Geometrically determining the leaky bucket parameters for video streaming over constant bit-rate channels

Abstract

For streaming of pre-encoded bitstreams over constant bit rate (CBR) channels, the channel bandwidth, the receiver buffer capacity as well as the latency requirement vary greatly from application to application. In this paper, we attempt to determine the minimum buffer size and the minimum start-up delay required for streaming a pre-encoded bitstream over CBR channels at any specific bit rate. The proposed method employs geometric operations to derive the optimal determination for low or high bit rates and sub-optimal determination for medium bit rates. The algorithm developed requires little extra information from the encoder and is easy to implement. Our algorithm is implemented in a H.264/AVC video encoder and its performance is compared with that of H.264/AVC hypothetical reference decoder. Our approach provides new theoretical insight and an excellent solution for determining the leaky bucket parameters for video streaming over CBR channels.

Introduction

In most video encoders, bits used to encode the individual video frames fluctuate with varying frame complexities. An encoder buffer is usually used to smooth out the bit rate variation so that the encoded bitstream is suitable for transmission over channels of limited bandwidth. A large buffer size allows large bit rate variation and is helpful for the video quality.¹ However, long start-up delay and large decoder buffer may be required in this case for video streaming. This phenomenon prevents us from using arbitrary encoder buffer size even if the available buffer capacity of the video encoder may be infinite.

The video buffering verifier (VBV) [1], [2] mechanism in MPEG video coding standards and the hypothetical reference decoder (HRD) [3], [9] model in H.264/AVC [10] address the relation between buffer size, channel bandwidth and start-up delay. A VBV/HRD-compliant encoder guarantees that the bitstream can be delivered to a receiver with certain buffer capacity at the given transmission bit rate and then decoded with the given start-up delay. A closely related concept is the leaky bucket model [4], which is a direct metaphor of the encoder buffer and can be characterized by $(R,B,\Gamma )$, where $B$ is encoder buffer size, R is channel bandwidth, and $\Gamma$ is start-up encoder buffer delay, which means the bit transmission from the encoder buffer starts $\Gamma$ seconds after the bits for the first frame are pushed into the buffer.

An encoder usually creates a bitstream according to a specified leaky bucket. For streaming of a bitstream encoded using leaky bucket $(R,B,\Gamma )$, if the channel bandwidth, the buffer capacity and the delay requirement for the video application happen to be equal to those specified by the given leaky bucket, there will be not any problem in delivering and decoding the bitstream. However, if the available channel bandwidth, decoder buffer size or the expected delay is smaller than that specified by the leaky bucket, problem will occur if we do not adjust the transmission parameters appropriately.

Thus, it is necessary to determine the minimum resource (leaky bucket parameters) required for a video streaming system. In [8], a generalized hypothetical reference decoder (GHRD) for H.264/AVC is presented to tackle this problem. The idea of the GHRD is as follows: given a number of leaky buckets that are known to contain the bitstream, the decoder can determine which leaky bucket to use if the bit rate is known. If the available channel bandwidth equals the bit rate specified by any of the leaky buckets, the leaky bucket is used directly for streaming. Otherwise, a linearly interpolated or extrapolated leaky bucket that safely contains the bitstream is used.

This paper proposes a simple but effective algorithm that geometrically determine the optimal or sub-optimal leak bucket parameters for video streaming over constant bit rate (CBR) channels at any specific bit rates, based on our previous work [6]. The advantages of the proposed algorithm include: (1) achieving optimal determination for low and high bit rates, and sub-optimal determination for medium bit rates; (2) being easy to implement; (3) requiring little extra information from the encoder; (4) providing new theoretical insight on determining parameters for video streaming.

The rest of this paper is organized as follows. Section 2 introduces the leaky bucket model and the connection between the encoder buffer and decoder buffer in a video streaming system. Section 3 describes the proposed algorithm. Section 4 presents the experimental results. Section 5 concludes the paper.