QoS provisioning for video streaming over IEEE 802.11 wireless LAN

This is an abridged version of a paper that will be presented at the IEEE International Conference in Consumer Electronics, which opens in Los Angeles.

During last several years, there has been a significant increase in the interest in the use of packetized video over the wireless networks. With rapid growth of emerging demand and deployment of wireless LAN (WLAN), much of IP traffic including multimedia traffic travels from wireless networks to a traditional wired LAN. Such a change in networking environments brings a necessity to refine conventional video rate control schemes for the wireless users in home, office, and public hot-spots.

The video rate control uses UDP as its transport layer protocol because UDP incurs no retransmission delay and jitter. The video traffics with UDP, so called unresponsive flows, do not reduce their load on the network when subjected to packet drops. That is to say, they simply transmit data at the rate at which it was encoded, regardless of the congestion state of the network. Thus, multimedia flows do not consume the same bandwidth as TCP flows, at least in a long-term average sense. It is necessary to have an application-layer flow control scheme for the adaptive transmission of multimedia over UDP. TCP-friendly rate-adaptation behavior for streaming is currently an important IETF requirement (RFC 3448).

The TCP-friendly flow control estimates the recent loss event rate of a connection at the receiver. The receiver communicates this loss rate back to the sender, which adapts its transmission rate to the degree of congestion estimated from the loss rate. To behave in a TCP-friendly manner, the sender adapts according to an equation that models the TCP response function in steady-state, but does so with significantly less fluctuation in the sending rate than the standard TCP congestion control algorithm. As a result, streaming applications can both smoothly and fairly react to congestion over longer time periods. That is to say, the conventional TCP-friendly flow control such as additive increase & multiplicative decrease (AIMD) control and TCP-friendly rate control (TFRC) evaluates the congestion degree and matches the transmission to the available bandwidth in the wired IP network.

However, end-to-end packet losses in a wireless-to-wired network can be caused by network congestion and unreliable error-prone wireless links. AIMD and TFRC schemes using the packet loss as a congestion measure cannot be directly applicable to a wireless network because there is no way to distinguish congestion losses from wireless losses. Thus, an end-to-end loss differentiation algorithm (LDA) is needed for congestion-sensitive video transport protocols over the wireless-to-wired link. Video transport protocols can take advantage of loss differentiation in two key ways.

First, wireless losses do not restrict the sending rate. Second, the useful feedback is provided to the video encoder. For example, if wireless losses are dominating, the encoder can transmit the video streaming under a lossy environment in cooperation with the error-resilience coding methods.

Previous research shows that congestion and wireless losses are differentiated by using a proxy server in an access point (AP) or an explicit congestion notification (ECN) mechanism. The proxy-based TCP-friendly streaming method uses a snoop protocol running on a proxy agent that is implemented in the AP. Snoop protocol intercepts streaming packets, analyzes them, and retransmits them to a client if necessary. Thus, the protocol improves the performance of communication over wireless links. However, since it assumes that a video client is on the last hop wireless link, the proxy-based TCP-friendly streaming is not applicable to the wireless-to-wired network.

On the other hand, ECN-based TFRC calculates TCP-friendly rate using ECN-marked packet loss ratio (PLR) instead of end-to-end PLR including the packet losses occurred in the wireless link. An ECN-capable random early detection (RED) router marks ECN-bit in incoming packets' IP header in a probabilistic manner when it detects congestion. The loss information of ECN-marked packets is collected in a video client and transmitted to the sender by using real time control protocol (RTCP) packets. As a result, the effect of wireless losses in flow control is effectively eliminated. However, the ECN-based TFRC approach has shortcomings. This approach assumes that all routers in the transmission path are ECN-capable and equipped with the RED queue management scheme.

Loss distinctions

Our LDA was designed for a video server station over the first hop wireless link of IEEE 802.11 WLAN. To prevent throughput degradation of multimedia flows traveling through wireless links, the wireless PLR occurred in the wireless link should be estimated. It is assumed that the most packet losses in the IEEE 802.11 wireless link are caused by the unreliable wireless channel features, when media access control (MAC) with request to send/clear to send (RTS/CTS) protocol is used in the infrastructure networks. Thus, the wireless losses over the first hop link should be distinguished from packet losses in the wired network.

In the wireless hop, both the video server and AP have the additional module so called wireless adaptation layer (WAL) witch can be seen as an OSI 2.5 shim layer. This layer is an abstraction used for service provisioning at the link layer. The WAL monitors incoming traffics for all video sessions and transfers the feedback messages including the number of received packets to senders periodically. Consequently, wireless video sender can estimate the wireless PLR because the sender is aware of the number of packets transmitted before the arrival of the returned WAL packet. On the other hand, end-to-end loss information is obtained from the RTCP packets returned from the client. Using two PLRs, wireless PLR and end-to-end PLR, the congestion PLR occurred by the congestion in the wired network is estimated.

Finally, the wireless video server can calculate TCP-friendly rates by substituting congestion PLR for end-to-end PLR. Thus, the target bit rates of the streaming video are not reduced by the wireless losses and the throughput degradation of a multimedia flow can be avoided. Our LDA system using WAL and RTCP feedback messages does not need to modify routers to be ECN-capable and equipped with RED queue management scheme.

After the TCP-friendly rates are determined, video coding methods to guarantee the quality of service (QoS) are considered. The video quality degradation due to the packet losses can be tolerated using error resilient coding methods at the application layer. Moreover, channel-adaptive video resilient coding can promptly react to the wireless losses, since the video streaming system is located in the first-hop wireless link.