Quality-optimization algorithm based on stochastic dynamic programming for MPEG DASH video streaming

Abstract

Abstract-In contrast to traditional push-based protocols, adaptive streaming techniques like Dynamic Adaptive Streaming over HTTP (DASH) fix attention on the client, who dynamically requests different-quality portions of the content to cope with a limited and variable bandwidth but aiming at maximizing the quality perceived by the user. Since DASH adaptation logic at the client is not covered by the standard, we propose a solution based on Stochastic Dynamic Programming (SDP) techniques to find the optimal request policies that guarantee the users' Quality of Experience (QoE). Our algorithm is evaluated in a simulated streaming session and is compared with other adaptation approaches. The results show that our proposal outperforms them in terms of QoE, requesting higher qualities on average.

Evaluation of Q-Learning approach for HTTP adaptive streaming

“…As can be observed in the Table I, the QoE values obtained by both approaches are quite similar. This result validates our formulation since we have obtained a comparable result to that of [6]. Actually, DASH-QL obtains a slightly higher QoE value, due to the fact that our algorithm reduces the number of quality switches and their switch depth, hence a lower S value is obtained.…”

“…However, results have been measured over the last 50 episodes in order to evaluate the performance of the DASH-QL algorithm when its convergence has been achieved. We have used the following parameter values: a typical segment duration T seg = 2 sec and each segment has been encoded at N q = 14 quality levels with bitrates Q¿ distributed between 100 and 4500 Kbps following a similar scheme to the distribution used in [6]. The buffer size (B max ) is 6 sec and we have determined experimentally that a reasonable target value to keep during the streaming session is B op t = 2/Z-B max = 4 sec.…”

“…In that sense, Garcia et al [6] propose an algorithm based on Stochastic Dynamic Programming (SDP), known as DASH-SDP. They compute off-line optimal control policies based on a priori models of the system.…”

What happens when stochastic adaptive video streaming players share a bottleneck link?

“…The model obtains an optimal MDP solution using dynamic programming. Researchers in [12], [2], [21], [3] and [6] propose stochastic dynamic programming (SDP) approaches for rate adaption in DASH players, where player buffer occupancy and bandwidth conditions determine the system rate. In [22] researchers employ a two-state Markov channel model to analyze buffer underflow-delay trade-offs for adaptive playout strategies.…”

“…This policy would indicate the optimal quality to request from the server for the duration of the streaming session (see Figure 3). However, in present MDP solutions the policy is global and is used by all players [28], [4], [12]. This may result in viewers receiving sub-optimal QoE characteristics.…”