Set Up the Compiler

The first step is to set up a supported C compiler. MATLAB Coder automatically locates and uses a supported installed compiler. You can change the default compiler using mex -setup. For more details, see Change Default Compiler. For a current list of supported compilers, see Supported and Compatible Compilers.

Break Out the Computational Part of the Algorithm into a MATLAB Function

To generate C code, the entry point must be a function. You do not have to generate code for the entire MATLAB application. If you have specific portions that are computationally intensive, generate code from these portions in order to speed up your algorithm. The harness or the driver that calls this MATLAB function does not need to generate code. The harness runs in MATLAB and can contain visualization and other verification tools that are not actually part of the system under test. For example, in the Construct a Sinusoidal Signal Using High Energy FFT Coefficients example, the plot functions plot the input signal and the reconstructed signal. plot is not supported for code generation and must stay in the harness. To generate code from the harness that contains the visualization tools, rewrite the harness as a function and declare the visualization functions as extrinsic functions using coder.extrinsic (MATLAB Coder). To run the generated code that contains the extrinsic functions, you must have MATLAB installed on your machine.

The MATLAB code in the for loop that reconstructs the original signal using high-energy FFT coefficients is the computationally intensive portion of this algorithm. Speed up the for loop by moving this computational part into a function of its own, GenerateSignalWithHighEnergyFFTCoeffs.m.

L = 1020;
Sineobject = dsp.SineWave('SamplesPerFrame',L,...
    'SampleRate',44100,'Frequency',1000);
rng(1);
numIter = 1000;
for Iter = 1:numIter
    Sinewave1 = Sineobject();
    Input = Sinewave1 + 0.01*randn(size(Sinewave1));
    [ReconstrSignal,numCoeff] = GenerateSignalWithHighEnergyFFTCoeffs(Input);
end
max(abs(Input-ReconstrSignal))
figure(1);
plot(Input)
hold on;
plot(ReconstrSignal,'*')
hold off

function [ReconstrSignal,numCoeff] = GenerateSignalWithHighEnergyFFTCoeffs(Input)

ft = dsp.FFT('FFTImplementation','FFTW');
ift = dsp.IFFT('FFTImplementation','FFTW','ConjugateSymmetricInput',true);

FFTCoeff = ft(Input);
FFTCoeffMagSq = abs(FFTCoeff).^2;
L = size(Input,1);
EnergyF = (1/L)*sum(FFTCoeffMagSq);
[FFTCoeffSorted, ind] = sort(((1/L)*FFTCoeffMagSq),1,'descend');

CumFFTCoeffs = cumsum(FFTCoeffSorted);
EnergyPercent = (CumFFTCoeffs/EnergyF)*100;
Vec = find(EnergyPercent > 99.99);
FFTCoeffsModified = zeros(L,1);
FFTCoeffsModified(ind(1:Vec(1))) = FFTCoeff(ind(1:Vec(1)));
numCoeff = Vec(1);
ReconstrSignal = ift(FFTCoeffsModified);
end

Make Code Suitable for Code Generation

Before you generate code, you must prepare your MATLAB code for code generation.

codegen -args {Input} GenerateSignalWithHighEnergyFFTCoeffs 

??? The left-hand side has been constrained to be non-complex, but the right-hand side 
is complex. To correct this problem, make the right-hand side real using the function 
REAL, or change the initial assignment to the left-hand side variable to be a complex 
value using the COMPLEX function.

Error in ==> GenerateSignalWithHighEnergy Line: 24 Column: 1
Code generation failed: View Error Report
Error using codegen
 
 

FFTCoeffsModified = zeros(L,1)+0i;

codegen -args {Input} GenerateSignalWithHighEnergyFFTCoeffs 

[ReconstrSignal,numCoeff] = GenerateSignalWithHighEnergyFFTCoeffs(Input);

[ReconstrSignalMex,numCoeffMex] = GenerateSignalWithHighEnergyFFTCoeffs_mex(Input);

L = 1020;
Sineobject = dsp.SineWave('SamplesPerFrame',L,...
    'SampleRate',44100,'Frequency',1000);
rng(1);
numIter = 1000;
for Iter = 1:numIter
    Sinewave1 = Sineobject();
    Input = Sinewave1 + 0.01*randn(size(Sinewave1));
    [ReconstrSignalMex,numCoeffMex] = GenerateSignalWithHighEnergyFFTCoeffs_mex(Input,L);
end
max(abs(Input-ReconstrSignalMex))
figure(1);
plot(Input)
hold on;
plot(ReconstrSignalMex,'*')
hold off

Compare the MEX Function with the Simulation

Notice that the harness runs much faster with the MEX function compared to the regular function. The reason for generating the MEX function is not only to detect code generation and run-time issues, but also to speed up specific parts of your algorithm. For an example, see Signal Processing Algorithm Acceleration in MATLAB.

You must also check that the numeric output results from the MEX and the regular function match. Compare the reconstructed signal generated by the GenerateSignalWithHighEnergyFFTCoeffs.m function and its MEX counterpart GenerateSignalWithHighEnergyFFTCoeffs_mex.

max(abs(ReconstrSignal-ReconstrSignalMex))

ans =

     2.2204e-16

The results match very closely, confirming that the code generation is successful.

