Ali C. Begen

Multi-path selection for multiple description encoded video streaming

This paper presents a new framework for multimedia streaming that integrates the application and network layer functionalities to meet such stringent application requirements as delay and loss. The coordination between these two layers provides more robust media transmission even under severe network conditions. In this framework, a multiple description source coder is used to produce multiple independently-decodable streams that are routed over partially link-disjoint (non-shared) path to combat bursty packet losses. We model multi-path streaming and propose a multi-path streaming and propose a multi-path selection method that chooses a set of paths maximizing the overall quality at the client. Overlay infrastructure is then used to achieve multi-path routing over these selected paths. The simulation results show that the average peak signal-to-noise ratio (PSNR) improves by up to 8.1 dB, if the same source video is routed over intelligently selected multiple paths instead of the shortest path or maximally link-disjoint paths. In addition to PSNR improvement in quality, the end-user experiences a more continual steaming quality.

Fast heuristics for multi-path selection for multiple description encoded video streaming

In a previous work [A. C. Begen, et al., 2003], we proposed an optimal multi-path selection method for multiple description (MD) encoded video streaming. To do so, we first modelled multi-path streaming and then developed an expression, i.e., an objective (cost) function, that estimated average streaming distortion in terms of network statistics, media characteristics and application requirements. Naturally, the ultimate goal was to find the set of paths that minimized this cost function. However, finding such sets of paths turned out to be intractable in large topologies. Hence, in this paper, we provide a fast heuristics-based solution by exploiting the infrastructure features of the Internet. The simulations run over various random Internet topologies show that the proposed heuristic is able to find a good solution in a much shorter time than the brute-force approach. Particularly, this heuristic is best suited to such interactive multimedia applications as video-conferencing and VoIP, where multi-path computation is a time-critical process. In addition, it is also suitable for the clients whose processing power capabilities are limited.

Rate-distortion optimized on-demand media streaming with server diversity

This paper studies the streaming of packetized media from multiple servers to a client over a lossy network. In particular, we propose a client-driven rate-distortion optimal packet scheduling algorithm that decides which packet(s) to be requested from which server(s) at a given request opportunity. In doing so, the proposed scheduling algorithm not only attains the maximal presentation quality but also conforms to the rate constraints dictated by the flow, window and congestion control mechanisms. The simulation results clearly demonstrate the efficacy of the proposed algorithm over the single-server rate-distortion optimized streaming.

An Adaptive Media-Aware Retransmission Timeout Estimation Method for Low-Delay Packet Video

Time-constrained error recovery is an integral component of reliable low-delay video applications. Regardless of the error-control method adopted by the application, unacknowledged or missing packets must be quickly identified as lost or delayed, so that necessary actions can be taken by the server/client on time. Historically, this problem has been referred to as retransmission timeout (RTO) estimation. Earlier studies show that existing RTO estimators suffer from either long loss detection times or a large number of spurious timeouts. The goal of this study is to address these problems by developing an RTO estimation method specifically tailored for low-delay video applications. In the media-unaware mode, this method exploits the temporal dependence in packet delay to optimally manage the tradeoff between the amount of overwaiting and redundant retransmission rate. As opposed to existing methods, our approach is completely adaptive to the source video characteristics and time-varying network conditions, and does not use any preset parameters. In the media-aware mode, on the other hand, the timeout estimates are jointly optimized based on the importance and urgency of the video packets such that the rendering quality is maximized under the given rate constraints. With a comprehensive set of simulation and experimental results, we show that both the media-unaware and media-aware RTO estimators detect lost packets faster and more accurately than their rivals. Furthermore, our results also substantiate the fact that the media-aware RTO estimator outperforms all other RTO estimators in terms of video quality

Proxy-assisted interactive-video services over networks with large delays

While emerging broadband access technologies such as DSL and cable are making multimedia services feasible and economically attractive for end-users, there still exist several hurdles in terms of service sustainability and reliability. Unfortunately, without the desired quality-of-service (QoS) support, tackling these hurdles with traditional solutions is an insuperably difficult task. Yet, novel designs that are proven to be useful in various scenarios may easily fail when the underlying network experiences severe packet loss or delay. Such circumstances are unavoidable in todays best-effort Internet and will likely prevail in the near future as well. A promising approach in satisfying the stringent requirements of delay-intolerant video applications is to benefit from configurable proxies. In this study, we introduce a versatile proxy-based solution to enhance the performance of such applications running over networks with large delays. We first propose a methodology that accurately identifies lost packets in real time. This methodology is then used by the proxy and end-users to improve the error-control/protection capability of the video applications. By Internet experiments between the US and Europe, we demonstrate the effectiveness and potential benefits of the proposed approach.

