ABSTRACT

Accurate path loss modelling and prediction will provide realistic information on the level of signal attenuation in a service area and contribute positively to a better performance of cellular radio frequency network. A better predictive path loss model that will facilitate superb cellular network planning process will be of great value to cellular radio network planners, stakeholders and end users. In this study, the development of a model that identifies path loss in a channel ofmobile cellular network was undertaken for Uyo capital city in Akwa Ibom State, Nigeria. The adaptive hybrid model which combines multi-resolution Discrete Wavelet Transform (MDWT) and Long Short Term Memory (LSTM) Recurrent Neural Network in MATLAB platform was used for modelling and prediction of the signal coverage path loss in Uyo capital city. Field drive test based experimental set up with the combination of test mobile system technology, map info.pro.software and Global positioning system aided the measurement of Reference Signal Received Power (RSRP) with respect to distance around four long Term Evolution (LTE) eNode B antenna sites, all which operates at 10MHz band width across Uyo city. The measured signal data was routed through a wavelet based decomposition process with three decomposition levels. The decomposed measured signal data was converted into path loss values and then utilized as input data to long short term memory model where relevant extracted information were captured and trained for robust predictive adaptive learning and prediction. The predicted outcome of the hybrid model was compared to the conventional prediction models. Some essential performance evaluation indices were employed for accuracy evaluation and validation of the result obtained. The result obtained in terms of the hybrid (MDWT-LSTMRNN) approach improves the mean prediction efficiency by 50% and trains with less time