On Capacity-Quality Tradeoffs in HTTPAdaptive Streaming over LTE Networks

Ozgur OymanIntel Labs

Santa Clara, CA 95054Email: ozgur.oyman@intel.com

Sarabjot SinghUniversity of TexasAustin, TX 78712

Email: sarabjot@utexas.edu

Abstract—The growing consumer demand for mobile videoservices is one of the key drivers of the evolution of newwireless multimedia solutions requiring exploration of new waysto optimize future wireless networks for video services towardsdelivering enhanced capacity and quality of experience (QoE).One of these key video enhancing solutions is HTTP adaptivestreaming (HAS), which has recently been spreading as a formof internet video delivery and is expected to be deployed morebroadly over the next few years. This paper summarizes ourproposed capacity and QoE evaluation methodology for HASservices based on the notion of rebuffering percentage as thecentral indicator of user QoE, and associated empirical databased on simulations conducted over 3GPP LTE networks.Further details on our work can be found in the papers listedin the references.

I. KEY IDEAS AND SUMMARY OF RESULTS

Adaptive streaming is an increasingly promising methodto deliver video to end users allowing enhancements in QoEand network bandwidth efficiency. Adaptive streaming aimsto optimize and adapt the video configurations over time inorder to deliver the best possible quality video to the userat any given time. Most of the expected broad adoption ofadaptive streaming will be driven by new deployments overthe existing web infrastructure based on the hypertext transferprotocol (HTTP), and this kind of streaming is referred here asHTTP adaptive streaming (HAS). HAS follows the pull-basedstreaming paradigm, where the client plays the central roleby carrying the intelligence that drives the video adaptation,i.e., since HTTP is a stateless protocol. Accordingly, thebroad deployment of HAS technologies will serve as a majorenhancement to (non-adaptive) progressive download methods,allowing for enhanced QoE enabled by intelligent adaptationto different link conditions, device capabilities and contentcharacteristics.HAS has already been spreading as a form of internet

video delivery with the recent deployments of proprietarysolutions such as Apple HTTP Live Streaming, MicrosoftSmooth Streaming and Adobe HTTP Dynamic Streaming . Inthe meantime, standardization of HAS has also made greatprogress with the recent completion of technical specifica-tions by various standards bodies including Third GenerationPartnership Project (3GPP), Moving Pictures Experts Group(MPEG) and Open IPTV Forum (OIPF).

As a relatively new technology in comparison with tradi-tional push-based adaptive streaming techniques, deploymentof HAS services presents new challenges and opportunities forcontent developers, service providers, network operators anddevice manufacturers. One of these important challenges isdeveloping evaluation methodologies and performance metricsto accurately assess user QoE for HAS services, and effectivelyutilizing these metrics for service provisioning and optimizingnetwork adaptation.Motivated by these goals, we developed a novel QoE-

based evaluation methodology for assessing video capacityof HAS services in the context of the 3GPP Long-TermEvolution (LTE) systems in terms of the number of unicastvideo consumers that can be simultaneously supported fora given target QoE. The QoE metric of interest here is therebuffering percentage, which is defined as the percentageof the total presentation time in which the user experiencesrebuffering due to buffer starvation. It is worth noting herethat in a recent study conducted by Conviva, rebuffering hasbeen identified the single most dominating QoE impairment.We define and use the notion of rebuffering percentage toquantify the video service capacity for HAS-based streamingover a HTTP/TCP/IP protocol stack. In particular, our LTEcapacity evaluation counts the number of users simultaneouslysupported via HTTP-based unicast video streaming sessionswhere the users are ”‘satisfied”’ Acov percentile of the time,with a user being counted as satisfied if and only if therebuffering percentage in its video streaming session is lessthan Aout.For our capacity evaluation, we use a dynamic system-level

simulator for the LTE air-interface based on a MATLAB-basedsoftware platform with detailed abstractions of application,transport, MAC and physical layers (details in [1]-[2]). Figure1 shows the LTE unicast video capacities of HTTP-basedfixed-rate streaming (i.e., progressive download) and HAS-based adaptive streaming, with Acov = 95% and Aout = 5%

subject to different target peak-to-signal ratio (PSNR) valuesof 32 dB and 37 dB, respectively. This empirical data clearlydemonstrates that HAS-based adaptive streaming allows forsupporting a significantly larger number of video users in com-parison with HTTP-based progressive download techniques,and finds the QoE-optimal capacity-quality tradeoff. Figure 2

Target PSNR = 32 dB Target PSNR = 37 dB0

20

40

60

80

100LTE Unicast Video Capacity

HTTP Progressive DownloadHTTP Adaptive Streaming

Fig. 1. LTE unicast video capacity comparison.

Fig. 2. Rebuffering percentage distribution comparison.

shows the distribution of rebuffering percentage across userswith HTTP-based fixed-rate streaming and HAS-based adap-tive streaming. The QoE enhancement from adaptive streamingis apparent from the plot in which, with fixed rate streamingover LTE at a target PSNR of 37 dB and only with 20 users inthe system, the 95-th percentile value of rebuffering percentis 5% whereas the corresponding value for an LTE systemwith HAS-based adaptive streaming and twice as load, i.e.,with 40 users, is less than 1%. This is an intuitively expectedoutcome, given the significantly varying link quality amongthe users in the LTE network, leading to frequent occurrencesof rebuffering with HTTP-based progressive download in theabsence of any video quality/bitrate adaptation, especiallywhen the network is unable to support the fixed bitrateduring moments of low throughput caused by unfavorable linkconditions. In contrast, with HAS-based adaptive streaming,each client device can dynamically select the quality/bitratelevels of the fetched videos to ensure continuous playbackwhile also optimizing quality that could be achieved for the

given link throughput, and such adaptation capability ensuresfinding the best possible compromise between high videoquality and minimal occurrences of rebuffering events anddelivering enhanced QoE to a larger number of LTE clients.

REFERENCES[1] O. Oyman and S. Singh, “Quality of experience for HTTP adaptive

streaming services,” IEEE Communications Magazine, Oct. 2011, sub-mitted.

[2] S. Singh, O. Oyman, A. Papathanassiou, D. Chatterjee, and J. Andrews,“Video capacity and QoE enhancements over LTE,” in IEEE InternationalConference on Communications, Jan. 2012, submitted.