Platform Features

Deploy at the right scale for the mission. All tiers share varta_core for identical classification logic — the same detection engine runs on a compact field unit and a full command center.

Matches detected signals against known drone RF fingerprints to identify the platform type. All tiers detect the same drone categories — the difference is classification depth. Mini identifies 10 generic frequency bands (e.g., "2.4 GHz drone activity"). Pro/Max identify 43+ specific platforms (e.g., "DJI Phantom 4, OcuSync, hovering"). No other sub-$100K system carries military FHSS/OFDM signatures.

• Orlan-10 — FHSS control (850-930 MHz), analog/digital video, 2+ hour loiter reconnaissance
• Lancet — Dual-band LoRa CSS (868 + 902 MHz), S-band video, 40-60 hops/sec loitering munition
• Supercam S350 — Proprietary FHSS (976.5-1021.5 MHz), 10-stripe waterfall pattern ISR platform
• Shahed-136 (Geran-2) — RF-silent GPS/INS baseline with optional RFD900x/LTE comms variant
• Consumer fleet — DJI (19 models), Parrot, Autel, FPV racing (ELRS, Crossfire, Tracer)

Multi-stage classification pipeline combining gradient-boosted noise gating, CNN spectrogram analysis, and rule-based heuristics in a HybridDecisionEngine that breaks the 85.9% heuristic false-positive ceiling.

Correlates RF detections with FAA-mandated RemoteID broadcasts and DJI DroneID telemetry to identify compliant vs "dark" (non-broadcasting) drones. Each sensor node acts as a distributed RemoteID receiver, publishing correlation data to Max where it is aggregated in the tactical dashboard.

• Distributed BLE scanning — Every Mini v2, Pro, and Max node with a USB BLE adapter scans independently and publishes to Max
• WiFi RemoteID Bridge — Dedicated WiFi receiver extends RemoteID range beyond BLE limits via WiFi NaN and Beacon frames
• DJI DroneID / OcuSync Decoder — Extracts serial number, operator position, altitude, speed, and heading directly from I/Q samples. Blue markers on tactical display with telemetry popup
• DroneID tactical display — Blue markers for identified drones, operator position overlay, real-time telemetry popup on the tactical map
• Dark drone escalation — RF signature present + no RemoteID match within 10s = automatic threat level increase
• Anti-spoof validation — RemoteIDAntiSpoof class with RF band consistency check, motion plausibility analysis, operator distance verification. Score < 0.4 triggers escalation
• ASTM F3411-22a — Bluetooth 4/5 Legacy Advertising and WiFi NaN (Neighbor Awareness Networking)
• Hardware — nRF52840 USB dongle recommended. Optional WiFi NaN requires monitor mode + root

Identifies when multiple detected drones are operating as a coordinated swarm rather than independent threats. The system analyzes temporal clustering, spatial convergence (within 1 km), behavioral fingerprinting (same RF band and channel spacing), and formation geometry (triangle, line, or cluster patterns). Escalation is automatic: 1 drone = THREAT, 3+ coordinated = EMERGENCY. "Suspected" vs "confirmed" swarm tiers with tracking_id deduplication prevent multi-sensor double-counting.

• Temporal clustering — Groups detections arriving within tight time windows across multiple sensors
• Spatial convergence — Identifies multiple tracks converging on a common point with prediction overlay
• Behavioral fingerprinting — Synchronized FHSS patterns, coordinated band allocation, formation geometry detection
• Tactical display — Swarm boundary polygons, convergence vectors, and automatic threat escalation on the map
• Validation — 40 tests (31 unit + 9 E2E) covering all swarm detection scenarios

Five geolocation methods in priority order, automatically selecting the best available based on sensor capabilities and synchronization status. From nanosecond-precision TDOA to always-available RSSI fallback, the system adapts to the deployed sensor mix.

Converts field-captured unknown signals into new detection signatures and pushes them to the entire fleet. On Pro, automatic I/Q capture occurs locally when UNKNOWN, CRITICAL, or EMERGENCY threats are detected. The full review-and-deploy workflow — capture, feature extraction, operator review, approval, and signed fleet push — requires Varta Max. Closes the loop from "unknown signal in the field" to "fleet-wide detection capability" in under an hour.

• Auto-capture — I/Q recording triggered on UNKNOWN, CRITICAL, or EMERGENCY classification events with forensic capture
• Feature extraction — Automated spectral analysis generates a candidate signature template
• Operator review UI — Web-based panel for signature validation, labeling, and approval
• Fleet push — Signed config distributed via MQTT to all deployed sensors
• Auto-generated test case — End-to-end regression test auto-generated from each new signature for continuous validation

Every failure mode defaults to the safe state. Invalid data never produces a CLEAR result — the system assumes threat presence until proven otherwise.

• Invalid sensor data → SENSOR_FAILURE — Never returns CLEAR on bad input
• Baseline freeze during anomalies — Anti-jamming protection prevents attackers from shifting the noise floor
• SNR gate (10 dB) — Classification suppressed below minimum signal-to-noise ratio
• TDOA eligibility gating — Unsynchronized clocks rejected to prevent false geolocation
• Health-gated fusion — Degraded sensors penalized 50% in confidence weighting
• Signed config verification — HMAC-SHA256 + Ed25519 signature required for all configuration updates

Detection events map to STIX 2.1 structured threat intelligence objects for machine-readable sharing with partner systems and government agencies. Satisfies DHS LRBAA RO3 requirement.

