PATH LOSS ESTIMATION ON SUB 6 GHZ AND MILLIMETER WAVE FREQUENCY USING FINE-TUNING

• Main Author: Arifiansyah Brahma B, Muhammad
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/71424
• Institution: Institut Teknologi Bandung

One of the important factors that affect the performance of a wireless network is the received power level which is determined by several parameters. One of these parameters is path loss whose value is random with a large deviation, even though it is measured at the same distance from the transmitter. In fact, an accurate path loss value is crucial in wireless communication network planning. The determination of the path loss value based on the empirical model has considered its fluctuating value by adding random variables which are assumed to be log- normal distribution. However, it requires an intensive, time-consuming and expensive measurement campaign to generate an empirical path loss model. In addition, ray tracing techniques for path loss estimation require high computational time because the required resolution of the terrain data must be at a scale proportional to the wavelength. Therefore, in this study, we developed a deep learning-based method to estimate path loss at sub-6 GHz and mmWave frequencies. Specifically, we use a convolutional neural network (CNN) model with fine-tuning to obtain an accurate path loss estimate even though the number of image datasets of the propagation area is limited. The results of this study indicate that the proposed method is able to improve prediction errors in terms of root mean squared error (RMSE) and mean absolute error (MAE) up to 47.4141% and 47.8494% respectively when compared to the empirical model 3GPP 38.901 which is used as a benchmark in this research. The use of the CNN model with optimal fine-tuning is also able to reduce the duration of the model training time, as indicated by the convergence of the minimum loss value achieved during the 1st epoch, while the CNN model without fine-tuning achieves the convergence of the minimum loss value during the epoch 10th.