| In this thesis, the research and development of a real-time speech encoder for hand hold terminal application chip (Garfield), and develop a method for system estimation of performance and power base on ARM ISS model.Garfield4 project is a System-On-a-Chip (SoC) with ARM7TDMI as core processor without cache and includes other on-chip modules like high-speed embedded memory (eSRAM) and Multimedia Accelerator (MMA), etc.In the paper, firstly, introduce some algorithms for speech encoder, and compare these algorithms of performance and complexity. Then describe G.729A and G.726 in details. We choose fixed-point algorithm, which makes FPU or DSP unnecessary, enhances the encoding performance and reduces the system cost, describe how to converse float-point algorithm to fixed-point. Secondly, we do optimization on every level, like global, functional, algorithm, and assembly of software itself; after this, we do optimization base on processor and Garfield chip, such as, using embedded SRAM and hardware Multimedia Accelerator. When done this, we do system performance estimation (ARM7TDMI, ARM720T, ARM9) using high level C model of EMI and SDRAM base on ARM ISS model, then compare the result of G.726 base on ARM7TDMI and real chip-Garfield, we find out that the difference is 10% around, so it is useful for developing a real system; At the same time, we analysis the encoded speech quality after optimization of these two algorithms; Also we explore the power consumption of G.729A using different memory location and get a good memory placement for low power consumption.By now, G.726 has been imported to Garfield4 chip, and completes the speech coding for real time implantation. When system clock is 70MHz, the speech coding capability is 36.08 Kbytes/s. Because of the low process ability of Garfield and the complexity of G.729A, G.729A can not reach real time implementation; but it can easily run on the next generation Garfield chip base on ARM720T and ARM9. |
PDF Full Text Request