Terrain-Aware Counter-UAS System

Where it stands · October 2026

Status

Done

Physics computed from ITU-R and IEC standards
Software tested in simulation: bearings recovered to within 0.1°
Full-scale acoustic prototype built and tested at sea level

Not yet done

No altitude, chamber or flight results yet
Corridor gate designed, not built
Next: refine the prototype and test it at high altitude

Limits, stated plainly

Acoustic range is hundreds of metres, not kilometres
The gate gives crossings, not tracks
Warning time is about 20 s

Performance numbers on this page are targets; costs and schedule are estimates; the tone shift, bearing biases and capability forecast are predictions.

01 · The problem at 4,500–5,000 m

Problem

01Degradation

Cold, thin air, icing and dust wear down the hardware. Hardening fixes this.

02Occlusion

Ridges block line of sight; fog and snow blind the optics. Hardening cannot fix this, and in mountains it dominates.

Key insight: sense at wavelengths that terrain and weather don’t block.

Situation	Our sensing channel	Camera / optics	Source
Over a 100 m ridge, 2 + 2 km path	≈ 24 dB loss for a 200 Hz rotor tone	≈ 89 dB	ITU-R P.526-16
Dense fog	GNSS L1 ≈ 0.0012 dB/km (77 GHz radar ≈ 1.8 dB/km)	≈ 315–340 dB/km	ITU-R P.840-9

Computed, and independently recomputed to within 0.5 dB.

02 · Terrain-aware C-UAS

System

The terrain picks the sensing wave, and the elevation map (CartoDEM) is an input to fusion. The chain adds to DRDO’s existing D4-class system; it does not replace it.

Corridor gate

Drone crossing at a pass
Doppler signature

Acoustic array

Bearing past the ridge
3-node triangulation for range

DEM terrain mask

Flags whether the drone is in line of sight

Cue

Line of sight: D4 EO/IR and laser
No line of sight: Gate jammer (Idea B) or crest head (Idea C)

Detect and cue. Neutralisation means re-siting and cueing existing soft-kill, not building a new weapon. The gate jammer denies the route; it does not down the drone.

Five contributions

01Diffraction as the design point

Sound bending over the ridge is what the system relies on, not an impairment it fights.

02Two altitude errors, found and fixed

A sound-speed bearing bias of up to ~12°, and a ×1.29 rotor-tone shift.

03Nested array Built

A compact core rules out ghost targets; far pods on unequal arms give precision; all on one clock. Self-survey and self-measured wind are on the roadmap.

04DEM-aware localisation Roadmap

Treats the terrain as a known lens, predicting how the bearing bends over a crest.

05Silent corridor gate Designed

A receive-only gate that keeps working alongside our own jammer.

03 · Idea 1 (primary) · Ridge-diffraction acoustic array

Array

Built prototype · the “nested array”

8Microphones, two boards, one clock

Built · sea-level tested

Hub core	0.30 m regular tetrahedron of 4 microphones on a 0.8 m mast
Outrigger pods	4 pods on unequal arms of 1.1 / 1.7 / 2.3 / 3.1 m
Microphones	8 × INMP441 I²S MEMS, digitised at the capsule (no analog cable)
Boards	2 × ESP32-S3 (4 capsules each) on a shared clock
Materials	Al 6061-T6 frame and enclosure · graphene clear coat · closed-cell PU insulation foam

No performance figures are claimed for the graphene coat.

The full-scale prototype.

Nested array geometry.

Shared clock between boards

Board 1 is the clock master; board 2 runs as a slave
A READY handshake: no clock edge exists until both boards are listening
Every packet carries an absolute frame index; the host merges the streams on it

Long runs to the pods

Plain I²S can’t run 3 m, so the pods use differential RS-422 over Cat5e
Blue pair BCLK · orange WS · green SD/READY · brown power/ground
AM26LV31/32 or MAX3490 line drivers (MAX485 is too slow); 74LVC157 multiplexer at the hub

Processing

GCC-PHAT time-delay estimation, band-limited on the cross-spectrum over 80 Hz–4 kHz. A known bug is fixed: filtering before PHAT inverts the delay
Harmonic-comb detector finds any rotor or propeller harmonic series, and rejects generator hum because hum is too stable
Temperature-corrected sound speed
Echoes off rock faces show up as separate lag peaks; only lags consistent across all microphone pairs are kept

Built-in

Self-test (BIT) every 10 minutes
Jammer BLANK input mutes processing while our own jammer fires

Range: triangulation across 3 nodes, 150–200 m apart

≈ 3%range error at 200 m

≈ 9%range error at 500 m

≈ 17%range error at 1 km

Targets, assuming 2° bearing accuracy. Spectral-tilt ranging was dropped because it saturates.

