Yazar "Bal, Mustafa" seçeneğine göre listele

Listeleniyor 1 - 3 / 3
  • Küçük Resim
    Öğe
    Path loss prediction from heightmap using deep learning
    (İstanbul Medipol Üniversitesi Fen Bilimleri Enstitüsü, 2021) Bal, Mustafa; Güntürk, Bahadır Kürşat
    Wireless channel parameters of a region are required for a successful network planning. Sufficient information about those parameters can be obtained by either actual measurements or ray-tracing simulations that use 3D model of the target area. However, measurements are costly and time consuming, and ray-tracing simulations have high computational cost. This thesis recommends various methods of estimating path loss or its parameters using deep learning. We give two solutions for estimating path loss or path loss parameters. Firstly, regression modeling is shown for estimating path loss exponent and shadowing factor of the wireless channel by using deep learning methods with satellite images or height map. Path loss dataset that is needed for training the deep neural network is produced by ray-tracing simulations. The deep network takes satellite image or height map as input and applies regression to estimate the desired channel parameters. Since the path loss is a critical value as well as its parameters in wireless communication, our second problem is to estimate the point-wise excessive path loss values of a region using the conditional general adversarial network. Ray-tracing simulations are also taken as the ground truth for this problem. With this method, we aim to find the path loss value of the receiver directly at each region. Even though it is not a perfect model for point-wise prediction but it can give us more reliable general information for the region than the convolutional networks. The results obtained are shown and analyzed as point-wisely for each region and probability distributions in detail.
  • Küçük Resim
    Öğe
    PL-GAN: Path loss prediction using generative adversarial networks
    (IEEE-Institute of Electrical and Electronics Engineers Inc., 2022) Marey, Ahmed; Bal, Mustafa; Ateş, Hasan Fehmi; Güntürk, Bahadır Kürşat
    Accurate prediction of path loss is essential for the design and optimization of wireless communication networks. Existing path loss prediction methods typically suffer from the trade-off between accuracy and computational efficiency. In this paper, we present a deep learning based approach with clear advantages over the existing ones. The proposed method is based on the Generative Adversarial Network (GAN) technique to predict path loss map of a target area from the satellite image or the height map of the area. The proposed method produces the path loss map of the entire target area in a single inference, with accuracy close to the one produced by ray tracing simulations. The method is tested at 900MHz transmission frequency; the trained model and source codes are publicly available on a Github page.
  • Küçük Resim
    Öğe
    Regression of large-scale path loss parameters using deep neural networks
    (IEEE-Institute of Electrical and Electronics Engineers Inc., 2022) Bal, Mustafa; Marey, Ahmed; Ateş, Hasan Fehmi; Baykaş, Tunçer; Güntürk, Bahadır Kürşat
    Path loss exponent and shadowing factor are among important wireless channel parameters. These parameters can be estimated using field measurements or ray-tracing simulations, which are costly and time-consuming. In this letter, we take a deep neural network-based approach, which takes either satellite image or height map of a target region as input, and estimates the desired channel parameters. We use the well-known VGG-16 architecture, pretrained on the ImageNet dataset, as the backbone to extract image features, modify it as a regression network to produce channel parameters, and retrain it on our dataset, which consists of satellite image or height map as input and channel parameters as target values. We demonstrate that deep networks can be successfully utilized in estimating path loss exponent and shadowing factor of a region, simply from the region's satellite image or height map. The trained models and test codes are publicly available on a Github page.