Rate control requirements in video communication systems: In many video communication systems, the compressed video bitstream needs to be transmitted over a constant bit rate (CBR) channel. However, due to the diversity of frame content, the output bit rate of the video encoder varies, so a buffer is needed to process the video stream at variable bit rate (VBR) and a rate control scheme is employed to prevent buffer overflow or underflow.
Challenges to low-latency video communication: For interactive video applications, end-to-end latency limits are very strict, which limits the size of the buffer you can use. Small buffers are prone to overflow and underflow, so they require more precise rates.Control algorithm。
Rate-controlled particle size: Rate control can be applied to the frame level or macroblock (MB) level. Compared to frame-level rate control, MB-level rate control can more accurately match the target bit budget and improve buffer adjustment, although this will reduce encoding efficiency.
Limitations of existing algorithms: Several MB-level rate control algorithms have been introduced in the literature, but these algorithms occasionally have large errors in bit rate estimation due to inaccurate source models, so they require a larger buffer.
ρ domain rate control model: In literature [4], there is a linear relationship between the bit rate (R) and ρ (the percentage of the zero transformation coefficient after quantization). This linear model has been used for rate control of H.263 and MPEG4 and can produce more accurate bit rate estimation.
The proposed algorithm: This paper proposes a MB-level rate control algorithm based on the ρ domain rate model for low-latency video communication. The algorithm uses an exponential model to describe the relationship between ρ and quantization step (Qstep) at the MB level, and introduces a switching QP calculation scheme to avoid large deviations between the frame size and the target bit rate.
Algorithm details: If the remaining bit budget is greater than the threshold, QP is calculated from the exponential model; otherwise, use the QP of the previous MB to add a constant as the QP of the current MB.
Paper structure: The next part will overview the original ρ domain rate control, describe the proposed exponential model, introduce the proposed MB-level rate control scheme in detail, and present the experimental results, and finally draw a conclusion.