ML-Based Preamble Collision Detection in the Random Access Procedure of Cellular IoT Networks

TL;DR

This work tackles preamble collisions in the RACH of cellular IoT by training supervised ML models to detect collisions from Power Delay Profiles after the initial access attempt. The authors generate a large, labeled dataset from MATLAB-based simulations across diverse channel and mobility scenarios, compare nine classifiers, and identify a neural network as the top performer, achieving near 0.99 Balanced Accuracy in-distribution and about 0.95 out-of-distribution. They further optimize the model with post-training quantization, achieving ultra-low inference times (down to 0.3 ms) with negligible accuracy loss, enabling practical deployment on base-station hardware. The results demonstrate a scalable, real-time collision-detection solution suitable for dense CIoT/mMTC networks, reducing wasted resources and improving RA efficiency.

Abstract

Preamble collision in the random access channel (RACH) is a major bottleneck in massive machine-type communication (mMTC) scenarios, typical of cellular IoT (CIoT) deployments. This work proposes a machine learning-based mechanism for early collision detection during the random access (RA) procedure. A labeled dataset was generated using the RA procedure messages exchanged between the users and the base station under realistic channel conditions, simulated in MATLAB. We evaluate nine classic classifiers -- including tree ensembles, support vector machines, and neural networks -- across four communication scenarios, varying both channel characteristics (e.g., Doppler spread, multipath) and the cell coverage radius, to emulate realistic propagation, mobility, and spatial conditions. The neural network outperformed all other models, achieving over 98\% balanced accuracy in the in-distribution evaluation (train and test drawn from the same dataset) and sustaining 95\% under out-of-distribution evaluation (train/test from different datasets). To enable deployment on typical base station hardware, we apply post-training quantization. Full integer quantization reduced inference time from 2500 ms to as low as 0.3 ms with negligible accuracy loss. The proposed solution combines high detection accuracy with low-latency inference, making it suitable for scalable, real-time CIoT applications found in real networks.

Figures (7)

• Figure 1: Structure of the PRACH preamble transmitter
• Figure 2: Strcuture of the PRACH Receiver
• Figure 3: Block diagram of the PDP peak detection process
• Figure 4: the 4-step and 2-step contention-based Random Access Procedures
• Figure 5: Pipeline of the proposed machine learning framework for detecting preamble collisions in the RA procedure.