MONITORING DAN PREDIKSI DINAMIKA SPEKTRUM LTE BAND 8 MENGGUNAKAN LONG SHORT-TERM MEMORY BERBASIS RTL-SDR DI LINGKUNGAN UNIVERSITAS PENDIDIKAN INDONESIA

- Mutiara Fatmadewi, Arjuni Budi Pantjawati¹

1. ¹ Universitas Pendidikan Indonesia

Repository at Universitas Pendidikan Indonesia (Universitas Pendidikan Indonesia), 2026

ABSTRAK Peningkatan penggunaan layanan mobile broadband berbasis LTE menyebabkan kepadatan pemanfaatan spektrum frekuensi radio semakin tinggi. Pemantauan dan pemodelan dinamika spektrum secara temporal diperlukan untuk memahami pola penggunaan kanal dan mengantisipasi potensi interferensi. Penelitian ini bertujuan menganalisis karakteristik spektrum LTE Band 8 (925 – 960 MHz) dan membangun model prediksi berbasis LSTM di lingkungan kampus Universitas Pendidikan Indonesia. Pengukuran dilakukan di sembilan gedung perwakilan fakultas pada kondisi indoor dan outdoor menggunakan RTL-SDR. Parameter yang dianalisis meliputi rata-rata daya spektrum, daya puncak, noise floor, Signal to Noise Ratio, dan okupansi spektrum. Uji stasioneritas Augmented Dickey-Fuller (ADF) menunjukkan 60% data bersifat tidak stasioner. Nilai ini mendukung pemilihan LSTM sebagai model prediktif. Data diproses menggunakan normalisasi Z-score dan sliding window berukuran 15. Data selanjutnya dibagi menjadi train (70%), validation (15%), dan test (15%). Model dibangun dalam dua skenario, yaitu Stacked LSTM 128 neuron (indoor) dan 256 neuron (outdoor). Hasil menunjukkan rata-rata okupansi spektrum 58,9% (indoor) dan 68,8% (outdoor), dengan SNR homogen di rentang 24,79-27,11 dB. Model outdoor menghasilkan RMSE 0,1821; MAE 0,1303; dan WAPE 19,31%, sedangkan model indoor menghasilkan RMSE 0,2598; MAE 0,1998; dan WAPE 20,65%. Evaluasi per gedung menunjukkan seluruh Lokasi menghasilkan WAPE di bawah 11%. Penelitian ini membuktikan bahwa data RTL-SDR tetap valid untuk pemodelan prediktif time series. Kata Kunci : LTE Band 8, RTL-SDR, LSTM, Spectrum Sensing, Time Series. ABSTRACT The growing use of LTE- based mobile broadband services has led to increasingly dense utilization of radio frequency spectrum. Temporal monitoring and modeling of spectrum dynamics are essential for understanding channel usage pattern and anticipating potential interference. This study aims to analyze the characteristics of LTE Band 8 spectrum (925-960 MHz) and to develop an LSTM-based prediction model at Indonesia University of Education. Measurements were conducted across nine faculty representative buildings under indoor and outdoor conditions using an RTL-SDR device. The parameters analyzed include average spectrum power, peak power, noise floor, Signal to Noise Ratio, and spectrum occupancy percentage. The Augmented dickey-fuller stationarity test revealed 60% of the data is non-stationary, which supports the selection of LSTM as the predictive model. Data were preprocessed using Z-score normalization and a sliding window of size 15, the split into train (70%), validation (15%0), and test (15%) sets. Two model were developed based on environmental scenarios a Stacked LSTM with 128 neuron for indoor and 256 neuron for outdoor conditions. Results show an average spectrum occupancy of 58,9% (indoor) and 68,8% (outdoor), with homogeneous SNR value ranging from 24,79 – 27,11 dB. The outdoor model achieved RMSE of 0,1821; MAE of 0,1202; and WAPE 0f 19,31%, while the indoor model achieved RMSE of 0,2598; MAE of 0,1998; and WAPE of 20,655. Building evaluation confirmed that all locations produced WAPE below 11%. This study demonstrates that relative RTL-SDR spectrum data remans valid for time series predictive modeling. Keywords : LTE Band 8, RTL-SDR, LSTM, Spectrum Sensing, Time Series.