Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Windex: Realtime Neural Whittle Indexing for Scalable Service Guarantees in NextG Cellular Networks

Archana Bura, Ushasi Ghosh, Dinesh Bharadia, Srinivas Shakkottai

TL;DR

This work addresses per-user service guarantees in NextG cellular RANs by formulating resource allocation as a constrained MDP and exploiting Whittle indexability to decouple decisions across users. It introduces Windex, a lightweight, neural-network-based approach that learns Whittle indices for each service class and deploys a linear-time scheduler that allocates resources to the top-$R$ users per slot, with index computation under $20\,\mu$s and 1 ms scheduling. The authors prove indexability for the constrained scheduling problem, train four class-specific Whittle networks using a NeurWIN-inspired workflow, and integrate the scheduler into the EdgeRIC realtime RIC on a 5 MHz link. Extensive simulations, emulations, and over-the-air traces show Windex substantially improves per-user service guarantees over slicing and traditional schedulers, while maintaining robustness to channel dynamics and scaling to larger UE counts. The approach offers a scalable, real-time mechanism to deliver heterogeneous QoS in NextG networks, with practical implications for Open RAN deployments.

Abstract

We address the resource allocation challenges in NextG cellular radio access networks (RAN), where heterogeneous user applications demand guarantees on throughput and service regularity. We leverage the Whittle indexability property to decompose the resource allocation problem, enabling the independent computation of relative priorities for each user. By simply allocating resources in decreasing order of these indices, we transform the combinatorial complexity of resource allocation into a linear one. We propose Windex, a lightweight approach for training neural networks to compute Whittle indices, considering constraint violation, channel quality, and system load. Implemented on a real-time RAN intelligent controller (RIC), our approach enables resource allocation decision times of less than 20$μ$s per user and efficiently allocates resources in each 1ms scheduling time slot. Evaluation across standardized 3GPP service classes demonstrates significant improvements in service guarantees compared to existing schedulers, validated through simulations and emulations with over-the-air channel traces on a 5G testbed.

Windex: Realtime Neural Whittle Indexing for Scalable Service Guarantees in NextG Cellular Networks

TL;DR

This work addresses per-user service guarantees in NextG cellular RANs by formulating resource allocation as a constrained MDP and exploiting Whittle indexability to decouple decisions across users. It introduces Windex, a lightweight, neural-network-based approach that learns Whittle indices for each service class and deploys a linear-time scheduler that allocates resources to the top- users per slot, with index computation under s and 1 ms scheduling. The authors prove indexability for the constrained scheduling problem, train four class-specific Whittle networks using a NeurWIN-inspired workflow, and integrate the scheduler into the EdgeRIC realtime RIC on a 5 MHz link. Extensive simulations, emulations, and over-the-air traces show Windex substantially improves per-user service guarantees over slicing and traditional schedulers, while maintaining robustness to channel dynamics and scaling to larger UE counts. The approach offers a scalable, real-time mechanism to deliver heterogeneous QoS in NextG networks, with practical implications for Open RAN deployments.

Abstract

We address the resource allocation challenges in NextG cellular radio access networks (RAN), where heterogeneous user applications demand guarantees on throughput and service regularity. We leverage the Whittle indexability property to decompose the resource allocation problem, enabling the independent computation of relative priorities for each user. By simply allocating resources in decreasing order of these indices, we transform the combinatorial complexity of resource allocation into a linear one. We propose Windex, a lightweight approach for training neural networks to compute Whittle indices, considering constraint violation, channel quality, and system load. Implemented on a real-time RAN intelligent controller (RIC), our approach enables resource allocation decision times of less than 20s per user and efficiently allocates resources in each 1ms scheduling time slot. Evaluation across standardized 3GPP service classes demonstrates significant improvements in service guarantees compared to existing schedulers, validated through simulations and emulations with over-the-air channel traces on a 5G testbed.
Paper Structure (18 sections, 6 theorems, 42 equations, 8 figures, 5 tables, 2 algorithms)

This paper contains 18 sections, 6 theorems, 42 equations, 8 figures, 5 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Open Radio Access Networks (O-RAN) and RAN Intelligent Controllers (RICs)
  4. Network Slicing and Scheduling in 5G
  5. Reinforcement Learning in wireless networks
  6. Whittle Index Based Scheduling
  7. Service Classes and Desired Guarantees
  8. Optimization Problem Formulation
  9. Whittle Indexability
  10. Training Whittle Networks
  11. Whittle-Index Based Scheduler
  12. Windex System Implementation
  13. Windex Evaluation
  14. Training Windex Neural Networks
  15. Evaluating Windex in a heterogeneous application environment
  16. ...and 3 more sections

Key Result

Theorem 1

The optimal resource allocation problem defined in equation eqn: singleUEproblem, for a given $\mu_1$ and $\mu_2$ is indexable.

Figures (8)

  • Figure 1: O-RAN and RIC overview
  • Figure 2: System overview
  • Figure 3: Windex Training for all service classes
  • Figure 4: In all the scenarios with similar over-the-air channel traces, Windex achieves good performance in terms of throughput and tsls violations for all the service models.
  • Figure 5: In various over the air channel traces, Windex has better overall performance in terms of throughput and tsls constraint violations.
  • ...and 3 more figures

Theorems & Definitions (11)

  • Theorem 1
  • Lemma 1
  • proof
  • Lemma 2
  • proof
  • Lemma 3
  • proof
  • Lemma 4
  • proof
  • Theorem 2
  • ...and 1 more