Principal Component-Based Approach for Profile Optimization Algorithms in DOCSIS 3.1

• Published in: IEEE eTransactions on network and service management Vol. 15; no. 3; pp. 934 - 945
• Main Authors: Ben Ghorbel, Mahdi, Berscheid, Brian, Bedeer, Ebrahim, Hossain, Md. Jahangir, Howlett, Colin, Cheng, Julian
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.09.2018; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Adaptive modulation; Algorithms; Cable modems; Clustering; Clustering algorithms; Complexity theory; Computer simulation; data over cable networks; OFDM; Optimization; Principal components analysis; profile optimization; Quadrature amplitude modulation; Signal to noise ratio; Subcarriers
• ISSN: 1932-4537; 1932-4537
• DOI: 10.1109/TNSM.2018.2828704

Data over cable service interface specification (DOCSIS) introduced the possibility of a variable bit-loading over the subcarriers within a channel in its release DOCSIS 3.1. This variable bit-loading will improve the data rates. However, to limit the encoding processing overhead, the concept of profiles was introduced. Each profile defines the modulation per subcarrier for a given channel while the number of allowed profiles is limited. Thus, an efficient profile assignment scheme, which determines the best set of profiles based on the users' channel conditions, is needed. Although various profile assignment algorithms have been proposed in the literature, realistic evaluation of these schemes has been difficult, as channel quality measurements of real DOCSIS 3.1 systems has not previously been available. In this paper, we exploit DOCSIS 3.1 measurement data to evaluate performance of the proposed algorithms. We propose to employ principal component analysis to derive low-dimensional clustering variables in order to ensure efficient profile optimization. We show how this technique can be employed with different clustering algorithms to improve the spectrum efficiency of the profiles by extracting the most important information of the channels in low-dimensional vectors. This not only reduces the complexity of the clustering, but also ensures better throughput. Moreover, we adapt the clustering algorithms to tailor them to the profile optimization problem. Finally, we present an exhaustive simulation-based performance analysis to compare the different algorithms for various scenarios using extrapolation of the measurements data.