O-RAN

Interference Classification (InterClass) xApp is an AI-driven extended application (xApp) deployed within Near-real-time RIC of the O-RAN architecture. Leveraging either KPIs or spectrograms, it is designed detects and classify interference in the cellular environment, identifying potential jammers or other types of interferers. Once an interferer is detected, the IC xApp analyzes its nature and impact, allowing for dynamic adjustments to various 5G base station parameters, such as, MCS selection, transmit power, and channel hopping. These adjustments help mitigate the adverse effects of interference, ensuring optimal network performance. 

 This InterClass xApp offers significant benefits in both  commercial 5G networks and defense environments, enabling robust, adaptive interference management in near-real-time, i.e., tens of milliseconds.

• InterClass-Spec xApp (based on Spectrograms)

  • Pros: Highly accurate

  • Cons: Large overhead, security issue

• InterClass-KPM xApp (based on KPMs)

  • Pros: Fast, secure, less overhead

  • Cons: Less accurate, can not detect interference types

Dataset: ~35K samples collected using our over-the-air O-RAN testbed (Available at www.nextgwirelesslab.org/datasets)

The IC xApp demo recognized as the best demo at IEEE MILCOM 2023. 

For more details, refer to the following relevant papers.

• N., Stephenson, A. Chiejina, N. Kabigting, and V. K. Shah, Demonstration of Closed Loop AI-Driven RAN Controllers Using O-RAN SDR Testbed, in IEEE MILCOM Demos 2023. [Link]

• A. Chiejina, B. Kim, K. Chowdhury, and V. K. Shah, System-level Analysis of Adversarial Attacks and Defenses on Intelligence in O-RAN based Cellular Networks, in ACM WiSec 2024. [Link]

Our team, in collaboration with Northeastern University, developed a novel SenseORAN framework that empowers 5G base stations to serve as spectrum sensors, enabling the precise detection of radar pulses within the CBRS band. The underlying concept is to deploy a ML module as a Radar Detection xApp within the Near-RT RIC in O-RAN systems. The 5G base station (i) utilizes a "You-Only-Look-Once" (YOLO) based ML module, finely tuned to recognize radar signals within spectrograms generated from I/Q samples collected during routine uplink cellular operations, and (ii) maintains an up-to-date inventory of "occupied" channels within CBRS band, indicating the presence of radar signals. 

Our exhaustive experiments using over-the-air LTE/5G O-RAN testbed demonstrate remarkable capabilities of SenseORAN framework, achieving 100% radar detection accuracy under SINR conditions of 12 dB or higher. Furthermore, it significantly reduces the end-to-end response time to <1s, representing a multi-fold improvement compared to the current FCC mandate of 1 minute (60 sec). 

For more details, refer to the following paper.

• G. Reus-Muns, P. S. Upadhyaya, U. Demir, N. Stephenson, N. Soltani, V. K. Shah, Kaushik R. Chowdhury, SenseORAN: O-RAN based Radar Detection in the CBRS Band, in IEEE Journal on Selected Areas in Communications SI on Open RAN, 2023

In collaboration with UC San Diego and TAMU, we have developed SPARC (Spatio-Temporal Adaptive Resource Control) framework for multi-site spectrum management in Private 5G/NextG Cellular Networks. Specifically, SPARC leverages a multi-timescale RIC architecture featuring an xApp for near-real-time (10 ms - 1s) interference detection and localization and a μApp (micro App) for real-time (<1 ms) intelligent resource allocation.

Comprehensive evaluations, including over-the-air experiments and emulations, demonstrate the significant performance gains achieved through SPARC, showcasing it as a promising solution for optimizing resource efficiency and network performance in private 5G/NextG networks.

For more details, refer to the following paper.

• U. Ghosh, A. Chiejina, N. Stephenson, V. K. Shah, S. Shakkottai, and D. Bharadia. SPARC: Spatio-Temporal Adaptive Resource Control for Multi-site Spectrum Management in NextG Cellular Networks. Proceedings of the ACM on Networking, 2(CoNEXT4), 1-18, 2024.

ORANSight is a multi-modal LLM framework designed for O-RAN specification compliance queries, code generation and visual understanding. We leverage both Retrieval Augmented Generation (RAG) as well as fine-tuning techniques pertaining to O-RAN specifications, 3GPP specifications, and srsRAN project codebase. Furthermore, we also incorporate a custom Reranker for high LLM recall, and an FAISS database for accelerated context retrieval.

Our initial experiments with an open-source ORAN-Bench-13K dataset, demonstrate that ORANSight significantly outperforms popular LLMs, such as, OpenAI's ChatGPT, Google's Gemini, and open-source Mistral. 

For more information on ORAN-Bench-13K, refer to the following paper.

P. Gajjar and V. K. Shah, ORAN-Bench-13K: An open source benchmark for assessing LLMs in open radio access networks, in IEEE CCNC 2025. [Link]

Our spin off company, WiSights Lab, aims to build domain specific LLMs and GenAI solutions for your enterprise needs. Do check out the website www.wisightslab.com.