多上下文MQ编码器优化与VLSI实现

Optimization of Two-Context MQ-Encoder and Implementationof VLSI Architecture

MQ(Multiple Quantization)编码器由于效率低下已经成为JPEG2000的性能瓶颈.本文对MQ编码算法中的上下文关系进行了提取,对索引表中的启动态和非暂态进行了分离,并提出一种用于预测索引值的方法.同时,对重归一化运算中出现的大概率事件和小概率事件进行分离,使其可并行对2个上下文完成编码.依据该算法,本文提出了一种多上下文并行处理的MQ编码器VLSI结构.实验结果表明,本文提出的MQ编码器能够工作在286.80MHz,吞吐量为573.60 Msymbols/sec,相比Dyer提出的Brute Force with Modified Byteout结构,本文的吞吐量提升约35%,且面积减小78%.

MQ-encoder is a key bottleneck in the JPEG2000 image compression system.In this paper,the dependence between contexts in the MQ-encoder has been extracted;and an improved mechanism has been proposed to acquire indexes by separating the ＂start-up＂ states and remaining states.Besides,a prediction method has been proposed to resolve the stall problem when the current context is the same as the previous context.Also the probabilities of MPS and LPS coding events in the renormalization procedure have been analyzed;and either of the MPS and LPS coding events has been optimized.Based on this improved arithmetic,a VLSI architecture for Two-context MQ-encoder is proposed.Synthesis result shows that the processing speed of the MQ-encoder could reach as high as 286.80MHz with a throughput of 573.60Msymbols/sec.Compared with the ＂Brute force with modified byteout＂ architecture,the throughput obtained in this work has been improved by 35% and the area has been reduced by 78%.