Machine Learning and Friends Lunch
Modeling Frame-level Errors in GSM Wireless Channels
Ping Ji
UMass
Abstract
There has been substantial recent interest in the development and deployment of wireless data communication systems such as 3G wireless networks, IEEE 802.11, and Bluetooth. The characteristics of such wireless channels (in particular, with respect to frame-level errors) are quite different from those of traditional wired links. Most wireless channel modeling efforts have focused on physical layer properties such as the signal to noise ratio. Communication protocols at the network layer and above, however, operate on the basic unit of a frame (or packet); and the choice of error model matters for the behavior of higher layer protocols. Frame-level wireless channel models are thus of particular interest in the design and evaluation of such protocols in wireless networks.
In this work, we compare four different approaches to modeling frame-level errors in a GSM wireless channel. One of these, the Markov-based Trace Analysis model (MTA), was developed explicitly for this purpose. The next two, k-th-order Markov models and hidden Markov models (HMMs), have been widely used to model loss in wired networks. All three of these have difficulty modeling empirical GSM frame-level error traces. The MTA model and HMM do not predict the frame error rates that measured from the trace very well, and all three models have difficulty capturing the long term temporal correlation structure. We propose a fourth model, the extended on/off model, which alternates between an ON (error-free) and an OFF (error-filled) state. The state holding times are taken from mixtures of geometric distributions. We show that this model, with mixtures of four to seven geometric distributions captures first order and second order statistics significantly better than the preceding three approaches.