5g high performing stand alone heterogeneousnetwork deployment employing machine learning techniques

Abstract

Fifth Generation mobile Wireless Network practically aims not only to connect everyone but everything as well, such as the IoT devices, self-driven cars etc. virtually on the same platform. The demand for higher data rates, lower latency, increased network capacity and uninterrupted user experience marks the definition for 5G wireless network. An important characteristic of 5G network is the Heterogeneous Networks. Any practical environment will include open outdoor – a direct Line Of Sight environment to office buildings –portraying the Non-Line Of Sight environment as well. Also there will be static users and dynamic users demanding for varying ranges of data rates. All these characterize the concept of Heterogeneous Network. Any new generation usually ensures backward compatibility to previous generations. 5G (IMT 2020) also ensures backward compatibility with 4G(IMT-Advanced) or 3G (IMT 2000) devices. Any 4G or 3G device will work seamlessly even when 5G is fully operational across the globe. As the name suggests multiple types of access nodes in the wireless network acts as the core of any Heterogeneous Network which include densely deployed femto/small cells under laid traditional macro cell network. This is one of the leads to the promising solution of high data rate, more capacity, lower latency etc. of the 5G communication. But small cell deployment is no easy task. Incorrect deployment can lead to poor data quality and higher interference. Machine Learning techniques can induce accurate and automated approach to small cell activation. This can resolve the issue of any incorrect deployment. Also Automation will lead to Rapid-to-Market of 5G services. In our current thesis work we thus implement the Kohonen’s Self-Organizing Map (SOM) as an automatic data-analysis method. It is extensively applied to clustering problems and data exploration in industry. This unsupervised neural network is popular for its topological character mapping where input data can be multi-dimensional. Hence we present how Kohonen’s SOM can be efficiently used to proceed with the small cell deployment with maximum achievable SINR. Mobile user of wireless services increases each day. Hence the concept of Ultra Dense Network ~ which characterizes the much denser small cell network as compared to actual number of active users ~ brings an important solution to the seamless connectivity to the increasing tele density. We show that with increasing user density, on applying SOM for the small cell deployment leads to enhanced Area Spectral Efficiency targeting for Maximum Coverage. We also demonstrate the concept of Macro User Offloading which essentially leads to traffic offloading from Macro Cell network to Small Cell Network leading to lesser load on existing Macro base stations with increasing tele density. A popular unsupervised algorithm is K-Means. In order to establish the better characteristics of Kohonen’s SOM we also establish a comparative study with the popularly implemented K-Means unsupervised learning algorithm. 5G cell edge spectral efficiency is represented by 5 percentile spectral efficiency. This is an important feature of any system’s performance index. We hence produce the comparison of system performance when deployed using SOM and while using K-Means. This further demonstrates the better clustering characteristics of SOM. Our entire analysis of SOM’s performance is extended from static to mobile user-equipment exhibiting stable and enhanced system performance with increasing successful handover percentage. K-Nearest Neighbors Algorithm is a supervised learning algorithm. We utilized K-NN intelligently in our work in between the training phases of our unsupervised learning techniques to extract with ease the interfering base stations while calculating the SINR of the system.