An efficient handover mechanism for 5G networks using hybridization of LSTM and SVM

• Published in: Multimedia tools and applications Vol. 81; no. 26; pp. 37057 - 37085
• Main Authors: Kaur, Gaganpreet, Goyal, Raman Kumar, Mehta, Rajesh
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.11.2022; Springer Nature B.V
• Subjects: 1211: AIoT Support and Applications with Multimedia; 5G mobile communication; Accuracy; Cellular communication; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Electronic devices; Interoperability; Multimedia Information Systems; Network latency; Predictions; Reference signals; Signal quality; Special Purpose and Application-Based Systems; Support vector machines; User interfaces; Websites; Wireless communications; Deep learning; Handover; Network selection; LSTM; SVM; Mobility management
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-021-11510-x

Mobile devices can access the internet through different wireless interfaces such as wireless fidelity (WiFi), worldwide interoperability for microwave access (WiMAX), and cellular networks like long-term evolution (LTE), fifth-generation networks (5G), etc. The main objective of the handover technique is to select the best network with minimum handover latency to provide seamless connectivity to the user. This paper proposes a hybrid handover technique for predictive handover based on long-short term memory (LSTM) and support vector machine (SVM). LSTM is used to predict the parameters of mobile devices such as location coordinates, speed, reference signal received power (RSRP), and reference signal received quality (RSRQ) at the next time step based on their values at previous time steps. The output of LSTM is supplied as input to the SVM for the selection of the most appropriate network. The mechanism proposed in this work significantly reduces the handover latency for predictive handover along with high prediction accuracy. The experimental results revealed that proposed approach can achieve accuracy up to 99.99% as compared to 85.76% (by using Stacked-LSTM) on dataset1 and improvement in validation and testing accuracy on whole dataset2 upto 76 and 75.92% relative to the accuracy 49.11 and 47.09% achieved by existing method as discussed in experimental and results analysis section.