Published: Sep 1, 2025
Authors: Xiaochan (Luna) Xue , syu , Saurabh Parkar , Yao Zheng

O-RAN/AI-RAN Series

This post is part of a series all about O-RAN/AI-RAN.

Fundamental

O-RAN/AI-RAN Testbed

Advanced

NextG UAV Detection Using an O-RAN-Controlled Reconfigurable Intelligent Surface (RIS)

A RIS- and O-RAN–assisted Integrated Sensing and Communication (ISAC) framework is presented for high-speed UAV detection and tracking in the 3.7 GHz band. The system integrates composite OFDM–FMCW waveforms, reconfigurable intelligent surfaces (RIS), and O-RAN distributed intelligence to enable scalable, low-latency, and adaptive sensing under spectrum-sharing constraints.

Highlights

Joint RIS and O-RAN coordination for wide-area UAV sensing
Composite OFDM–FMCW waveform enabling range–Doppler detection
Hierarchical, state-driven sensing strategy
GRU-based predictive control for proactive RIS adaptation

System Architecture

Operation in the CBRS band (3.7 GHz) balances coverage and sensing resolution
RIS enhances signal observability and mitigates LoS limitations
O-RAN provides distributed intelligence through Near-RT and Non-RT RICs
ISAC enables spectrum and infrastructure sharing between sensing and communication

Fig. 1. System Architecture

Multi-Stage UAV Detection Strategy

A four-stage state machine governs sensing resolution and resource allocation:

Stage 0 – Idle
- Continuous wide-area surveillance
- Minimal sensing and RIS resources
Stage 1 – Initial Detection
- Coarse FMCW sweeps
- Broad beams for rapid target discovery
Stage 2 – Classification
- Refined angle and velocity estimation
- Classification between UAVs and clutter
Stage 3 – Identification
- Narrow beams and high SNR
- Fine-grained trajectory and motion characterization

State transitions depend on detection confidence, SNR, QoS constraints, and resource availability.

Fig. 2. State machine for adaptive UAV sensing. Equipped with backward transitions enabled for re-verification or uncertainty handling.

RIS-Assisted Measurement Model

RIS configuration: 8x2 RIS elements array separated at $\lambda/2$ distance.
Sensing resource allocation managed at each stage.
Enables efficient trade-off between sensing precision and communication QoS

Fig. 3. An 8x2 RIS array in 3D. Simulated directivity patterns for different azimuth/elevation combinations, illustrating beamforming capabilities.

UAV position and velocity estimated in 3D ENU coordinates
Range Velocity estimations
- Beat frequency enables range estimation
- Doppler shift provides radial velocity
2D Multiple Signal Classification (MUSIC) for azimuth and elevation angles extracted via array processing

Fig. 5.UAV trajectory in a 3D East-North-Up (ENU) coordinate system.

Path Prediction and state change using GRU

SNR thresholding for UAV Detection confidence
Resource allocation for finer sensing
Gated Recurrent Unit (GRU) predicts next co-ordinate in 3D-space
- Inputs: Co-ordinate window of last 64 CPIs
Prediction latency compatible with Near-RT RIC requirements
Enables proactive beam steering for fast UAV maneuvers
the GRU model achieves a training RMSE of 0.0663 and a validation RMSE of 0.1363