TY  - JOUR
T1  - Deep Learning-Based Performance Improvement of Statistical Estimation-Based ABR Algorithm
AU  - Moon, Ie-bin 
AU  - An, Donghyeok 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2025
DA  - 2025/1/1
DO  - 10.7840/kics.2025.50.4.616
KW  - ABR Algorithm
KW  - Deep Learning
KW  - LSTM
KW  - MPC
KW  - Quality of Experience
KW  - Transformer
AB  - With the increasing demand for OTT platforms such as Netflix and the growth of the video streaming market, enhancing the performance of Adaptive Bitrate (ABR) algorithms, a key component of streaming services, and improving the Quality of Experience (QoE) for users have become more critical. The network bandwidth prediction algorithms in conventional Model Predictive Control (MPC) and Robust MPC-based ABR algorithms rely on statistical estimation methods, which can lead to performance degradation due to prediction errors, especially in scenarios with highly fluctuating bandwidth. In this study, we propose improvements by employing LSTM (Long Short-Term Memory) and Transformer-based time series prediction models for network bandwidth forecasting and applying them to the MPC algorithm. By quantitatively comparing the ABR performance measured through QoE metrics using an ABR algorithm simulation framework, we observed that both LSTM and Transformer models achieved superior performance, improving the conventional MPC-based ABR algorithm by 8.71% and 8.91%, respectively.