TechZine – Alcatel-Lucent

Technology

Optimizing Video at the Base Station with Adaptive Guaranteed Bit Rate

Highlights

  • HAS video flows can be stabilized in today’s unpredictable wireless environment.
  • More video flows can be supported without starving best-effort traffic.
  • Aggregate Subscriber Quality of Experience (QoE) with video is improved with fair resource distribution.

By building “coding awareness” into their networks, mobile network operators (MNOs) can allocate HTTP Adaptive Streaming (HAS) video resources more efficiently, while delivering a better quality of experience.

As mobile video traffic continues to expand, MNOs are looking for cost-effective video optimization methods that can help them meet demand. “Out-of-network” and “in-network” solutions both have their limitations. But an innovative resource allocation solution for video flows enables more consistent performance for HAS video streams across rapidly fluctuating network conditions. In addition, it gives MNOs more granular control over how resources are allocated in real time.

Implemented at the base station, this method of optimization uses an intelligent algorithm to dynamically adapt the network’s Guaranteed Bit Rate (GBR), in response to increasing or decreasing network congestion. This dynamic adjustment of GBR is the key to “coding awareness” in the radio access network. It allows more stable bandwidth to be delivered to each HAS client, which is crucial for good HAS video quality in unpredictable wireless environments.

In addition, Adaptive GBR (AGBR) provides an automated “control knob” for balancing video quality and resource utilization. It provides aggregate QoE optimization in a heterogeneous environment of different screen sizes, video intensities, radio conditions — and data users. Resources are allocated to video users according to the real-time level of contention for available resources, with a goal of optimizing the experience of video users , without ever starving best-effort traffic.

In simulated comparisons, AGBR supported more HAS video clients per sector than GBR, without starving best effort flows. It also supported 5% more video flows, with ten times as much best-effort throughput, compared to GBR.

Limitations to today’s video optimization approaches

Currently, there are two main approaches to video optimization—out-of-network solutions, such as transcoding and transrating, and in-network solutions that can make use of the Guaranteed Bit Rate defined in 3GPP standards. The primary advantage of out-of-network methods is their in-depth knowledge of the video characteristic requirements. However, they cannot take into account the latest real-time radio conditions, nor can they affect RAN resource allocation for the various video flows.

HTTP Adaptive Streaming — HAS is one of today’s most popular technologies for streaming video. Implemented chiefly by content providers, it incorporates a network-aware approach to video optimization. That is, it adjusts the codec rate (which determines video quality), based on the bandwidth or data rate available between the video server and the client. It also optimizes performance of individual flows based on screen size and video complexity.

However, HAS is not able to take into account the rapidly changing radio conditions and congestion often found in wireless networks. The streaming rate adapts only to available bandwidth. As a result, users with similar radio conditions will get the same throughput. Hence those watching a “complex” video will have a lower QoE than users watching a “simple” video. And users with a small screen will have a higher-QoE than users viewing the same content on a bigger screen.

Guaranteed Bit Rate — The usual network-based method for addressing performance quality is to provide bandwidth stability using QoS mechanisms that are defined in industry standards. For video flows, the best available approach is to assign Guaranteed Bit Rate (GBR) bearers to the flows. This means that a base station’s resource allocation is modified to give more resources to its video traffic, enabling better than best-effort quality. Specifically, it helps deliver a more consistent, steady throughput to a HAS video client, which prevents quality fluctuations as a user moves through the network.

However, GBR has limitations. First, improvements in video performance often come at the expense of lower throughput for best-effort traffic. In addition, GBR doesn’t have the flexibility needed to respond rapidly to radio and congestion conditions that vary over relatively short time periods. As a result, a GBR value that seems optimal initially may not be desirable or achievable as radio conditions worsen, as the GBR bearer then consumes too much of the network’s resources.

Therefore, setting the GBR value can be problematic. Setting it too low makes it largely irrelevant. Setting it high makes resource consumption inefficient. Fairness is also an issue. When the GBR cannot be achieved, the throughput of video flows typically follows proportional fairness. That means some flows may provide very good quality, while others perform very poorly.

A “control knob” for quality and resource utilization

What’s needed to overcome the limitations of HAS and GBR is an intelligent method of resource allocation at the base station. It should be able to implement principles of aggregate fairness, with the goal of improving video performance across multiple flows — while balancing video quality against the needs of all traffic types, when network resources are scarce. This approach to optimization should incorporate the following objectives:

  • Average throughput to HAS clients should be maintained, as much as possible, when radio and congestion conditions change — only allowing gradual changes to the average throughput.
  • When devices with different screen sizes experience the same radio conditions, they should get different rates.
  • The priority level for video over data clients should be configurable.
  • To ensure that data clients are not starved by video flows, resource partitioning between data and video clients should be adaptable.
  • When resources available for video diminish, throughputs for each video flow should be set to reflect its radio conditions.
  • Throughputs for video clients should be chosen to maximize an aggregate throughput measure, with fairness criteria more suited for video and different from best effort data criteria.
  • When resources are scarce, each client’s throughput should be scaled down proportionally to its minimum target value, making sure that resources remain for best effort data clients.
  • When some users’ throughputs fall to the minimum acceptable level, additional resources should be borrowed from flows with better radio conditions, as needed to keep as many flows as possible above the minimum.

