A Machine Learning Assisted Approach for Predicting Handovers in Wireless Software Defined Networks

Machine Learning and Deep Learning algorithms are continually being explored for network optimization typically for software defined wireless networks. With the increasing number of users in cellular networks and need for increased bandwidth requirement owing to multimedia applications, the choice and utilization of effective multiplexing techniques for 6G onwards has become mandatory. Orthogonal High data rates, low bit error rate (BER), low latency and packet loss are fundamental requirements of wireless communications which may be challenging to achieve under fading channel conditions with mobile users. One of the technologies may perform better than the other under practical channel conditions and user attributes which points to the fact that co-existence and vertical handover among the technologies would increase the Quality of Service (QoS) if a choice among multiple techniques is provided. Moreover, estimating the wireless channel characteristics would also facilitate the decision of picking a particular technology in real time scenarios. This work proposes a vertical handover mechanism among the possibly co-existing wireless technologies based on the estimated future Bit Error Rate (BER). A data driven model (Deep Neural Network Model) trained with features such as pilot bits, received bits, path loss factor and signal to noise plus interference ratio (SINR) is used to estimate channel response and estimate the error rate, which is chosen as the primary QoS metric governing the handover. Various channel conditions are simulated for a frequency fading channel to estimate handover characteristics.