IEEE Access (Jan 2021)
Adaptive DRX Method for MTC Device Energy Saving by Using a Machine Learning Algorithm in an MEC Framework
Abstract
Machine-type communication (MTC) is recognized as an enabling building block for constructing the Internet of Things (IoT) through cellular networks. Since MTC devices (MTCDs) are usually powered by batteries, reducing the energy consumption of an MTCD is an important requirement in the design of an IoT system. The third-generation partnership project (3GPP) specified discontinuous reception (DRX) to save energy in user equipment (UE) by making a UE device sleep whenever it is not involved in data transmission or reception. However, UE is used for human-type communications (HTCs), and MTCs are different from HTCs in many ways, namely, in terms of traffic arrival patterns, mobility, and information priority. Thus, using the DRX developed for UE to save the energy of an MTCD leads to inefficiency in terms of the amount of power saved and the data transmission delay. In the literature, many efforts have been made to dynamically vary the DRX cycle based on traffic activities to reduce the energy consumption of an MTCD. These assume a specific distribution for the traffic arrival process and predict the next packet arrival time to control the DRX cycle. However, since the characteristics of the MTC traffic arrival pattern vary widely according to the types of applications and the network status, previous approaches are limited in the achievable energy efficiency and the range of MTC applications that they can support. To address these issues, we devise adaptive DRX (ADRX) by adopting a machine learning algorithm called bank-of-experts. Without making any assumption regarding the underlying traffic process, ADRX predicts the next packet arrival time using only the observed packet arrival times. Using the predicted next packet arrival time, ADRX dynamically adjusts the sleep interval. In terms of implementation, ADRX is executed in an eNB by following the Multi-access Edge Computing (MEC) framework. Thus, an MTCD avoids consuming the energy required to execute a sophisticated energy-saving algorithm while obtaining the benefits provided by ADRX. Since different IoT services have different requirements for the amount of saved energy and data transmission delay, we design a loss function by combining these two performance metrics so that ADRX can support a wide range of application services. We first analyze the bound of the cumulative regrets of ADRX, and then through simulations, we compare the performance of ADRX with the standard DRX with optimal sleep duration (OSDRX) and a generalized and autonomous DRX (GDRX) scheme. The results show that ADRX outperforms GDRX in terms of the loss function and has almost the same performance as OSDRX.
Keywords
