A logistic regression approach to location classification in OFDMA-based FFR systems

In this paper, the problem of location classification in OFDMA-based FFR systems is considered for two FFR schemes, namely Strict FFR (FFR-A) and Soft Frequency Reuse (FFR-B). The FFR systems considered in the literature are mostly studied with static resource partitioning where a greater portion of the sub-carrier resources are reserved for users in the cell centre and a smaller portion for users near the edge. Once the resources are statically partitioned, they can then be allocated dynamically. In order to ensure proper resource allocation, the base station has to classify the location of the user as either cell-centre or cell-edge. A common practice is to use a one-dimensional threshold such as SINR, which is assumed to be a good indicator of distance. In this paper, the impact of misclassifying the user location on the overall system performance is studied. It is shown that in an urban environment with shadowing, the one-dimensional threshold approach gets the location classification right only 67% of the time and when compared to an ideal system based on accurate location, the overall cell throughput drops by 38% for FFR-A and 14% for FFR-B. Similarly, service rate drops of 24% and 28% are also observed for FFR-A and FFR-B respectively. A better technique based on a combination of two measurements, namely received power and SINR, is proposed where the higherdimensional threshold is determined through Logistic Regression. This new approach is shown to have a classification accuracy of more than 80%. As a result, the system performance is shown to be much better than the one-dimensional threshold approach and comparable to that based on accurate location.