Conversion of dlmread() using MatLab Coder - Not supported for code generation to C

I have a compensation function that reads in a 2-column csv file. When I run this fuction in the MatLab Coder Generator to convert it to a standalone C function, it states that my dlmread() call, which reads in the csv file, is not supported. I saw that dlmread() should be replaced with readmatrix(), but that is not supported either.

How do I go about converting this call, dlmread() or readmatrix(), over to Matlab code that the MatLab Coder app will recognize? Below is my function:

function x = compensation(band, x, fc)

% Algorithm parameters
  fs = 4.5e9;             % Sample rate
  [Nf, K] = size(x);      % Number of SHAPE vectors in the file

  if strcmp(band, 'lb')
    comp_fn = 'low_band_comp.csv';
  elseif strcmp(band, 'hb')
    comp_fn = 'high_band_comp.csv';
  end

% Calculate the average power spectral density
  Xf = fftshift(fft(x, [], 1), 1);
  f = [0 : Nf-1]' * fs/Nf - fs/2;

% Read the compensation vector data
  F = dlmread(comp_fn);
  frq = F(:,1)*1e6;       % Convert MHz data to Hz
  mag = F(:,2);

% Translate compensation data in frequency
  if strcmp(band, 'lb')
    frq = frq - fc;         % Digital LO
  elseif strcmp(band, 'hb')
    frq = fc - frq;         % f_out = f_lo - f_bb   
    frq = flipud(frq);      % Digital LO 
  end                       % Put in ascending order for interpolation

% Normalize compensation data in magnitude
  mag = (10.^(mag/10));
  mag = mag/mean(mag);
  mag = 10*log10(mag);
  M = length(mag);

% Interpolate the compensation data
  mag_m = ones(Nf,1);
  k = 1;
  for m = 1 : Nf
    if f(m) >= frq(1) && f(m) < frq(M)
      if f(m) < frq(k+1) && f(m) >= frq(k)
        mag_m(m) = (mag(k+1) - mag(k))*(f(m) - frq(k))/(frq(k+1) - frq(k)) + mag(k);
      else
        mag_m(m) = (mag(k+1) - mag(k))*(f(m) - frq(k))/(frq(k+1) - frq(k)) + mag(k);
        k = k + 1;
      end
    end
  end
  figure(3);
  plot(frq/1e6, mag, 'b.-');
  hold on;
  plot(f/1e6, mag_m, 'r.-');
  hold off;
  title('Compensation Spectrum');
  xlabel('Frequency - MHz');
  ylabel('Magnitude - dB');
  grid on;

% Convert to magnitude scaling values
  mag_vec = 10.^(+mag_m/20);

% Apply the compensation vector to frequency domain data
  Xf = Xf .* (mag_vec * ones(1,K));

% Recalculate the SHAPE vector from frequency scaled input
  x = ifft(fftshift(Xf, 1), [], 1);

end

Answer 1

[Yash]

Hi @Scotty Mac

In MATLAB Coder, the dlmread and readmatrix functions are not supported because they are part of the MATLAB base workspace and are not compatible with code generation. To read CSV files in generated C code, you can use the fscanf function.

Here's how you can modify your compensation function to use fscanf instead of dlmread or readmatrix:

function x = compensation(band, x, fc)
    % Algorithm parameters
    fs = 4.5e9;             % Sample rate
    [Nf, K] = size(x);      % Number of SHAPE vectors in the file
    if strcmp(band, 'lb')
        comp_fn = 'low_band_comp.csv';
    elseif strcmp(band, 'hb')
        comp_fn = 'high_band_comp.csv';
    end
    
    % Calculate the average power spectral density
    Xf = fftshift(fft(x, [], 1), 1);
    f = [0 : Nf-1]' * fs/Nf - fs/2;
    
    % Open the CSV file for reading
    fid = fopen(comp_fn, 'r');
    if fid == -1
        error('Failed to open the compensation file.');
    end
    
    % Read the compensation vector data
    F = fscanf(fid, '%f,%f', [2, inf]);
    F = F';
    fclose(fid);
    
    frq = F(:, 1)*1e6;  % Convert MHz data to Hz
    mag = F(:, 2);
    
    % Translate compensation data in frequency
    if strcmp(band, 'lb')
        frq = frq - fc;         % Digital LO
    elseif strcmp(band, 'hb')
        frq = fc - frq;         % f_out = f_lo - f_bb   
        frq = flipud(frq);      % Digital LO 
    end
    
    % Put in ascending order for interpolation
    [frq, idx] = sort(frq);
    mag = mag(idx);
    
    % Normalize compensation data in magnitude
    mag = (10.^(mag/10));
    mag = mag/mean(mag);
    mag = 10*log10(mag);
    M = length(mag);
    
    % Interpolate the compensation data
    mag_m = interp1(frq, mag, f, 'linear', 'extrap');
    
    % Rest of your code...
    
    figure(3);
    plot(frq/1e6, mag, 'b.-');
    hold on;
    plot(f/1e6, mag_m, 'r.-');
    hold off;
    title('Compensation Spectrum');
    xlabel('Frequency - MHz');
    ylabel('Magnitude - dB');
    grid on;
    
    % Convert to magnitude scaling values
    mag_vec = 10.^(+mag_m/20);
    
    % Apply the compensation vector to frequency domain data
    Xf = Xf .* (mag_vec * ones(1,K));
    
    % Recalculate the SHAPE vector from frequency scaled input
    x = ifft(fftshift(Xf, 1), [], 1);
end

In the fscanf function, [2, inf] is used to specify the format and size of the data that you want to read from the file.

• The 2 in [2, inf] indicates that you expect to read two values from each line of the CSV file: one for the frequency and one for the magnitude.
• The inf in [2, inf] specifies that you want to read as many rows of data as are available in the file. In other words, it tells MATLAB to read until the end of the file is reached.

This code uses fopen to open the CSV file, fscanf to read its contents, and fclose to close the file when you're done. It also sorts the data by frequency to ensure that it's in ascending order for interpolation, as you did in your original code.

I hope this helps.

[Scotty Mac]

@Yash,

Thank you so much for the help. Yes, this helps alot. Much appreciated!