A consensus approach to PI gains tuning for quality‐fair video delivery

Abstract

In this paper, we provide necessary and sufficient conditifor the quality-fair delivery of multimedia contents to mobile users. We control the encoding rates and the transmission rates of the video streams, delivered through a limited capacity channel. This problem is cast into a problem of consensus among identical discrete-time linear systems, connected through a network with fixed and fully connected topology. The information exchanged over the communication network is the measure of the quality of the encoded videos. Based on a consensus result for identical linear systems, we reduce the problem of designing the proportional and integral gains of the encoding rate and transmission rate controllers to a linear static output feedback. We propose an iterative design technique based on linear matrix inequalities to solve the corresponding nonconvex problem, thereby providing a constructive optimality-based approach to the proportional and integral gains tuning problem. We demonstrate the effectiveness of our method in simulations, where we compare it with pre-existing approaches.L. DAL COL ET AL. of stored videos. Lagrangian optimization frameworks are considered in [4,5]. In [4], the sum of the achievable rates is maximized while minimizing the distortion difference among streams. In [5], the average quality is maximized under total rate constraints, and constraints on the peak signal-tonoise ratio (PSNR) discrepancy between streams. Both approaches require to gather all RUCs of the streams at the control unit. The user experience is accurately modeled in [6] using the empirical cumulative distribution function of the predicted video quality. In [7,8], a heuristic control approach based on proportional integral (PI) controllers is used to regulate the encoding rate of video servers and the transmission rate of the video streams towards the channel. In fact, if the videos are encoded at a constant bit-rate, the quality may change significantly with the variations of the characteristics of the contents. On the other hand, if the contents are encoded at a variable bit-rate, the buffering delay may increase significantly, leading to unacceptable delivery fluctuations. The joint actions of the controllers provide the desirable effect in terms of video fairness: Videos with utility below the average are drained faster through the corresponding buffers, and the encoding rate of such streams is then increased to improve the quality. In the scheme [7,8], the control is performed in some media-aware network element (MANE) [9] at the bottleneck of the links between the remote servers and the communication channel. The MANE is located close to the base station, to which the clients are connected. In this fully centralized version of the controller, the MANE is in charge of sending the encoding rate target to each video server, based on the measure of the buffer level. On the contrary, in a partially distributed control architecture, the servers receive only the individual buffer discrepancies, and they are in charge of comp...

Guest editorial: special issue on Consensus‐based Applications in Networked Systems

“…Dal Col et al . propose a consensus‐based approach to PI gains tuning for quality‐fair video delivery. The PI gain tuning problem is reduced to a linear static output feedback problem with linear consensus results.…”