USE OF DISTANCE INFORMATION TO IMPROVE PATHLOSS ESTIMATION PERFORMANCE AT SUB-6 GHZ AND MMWAVE FREQUENCIES WITH DEEP LEARNING
The rapid development of technology and the increasing need for community connectivity demands very fast and stable internet network access. In recent years, mobile operators have been racing to launch 5G services, which requires accurate channel models to estimate path loss, as an important comp...
Saved in:
| Main Author: | |
|---|---|
| Format: | Theses |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/84840 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | The rapid development of technology and the increasing need for community
connectivity demands very fast and stable internet network access. In recent years,
mobile operators have been racing to launch 5G services, which requires accurate
channel models to estimate path loss, as an important component in the design of
5G systems and future mobile communications systems. Path loss refers to the loss
of signal strength transmitted over long distances, so more detailed and accurate
path loss estimates are needed. Various models have been proposed, including
conventional empirical models, but these methods are often too simple and only
provide one-shot solutions that are less adaptive.
Alternatively, machine learning, especially deep learning, has been proven to
provide faster and more accurate results in wireless communications. Previous
research using the VGG architecture without additional features showed low
correlation: 0.091 at 2.6 GHz, 0.156 at 28 GHz, and 0.116 at 38 GHz, and was
unable to capture changes in path loss with distance well. To overcome these
limitations, we propose three new architectures: VGG+D, VGG+DFC, and
VGG+EM, which are developed for sub-6 GHz and mmWave frequencies. The
results show that the VGG+EM architecture, which integrates empirical data with
CNN output, shows the best performance with correlations of 0.617 at 2.6 GHz,
0.593 at 28 GHz, and 0.621 at 38 GHz, far surpassing other architectures.
This research confirms that the VGG+EM architecture is not only superior in terms
of correlation and prediction accuracy compared to deep learning models, but also
more effective than traditional empirical models. Therefore, this architecture offers
a more reliable and adaptive solution for planning and optimizing 5G networks,
especially at mmWave frequencies. |
|---|