Packet scheduling for multiple description video streaming in multipoint-to-point networks

This paper presents a client-driven rate-distortion optimized packet scheduling algorithm for streaming multiple description (MD) encoded video, where the descriptions are distributed among different video-on-demand servers. Previously, multiple description coding (MDC) has been proposed to provide reliable video communication within a content delivery network, where the main goal was to imitate path diversity by using multiple servers concurrently. Although this approach can alleviate the severe effects of bursty packet losses, transmitting video packets in a time-sensitive and network-adaptive manner is fundamental to the success of MD streaming. Hence, in this paper we propose a packet scheduling algorithm that maximizes the quality of the video rendered at the client under the given rate constraints. This algorithm jointly considers the timeliness requirements of the streaming application, dependency structure of the streamed video, network conditions as well as the error-resiliency features of MDC.

Predictive Modeling of Video Packet Delay in IP Networks

This paper studies linear prediction algorithms for packet-delay modeling. A detailed examination of the delay traces collected from video streams encoded at different bitrates, suggests that autoregressive (AR) models can exploit the correlation among the delay samples and produce the best estimates in terms of the mean-squared error criterion. Simulation results show that AR prediction can reduce the average prediction-error power significantly as compared to the exponentially-weighted moving average prediction as well as the recursive weighted median filtering. This is a promising result since many layers in the multimedia communication protocol stack, e.g., rate control, error control and network adaptation, can greatly benefit from accurate packet-delay prediction.

Estimating Packet Arrival Times in Bursty Video Applications

In retransmission-based error-control methods, the most fundamental yet the paramount problem is to determine how long the sender (or the receiver) should wait before deciding that an unacknowledged (or a missing) packet is lost. This waiting time is generally referred to as retransmission timeout (RTO). An accurate RTO estimation has two main advantages: First, the lost packets can be identified earlier, and hence, can be recovered faster. Second, redundant retransmissions can be avoided, which subsequently not only saves the network resources, but also helps existing network congestion alleviate sooner. Although it is statistically possible to prevent any unnecessary retransmission at the expense of long error-recovery times, such an approach can only be justified for data applications; it is not well-suited for delay-sensitive applications, for which the agility in recovering the lost packets is as important. With this motivation, we recently introduced an RTO estimation algorithm for delay-sensitive applications (A. C. Begen et al., 2004). Provided that the packets are transmitted at equal intervals, this technique successfully estimates the arrival times based on the interarrival-time observations. In this study, we relax the requirement of equal transmission intervals and generalize our technique to handle bursty video applications

Redundancy-controllable adaptive retransmission timeout estimation for packet video

Time-constrained error recovery is an integral component of reliable low-delay video applications. Regardless of the error-control method adopted by the application, unacknowledged or missing packets must be quickly identified as lost or delayed, so that necessary timely actions can be taken by the server/client. Historically, this problem has been referred to as the retransmission timeout (RTO) estimation. Earlier studies show that existing RTO estimators suffer from either long loss detection times or a large number of pre-mature timeouts. The goal of this study is to address these problems by developing an adaptive RTO estimator for high-bitrate low-delay video applications. By exploiting the temporal dependence between consecutive delay samples, we propose an adaptive linear delay predictor. This way, our RTO estimator configures itself based on the video characteristics and varying network conditions. Our approach also features a controller that optimally manages the trade-off between the amount of overwaiting and redundant retransmission rate. The skeleton implementation shows that the proposed RTO estimator discriminates lost packets from excessively-delayed packets faster and more accurately than its rivals, which consequently enables the applications to recover more packets under stringent delay requirements.

Videoconferencing over an intermediate-proxy

While the recent developments in access technologies such as DSL and cable enable end-users to communicate with each other in the means of video and voice, there still exist several hurdles in the sustainability and reliability of Internet communication services. In this paper, we focus on improving the performance of the delivery solutions for interactive media in the best-effort Internet. A promising approach in satisfying the stringent delay/loss requirements of interactive media transmission is to benefit from configurable proxies. In this study, we introduce an intermediate-proxy solution for videoconferencing applications running over the networks with large delays. By the Internet experiments between the U.S. and Europe, we demonstrate the effectiveness and potential benefits of the proposed approach.