SourceForge.net Logo
   		firfilter

Download

Examples

Documentation

Contact Me

Introduction

The firfilter package is a set of C++ classes which provide commonly used building blocks for constructing a FIR filter system.
The current set of building blocks are:


In addition to these building blocks, there are also several functions for stimulating and measuring the system response:

  • runFilters - unit impulse, single frequency, frequency sweep or N-equally spaced (i.e., xn of DFT)

  • spectralAnalysis - DFT analysis


An example

A 10-band equalizer system was constructed and analyzed using the firfilter package and documented here.


Running the equalizer example

  1. Download the firfilter files.
  2. unzip and untar (where x.y is the version):

     > tar zxf firfilter-x.y.tar.gz

  3. cd to the filter/examples/equalizer directory and make a non-debug executable:

     > cd /.../firfilter-x.y/filter/examples/equalizer
    > make ndebug

  4. If successful, you should see similar (value of linux_xxx depends on uname -ps) to:

     > ls linux*/ndebug/*exe
    linux_x86_64/ndebug/equalizer.exe

  5. Run the equalizer without any arguments:

     > linux*/ndebug/equalizer.exe

    creates the filters and system described here.  This run also outputs several text files which can be viewed using graph.  The files are prefixed with lpf, bpf and hpf for the lowpass, bandpass and highpass filters, respectively.  The cutoff frequencies are also specified in the basename, too.  For example, during design/generation, DFTs of the (windowed) filters are created:

    lpf_fc45.dft.txt DFT of lowpass filter with (design constraint) cutoff of 45 Hz.
    bpf_fl45_fh75.dft.txt DFT of bandpass filter with (design constraint) cutoffs of (lower-band) 45 Hz and (upper band) 75 Hz.
    hpf_fc10665.dft.txt DFT of highpass filter with (design constraint) cutoff of 10.665 kHz.

    After the filters are generated, spectral analysis (DFT) of the entire system is done creating a series of spectralMagDb files.  Each of these files corresponds to an endpoint in the system.  These endpoints are annotated on the original schematic and shown here as Ei labels (where i is [0-13]). 


    spectralMagDb_10g.txt
         		DFT of summation point of the lowest 5 equalizer bands with fc [30,60,120,250,500] Hz.
    spectralMagDb_11g.txt DFT of summation point of the middle 2 equalizer bands with fc [1, 2] kHz.
    spectralMagDb_12g.txt DFT of summation point of the upper 3 equalizer bands with fc [4, 8, 16] kHz.

    6. These DFTs can be viewed using the     viewSpectral.rb script:
> ./viewSpectral.rb

That script requires Ruby and graph.  You can also view the results at these links:
  1. Next, we'll run some single and frequency sweep runs, using a (Ruby) script runFilters.rb:

    > ./runFilters.rb |& tee runFilters.log
First, the single frequency sweeps are done starting with an impulse response.  The sweep results are dumped into the N64 subdirectory (N64 indicates filter order 64).  The single frequency results are named fsweep_fx_y.txt, where x is the single frequency.  The value y can be one of:
  • xn, which is the x[n] input stimulus to the filter system
  • yig, where i is a value 1...13 which indicates the endpoint in the system.  For now, we'll just focus on 10...13, where 10...12 are the summation points shown here and 13 is the filter output y[n].  The yig indicates gold reference, which is the double precision result.  (The yir refer to the revised results, which are fixed precision.  More on these later!)
The impulse response results can be viewed as:


> graph -T X -C -x 0 500 N64/fsweep_f0_y{10,11,12,13}g.txt

and vary the -x arguments, etc. for more/less detail.

There are 4 single sweeps at frequencies: 0 (impulse), 120, 1500, 8000.  The latter 3 correspond to middle points of each of the 3 banks.  Thus, as you view the different outputs note the attenuation at all but the output (y13) and the appropriate bank output: y10, 11 or 12.

For example, view the f=120 response:

> foreach i (10 11 12 13)
    graph -T X -C -x 1000 2000 -L "i=${i}" N64/fsweep_f120_{xn,y${i}g}.txt
  end

Then, repeat for the other frequencies:

> foreach i ( 10 11 12 13 )
    graph -T X -C -x 1000 1100 -L "i=${i}" N64/fsweep_f1500_{xn,y${i}g}.txt
  end

> foreach i ( 10 11 12 13 )
    graph -T X -C -x 1000 1050 -L "i=${i}" N64/fsweep_f8000_{xn,y${i}g}.txt
  end

You should verify that the expected attenuations occur at the expected endpoints (of the corresponding graph results)!

  1. Notice that a series of N64/*r.txt files were also created during these runs.  These *r files are the revised results, and the *g files are the golden results.  The revised are simulation using fixed point arithmatic and the golden are simulations using (full) double precision.

    The (current) revised precision is 18-bit: using 14 fraction and 4 integer (2's complement) bits.

    The bit-accurate (revised) simulation capability will be utilized here to co-simulate/verify a synthesizable Verilog HDL model of the 10-band graphic equalizer.