外文文獻(xiàn)及翻譯-使用基于重構(gòu)計(jì)算機(jī)平臺的fpga分析高性能功率譜（英文）

1、High Performance Power Spectrum Analysis Using a FPGA Based Reconfigurable Computing PlatformYogindra Abhyankar, Sajish C, Yogesh Agarwal Hardware Technology Development Group Centre for Development of Advanced Computing

2、 Pune 411 007, India yogindra@cdac.inC.R. Subrahmanya, Peeyush Prasad Department of Astronomy and Astrophysics Raman Research Institute Bangalore 560 080, India crs@rri.res.inAbstractPower-spectrum analysis is an importa

3、nt tool providing critical information about a signal. The range of applica- tions includes communication-systems to DNA-sequencing. If there is interference present on a transmitted signal, it could be due to a natural

4、cause or superimposed forcefully. In the latter case, its early detection and analysis becomes important. In such situations having a small observation window, a quick look at power-spectrum can reveal a great deal of in

5、formation, including frequency and source of in- terference. In this paper, we present our design of a FPGA based reconfigurable platform for high performance power- spectrum analysis. This allows for the real-time data-

6、 acquisition and processing of samples of the incoming sig- nal in a small time frame. The processing consists of com- putation of power, its average and peak, over a set of input values. This platform sustains simultane

7、ous data streams on each of the four input channels.1. IntroductionThe concept and use of power spectrum of a signal is fundamental in engineering - in communication systems, microwave and radars. Recently, it is also be

8、ing used in diverse applications such as gene identification. In a typ- ical transmit-receive system, if the received signal is pure and as expected, no filtering is required. However, on the other-hand, any interference

9、 overriding the received signal may require certain analysis in order to know more about the interference. As the interference tends to pump addi- tional power in the received waves, the power becomes a useful criterion

10、for such an analysis. Using the reverse- engineering techniques, the excess power information with the incoming signal may help in finding the characteristicsof the interface such as frequency, source etc.A power spectru

11、m [5] is a representation of the magni- tude of the various frequency components of a signal. By looking at the spectrum, one can find how much energy or power is contained in the frequency components of the sig- nal. An

12、alysis or evaluation of the power spectrum is one of the ways of isolating noise.There are a couple of techniques for generating the power spectrum. The most common one is by using the Fourier transform [6]. The other te

13、chniques such as the wavelet transform or the maximum entropy method can also be used.Experimentally, power spectrum can be determined in three ways: (1) Using a spectrum or signal analyzer - a commercial instrument [2]

14、dedicated for displaying the real time power spectra (2) Using a microcomputer based add- on signal analyzer card, or (3) by digitizing experimental data and performing a Fast Fourier Transform (FFT) on a desktop machine

15、.In terms of cost and complexity, the above-mentioned three options are in the descending order, while considering the flexibility, they are in the ascending order. Dedicated analyzers are some times used, however they m

16、ay not be cost effective, flexible or competent enough, to extract the interference related information when the observation win- dow is short.In general, the second option provides additional flexi- bility, especially w

17、hen the Field Programmable Gate Array (FPGA) is used.In this paper we present our design of a very powerful reconfigurable computing based design for solving com- plex signal functions and real-time analysis. Although th

18、is works as an add-on card for a workstation, it is extremely powerful, flexible and relatively cost effective. The power spectrum analysis uses modules developed by us for multi- channel data acquisition and several sig

19、nal processing oper- ations performed simultaneously on four data channels.1-4244-0690-0/06/$20.00 ©2006 IEEE.Figure 2. Power spectrum analyzer implemented on compute FPGAhost performs post-processing and other oper

20、ations on the processed data generated by the RC. This is required to complete the power spectrum analysis. As shown in figure 1, the input LVDS data-streams are handled by the on-board receivers to provide compatible si

21、gnals for the compute engine. The power spectrum computation block that resides in the XCV800 compute FPGA is shown in figure 2. It consists of six main components: Input Sampler and buffers, Multi-channel FFT units, Cha

22、nnel Separator and Power computation unit, Average and Peak Power Compu- tational unit, Time-stamping and control, and the XCV800- XCV300 interface. In the following subsections, we de- scribe these components of the app

23、lication.4.1. Input Sampler and buffersThe spectrum analyzer application requires four LVDS channels as inputs, each having a 4-bit data width. How- ever, there are only eight dedicated differential lines for the channel

24、s. The channels are time-multiplexed in pairs, i.e. channel-1 and channel-2 goes on four lines, while channel- 3 and channel-4 on the remaining four lines. A clock, serv- ing as a strobe is provided. The data to the samp

25、ler unit, changes on positive and negative edges of this clock. The channel-multiplexed input data is passed to the Sam- pler unit, de-multiplexed and forwarded to channel buffers as well as to the input-data-buffers. Th

26、e data from the chan-nel buffers are input to the FFT block, while the data from the input-data-buffers are stored in the SDRAM.The channel buffering is necessary to collect a block of data before the FFT computation. By

27、 using buffer pairs at each FFT input, the data is read and processed by the FFT unit in parallel to the input data streamed by the host in the other buffer. When the FFT core finishes processing the current input data,

28、the memory banks are swapped and the data load and computation continues on the alternate memory bank.4.2. Multi Channel FFTThis block uses two, 256-point complex-FFT units from Xilinx CoreGen library, working in paralle

29、l on the four in- put data channels. Instead of using them for complex FFT computation having real and imaginary inputs, they are used for processing two real data streams. The units calculate complex FFT according to th