Smart Scan strategy for Electronic Warfare
DRDO · Clean & Green Technology · Software
A clean sequential decision problem with a fair baseline and simulation explicitly sanctioned, which is a rare position for a defence statement — the risk is presentational rather than technical, so plan how you will make a general panel feel why dwelling in the right band at the right moment is hard.
Data: J. C. Wise, Radar Emitter Database, 2024 — https://huggingface.co/datasets/alan-turing-institute/turing-synthetic-radar-dataset
What it actually is
To find hostile radar and radio transmissions you have to sweep a receiver across a very wide band, but the receiver can only listen to a small slice at a time, so where it points and when decides what it catches. Current practice sweeps on a fixed plan drawn up before the mission, which wastes time on harmless transmitters and misses new ones. The ask is a scheduler that learns where to listen next from what it has already heard.
What to build
Two connected pieces: a simulated radio environment holding ground truth on which emitters are transmitting in which band at each time step, including frequency-agile and spatially scanning emitters, together with a receiver model whose instantaneous bandwidth is much narrower than the band it must cover; and a learned scheduler deciding which band to dwell on at each step to minimise intercept time and maximise interception ratio, trained on the hits and misses the simulator returns, evaluated against the figures of merit the statement names — probability of detection and false alarm, sensitivity, average intercept rate, average reward, percentage of correct predictions and average intercept time error — and benchmarked against the conventional open-loop sweep it is meant to beat, with a separate treatment of the periodically scanning emitter case.
Smallest thing that wins the room
Run your scheduler and a fixed open-loop sweep side by side over the same simulated environment and show the intercept time curve separating, with the learned receiver locking onto a newly appearing agile emitter while the sweep is still working through empty bands.
How crowded this one gets
A guess, projected from the 2025 statements — the last year where both the submission counts and the winners were published.
Quieter than 59% of the 226 · #93 of 226 by expected field
A normal-sized field. Your idea has to be good, not miraculous.
Why: defence, intelligence and space bodies drew small fields.
This is a guess, not a fact
Nobody has published 2026’s numbers yet. This is an analysed estimate from last year’s pattern, so please do not take it as the truth — check the live counter on the SIH portal before you decide anything. The range covers the middle half of likely outcomes, so one statement in two lands outside it. Entry closes at 500 ideas per statement, so no range goes past that — a statement that reaches the cap fills and shuts rather than drawing an unlimited crowd. The model reads only three things a team can see before choosing — software or hardware, the theme, and what kind of body posted it — and those explain about a quarter of the variation in last year’s field sizes (R² 0.25 on held-out statements). Trust the band more than the number, and the ordering more than either. It cannot see how good your idea is, which is the part that actually decides it.
The scores
The number is the shorthand. The line under it is the reason.
Acceptance potential
4/5Genuinely underrated — the statement sanctions simulation as the ground truth so the circularity objection that sinks most defence statements does not apply, the objective is a clean sequential decision problem where you can beat a stated baseline numerically, the figures of merit are given, and electronic warfare is a domain almost no student team will enter.
Feasibility
4/5Unusually for a defence statement the ground truth is meant to be simulated — the statement itself asks you to build an RF environment model with known emitter status per band and time slot — so the usual objection that you have no real data does not apply here, and a synthetic radar emitter dataset is linked to seed the emitter behaviours.
Innovation scope
5/5No method is prescribed anywhere — the statement defines the objective and the figures of merit and leaves the entire approach open, so whether you frame this as a restless multi-armed bandit, a partially observable decision process or a reinforcement learning policy is your call and the framing itself is the contribution.
Clarity
3/5The figures of merit are named with unusual precision and the two-dimensional nature of the search problem is explained well, but the receiver parameters, the band count, the emitter behaviour models and the evaluation protocol are all left undefined, and the dense domain vocabulary makes the statement hard to parse without prior exposure to electronic support.
Effort
HeavyA defensible RF environment simulator, a receiver model, a learned scheduler, baseline implementations to compare against and an evaluation harness across seven figures of merit is four workstreams, with the simulator quietly the most important because everything else is measured against it.
Demo-ability
MediumA spectrum waterfall with the receiver's dwell decisions overlaid and an intercept-rate curve pulling away from the open-loop baseline is genuinely persuasive to anyone who knows the domain, but it is abstract and a general judge will need the problem explained before the result means anything.
In its favour
- Green flag: The statement explicitly asks for a simulated RF environment with truth information, which means simulation is the sanctioned methodology rather than a workaround — this removes the data objection that undermines nearly every other defence statement here
- Green flag: The figures of merit are enumerated, so your evaluation protocol is specified for you and results are comparable rather than self-defined
- Green flag: The open-loop sweep gives you an obvious, fair and unarguable baseline, so you can demonstrate improvement as a number rather than as a claim
- Green flag: Electronic warfare is intimidating enough that very few teams will attempt this, and a well-formulated bandit or POMDP treatment would stand out sharply
Against it
- Red flag: Your scheduler's performance is a property of the emitter model you wrote, so a policy tuned to your simulator's regularities may be learning your assumptions rather than a general strategy — vary the emitter behaviours adversarially and report across them
- Red flag: The domain vocabulary is dense and a general judging panel may not follow why the problem is hard, so budget real time on explaining the two-dimensional search before you present any result
- Red flag: Frequency-agile emitters that hop unpredictably have an information-theoretic ceiling on interception, and a submission claiming near-perfect interception against them has a modelling error somewhere
- Red flag: The statement asks separately for an approach to optimally intercept a periodic scan receiver, which is a distinct analytical sub-problem teams will read past and skip entirely
What you will be writing
- restless multi-armed bandit scheduling
- POMDP belief-state tracking of emitter activity
- reinforcement learning with intercept reward shaping
- RF environment simulator with per-band truth states
- Thompson sampling / UCB baselines
- intercept time and interception ratio evaluation harness
- Electronic warfare
- Sequential decision making
- Spectrum surveillance
Prior art to read before you start
adaptive receiver scan scheduling · reinforcement learning for search allocation · emitter interception probability modelling
Analysed by Claude Opus. Every score above is a judgment call with its reasoning attached — kindly cross-check this against the official statement on the SIH portal before your team commits to it.