Adaptive GBR for coding-aware resource allocation

What can provide the fine-tuning required to meet these resource allocation objectives at each base station? Adaptive GBR (AGBR) uses an intelligent algorithm to dynamically adjust the effective GBR value, based on the congestion level in the local eNodeB. This approach provides a stable wireless channel for HAS video, which enables the consistent code rate needed to provide a better quality of experience for video viewers.

In other words, this coding-aware solution improves the flexibility of GBR, so that it can compensate for the limitations of HAS in a wireless environment. It also addresses the question: What is the “right” guaranteed rate for HAS video? AGBR’s adaptive capabilities allow high rates, such a 1 Mbps, for better video quality when a cell is largely idle. But it can quickly restrict the video flows to lower rates, such as 250 kbps, if a growing numbers of users create network congestion.

The AGBR algorithm — A desired rate and acceptable rate are set for video flows. Then this intelligent algorithm computes total video resources according to the proportion of video among all the flows, with a slight bias toward video flows to ensure better than best-effort delivery. Resource allocation is adapted with each arrival and departure from a cell, and a resource cap for video flows ensures that best-effort traffic is not starved, even when there is congestion.

Available video resources are all allocated across different video flows, according to a “constrained” optimization. This approach maximizes an aggregate throughput objective that is subject to the minimum and maximum constraints for each flow. The AGBR algorithm applies separate and different fairness criteria than the proportional fairness typically used in the eNodeB scheduler.

AGBR compared with best effort — The results of a simulation comparing AGBR with best-effort video are shown in Figure 1. With best-effort video, throughput is a function of SINR (geometry). So, as shown in this example at the lowest HAS rate (150 kbps), video would stall for 13% of users in the simulation. AGBR, in contrast, reapportions eNodeB resources. This approach reduces the probability of a stall and maintains greater video rate stability—resulting in a better quality of experience for viewers. (In this simulation, the same resources were allocated to data users in both the AGBR and best-effort cases.)

Figure 1. AGBR and best-effort: a comparison

Figure 1. AGBR and best-effort: a comparison

Implementation of AGBR

AGBR is implemented in coordination with a base station’s existing scheduling algorithm, allowing only small changes to the GBR at defined intervals. Optimization is executed in real-time using a virtual scheduler approach. Figure 2 shows a sample infrastructure that can support AGBR.

Figure 2. Sample end-to-end AGBR architecture

Figure 2. Sample end-to-end AGBR architecture

The algorithm for GBR adaptation is implemented in the eNodeB for video flows identified as AGBR bearers. This handling of specific video flows must be triggered by an outside source, as indicated by the functions within the red dotted rectangle. These include a Video Application, a Traffic Detection Function which performs deep packet inspection to identify target video flows, or a CDN tasked with delivering HAS content to the UE. The triggering from one or more of these sources may be sent directly to a 3GPP Standardized Policy Charging Resource Function (PCRF) using the 3GPP Rx or Sy interface. Or, for application specific tailoring of the bearer setup in the RAN and EPC, it may be sent to an Application Policy Server (APS) via a REST API interface.

The APS provides additional capabilities to tailor the AGBR setup to the requirements of the device, the HAS application and operator policy. In the 3GPP core, the PCRF uses standardized protocols to trigger setup of Dedicated GBR Bearers for the video flows. Key parameters in the setup process are used by the eNodeB to trigger AGBR rather than GBR processing.

The architecture shown in Figure 2 can be adapted to meet specific MNO requirements, using different triggers or policy-aware network elements to set up the bearers.

Improved end-user satisfaction, plus increased efficiency

This innovative approach to HAS video optimization promotes a better quality of experience for end users, while giving MNOs more granular control over resource allocation. It resolves the fairness issues raised by GBR with a balanced trade-off between video and data throughputs — supporting more video flows and distributing resources fairly among video users, without starving best-effort data traffic. The concept used in AGBR can also be applied to other content streamed in a fashion similar to HAS.

To contact the authors or request additional information, please send an e-mail to techzine.editor@alcatel-lucent.com.

Editor’s note: This article is based on Bell Labs research conducted by Andre Beck, Steve Benno, Mark Clougherty, Danny De Vleeschauwer, Gang Li, Ray Miller, Dave Robinson and the authors.