically 5–20 km, prototypes 40–50 km |
| In-fiber attenuation | ~0.35 dB/km @1310 nm · ~0.20–0.25 dB/km @1550 nm |
Ranked by the probability of a working, affordable, low-false-alarm detector within 1–3 years.
Physics. A taut fiber against bright sky is a high-contrast straight line; even a sub-pixel-wide line dims its pixel (partial occlusion), and a moving line is ideal for event cameras that fire on per-pixel brightness change — turning the line into a clean “time surface.”
Prior art. Skyshield (arXiv 2508.09397, MobiCom’25) does literally sub-millimetre thin-obstacle detection via event cameras — mean F1 ≈ 0.71 at 21.2 ms. UCorr handles “single-pixel/sub-pixel” wires (AUC ≈ 0.99). Decades of helicopter wire-strike CV/LIDAR transfer directly.
Why #1. Closest to existing working systems, cheapest, and uniquely yields the bearing of the tether → operator direction. Weakness (clutter false alarms, short range) is exactly what fusion fixes.
Physics — the most interesting idea. The video/control is light inside the glass. It escapes at macrobends (1550 nm is far more bend-sensitive), at the kilometer-long spool on the drone, and at connectors. A 1310/1550 nm-tuned IR/SWIR sensor could see a line that is self-luminous in exactly the telecom band — an almost-unspoofable signature.
Prior art. Clip-on fiber identifiers & VFLs detect live traffic by catching macrobend leakage with a photodiode every day — the physics is rock-solid. The only open question is magnitude at standoff.
Honest hard part. Launch power is modest, leakage is small, it spreads into ~2π then drops as 1/r². Daylight detection at hundreds of meters is very hard; likely near-field, favorable-geometry, or night/narrowband. But if detected, false-alarm rate is the lowest of any avenue.
Physics. A 0.1–0.9 mm glass cylinder is large vs. wavelength → Mie/geometric scattering with a forward-scatter lobe, specular glints, and ~3.4% Fresnel reflection per glass-air surface — a detectable return at the right angle.
Prior art. HOWARD helicopter LIDAR detects 1-inch (25 mm) wires at ≤200 m; our fiber is 30–100× thinner, so naive scaling cuts range hard. Encouragingly, lab laser scattering maps single ~µm spider-silk fibers with <3.5 mW — sub-10-µm transparent filaments do scatter detectably.
Best use. A short-range laser tripwire/fence the drone-fiber must cross — far more favorable than volumetric search.
Physics. A fiber under flight tension becomes stress-birefringent — it retards/rotates polarized light in a way twigs, dust and most clutter do not. Industry already measures 5–200 g tension in 125 µm fibers at ≤9 m/s via this retardation.
Best use. A cheap snapshot polarization camera bolted onto the Rank-1 pipeline, extracting a degree/angle-of-polarization channel that turns a generic “line” into a confident “glass fiber under tension.” It directly attacks Rank 1’s #1 weakness — false alarms.
The fiber removes only the radio link — the drone is still physical, noisy, warm, magnetized and visible. Each layer with its honest limit:
Radar / micro-Doppler — mass/motion + rotor-blade frequency shifts; works regardless of the silent link (Spotter Global claims dark/fiber tracks ~250–1,050 m). Limit: small low-flyers hide in clutter. Acoustic — a fiber drone sounds like any drone, so this is one of the most reliable layers when RF is silent (FPV ~200–300 m, bearing from a mic array). Limit: wind/ambient noise, quiet props. Thermal/IR — warm motors/battery, 24/7. Limit: dawn/dusk thermal crossover, weather. RF / motor EMI — the C2 link is gone; only faint motor EMI remains, catchable at metres–tens of metres. Verdict: demote to occasional confirmer. Magnetic — the drone is a tiny moving magnetic dipole; limit: field falls as ~1/r³ → single-metre, last-ditch close-pass confirmer only. EO + AI — see the airframe directly; the substrate the line-detector rides on.
AI is the brain. Computer vision spots/tracks the speck and rejects birds; sensor fusion combines these individually-weak channels into one calibrated confidence + bearing. AI raises awareness — it does not pull a trigger; it needs calibrated uncertainty, hard negatives, and a human in the loop.
Why it matters here. You cannot jam what does not transmit, so against FOG-D the counter is detection, not jamming. The tether detector's unique gift — operator bearing and “fiber-guided” classification — is the missing fusion layer. Design tether detection as a new layer cued by airframe sensors: radar/acoustic says “drone, bearing X” → point the line/leakage/polarization camera there.
Power-line / cable detection (LIDAR & vision — helicopters, UAV inspection): mature thin-line CV, catenary fitting, and the documented truth that thin lines blend into terrain/sky and are caught only at limited range. Spider-silk optics: µm filaments detected by low-power laser scattering; model silk as scatterers, not reflectors. Distributed fiber sensing (OTDR/OBR/Φ-OTDR): sub-mm Rayleigh interrogation, and the reverse trick (VFL/clip-on identifiers prove leakage is catchable). Event-camera thin-obstacle work transfers wholesale — it is Rank 1.
Empirically map detection range/contrast of a sub-mm line vs. background & lighting — directly de-risks the Rank-1 winner with hardware you may already own.
Success = a range-vs-background detection curve. One afternoon tells you if Rank 1 is real for your optics.
Measure macrobend/spool leakage at 1310/1550 nm and find the max standoff — the key unknown gating Rank 2.
Success = the first number for “leakage photons vs. range.”
Success = a polarization feature that turns a generic “line” into a confident “fiber” — the cheap false-alarm killer for Experiment A.
Full writeup with worked diffraction math and the complete source list: OPTIC-TETHER-DETECTION-RESEARCH.md in the Project Noam repository.
לזכר סמל נהוראי לייזר ז״ל — אור ולייזר, כדי לראות את הקו הסמוי.
לזכרה של עלמה ז״ל — iddoperez.ai — iddoperez.ai