Pre-processing of Data in RSS Signature-Based Localization

Most applications for location based services rely on a precise determination of the mobile terminal (MS) location. This paper proposes a method of locating a MS in a GSM cellular system based on the radio signal strength (RSS) and use of neural network (NN). The proposed algorithm is best suited for urban and suburban environments. Due to the big effort during RSS collection from the streets to realize a fingerprint database, predicted RSS-data of the area of concern are rather used for this experiment. An exact modelling of all effects in prediction formula in a complex environment is not appropriate in a practical implementation. Therefore, the predicted data differ from the real one. A system is developed which calibrates the radio propagation prediction data on the basis of sample real collected measurements. During this correction procedure, the instability or stochastic behavior of the RSS is taken into account. An algorithm which reduces the noise in the collected RSS is applied. Having the corrected predicted data, there is no need of collecting the RSS from all the streets to train the NN. The positioning accuracy is compared for different pre-processing methods and calibration shows to reduce the positioning error.