readme

system C源码 一种替代verilog的语言

========================================================================
				README for FFT
========================================================================
 
  This example demonstrates a simple synchronous system for a 16-point 
  FFT computation. This example was developed and tested on Solaris2.5.

  Following are the components of the system:

  a. source : Reads in real and imaginary samples from the files "in_real"
 	      and  "in_imag". Interacts with the process "fft". The input 
	      files need to be in ASCII format.

  b. fft : The main block in the system. Computes 16-point fft. More about
	   this block later.

  c. sink : Reads the real and imaginary components of the output transform
 	    values, and writes those to the files "out_real" and "out_imag" 
            respectively. Output files are produced in ASCII format.

      	 +--------------------------------------------------+
      	 | +-----------+     +-----------+   +-----------+  |
      	 | |           |     |           |   |           |  |
      	 | | source    |-----| FFT       |---| sink      |  |
      	 | |           |     |           |   |           |  |
      	 | +-----------+     +-----------+   +-----------+  |
      	 |     main                                         |
      	 +--------------------------------------------------+

  You can find the floating point and fixed point versions in the following 
  directories, respectively:

   ./fft_flpt/
   ./fft_fxpt/

Steps: 
------
i.  Variable "SYSTEMC" within the file Makefile.defs should points to your 
    installation of SystemC.

ii. Choose the appropriate Makefile, i.e.,

       Makefile.gcc   for gccsparcOS5
       Makefile.hp    for hpux10
       Makefile.linux for linux
       Makefile.sun   for sparcOS5

iii. Build the executable, using the Makefile. For instance,

       > make -f Makefile.gcc

iv.  Prepare complex input sequence in the files "in_real" & "in_imag" and 
     run the simulation.

       > run.x

Verification:
-------------

  As a sanity check, the block was tested with a DC signal and an Impulse
  wave.
  
  A COSSAP testbench, using cossap block "DFT_QC", was used to verify the 
  results in a general case. Results of the fixed point block are within 
  a certain range of the precise result, as allowed by the finite precision
  arithmetic.

  Some sample input files are included in each directory. Corresponding 
  result files are also provided with the extension ".golden".

The FFT Block:
--------------
                           ---------------
           in_real  ----->|               |-----> out_real 
                          |               |
                          |               |
           in_imag  ----->|               |-----> out_imag 
                          |               |
                          |   FFT BLOCK   |
           data_req <-----|   (16-point)  |-----> data_ready
                          |               |
                          |               |
         data_valid ----->|               |<----- data_ack
                          |               |
                           ---------------
    Data Read:

    The block initiates the reading of a sample by sending a "data_req" 
    signal to the source of input data. It waits for "data_valid" signal
    from the data source to become high; then it lowers the "data_req" 
    signal and  reads the real and imaginary data samples from it's input 
    ports "in_real" and "in_imag" respectively. 

    The block reads in 16 samples of input data in this fashion.

    FFT Computation:

    This block computes 16-point FFT on a sequence of complex inputs, 
    using radix-2 decimation in frequency algorithm.

    This block performs finite precision arithmetic. The input data is
    read as a signed 16-bit fixed point number, with 10 fractional bits. 
    Twiddle factors and output values have the same representation. 

    Internal to the block, computation is performed using fixed point 
    arithmetic. The input samples and output transforms, are externally 
    inferred as 16-bit integers.

    Data Write:

    Once the FFT calculation is performed, the block writes the tran-
    sform values to a sink. It puts the real and imaginary components 
    of transform value on it's output ports "out_real" and "out_imag"
    respectively, and sends a "data_ready" signal to the sink. It waits
    for "data_ack" signal from the sink to go high and then it writes 
    the next value.

    The block writes the 16 transform values to the output ports in 
    this manner and then it goes to the next Read-Compute-Write cycle.

Floating Point Version
----------------------
  The floating point version is the first step, in modeling this block, 
  where the emphasis is to prove the algorithm and verify the results, 
  while working at the highest level of abstraction. This does not reflect
  anything about the target architecture. 

Fixed Point Version
-------------------
  The fixed point version is the next step, taking closer to the RTL 
  level. This is refined from the floating point version, where all 
  data are refined from infinite precision representation to finite 
  bitwidth representation. All computation is performed using the 
  fixed-point arithmetic.

Future Extensions/Limitations:
------------------------------

  * Control refinement would be the next step, leading to the RTL 
    version of the block.

  * This example does not handle overflow and underflow in the fixed 
    point version.

  * The current block is a 16-point FFT block. It could be extended 
    to a general N-point FFT, with N as a parameter.

  * In the present model, not much attention is paid to the quality of 
    the target hardware. For instance: instead of computing the w-values
    within the block, it could be precalculated outside, and stored in a 
    ROM. Several HW optimizations(for area and speed) can be tried out, 
    once the RTL version of the block is ready.

========================================================================