Research On Resource Optimization In NB-IoT Based On Fog Computing

The emerging fog computing architecture offers cloud services by deploying fog computing servers at the mobile base stations(BSs)at the edge of wireless networks,which greatly reduces network load and transmission delay.Massive amount of data generated by Internet of Things(IoT)devices can be offloaded to Fog computing servers for storage and compute intensive processing.Narrowband Internet of Things(NB-IoT)is a new cellular technology for the transmission of IoT data to the BS.Due to the resource constraint of IoT devices and NB-IoT network in terms of communication and computing resources,efficient scheduling is needed to meet the requirements of system performance.In this paper,a joint computation offloading and multi-user scheduling algorithm is proposed for IoT edge computing system to minimize the long-term average weighted sum of delay and power consumption under stochastic traffic arrivals.The dynamic optimization problem is formulated as an infinite-horizon average reward continuous time Markov decision process model.One critical challenge in solving this Markov decision process(MDP)problem for the multi-user resource management is the curse-of-dimensionality problem,where the state space of the MDP model and the computation complexity increase exponentially with the growing number of users or IoT devices.In order to deal with it,two algorithms are proposed.The first algorithm uses the Approximate Dynamic Programming(ADP)techniques,i.e.,the linear value-function approximation and Temoral-Difference(TD)learning with post-decision state and semi-gradient descent(SGD)method,to derive a simple algorithm for the solution of the Continuous-time Markov Decision Process(CTMDP)model.The second algorithm proposes a Convolutional Neural Network(CNN)architecture to approximate the value functions for the post-decision system states.Two algorithms designed to solve the CTMDP problem are semi-distributed,where the IoT devices submit bids to the BSs to make the resource control decisions centrally.The simulation results show that the two algorithms have significant performance improvement over the baseline algorithm and the Multi-user Multi-traffic offloading algorithm which is designed based on the deterministic task model.The performance of the DRL algorithm based on the approximation architecture of the value function of the CNN is better than the ADP algorithm.