Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ProSAS: An O-RAN Approach to Spectrum Sharing between NR and LTE

Sneihil Gopal, David Griffith, Richard A. Rouil, Chunmei Liu

TL;DR

ProSAS tackles the challenge of sharing spectrum between LTE and NR in an O-RAN setting by combining time-series demand forecasting with convex optimization for resource allocation over a shared pool $N_R$. The framework operates with non-RT training and near-RT inference, and employs an O-RAN-compatible deployment with data gathered and processed through open interfaces. It introduces two allocation strategies, OPT$_{max}$ and OPT$_{avg}$, and analyzes fairness via Jain's index, validated on real LTE data and GAN-generated NR data. Results show OPT$_{avg}$ improves fairness and reduces starvation in scarce-resource regimes, while OPT$_{max}$ can exploit larger pools to yield surpluses, demonstrating the practical viability of proactive spectrum sharing in O-RAN environments.

Abstract

The Open Radio Access Network (O-RAN), an industry-driven initiative, utilizes intelligent Radio Access Network (RAN) controllers and open interfaces to facilitate efficient spectrum sharing between LTE and NR RANs. In this paper, we introduce the Proactive Spectrum Adaptation Scheme (ProSAS), a data-driven, O-RAN-compatible spectrum sharing solution. ProSAS is an intelligent radio resource demand prediction and management scheme for intent-driven spectrum management that minimizes surplus or deficit experienced by both RANs. We illustrate the effectiveness of this solution using real-world LTE resource usage data and synthetically generated NR data. Lastly, we discuss a high-level O-RAN-compatible architecture of the proposed solution.

ProSAS: An O-RAN Approach to Spectrum Sharing between NR and LTE

TL;DR

ProSAS tackles the challenge of sharing spectrum between LTE and NR in an O-RAN setting by combining time-series demand forecasting with convex optimization for resource allocation over a shared pool . The framework operates with non-RT training and near-RT inference, and employs an O-RAN-compatible deployment with data gathered and processed through open interfaces. It introduces two allocation strategies, OPT and OPT, and analyzes fairness via Jain's index, validated on real LTE data and GAN-generated NR data. Results show OPT improves fairness and reduces starvation in scarce-resource regimes, while OPT can exploit larger pools to yield surpluses, demonstrating the practical viability of proactive spectrum sharing in O-RAN environments.

Abstract

The Open Radio Access Network (O-RAN), an industry-driven initiative, utilizes intelligent Radio Access Network (RAN) controllers and open interfaces to facilitate efficient spectrum sharing between LTE and NR RANs. In this paper, we introduce the Proactive Spectrum Adaptation Scheme (ProSAS), a data-driven, O-RAN-compatible spectrum sharing solution. ProSAS is an intelligent radio resource demand prediction and management scheme for intent-driven spectrum management that minimizes surplus or deficit experienced by both RANs. We illustrate the effectiveness of this solution using real-world LTE resource usage data and synthetically generated NR data. Lastly, we discuss a high-level O-RAN-compatible architecture of the proposed solution.
Paper Structure (10 sections, 15 equations, 5 figures, 1 table)

This paper contains 10 sections, 15 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. The ProSAS Framework
  4. Radio Resource Demand Prediction
  5. Optimal Resource Allocation
  6. Deployment Architecture
  7. Numerical Results for ProSAS
  8. Conclusions
  9. Proof of Convexity for $\text{OPT}_\text{max}$
  10. Proof of Convexity for $\text{OPT}_\text{avg}$

Figures (5)

  • Figure 1: High-level structure illustrating the deployment of ProSAS within the architecture ref:oran_tr_wg1.
  • Figure 2: Average PRB usage vs. time for , and +, i.e., aggregate, corresponding to time granularity of 1-hour and 1-minute.
  • Figure 3: CDF of average PRB demand for and synthetically generated dataset for time granularity 1-hour and 1-minute.
  • Figure 4: Evaluation of prediction models in terms of . -I and -II refer to - and , respectively.
  • Figure 5: Results for $\text{OPT}_\text{max}$ versus $\text{OPT}_\text{avg}$ for resource pool size $N_r = 10$ resources.