Detects RF-silent drones using motor/engine acoustic signatures via USB MEMS microphone with MFCC + spectral feature extraction. Trained on a 13,997-segment corpus with balanced retraining achieving 88.9% overall accuracy. Supplementary modality only — acoustic detection never overrides an RF CLEAR determination.

• Training corpus — 13,997 labeled segments with balanced retraining progression and class-weighted sampling
• Class-specific performance — Shahed 92.3% recall, quadcopter 90.9% recall, overall 88.9% accuracy
• 3-tier pipeline — MFCC fingerprint matching → class-generic threshold gates → ResNet-18 class inference
• Shahed-136 terminal phase — GPS/INS-guided with no RF emissions; acoustic rotor signature is the primary detection vector
• Quadcopter rotor profiles — MFCC fingerprints for multi-rotor acoustic signatures at various ranges
• Supplementary only — Acoustic THREAT can escalate but never downgrades an RF-based assessment

Correlates detections from multiple geographically distributed sensors to increase confidence and reduce false alarms. Mini and Pro act as Nodes — publishing detections (and DF bearing data from equipped stations) to the network. Max acts as the Hub — aggregating, fusing, and triangulating across all sensors using the 5-method geolocation priority chain. Three transport layers with encryption, edge buffering for disconnected operation, and full provenance tracking on every detection event.

Browser-based operational interface with PMTiles offline mapping, real-time WebSocket threat streaming, and integrated operator tools for the full detection-to-response workflow.

• PMTiles offline mapping — No internet required; pre-loaded map tiles for field and SCIF deployments
• Real-time WebSocket display — Live threat tracks, status updates, and alert streams with sub-second latency
• Operator position overlay — RemoteID-derived operator locations rendered alongside drone tracks
• Swarm boundary polygons — Visual overlay for coordinated threat clusters with convergence prediction
• Signature review panel — Integrated UI for the rapid signature pipeline — capture, label, approve, push

View the Tactical Command UI demo →

Five independent jammer detection paths ensure no jamming technique evades classification. Each path targets a distinct electronic warfare strategy.

Monitors sensor CPU, temperature, battery, scan rate, noise floor, and clock sync to detect degradation before it causes missed threats or false alarms. On Mini, health monitoring runs locally and triggers local alerts. On Pro, local monitoring plus MQTT health telemetry to the network. On Max, fleet-wide health status is displayed on the dashboard with real-time toast notifications for state changes. Fleet authentication via HMAC-SHA256 ensures only trusted sensors contribute to fusion.

• Auto-degradation triggers — CPU usage, temperature, battery level, scan rate deviation, noise floor shift, time sync drift
• Automatic recovery — Sensor returns to HEALTHY after N consecutive healthy heartbeats
• Health tooltips — UI displays real-time sensor health status with detailed metrics on hover
• Alert toasts — Immediate operator notification on sensor degradation or fault events
• Fleet authentication — HMAC-SHA256 signed heartbeats prevent unauthorized sensor injection

Full lifecycle tracking from discovery through decommission with SensorLifecycleManager managing 25+ state transitions and complete audit history.

• State machine — 25+ state transitions covering discovery, pairing, activation, degradation, fault, decommission
• Audit log — SensorEvent records every state change with timestamp, source, and metadata
• MQTT commands — Remote sensor management via encrypted command topics
• Sensor pairing workflow — Certificate exchange, capability negotiation, and trust establishment
• Unpaired tracking — Monitors unknown sensors appearing on the network for security awareness
• UI detail modal — Per-sensor lifecycle view with full event history and current state

Distributes updated signature databases to all sensors with cryptographic verification and rollback protection. Mini and Pro receive OTA updates and verify signatures before applying. Max acts as the Admin — managing releases, hosting a local mirror for air-gapped deployments, and tracking per-sensor registry compliance across the fleet. If a signature update fails verification, the sensor automatically falls back to its last-known-good database.

• Compliance tracking — Per-sensor version reporting ensures fleet-wide signature consistency
• Default taxonomy — 64-entry classification hierarchy covering all known drone/jammer categories

Deep learning spectrogram classifier trained on 3,804 real-world spectrograms, achieving 90.3% holdout accuracy across 14 drone/jammer classes. Noise false positives reduced from 89.7% to 9.6%. Deployed on Pro via TensorRT INT8 quantization and on Mini v2 via ONNX CPU inference.

End-to-end validation using real RF hardware (PlutoSDR TX → TinySA RX) with realistic channel models. 26 drone/jammer profiles tested through the complete detection pipeline.

• 26 profiles — Covers DJI, Lancet, Orlan, ELRS, Crossfire, GPS jammer, wideband jammer, and more
• Realistic channel — RealisticChannel model with Doppler shift, Rayleigh/Rician fading, phase noise, multi-drone interference
• Results — 23 PASS, 3 XFAIL (known edge cases with documented rationale)
• Determinism — Seeded random number generation for reproducible test runs across environments
• Pipeline coverage — Tests full path from RF transmission through detection, classification, and alert generation