Xilinx XAPP169 manual Performance, Error Handling

Page 21

MP3 NG: A Next Generation Consumer Platform

R

Device decodes from

IP Bus Controller

State

Machine

SYS_CLK FLASH_SEL_N

ACK_N WR_IN_N[3:0]

RD_IN_N

FL_CE_N[3:0]

FL_ALE

FL_WE_N

FL_RE_N

FL_SE_N

FL_WP_N FL_R/B_N

Figure 20: FLASH Controller Block Diagram

When the architecture for this system was being planned, one issue that needed to be addressed was whether to execute the program image directly from FLASH or to copy it to SDRAM. There were two key issues that needed to be considered when making this decision.

Performance

The narrow, 8-bit, interface used for both address and data is one performance issue, but it is not the biggest. The real problem is random access latency. Within a 512-byte memory page, data can be read with a 50 ns read cycle time. The problem comes when the processor accesses data on a different page. The time required to load a page into the page register, where it can be accessed, is 7 ∝s. With a 66 MHz processor frequency this represents 462 instruction times. This latency will adversely effect real-time performance.

Error Handling

Each 512-byte page in the FLASH has 16 bytes of spare storage for storing ECC information. The problem is that this is not enough storage for implementing ECC for small block sizes. For example, to correct single bit errors on an RC32364 cache line (16 bytes) using a Hamming code the following relationship must be satisfied:

N ≤ 2**K – K – 1

where: N is the number of data bits in the block K is the number of ECC bits

Solving for K:

16 * 8 = 256 ≤ 2K – K – 1

K = 9 ECC bits per cache line

Since there are 32 (512/16) cache lines per page, a total of 36 bytes are needed for ECC storage. Recall that 16 bytes are available.

In order to get around this problem, the block size could be increased to 32 bytes. At 32 bytes, ten bits of ECC are required per block, but there are now only 16 blocks per page which is consistent with the available ECC memory per page. Performing the block check over two cache lines could accommodate this larger block size. The down side to this is that every time a cache line is loaded, two would have to be checked with a corresponding increase in memory latency.

As a result of this the decision was made to copy the executable image to the SDRAM memory at boot time. This not only increases performance but also turns the ECC checking issue into a non real-time software exercise.

The estimated FPGA device resources used to implement this block include an estimated 100 CLBs, and the ten I/O pads listed in Table 4.

XAPP169 (v1.0) November 24, 1999

www.xilinx.com

21

 

1-800-255-7778

 

Page 20 Page 22
Image 21
Page 20 Page 22
Contents Summary MP3 NG a Next Generation Consumer PlatformIntroduction MP3 BackgroundSolution Overview MP3 TechnologyMP3 NG a Next Generation Consumer Platform IDT RC32364 RISController MP3 NG System Block DiagramRC32364 Block Diagram RC32364 Read Timing Crystal CS4343 Stereo DAC CS4343 Block DiagramControl Port Timing Samsung Flash Memory KM29U64000T Block DiagramKM29U64000T Read Timing Micron Sdram MemoryMT48LC1M16A1 Block Diagram Function Controller National SemiconductorSoftware Architecture System ImplementationMP3 Decoder and Audio ISR UI ManagerMemory Manager Xilinx SpartanBlock Mapping Code InitializationIP Bus Controller Fpga Logic Block DiagramCPU Interface CPU Interface Block DiagramCpumasterclk LCD ControllerLCD Controller Interface Signal Summary LCD Controller Block DiagramMemory Interface Memory Interface Block DiagramSdram Controller Interface Signal Summary Type Description Sdram ControllerFlash Controller Performance Error HandlingIrda Controller Irda Controller Block DiagramAudio DAC Interface Audio DAC Interface Block DiagramTouch Screen Interface Audio DAC Interface Signal Summary Type DescriptionConclusion Spartan Device SelectionTotal 500 137 References Date Version # Revision 11/24/99 Initial release Revision History

XAPP169 specifications

Xilinx XAPP169 is a pioneering application note that delves into the design and implementation of high-performance digital signal processing (DSP) systems. It serves as a reference guide for engineers and designers looking to leverage Xilinx Field Programmable Gate Arrays (FPGAs) for sophisticated DSP applications. The document provides a comprehensive overview of the techniques and methodologies necessary to harness the power and flexibility of FPGA technology in DSP design.

One of the main features of XAPP169 is its focus on the integration of various DSP functions, including filtering, modulation, and Fourier transforms. By utilizing the inherent parallelism of FPGAs, designers can achieve significant performance enhancements compared to traditional DSP implementations. This parallel processing capability allows for real-time processing of high-bandwidth signals, making XAPP169 ideal for applications such as telecommunications, aerospace, and medical imaging.

The application note emphasizes the use of Xilinx’s advanced tools and libraries, such as the Xilinx System Generator for DSP and the Xilinx Vivado Design Suite. These tools facilitate the modeling, simulation, and synthesis of DSP algorithms tailored to specific requirements, enabling a rapid development cycle. By providing pre-optimized building blocks and IP cores, XAPP169 streamlines the design process, reducing time-to-market for new products and innovations.

Additionally, XAPP169 highlights the ability to leverage high-speed serial transceivers present in Xilinx FPGAs. These transceivers enable reliable transmission of data across long distances with minimized latency and optimized bandwidth utilization. The application note outlines various techniques for managing signal integrity and maximizing throughput, ensuring that designs can meet the stringent requirements of modern DSP applications.

Another characteristic of XAPP169 is its attention to resource utilization and optimization strategies. The document discusses how to balance performance with area and power consumption, which is crucial in embedded applications where space and power are at a premium. By employing advanced synthesis strategies and leveraging the capabilities of Xilinx’s architecture, designers can create efficient and scalable DSP systems.

In summary, Xilinx XAPP169 serves as an invaluable resource for engineers seeking to harness the capabilities of FPGAs in DSP applications. With its focus on high-performance design, integration of advanced tools, and optimization strategies, it opens up new possibilities for innovation in various fields where digital signal processing is essential.
Top Page Image Contents