The Relevance Vector Machine Applied to the Modeling of Wireless Channels

Abstract

A good modeling of the radio propagation channel is essential to the design of high-performance wireless systems; therefore, the proper interpretation of the data acquired from the sounding process is a task of major importance in the construction of such models. The relevance vector machine (RVM) constitutes a learning algorithm, based on Bayesian Statistics, used in regression and classification problems. Recently, RVM was employed to filter the channel multipath components of simulated power delay profiles embedded with noise, enabling the determination of the paths arriving at the reception antenna, their arrival times and complex amplitudes. In this paper, the RVM algorithm is further studied, regarding its detection capabilities, but the power delay profiles were obtained from measurements carried out in an indoor channel. A comparison with the constant false alarm rate (CFAR) multipath identification scheme, based on computational simulation and real channel measurements, evidences the behavior of both detection schemes. Simulations also present the detection limits of the method, such as maximum multipath magnitude ratio and minimum interarrival time. Finally, the characterization of important parameters of a real wideband multipath indoor channel is presented, in terms of confidence intervals and probability distribution fittings.