MPEG−2 to H.263 Transcoder System Modeling and Implementation

In this paper, we discuss the implementation and modeling of an MPEG−2 to H.263 transcoder. With the emergence of video on demand in the cable television market, broadcasters are considering expanding this service to the internet. Due to the limited bandwidth of internet communication, the high quality and high bandwidth broadcasts must be reduced to a low bitrate suitable for video transmission over the internet. For this purpose, we focus on transcoding between the MPEG−2 and H.263 video standards. Our implementation focuses on decreasing the spatial resolution of a frame in the video sequence. Our model of the transcoder introduces parallelism to help facilitate real−time goals. Introduction High definition television (HDTV) and digital cable provide high quality digital video to consumers at home. However, the bitrates required can be fairly high. Both markets utilize the MPEG−2 standard, which supports bitrates as low as 1.5 Mbps. Current HDTV formats require between 16.9 and 18.8 Mbps depending on the resolution used, while current digital cable boxes support up to 30Mbps [2]. Video transmission over the internet will require much lower bitrates, with even broadband connections supporting typical bitrates of only 200 to 300 kbps. Transcoding is the discipline concerned with conversion from one standard to another efficiently. Two video coding standards, MPEG−2 and H.263 contain many similarities than can be leveraged to reduce the complexity of the conversion. Most significantly, they both use similar frame types. Each standard supports I−, P−, and B− frames. I−frames (intra−frames) apply the Discrete Cosine Transform (DCT) to non− overlapping blocks of pixels (macroblocks) and quantize the resulting coefficients. I− frames are self−contained; in other words, they do not depend upon any other frame to define their appearance. P−frames use blocks from the previous frame to define its blocks, while B−frames use the blocks from the preceding and following I− or P−frame. For enhanced compression, H.263 has a special PB frame type that combines a P− and B−frame [3]. In this paper, we incorporate parallel processing techniques with common methods utilized in transcoding implementations. System Models The high level SDF graph demonstrates the parallelism of the MPEG standard. Each X−Code block operates upon a set of frames which lie between two intra−coded frames in the input stream. The purpose of this model is to analyze the tradeoffs associated with a parallel processor implementation of our transcoder. Figure 1. SDF Model of System The model for the individual X−code block reuses the block which converts a group of frames containing two B−frames and one P−frame to a PB−frame to reduce code size. The details inner workings of each block will be discussed shortly. Figure 2. SDF Model of Transcoder MPEG File Reader X−Code