Machine Learning Based Sleep Interval Deciding Algorithm In Time Division Multiplexing Passive Optical Network (TDM-PON) / Dk. Siti Nur Amalina Binti Pg. Hj. Damit

The project report is submitted in fulfillment of the requirements for Master by Coursework in Information Security.

Abstract

The study of Passive Optical Network (PON) standard as a next generation broadband optical access network in providing large capacity of bandwidth and saving energy consumption capability compared to other access technologies (such as Point-to-point), has always been a favorite subject to look onto. Since Information Communication Technology (ICT) is widely arise, Time-Division Multiplexing Passive Optical Network (TDM-PON) appears to be a promising technology that can meet users demand requirement. However, to keep the offer of ability to cope with the demand in TDM-PON, making further effort towards necessary approach is obligatory, and based on research found in academia and industry as well; there are still possible ways to make an enhancement towards energy efficiency.

Popular saving energy approach wherein already made by TDM-PON, is to tum Optical Network Unit (ONU) into sleep state when nothing to transmit or receive. However, the approach could arise risk in the case of letting ONU to switch into sleep state in a not at a right moment. Optical Line Transmission (OLT) will not transmit the data when ONU is inactive or sleep. Therefore, this can made OLT to suffer and unfortunately causing delay, which then deteriorate the Quality of System (QoS) that could contradict PON's benefit that PON originally bring in. Hence, the sleep intervals can give impact towards delay in traffic. Delay experienced by downlink's traffic and energy it consumes can cause TDM-PON with sleep-mode method to be born with a compromise issue due to tradeoff problem.

Concerning matter mentioned above, our goal is then established which is to save energy while also meeting network operator's access delay requirement. There are several studies related to this and we realize that most of existing work's solution utilize average value in order to forecast traffic arrival. Whereas here, our approach is to use machine learning and real traffic traces. We determine a suitable machine learning of Auto Regression Integrated Moving Average (ARIMA) model. We do also have evaluated our solution in real traffic case based on prediction, delay and energy saving performance in this thesis. It is to the best of our knowledge that our contribution by taking the concept of machine learning and digital twin based traffic arrival behavior as the first proposed solution towards trying to tackle the issue arise in achieving appropriate feasible sleep time in TDM-PON sleep state approach within this research domain.

Thesis is also available in CD and is not for loan or reference use.