Code

`array_node.ino`	Firmware for 1 or 2 boards, with the READY handshake
`arraylib.py`	Square, tetra and nested geometries
`capture.py`	Merges the two streams and reports drops
`analyse.py`	Analysis; in simulation, bearings are recovered to within 0.1°

Planned field refinement · Build Guide v3

v3Two rigid tetrahedra

Roadmap · not what is built

The field node replaces core-plus-pods with two 0.25 m-edge tetrahedra, each with its own ESP32-S3 and LiFePO4 battery, joined by a shared RS-422 clock and an acoustically measured 4 m baseline. Sampling at 48 kHz.

Edge	Aliasing ceiling	Margin over 516 Hz
0.30 m	521 Hz	1%
0.25 m	625 Hz	21%

At −30 °C and 5,000 m the 400 Hz harmonic shifts to 516 Hz.

Wind from the array itself: two-way chirps between the halves, t_ab = d/(c+u), t_ba = d/(c−u); 39 µs per m/s at d = 2 m. Needs the shared clock.
Limit found: picking the right grating lobe in one snapshot works only up to a 4–5 m baseline; 10–20 m needs multi-hypothesis tracking.
Cost: node ≈ ₹24,700 (estimate).

Planned field node (render).

04 · Idea 2 (secondary) · Satellite forward-scatter corridor gate

Gate

Crossings at the choke points

Designed · not built

Principle. A drone crossing the satellite-to-receiver line scatters the signal strongly forward (Babinet’s principle). The sensor is receive-only and gives crossings, not tracks. Gate width is the Fresnel radius, √(λ·d).

Siting. Terrain that blinds radar also funnels a hidden drone into passes, nallahs and valley floors. Gates sit forward at those choke points, not on a perimeter: warning comes from where the gate is placed, not from how far it sees.

Gate siting at a mountain pass.

Configuration	Hardware	Notes
Field (operational)	Ku-band geostationary beacon, single channel, fixed baseline	≈ 49× the radar cross-section of GPS L1 at 11 GHz; the dish rejects ground jammers; the LNB needs 13–18 V but the RTL-SDR v4 bias-tee gives 4.5 V, so a separate inserter is required
Bench / development	One patch antenna, one RTL-SDR v4, GNSS-SDR	One channel is enough because the reference is a software replica of the public C/A code

Signal and discrimination

Signature: a Doppler chirp of ≈ 17 Hz/s, from v²/(λ·d_r). A rate, not a frequency, so it is matched-filterable
A real crossing appears on one satellite; faults, scintillation or cloud appear on all. The receiver processes ≥ 4 satellites and rejects anything common
A jump in the GNSS noise floor on every satellite at once is reported as a hostile-EW alert, not a failure
Receiver thermal drift cancels in the cross-correlation

Reach and cost

50–80 mrealistic reach

₹20,300per gate (estimate)

Roadmap Coverage planner: DEM + satellite orbits → fraction of each hour each corridor is covered.

05 · Neutralisation strand

Deny

“Deny at the choke point, not through the hill.” All numbers are computed, not measured.

1B · Gate-triggered choke-point jammer

The lead mode, keyed only on a gate crossing
10 mW nominal (100 mW ceiling) on L1; 0.1–1 W per C2 band (2.4 / 5.8 GHz) into a sector antenna down the pass
≈ 14 µW breaks L1 lock at 80 m; denial bubble ≈ 2 km in-beam at 10 mW, contained by the valley walls
1 W jammer vs a 100 mW controller: C2 J/S ≈ 41 dB. Losing C2 usually triggers return-to-home

2C · Ridge-crest slave effector

The fallback: D4’s own jammer or laser on the crest to restore line of sight
L1 then needs ≈ 8.7 mW instead of 71 W
The crest head is the most exposed node

3A · Jamming through the ridge · rejected

≈ 71 W for GNSS alone, with ≈ 180 km of line-of-sight collateral denial
≈ 6.3 kW (2.4 GHz) or ≈ 15 kW (5.8 GHz) for C2

Why not spoofing?

