An Efficient Streaming and Decoding Architecture for Stored FGS Video

Yi-Shin Tung, Ja-Ling Wu, Po-Kang Hsiao, and Kan-Li Huang

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 12, NO. 8, AUGUST 2002

Outline

Introduction Multiple-Way Bitplane Decoding On-Demand Streaming and Decoding

Scheduling Refined FGS Decoding Architecture

and Experimental Results Conclusions

Introduction

Bitplane decoding’s need for accessing large frame buffer several times.

Enhancement data needs to be queued up until base layer no longer need to be referenced.

Multiple-Way Bitplane Decoding

352*240*5(bits)

352*240(bytes)Add all bitplanes

Inverse zigzag scanIDCT

Multiple-Way Bitplane Decoding (Cont.)

When the frame size is large, a cache miss will happen.

Treat all the decoding processes on a macroblock(MB)-basis.

All processes for one MB(VLD, combine bitplanes, idct) are down together and then proceed to the next MB.

Multiple-Way Bitplane Decoding (Cont.)

16*16*5(bits)

16*16(bytes)Add all bitplanes

Inverse zigzag scanIDCT

352*240(bytes)stored

On Demand Streaming and Decoding Scheduling

All FGS-VOPs depend only on their corresponding base-layer VOPs.

In general decoding procedure, we can’t immediately reuse the decoded frame buffer of the base layer.

I B B P

FGSVOP

FGSVOP

FGSVOP

FGSVOP

On Demand Streaming and Decoding Scheduling (Cont.)

The enhancement-layer decoding will be delayed.

The streaming schedule is also modified to save the required decoding buffer.

I B B P

FGSVOP

FGSVOP

FGSVOP

FGSVOP

On Demand Streaming and Decoding Scheduling (Cont.)

On Demand Streaming and Decoding Scheduling (Cont.)

Refined FGS Decoding Architecture and Experimental Results

Refined FGS Decoding Architecture and Experimental Results (Cont.)

Refined FGS Decoding Architecture and Experimental Results (Cont.)

Refined FGS Decoding Architecture and Experimental Results (Cont.)

Conclusions

Remove the bottleneck of cache traffic. The resulting decoding process can

effectively utilize the frame buffers. About 20%~40% performance gain

can be obtained.