Abstract
Real-time Transport Protocol (RTP) is a widely used protocol providing end-to-end transportation for various media in real-time applications. This paper describes the RTP encapsulation of H.264/SVC video data for the RTP/RTCP protocol. The sequence number and timestamp in the RTP packet header are fundamental to correct decoding; therefore, their format and encapsulation are analyzed in detail. An NS-2 network simulation model is designed to transmit real video streams and verify the RTP encapsulation of H.264/SVC video. The simulation results demonstrate the correctness and validity of the proposed RTP encapsulation program.
Introduction
Transmitting high-quality video over the internet in real-time requires a careful balance between speed and reliability. Traditional TCP is often unsuitable due to its "stop-and-wait" retransmission mechanics, which introduce unacceptable latency. Conversely, raw UDP lacks the timing and sequencing information necessary for synchronized playback. RTP serves as the industry standard to address these issues, yet researchers often struggle to simulate actual video streams within network environments like NS-2.
The primary challenge lies in encapsulation: packaging a complex H.264/SVC (Scalable Video Coding) bitstream into RTP packets that a network simulator can process. Specifically, video frames often exceed the Maximum Transmission Unit (MTU) of a network (typically 1500 bytes). Without proper handling of this fragmentation, simulation results fail to reflect real-world network behavior.
Methodology and Framework
The authors implemented a transmission framework that bridges raw video coding with simulated network environments. The process begins with H.264/SVC layering, using the Network Abstraction Layer (NAL) to interface between the video coder and the transport protocol.
To make the simulation realistic, the authors developed a method to generate "trace files" from the RTP packets. These files contain packet IDs, transmission times (derived from RTP timestamps), and sizes. A proxy in NS-2 reads these files to inject "virtual" traffic that accurately mirrors the characteristics of the real video stream.
RTP Encapsulation and Fragmentation
A critical component of this framework is the handling of large video units. While RTP offers several modes, this study focuses on Fragmentation Units (FU_A). This mode allows a single large NAL unit—such as a video frame or slice—to be split across multiple RTP packets.
By implementing FU_A fragmentation, the simulation accounts for the overhead of multiple headers and the increased risk of packet loss when a single video unit is split. The 12-byte RTP header is analyzed to ensure that sequence numbers and 32-bit timestamps are correctly utilized to detect packet loss and maintain synchronous playback at the receiver.
Experimental Results
The study utilized the classic "Foreman" video sequence, encoded at 30fps with a 16-frame test sequence. The simulation tested the video stream under varying conditions by introducing Constant Bit Rate (CBR) background traffic to simulate network congestion.
- Quantitative Results: Quality was measured using Peak Signal-to-Noise Ratio (PSNR). In a vacuum without background traffic, the video maintained a high PSNR of approximately 36-37 dB. When background UDP flows were added, the PSNR dropped to 33-35 dB, reflecting the impact of packet loss and delay.
- Qualitative Results: Visual comparisons revealed that network congestion led to visible artifacts and unsolvable screen areas in the reconstructed video, demonstrating the real-world consequences of packet loss on compressed bitstreams.
Conclusion and Future Work
The research highlights that the RTP timestamp is the critical link between the video world and the simulation world. By accurately modeling fragmentation through FU_A, the authors have provided a more realistic method for testing video performance in NS-2.
Future work will involve more extensive scalability testing. While H.264/SVC is designed for spatial and temporal scalability, this study did not deeply explore the protocol's ability to drop layers during congestion to maintain a stable, albeit lower-quality, stream. Additionally, longer video sequences will be necessary to observe long-term jitter and buffer behavior in complex network topologies.
Summary updated