Ruled out by cue accuracy (±2°, ~140 m cross-range at 4 km) and by collateral, not by power.

06 · High-altitude engineering findings · our own derivations

Altitude

313 m/sspeed of sound at −30 °C, not 343 m/s

+11.9°uncorrected bearing bias at 60°

×1.27–1.29rotor speed-up in thin air: 200 Hz → ≈ 255–258 Hz (prediction)

×1.8convection derating at 4,500 m (Rhee)

Finding	Numbers	Fix
Sound speed	Bearing off by +3.3° at 30° and +11.9° at 60°; no solution at all beyond 65.7°	Measure it: a ₹100 temperature sensor
Rotor tone shift	`f_alt = f_sea·√(ρ_sea/ρ_alt)`; drones also radiate ≈ 4.7 dB more quietly (prediction)	Processing band widened upward
Pointing	Micro-radian pointing	Designed out: no steered aperture
Cable rigidity	Cable stiffness at altitude	Removed by digitising at the capsule
Cooling	×1.8 convection derating at 4,500 m	Fanless conduction cooling, which also closes dust and snow paths
Vortex shedding	A 10 mm arm sheds at 200 Hz in a 10 m/s wind, inside our band	Helical strakes
Arm rigidity	8 × 1 mm 6061-T6 tube, iced, 30 m/s gust: 0.74 mm deflection, a tenth of a sample, safety factor ≈ 13 (a 6 mm tube is marginal)	8 × 1 mm tube arms
Enclosure	A sealed box sees ≈ 47 kPa, or ≈ 4.3 kN on a 0.3 m lid	IP67 with an ePTFE vent beats sealed IP68

Other hardening choices

Silicone elastomers only; gold M12 IP67 connectors
LiFePO4 cells kept warm by processor waste heat in the PU-foamed bay, with charge-inhibit below 0 °C
Two-position mast (0.8 / 1.6 m); rock-bag ballast
Rigid, heatable, hydrophobic microphone screen
Graphene coat applied with bond points masked, so the EMI cage stays continuous

Portable unit

≈ 1.7 kgper node

450 mmlongest part

≈ 6 kgwhole system

07 · Hardening DRDO’s existing D4-class hardware

Rugged

The problem statement puts special emphasis on DRDO’s existing EO/IR and laser gimbals, radar and jammer: cable rigidity, derating, structural deformation, wind, and drift in micro-radian pointing. The ruggedisation document (17 pp) opens with a clause-by-clause matrix.

Area	Computed numbers
Derating at altitude	IEC 60664-1 electrical clearance × 1.48 at 5,000 m; water boils at 83 °C
Gimbal loads (assumed 0.4 m turret)	Wind torque 0.42 N·m in a 30 m/s gust, so holding 10 µrad needs ≥ 42,000 N·m/rad of stiffness
Structural and optical deformation	Aluminium optical bench bends 354 µrad per 5 K gradient · steel bearing in an aluminium housing gains +44 µm clearance at −60 K · motor Kt/R ×1.40 at −40 °C · germanium lens defocuses 0.9 mm against a ±29 µm depth of focus
Material rule	Uniform soak: choose low expansion α (Invar, CFRP). Heat-flux gradients: choose low α/k; SiC is 7× better than aluminium, and titanium is worst
Qualification	MIL-STD-810H methods 500, 502, 503, 505 and 521, plus JSS 55555

Cold-start cable-torque map

A slow sweep learns the drag as τ_c(θ,T) and cancels it; the drag is also trended as a health signal.

Capability forecast

“How well will we perform tonight”, built from the compensation models (prediction).

All values are taken from public descriptions or assumed; none are DRDO data.

