Abstract
The entertainment industry's market is increasing, primarily with the introduction of technological advancement which includes cinematography equipment and high-resolution videos. Digital Video Broadcasting (DVB) systems have held the fort in streaming video content to homes, mostly using satellite, terrestrial, and cable communication. Customers subscribe periodically through top boxes (or loosely called decoders) to access premium content, both live and on-demand. However, there was a paradigm shift in internet-based video content. The internet has the advantage of readily-available, easy-to-access protocols and servers, with only an endless stream of information necessary to be transmitted to the user's equipment. Additionally, it is relatively cheap to maintain and less complicated to set up from both marketers' and customers’ perspectives - a major reason for the massive adoption. Hence, traditional broadcasting companies are pressured to adopt Internet streaming. However, the problem with the internet is the lack of Quality of Service (QoS).
Unlike DVB systems, where there are many procedural standards to test the durability of the system, the adoption of the internet limits overall quality assurance. Except for finding a presumably reliable network provider, content providers cannot control or measure the transmission quality. Even if they successfully have a stream of high-quality content at super transmission speed, there is no guarantee of how the customer receives the stream, due to the uncertainties such as broadband service and current network conditions. It is crucial to comprehend the customer’s experience of seeing video streams through the internet. Moreover, given the enormous amount of data from one hour of a video clip streaming, it is essential to compress videos before sending them to the client. While compressing videos can reduce redundant features that ease the strain on the network bandwidth, these videos can lose quality, deterring the user’s quality of experience. Rate adaption techniques have also been applied to improve user QoE in DASH-based streaming platforms, by regulating the buffer and checking the bandwidth.
For broadcasting companies to fully benefit from the internet streaming wave, we develop a DVB platform capable of live streaming and use different devices and network conditions to access the content from our platform. A Dektec, microcontroller, and E-probe are used to measure data such as bit and Signal-to-noise ratio and modulation error ratio, forming the QoS. The aim is to develop a monitoring unit that assesses video quality in terms of QoE, hence we design the QoE using natural scenes, hybridizing the NIQE and BRISQUE parameters. A restricted Boltzmann machine is developed to understand how QoS and QoE factors interact. The simulated results show the correctness of our framework, from the good
iv
correlation values among streamed videos including the real-time videos. RMSE and PLCC values from our proposed RBM technique show a better performance than base techniques.
Our second contribution covers the efficiency of internet streaming along a DVB setup. We consider a video streaming platform with two communication channels - radio access internet communication and a Low Earth Orbit mega-constellation satellite backchannel. We leverage the power of both two channels to serve a super-resolution video while developing dynamic adaptive streaming to send feedback information to the server, manipulate the video bitrate, or switch encoding strategies to improve customer QoE. A deep reinforcement learning technique is applied to learn policies that deliver the maximum quality of videos. For various coding standards, the results show good Peak-Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Metric (SSIM) values. A framework also has a 21% decrease in delay compared to DASH-only implementation.
To find a well-formulated QoE with minimal subjectivity - a current issue in video streaming, a multi-objective optimization (MOO) framework is developed to simulate the reward system such that we generate Pareto optimal answers. The Pareto solution allows for the selection of QoE values according to different human perceptions even after reward computation in the learning process. We validate the MOO using statistical metrics, and it shows a closely packed set of solutions that indicates a valid implementation.