Wireless Backhaul Network Optimization Using Automated KPIs Monitoring System Based on Time Series Forecasting

Owing to the advancement of a communication network to the fourth generation and soon moving towards the fifth generation, the capacity demand has substantially increased in recent years. Currently, `point-to-point microwave' an imperative technology is used in wireless backhaul networks around the globe that bequeaths interconnectivity amid the core network and the base stations. Because of this high demand, the wireless backhaul network optimization is a very important factor and has become the iterative process for better coverage and delivering high Quality of Service (QoS). To facilitate this process, traffic projections execute a significant job to enrich the network excellence resulted in better network capacity planning and frequency reuse. We proposed an adaptive solution for network Key Performance Indicators (KPIs) monitoring based on time series forecasting. The forecasting is grounded on Autoregressive Integrated Moving Average (ARIMA) models, Autoregressive (AR) and Multilayer perceptrons (MLP) neural network method. It analyzes the network utilization on specific time intervals (past and current data) and forecasts future requirements. It investigates observed unbalanced resources allocations in both ways underutilize and overutilize links based on real network utilization. Based on the proposed system forecasted results, the optimal solution for network capacity planning and frequency channel reuse will be suggested to reduce the resource wastage. We have observed that in the context of convergence time the AR is best and MLP with 10 hidden nodes takes longest. But when we discuss the results the MLP with 10 hidden nodes gives the best. The one analysis was done on 181 days observations, we used AR(16), SARIMA(1,0,0)(1,1,0) [7] and MLP 10 hidden nodes with RMSE 37.87, 40.00 and 13.54 respectively it indicates that MLP was best. Conclusively, we evaluated the performance of methods using RMSE. On the grounds of experiments, we perceived that MLP was good in order to predict future capacity.