08 · Judge stress test · nine attacks and our answers

Stress test

Attack	Answer
“You answered a different problem statement”	Performance is the probability of detecting, tracking and engaging in time. We harden DRDO’s hardware and fix terrain blinding. Cueing cuts the EO search area by ≈ 50× (±3° → a 6°×6° box), and the masking flag stops wasted dwell.
“Detecting what you can’t engage only buys seconds”	Yes, about 20 s. So the argument is track continuity through pop-up-and-peek and a pre-pointed effector, not lead time.
“Your own jammer kills the GNSS tripwire”	The Ku beacon is the field configuration and its dish rejects ground jammers; there is a BLANK input and look-through windows; GPS jamming becomes an alert.
“A satellite tripwire contradicts your occlusion argument”	It is a corridor gate. Terrain funnels hidden drones into passes, and the gates sit forward at those choke points.
“Knife-edge diffraction is the wrong model for sound”	ITU-R P.526 is used only for the radio-versus-optics comparison. The node measures wind itself and reports wind shadows in a live confidence map; range is stated as wind-dependent.
“Your bearing points at the ridge, not the drone”	DEM-aware localisation predicts the bending (roadmap). Arrivals at crest height are flagged as diffracted.
“Your threat model is assumed”	Terrain-hugging is how drones beat radar. The harmonic comb detects any rotor or propeller.
“You’ve multiplied failure points”	Nodes go only in the dead ground, so losing one loses one sector. Self-test runs every 10 minutes, and one person can set up a node.
“What’s actually new?”	Terrain-aware C-UAS, plus the five contributions above.

Why not a radar on the ridge?

A radar emits, so it can be direction-found. A valley post has several dead sectors, which means several radars on exposed ridges. One acoustic sector costs ≈ ₹25k; a radar costs crores.

Fixed-wing drones or piston-engined munitions?

The comb detector finds any harmonic series.

09 · Estimates

Cost

Item	Estimate
Field node	≈ ₹24,700
Gate	≈ ₹20,300
One valley post (3 nodes + 2 gates)	≈ ₹1.15 lakh, excluding the gate jammer
Cost-down variant (shared battery, cheaper processor, phased trial)	≈ ₹0.9 lakh for the full system; ≈ ₹56k for a 2-node + 1-gate trial set
1/5-scale bench demonstrator (optional)	≈ ₹6,650

≈ ₹1.15 lakhper valley post

Scales by terrain, not by perimeter: add nodes in dead ground and gates at passes.

10 · Validation plan

Validation

Stage 1 · Bench

Each subsystem on the bench first. Stage 3 of either waits until stage 1 of both works.
Stage 2 · Real instrument

Chamber programme to MIL-STD-810H, including thermal cycling −40/+60 °C × 50; cold-soak deploy in mitts; wind tunnel with and without strakes; dust, submersion, shaker sweep, ice accretion.
Stage 3 · Outdoors

Field experiments E1–E5 and F1–F3, after DGCA clearance for any target flight.

First resultE2, a drone behind an embankment: publishable, and needs no mountain.

Research gaps

Seven research gaps (G1–G7), identified in a literature sweep in September 2026, shape the experiment plan.

Open engineering items

Bench-verify ESP32-S3 slave full-duplex
If board 2 resets mid-session, frame alignment is lost: detected, not yet recovered automatically

11 · Context

Context

Built for Smart India Hackathon 2026, problem statement SIH26050: “High Altitude Performance Optimization and Robust Design of Anti-Drone System” (DRDO · Robotics and Drones · Hardware). I lead team GOODFELLASS (Team ID 152794) through the idea-submission round.

The system is the first and only product of Aeroscan Defence Systems Pvt. Ltd., my defence-tech startup, which is being proposed to the startup incubation cell at Sathyabama Institute of Science & Technology.

An earlier version fused RF, radar, EO/IR and acoustic sensors with weather-based weighting. It was dropped in favour of the 8-microphone acoustic array plus the corridor gate, which address occlusion directly.

Deliverables

Idea deck v3: 6 slides on the official SIH 2026 template
Walkthrough video (~4:15, 7 chapters), plus a demo that walks a speaker around a building corner to show diffraction live
Build Guide v3 (16 pp) and Ruggedisation document (17 pp)
5-page plain-English judge brief and a 16-page technical annex
Acoustic array firmware and Python host tools

Next project

Drone-Based Landfill Survey System

View project