| Abstract |
The integration of High Altitude Platform Station (HAPS), the Internet of Things (IoT), and Artificial Intelligence (AI) fields has the potential to generate remarkable solutions for addressing the complex challenges present in smart cities. A HAPS is a network component operating at an altitude of approximately 17–20 km in the stratosphere. Furthermore, in addition to communicating with each other, such HAPS must maintain continuous connectivity with other terrestrial entities like IoT enabled devices, base station, and humans. From this perspective, recent research is focused on predicting the network signal strength using ANN technique. This research work proposed a technique to evaluate signal strength from HAPS to IoT devices in smart cities, aiming to ensure network connectivity, and deliver the desired Quality of Service (QoS). The proposed solution employs an Artificial Neural Network (ANN), which considers various relevant factors such as HAPS altitude, transmitter height, receiver height, path loss, distance, signal frequency, and transmitted power to predict HAPS signal strength accurately and efficiently. Generative deep learning techniques, such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs), are employed to generate realistic datasets. The findings show that ANN-based methods can significantly enhance overall performance while decreasing signal distortion. The proposed ANN technique demonstrates strong agreement with validation data obtained from simulations across various activation functions such as Rectified Linear Unit (RELU), Tanh, Swish, etc. The effectiveness of the proposed method was analyzed using metrics such as Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). Hence, the proposed ANN method proves to be reliable, valuable, and fast in estimating signal strength, determining optimal HAPS flight paths, and predicting subsequent locations based on received signal strength. In our proposed ANN method, the predictions with Swish activation have the lowest MSE (5%) and RMSE (22.583%), which means that, on average, Swish predictions stay closest to the true values. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025. |
