Model Order Selection for Collision Multiplicity Estimation

Abstract

The collision multiplicity (CM) is the number of users involved in a collision. The CM estimation is an essential step in multi-packet reception (MPR) techniques and in collision resolution (CR) methods. We propose two techniques to estimate collision multiplicities in the context of IEEE 802.11 networks. These two techniques have been initially designed in the context of source separation. The first estimation technique is based on eigenvalue statistics. The second technique is based on the exponentially embedded family (EEF). These two techniques outperform current estimation techniques in terms of underestimation rate (UNDER). The reason for this is twofold. First, current techniques are based on a uniform distribution of signal samples whereas the proposed methods rely on a Gaussian distribution. Second, current techniques use a small number of observations whereas the proposed methods use a number of observations much greater than the number of signals to be separated. This is in accordance with typical source separation techniques.