Optimizing Power Consumption in Converged Networks with Novel Deep Learning

In the present-day world, high-speed internet demand is increasing day by day. Optical Fiber communication, which can cater to high bandwidth requirements, has a bottleneck for extending till the last mile for mobile User Equipment (UE). Converged networks, which use the combined advantage of Optical Fiber for the backbone network and wireless communication through Passive Optical Networks (PON) for the last mile, are the area of focus in this research paper. For effective utilization of available backbone bandwidth, it is essential to use Dynamic Bandwidth Allocation (DBA) techniques by the Optical Line Terminal (OLT) to allocate bandwidth to Optical Network Units (ONUs) to meet the requirements of UE devices. The growing density of UE devices demands power-saving techniques at the access network level. Significant research literature is available on the use of machine learning techniques for DBA in PON, but integration of spatial and temporal learning mechanisms is not widely explored for reducing power consumption in PON. A Hybrid dynamic bandwidth allocation technique is proposed in this work, which uses the temporal recognition capability of the Long Short-Term Memory (LSTM) algorithm and the spatial recognition capability of the Deep Q Network (DQN) Algorithm. The proposed hybrid model requires a considerably large dataset for training, which is achieved using the Generative Adversarial Network (GAN) method. The results of the proposed hybrid model are compared with the standalone Deep Q Network model, and it is verified that there is an 11% reduction in buffer occupancy at ONUs and a 20 % reduction in the Power consumption of the overall system.