Novel H.265 Video Traffic Prediction Models Using Artificial Neural Networks
Primary Faculty Mentor’s Name
Dr. Rami Haddad
Proposal Track
Student
Session Format
Poster
Abstract
By the year 2019, the Cisco Virtual Networking Index forecasts that video traffic of all types will constitute 80%-90% of global internet traffic, amounting to 1.24-1.44 zettabytes of data passing through internet infrastructure. Video bandwidth forecasts can significantly improve the video transport mechanisms by improving the efficiency of Dynamic Bandwidth Allocation. Especially with the increasing popularity of the newer, high-definition video resolution such as 4K and high compression ratio encoding standards such as H.265, the forecasting of video frames becomes necessary to meet network demand. While large data transfers are becoming more commonplace and video streaming demands continually increase, the physical internet infrastructure is oftentimes unable to be implemented or to be upgraded rapidly enough to provide the requisite bandwidth to satisfy user demand. To provide increased network throughput with the limitations imposed by hardware, the use of highly trained and adaptive neural-computational network models will allow internet service providers to effectively forecast the necessary bandwidth to provide uninterrupted video delivery and streaming. We propose using an artificial neural network to predict the necessary bandwidth requirements of H.265 video streaming. We have chosen to utilize the Feed Forward Backward Propagation neural network and the Recurrent neural network to predict bandwidth requirements, as neither method has previously been attempted with H.265 encoding. The trace files in this library are divided approximately in half, with one half to be used to train the artificial neural network and the other to be used to test the response of the network. This approach is used for both the FFBP and Recurrent neural network models. Preliminary results have been compiled with both neural network models. The results from the Feed Forward Back Propagation network were consistently above 92% accuracy in forecasting when tested after training. Similarly, the results from the Recurrent neural network model were consistently above 94% accuracy in prediction.
Keywords
Artificial Neural Network, FFBP, Recurrent, H.265
Location
Concourse and Atrium
Presentation Year
2015
Start Date
11-7-2015 10:10 AM
End Date
11-7-2015 11:20 AM
Publication Type and Release Option
Presentation (Open Access)
Recommended Citation
Daly, Collin, "Novel H.265 Video Traffic Prediction Models Using Artificial Neural Networks" (2015). Georgia Undergraduate Research Conference (2014-2015). 3.
https://digitalcommons.georgiasouthern.edu/gurc/2015/2015/3
Novel H.265 Video Traffic Prediction Models Using Artificial Neural Networks
Concourse and Atrium
By the year 2019, the Cisco Virtual Networking Index forecasts that video traffic of all types will constitute 80%-90% of global internet traffic, amounting to 1.24-1.44 zettabytes of data passing through internet infrastructure. Video bandwidth forecasts can significantly improve the video transport mechanisms by improving the efficiency of Dynamic Bandwidth Allocation. Especially with the increasing popularity of the newer, high-definition video resolution such as 4K and high compression ratio encoding standards such as H.265, the forecasting of video frames becomes necessary to meet network demand. While large data transfers are becoming more commonplace and video streaming demands continually increase, the physical internet infrastructure is oftentimes unable to be implemented or to be upgraded rapidly enough to provide the requisite bandwidth to satisfy user demand. To provide increased network throughput with the limitations imposed by hardware, the use of highly trained and adaptive neural-computational network models will allow internet service providers to effectively forecast the necessary bandwidth to provide uninterrupted video delivery and streaming. We propose using an artificial neural network to predict the necessary bandwidth requirements of H.265 video streaming. We have chosen to utilize the Feed Forward Backward Propagation neural network and the Recurrent neural network to predict bandwidth requirements, as neither method has previously been attempted with H.265 encoding. The trace files in this library are divided approximately in half, with one half to be used to train the artificial neural network and the other to be used to test the response of the network. This approach is used for both the FFBP and Recurrent neural network models. Preliminary results have been compiled with both neural network models. The results from the Feed Forward Back Propagation network were consistently above 92% accuracy in forecasting when tested after training. Similarly, the results from the Recurrent neural network model were consistently above 94% accuracy in prediction.
