| Summary: | 博士 === 國立臺灣科技大學 === 電子工程系 === 103 === Nowadays the combination of multi-media and handheld devices has become the mainstream consumptive product. Processing of multi-media signals such as video and stream on such devices is more and more frequently. How to realize smooth processing of multi-media signal has become a hot topic. A lot of hardware manufacturers will improve performance by adding more CPUs, while some others make use of use Graphic Processing Unit (GPU) or Digital Signal Processor (DSP) to accelerate graphic processing. In terms of the software, it will manage memory, control the process of programs and simplify the algorithms to achieve the objective of meeting the time requirements. To sum up, the independent processing of software and hardware can’t optimize the performance improvement. Therefore, the dissertation will focus on applying the characteristics of the hardware platform on the handheld devices and integrating with the software management to conduct appropriate performance allocation for the architecture of the multi-media processor.
There are two common methods to improve hardware performance. One is adding more CPU cores, in which the cores process is in a sequential way that makes it more difficult to write the multi-media players on application layer. Thus, we don’t choose such hardware platform. The other is a hardware platform composed of a CPU based on Reduced Instruction Set Computing (RISC) and a System On Chip (SOC) integrated with DSP, which is our choice due to the parallel processing of DSP. With scheduling algorithm, we resolve the assignment of parallel processing tasks for two different processors, which is different from the previous solution that the first makes the schedule and then distributes the tasks for RISC and DSP respectively in a fixed way. In this way, a certain chip will keep computing, so it fails to achieve the effectiveness of task partitioning. The common scenario is they can’t perform multiple multi-media computing tasks before processing the simple input/output, because all system resources are occupied by high-computing processor. In view of this, we change to adjust the time for task assignment by real-time monitoring resources, so as to maximize the performance of the resources. Besides, the core architecture varies in different characteristics, so does the data transmission unit. Regarding reducing the performance degradation due to these factors, Direct Memory Access (DMA) is added to improve the communication problem of time consumed between two different processors. Finally, in order to resolve image compression- the multi-media processing technology with the heaviest computing, the dynamic quality control algorithm is proposed. It mainly improves the encoding performance, achieving smooth stream playing. By monitoring the system performance, it knows when to assign more tasks to DSP, and when to reduce the encoding workload in DSP. Consequently, it avoids the system resources not being occupied by a certain computing.
The highlight of hardware scheme for the handheld device lies in the portability. However, to confess, as the multi-media will be applied more widely and the execution will consume chip computing, the battery power will become less in a proportional way with more energy consumed. If it fails to gain more battery power, it will be unable to achieve high-endurance power supply. Thus, we propose improving performance by software computing, which further allows the user to execute various multi-media in a long time on light and slim handheld devices.
|
|---|