Generate a Standalone Executable

If your goal is to run the generated code inside the MATLAB environment, your build target can just be a MEX function. If deployment of code to another application is the goal, then generate a standalone executable from the entire application. To do so, the harness must be a function that calls the subfunction GenerateSignalWithHighEnergyFFTCoeffs. Rewrite the harness as a function.

function reconstructSignalTestbench()
L = 1020;
Sineobject = dsp.SineWave('SamplesPerFrame',L,...
    'SampleRate',44100,'Frequency',1000);
rng(1);
numIter = 1000;
for Iter = 1:numIter
    Sinewave1 = Sineobject();
    Input = Sinewave1 + 0.01*randn(size(Sinewave1));
    [ReconstrSignal,numCoeff] = GenerateSignalWithHighEnergyFFTCoeffs(Input,L);
end

Log all 1000 frames of the input and reconstructed signal and the number of FFT coefficients used to reconstruct each frame of the signal. Write all this data to a binary file named data.bin using the dsp.BinaryFileWriter System object™. This example logs the number of coefficients, which are scalar values, as the first element of each frame of the input signal and the reconstructed signal. The data to be written has a frame size of M = L + 1 and has a format that looks like this figure.

N is the number of FFT coefficients that represent 99.99% of the signal energy of the current input frame. The meta data of the binary file specifies this information. Release the binary file writer and close the binary file at the end.

The updated harness function, reconstructSignalTestbench, is shown here:

function reconstructSignalTestbench()
L = 1020;
Sineobject = dsp.SineWave('SamplesPerFrame',L,...
    'SampleRate',44100,'Frequency',1000);
header = struct('FirstElemInBothCols','Number of Coefficients',...
    'FirstColumn','Input','SecondColumn','ReconstructedSignal');
bfw = dsp.BinaryFileWriter('data.bin','HeaderStructure',header);
numIter = 1000;

M = L+1;
ReSignalAll = zeros(M*numIter,1);
InputAll = zeros(M*numIter,1);
rng(1);

for Iter = 1 : numIter
    Sinewave1 = Sineobject();
    Input = Sinewave1 + 0.01*randn(size(Sinewave1));
    [ReconstrSignal,numCoeffs] = GenerateSignalWithHighEnergyFFTCoeffs(Input);
    InputAll(((Iter-1)*M)+1:Iter*M) = [numCoeffs;Input];
    ReSignalAll(((Iter-1)*M)+1:Iter*M) = [numCoeffs;ReconstrSignal];
end

bfw([InputAll ReSignalAll]);   
release(bfw);

The next step in generating a C executable is to create a coder.config object for an executable and provide a main.c function to this object.

cfg =  coder.config('exe');
cfg.CustomSource = 'reconstructSignalTestbench_Main.c';

Here is how the reconstructSignalTestbench_Main.c function looks for this example.

/*
** reconstructSignalTestbench_main.c
*
* Copyright 2017 The MathWorks, Inc.
*/
#include <stdio.h>
#include <stdlib.h>

#include "reconstructSignalTestbench_initialize.h"
#include "reconstructSignalTestbench.h"
#include "reconstructSignalTestbench_terminate.h"

int main()
{
    reconstructSignalTestbench_initialize();
    reconstructSignalTestbench();    
    reconstructSignalTestbench_terminate();
    
    return 0;
}

For additional details on creating the main function, see Generating Standalone C/C++ Executables from MATLAB Code (MATLAB Coder).

Set the CustomInclude property of the configuration object to specify the location of the main file. In this example, the location is the current folder.

cfg.CustomInclude = ['"',pwd,'"'];

Generate the C executable by running the following command in the MATLAB command prompt:

codegen -config cfg -report reconstructSignalTestbench

MATLAB Coder compiles and links the main function with the C code that it generates from the reconstructSignalTestbench.m.

If you are using Windows, you can see that reconstructSignalTestbench.exe is generated in the current folder. If you are using Linux, the generated executable does not have the .exe extension.

Read and Verify the Binary File Data

Running the executable creates a binary file, data.bin, in the current directory and writes the input, reconstructed signal, and the number of FFT coefficients used to reconstruct the signal.

!reconstructSignalTestbench

You can read this data from the binary file using the dsp.BinaryFileReader object. To verify that the data is written correctly, read data from the binary file in MATLAB and compare the output with variables InputAll and ReSignalAll.

The header prototype must have a structure similar to the header structure written to the file. Read the data as two channels.

M = 1021;
numIter = 1000;
headerPro = struct('FirstElemInBothCols','Number of Coefficients',...
    'FirstColumn','Input','SecondColumn','ReconstructedSignal');
bfr = dsp.BinaryFileReader('data.bin','HeaderStructure',...
headerPro,'SamplesPerFrame',M*numIter,'NumChannels',2);
Data = bfr();

Compare the first channel with InputAll and the second channel with ReSignalAll.

isequal(InputAll,Data(:,1))

ans =

  logical

   1

isequal(ReSignalAll,Data(:,2))

ans =

  logical

   1

The results match exactly, indicating a successful write operation.

Relocate Code to Another Development Environment

Once you generate code from your MATLAB algorithm, you can relocate the code to another development environment, such as a system or an integrated development environment (IDE) that does not include MATLAB. You can package the files into a compressed file using the packNGo function at the command line or the Package option in the MATLAB Coder app. For an example that illustrates both the workflows, see Package Code for Other Development Environments (MATLAB Coder). For more information on the packNGo option, see packNGo in RTW.BuildInfo Methods (MATLAB Coder). You can relocate and unpack the compressed zip file using a standard zip utility. For an example on how to package the executable generated in this example, see Relocate Code Generated from MATLAB Code to Another Development Environment.