Sir Syed University of Engineering and Technology, Karachi, Pakistan.
Politecnico di Milano, Milan, Italy.
Sci Rep. 2022 May 19;12(1):8378. doi: 10.1038/s41598-022-12349-4.
The physical random access channel (PRACH) is used in the uplink of cellular systems for initial access requests from the users. It is very hard to achieve low latency by implementing conventional methods in 5G. The performance of the system degrades when multiple users try to access the PRACH receiver with the same preamble signature, resulting in a collision of request signals and dual peak occurrence. In this paper, we used two machine learning classification technique models with signals samples as big data to obtain the best proactive approach. First, we implemented three supervised learning algorithms, Decision Tree Classification (DTC), naïve bayes (NB), and K-nearest neighbor (KNN) to classify the outcome based on two classes, labeled as 'peak' and 'false peak'. For the second approach, we constructed a Bagged Tree Ensembler, using multiple learners which contributes to the reduction of the variance of DTC and comparing their asymptotes. The comparison shows that Ensembler method proves to be a better proactive approach for the stated problem.
物理随机接入信道 (PRACH) 用于蜂窝系统的上行链路，用于用户的初始接入请求。在 5G 中通过实现传统方法来实现低延迟是非常困难的。当多个用户尝试使用相同的前导签名访问 PRACH 接收器时，系统的性能会下降，从而导致请求信号的碰撞和双峰值的出现。在本文中，我们使用了两种机器学习分类技术模型和信号样本作为大数据，以获得最佳的主动方法。首先，我们实现了三种有监督学习算法，决策树分类（DTC）、朴素贝叶斯（NB）和 K 最近邻（KNN），以便根据两类结果进行分类，标记为“峰值”和“虚假峰值”。对于第二种方法，我们构建了一个袋装树集成器，使用多个贡献者来减少 DTC 的方差，并比较它们的渐近线。比较表明，集成器方法对于所提出的问题是一种更好的主动方法。
