Product Roadmap

What We've Built

Varta has gone from concept to production deployment in under six months. These are the major capabilities delivered:

• ✓ 43+ drone/jammer signatures + 150+ platform coverage — Consumer, military, NATO, and EW systems validated on 762 real-world I/Q samples across 3 datasets + 13,997 acoustic segments
• ✓ Four-tier architecture shipped — Mini v1 (freq/power), Mini v2 (full I/Q + ONNX), Pro (full I/Q + TensorRT), Max (fusion + command center)
• ✓ Hybrid AI/ML pipeline — Cyclostationary + FHSS + CNN (90.3%) + RF Gate Classifier + benign signal rejector with protocol preamble correlators
• ✓ Acoustic ML detection pipeline in production — 3-stage acoustic flow (MFCC fingerprint → class-generic thresholds → ResNet-18 inference) with 13,997-segment corpus baseline and Shahed-class support
• ✓ 5-method geolocation + DF — TDOA, DF intersection, DF+RSSI hybrid, Hybrid TDOA+RSSI, RSSI fallback with 3-layer DF disambiguation
• ✓ Field capture + rapid signature workflow — Edge upload, Max capture inbox, feature/spectrogram APIs, signature-from-capture, operator review, and OTA push
• ✓ RemoteID + DroneID enrichment — FAA RemoteID correlation, dark-drone alerts, DJI OcuSync decode, and tactical map telemetry overlays
• ✓ Threat intelligence interoperability — STIX/TAXII 2.1 export path for SOC/SIEM and downstream intelligence workflows
• ✓ Fleet security + lifecycle controls — Signed config verification, pairing/auth, lifecycle management (25+ transitions), decommission blocklist, OTA rollback
• ✓ Operational platform hardening — GPU-accelerated DSP, TensorRT INT8 inference, nightly hardware regression, CI/CD deploys, and 26-profile hardware E2E tests
• ✓ Fish Ball GigE SDR — $99 production SDR with GigE interface as primary hardware alongside PlutoSDR
• ✓ Email/SMS/webhook notification pipeline — Configurable alert delivery to operators via multiple channels
• ✓ Sensor firmware OTA lifecycle — Over-the-air firmware updates with rollback and integrity verification
• ✓ Signal temporal persistence scoring — Duration-weighted signal tracking reduces transient false positives
• ✓ Site baseline fingerprinting — Per-site RF environment baseline for improved anomaly detection accuracy
• ✓ GPS-denied operations — 6 time-sync fallback sources ensure TDOA eligibility without GPS
• ✓ Friendly UAV whitelist + deconfliction — Known-friendly drone registry prevents false positive escalation
• ✓ RF-to-acoustic Bayesian fusion — Combined confidence scoring across RF and acoustic modalities
• ✓ Doppler velocity estimation — Radial velocity from RF signal Doppler shift for approach/recede detection
• ✓ ELRS/CRSF frame decoder — Protocol-level decode of ExpressLRS and Crossfire control frames
• ✓ Platt calibration for CNN confidence — Calibrated probability outputs from the spectrogram classifier

Training + Feature Status

Acoustic detection is part of the production detection stack and is being actively hardened for minority-class robustness in contested environments.

• ✓ All-tier acoustic detection path shipped — USB-mic input, RF-silent threat support, and fused confidence output in operational detections
• ✓ 3-stage acoustic decision flow shipped — MFCC fingerprint matching, class-generic threshold gates, then ResNet-18 class inference
• ✓ Acoustic corpus and training workflow established — 13,997 labeled segments with on-device Jetson training/evaluation workflow
• ✓ Balanced acoustic retraining delivered — 88.9% overall accuracy, Shahed 92.3% recall, quadcopter 90.9% recall. Weighted sampling and class-balance tuning achieved targets
• ⟳ Acoustic threshold hardening — Class-specific confidence calibration to reduce background false positives in live deployments

Hardening Tracks in Progress

Current roadmap work from the v3.37.1 program is focused on compliance and model-quality hardening for operational deployments.

• ⟳ FIPS 140-2/3 remediation — Gap closure and implementation updates for government-grade cryptographic compliance
• ⟳ Acoustic model retraining — Balanced sampling pipeline targeting >90% overall accuracy and >80% helicopter recall, plus threshold hardening for field noise resilience
• ⟳ CNN retraining on full RFUAV corpus — Jetson-focused retraining/evaluation for stronger real-world generalization
• ⟳ Multi-sensor field validation — Ongoing DF/TDOA quality checks and fusion stability benchmarking in live RF environments

v3.25.0 – v3.37.1 Program Additions

These capabilities moved from design into shipped workflow and now underpin current operational posture.

• ✓ RF Gate + Benign Rejector — 5-class ML RF gate and I/Q-based benign suppression with WiFi/BLE/LoRa preamble correlators
• ✓ Field capture sync pipeline — Edge upload to Max, capture inbox, feature extraction API, spectrogram generation, signature-from-capture workflow
• ✓ DroneID/RemoteID tactical enrichment — DJI OcuSync decode, operator-position overlays, dark-drone alerting, and map telemetry popups
• ✓ Acoustic 3-tier detector integration — MFCC + threshold + ResNet acoustic classifier added to live pipeline for RF-silent threat support
• ✓ STIX/TAXII 2.1 export workflow — Threat intelligence export capability for SOC/SIEM integration and sharing
• ✓ Sensor lifecycle + decommission controls — Pairing/authentication, 25+ state transitions, and decommission blocklist enforcement
• ✓ Hardware reliability automation — Nightly hardware regression runner, CI/CD auto-deploy, and cross-tier Mini v2/Pro/Max smoke coverage

Post-Q2 2026

• ⏳ Russian military drone field captures — Controlled collection program for Lancet/Orlan/Supercam validation (public datasets are not available)
• ⏳ SATCOM band monitoring — Ku/Ka expansion (12–40 GHz) for higher-end military and NATO links
• ⏳ Mini v2 deployment benchmarking — Wider field rollout with Pi 5 performance telemetry and thermal/power profiling
• ⏳ Federal variant — Variscite DART-MX8M-PLUS + ADALM-PLUTO pathway for NDAA-constrained deployments

Documented Limitations

Every detection platform has boundaries. We document ours openly and ship mitigations for each.

• DJI Mini — Invisible to cyclostationary OFDM (WiFi protocol, no CP). Mitigated by BW signature matching + CNN classification
• Flight mode modulation shift — DJI Inspire changes from FHSS_OFDM to OFDM by mode. Handled by temporal adaptation + hybrid ML
• Military signatures remain OSINT-seeded — Lancet, Orlan-10, and Supercam field captures are still a program objective because public datasets do not exist
• SATCOM visibility not yet in production — Ku/Ka monitoring (12–40 GHz) is a future track, so current coverage is strongest in sub-6 GHz control/video bands
• Acoustic class balance still improving — Retraining is underway to raise performance on underrepresented rotorcraft classes, especially helicopter-like signatures

Proprietary Platform

Varta Platform is proprietary technology. It is not open source and is not free to use.

• Commercial licensing: Provided by agreement based on deployment scope
• Pilot and evaluation access: Available on request
• Support and maintenance: Offered under active service plans

Get Involved

Contact us for deployment, licensing, and support details.

• Email: team@varta-systems.com
• Documentation: Start at Documentation Hub or go directly to Features, Technology, Architecture, API