Enhancing Training Efficiency and Accuracy of Radio Fingerprint-Based Indoor Localization Via OSGAN Data Augmentation and RBFKAN Classifier

Krisnawan, Aditya Bagus (2025) Enhancing Training Efficiency and Accuracy of Radio Fingerprint-Based Indoor Localization Via OSGAN Data Augmentation and RBFKAN Classifier. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022231051-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (12MB) | Request a copy

Abstract

The development of navigation systems is rapidly advancing both outdoors and indoors. A good localization system must support navigation systems. Indoor localization requires high accuracy. GPS is not suitable for indoor use due to its low accuracy. Radio fingerprinting is a promising indoor localization technique that offers high accuracy. However, its accuracy requires data acquisition that is not straightforward. This technique requires extra time and effort in data collection. The accuracy of fingerprint-based localization is influenced by the amount of data. The more samples collected, the more accurate the results. To save time and effort, a GAN model can be a solution for generating a new database. A one-stage GAN model (OSGAN) is proposed in this study due to its potential for stable training. We proposed RBFKAN to enhance accuracy, given its strong generalization capabilities. The research results demonstrate that Wasserstein OSGAN (W-OSGAN), which has the most stable training process, achieves the shortest training duration and highest localization accuracy. With the proposed method, the duration of fingerprint data collection becomes more efficient, reducing from 120 seconds to 67.55 seconds per reference point (RP). The accuracy increase after augmentation was 32.5% using MLP and 23% using RBFKAN. The RBFKAN increase was smaller because the RBFKAN accuracy before augmentation was already quite high compared to MLP. RBFKAN and MLP achieved the highest accuracy of 95.7% and 88.8%, respectively. Additionally, the efficiency achieved with GPU A100-based OSGAN reached 74%. Meanwhile, efficiency without the augmentation process only yielded 45%.

==========================================================================================================================================
Perkembangan sistem navigasi berkembang pesat baik di luar ruangan maupun di dalam ruangan. Sistem lokalisasi yang baik harus mendukung sistem navigasi. Lokalisasi di dalam ruangan memerlukan akurasi yang tinggi. GPS tidak cocok untuk penggunaan di dalam ruangan karena akurasinya yang rendah. Radio fingerprinting adalah teknik lokalisasi di dalam ruangan yang menjanjikan dengan akurasi tinggi. Namun, akurasinya memerlukan pengumpulan data yang tidak mudah. Teknik ini memerlukan waktu dan usaha ekstra dalam pengumpulan data. Akurasi lokalisasi berbasis fingerprint dipengaruhi oleh jumlah data. Semakin banyak sampel yang dikumpulkan, semakin akurat hasilnya. Untuk menghemat waktu dan usaha, model GAN dapat menjadi solusi untuk menghasilkan basis data baru. Model one-step GAN (OSGAN) diusulkan dalam studi ini karena potensinya dalam pelatihan yang stabil. Kami juga mengusulkan RBFKAN untuk meningkatkan akurasi, mengingat kemampuannya yang kuat dalam generalisasi. Hasil penelitian menunjukkan bahwa Wasserstein OSGAN (W-OSGAN), yang memiliki proses pelatihan paling stabil, mencapai durasi pelatihan terpendek dan akurasi lokalisasi tertinggi. Dengan metode yang diusulkan, durasi pengumpulan data sidik jari menjadi lebih efisien, berkurang dari 120 detik menjadi 67,55 detik per titik referensi (RP). Peningkatan akurasi setelah augmentasi mencapai 32,5% menggunakan MLP dan 23% menggunakan RBFKAN. Peningkatan akurasi RBFKAN lebih kecil karena akurasi RBFKAN sebelum augmentasi sudah cukup tinggi dibandingkan dengan MLP. RBFKAN dan MLP mencapai akurasi tertinggi masing-masing 95,7% dan 88,8%. Selain itu, efisiensi yang dicapai dengan OSGAN berbasis GPU A100 mencapai 74%. Sementara itu, efisiensi tanpa proses augmentasi hanya mencapai 45%.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Indoor Localization, Radio Fingerprint, OSGAN, RBFKAN
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5103.2 Wireless communication systems. Two way wireless communication T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors
Divisions:	Faculty of Electrical Technology > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Aditya Bagus Krisnawan
Date Deposited:	24 Jul 2025 02:58
Last Modified:	24 Jul 2025 02:58
URI:	http://repository.its.ac.id/id/eprint/121004

Actions (login required)

View Item