GoS Evaluation for OFDMA Cellular Networks

Analytical methods on grade of service (GoS, quantified by the blocking probability which is denoted as Pb) evaluation in orthogonal frequency division multiple access cellular networks for real-time traffic are important for cell-dimensioning. Existing works use simplified spectral efficiency (SE) models that lead to conservative outcomes. For computing cell-wide and cell-average Pb the exact but mathematically intractable Kaufman Roberts Algorithm (KRA) is indispensable. The Erlang approximation (EA) tractably computes only cell-average Pb but ignores bandwidth allocation granularity leading to an error of around 25% in typical scenarios. In this work, by using an accurate SE model we show a significant reduction in projected base station (BS) density requirement for maintaining satisfactory cell-average Pb. We show that cell-average Pb does not completely depict the GoS situation within a cell since there is significant variation of Pb inside it. Further, we enhance the EA in two ways. Firstly, we improve its accuracy to within 2% of KRA by accounting for bandwidth allocation granularity. Secondly, we extend it to compute cell-wide Pb like the KRA. The resulting method helps to explain properties such as the linearly increasing trend of region-wise Pb vs. region-index in case of low bit-rate streaming calls. We also use the aforementioned linearly increasing trend to arrive at another alternative to KRA. The development of these two alternatives to KRA highlights certain interesting properties of region-wise and cell-average Pb as well as the Subhendu Batabyal G. S. Sanyal School of Telecommunications Indian Institute of Technology, Kharagpur Tel.: +91-801-674-6110 Fax: +91-322-228-2266 E-mail: [email protected] Suvra Sekhar Das G. S. Sanyal School of Telecommunications Indian Institute of Technology, Kharagpur E-mail: [email protected] 2 Subhendu Batabyal, Suvra Sekhar Das EA. The error in estimated BS density requirement with these alternatives in case of 12.2 kbps voice in the interference limited situation is negligibly small. We employ discrete event simulations to validate our models.