URLLC • Edge Computing

URLLC-V2X

Joint Resource Allocation and Computation Offloading using Transformer-Based Learned Surrogate

Abstract

Ultra-reliable low-latency communications (URLLC) in vehicular edge computing environments demands stringent end-to-end delay guarantees with reliability bounds exceeding conventional Shannon-theoretic limits. This research addresses the joint optimization of resource block allocation, transmit power control, and computation offloading decisions for vehicular URLLC users under finite blocklength information theory constraints. We consider a comprehensive vehicle-to-everything (V2X) architecture comprising both cellular user equipment (CUE) vehicles communicating with road-side units and vehicle-to-vehicle (V2V) pairs utilizing direct device-to-device links.

Research Highlights

Transformer-Based Learned Surrogate

Approximation of optimal policies with sub-millisecond inference time, enabling real-time vehicular deployments.

Finite Blocklength Analysis

Optimization under short-packet information theory constraints for URLLC scenarios.

NP-Hardness Proof

Formal proof through polynomial-time reduction from 0-1 Knapsack problem characterizing computational complexity.

Queueing-Aware Allocation

Integrated M/D/1 models with dimensionally consistent service rate definitions for deterministic packet sizes.

Key Results

94.7%
Optimality Gap
vs branch-and-bound optimal
3 orders
Speed Improvement
computation time reduction
<1 ms
Inference Time
sub-millisecond deployment

Technologies

PythonPyTorchTransformerMINLPMonte CarloV2XEdge ComputingInformation Theory

Authors: AlHussein A. AlSahati, Dr. Houda Chihi

Status: Submitted to IEEE • Research Lab: InnovCOM Lab, Sup'COM Tunisia