Can I write this in a more compact way?

Hi!

Is there a way to write to following statements (the for loop) in a more compact way?

The function

signalSources.generateSineWave

returns a

(ts*fs)x1 double

array. I tried using arrayfun(), but this returns a cell array.

signal = 0;
square_multiples = 1:2:9;

fSquare = @(x)1/x*signalSources.generateSineWave(pitch*x,ts,fs,pi);

for n = square_multiples
    signal = signal + fSquare(n);
end

Thank you very much

Timo

[TADA]

I may be missing something but

signal = sum(arrayfun(fSquare, square_multiples));

seems to do it

[Timo Kuchheuser]

Hi!

Thanks, but I'm afraid that doesn't work:

>> signal = sum(arrayfun(fSquare,square_multiples));
Error using arrayfun
Non-scalar in Uniform output, at index 1, output 1.
Set 'UniformOutput' to false.
 
>> signal = sum(arrayfun(fSquare,square_multiples,'UniformOutput',false));
Error using sum
Invalid data type. First argument must be numeric or logical.

[Bruno Luong]

The for-loop looks already very compact to me.

What is the point to put it at all cost in a single line?

[Timo Kuchheuser]

Here's the full code to test:

%% Signal parameters
pitch = 440;            % Tone pitch
fs = 44100;              % Samplerate in Hz
ts = 2;                 % Duration in seconds
phi = 0;                % phase offset

signal = 0;
square_multiples = 1:2:9;

fSquare = @(x)1/x*generateSineWave(pitch*x,ts,fs,phi);

for n = square_multiples
    signal = signal + fSquare(n);
end

function sine = generateSineWave(pitchHz,tSeconds,fs,phi)
%generateSineWave returns sinusoidal curve with given parameters
%   pitchHz     frequency in hertz
%   tSeconds    duration in seconds
%   fs          sampling rate in hertz
%   phi         phase offset

    l = tSeconds*fs;        % length
    n = 0:l-1;              % normalized time
    w = (2*pi/fs)*pitchHz;  % normalized angular frequency
    sine = sin((w*n)+phi)';  % sine wave with phase offset
end

[Timo Kuchheuser]

@Bruno Luong

Just pure curiosity!

[Guillaume]

For the record, the way to do it with arrayfun would be:

signal = sum(cell2mat(arrayfun(fsquare, square_multiples.', 'UniformOutput', false)), 1);

since fsquare returns a vector, you're forced to store the outputs into a cell array. The cell array is then converted into a 2D matrix (for that the input array to arrayfun must be a column vector, hence the .'), then the matrix is summed across the rows. Note that the conversion from cell array to matrix involves data copy that you don't have with the explicit loop, so it's likely that cellfun is going to be slower.

Bruno is correct, the proper way to speed up the code is to change the original function so that it creates that 2D matrix directly, without going through a cell array.

[Timo Kuchheuser]

Thank you!

Answer 1

[Bruno Luong]

Cellfun/arrayfun is just disguished for-loop. They are more compact in syntax, less flexible and often slower.

Then the proper way is to vectorize your function as well wrt pitchHz, that returns a 2d array of signal, with the second dimension corresponds to pitchHz then just sum it.

%% Signal parameters
pitch = 440;            % Tone pitch
fs = 44100;              % Samplerate in Hz
ts = 2;                 % Duration in seconds
phi = 0;                % phase offset

square_multiples = 1:2:9;
harmonic_signal = signalSources.generateSineWave(pitch*square_multiples,ts,fs,pi) ./ square_multiples;
signal = sum( harmonic_signal, 2);

function sine = generateSineWave(pitchHz,tSeconds,fs,phi)
%generateSineWave returns sinusoidal curve with given parameters
%   pitchHz     frequency in hertz
%   tSeconds    duration in seconds
%   fs          sampling rate in hertz
%   phi         phase offset
    l = tSeconds*fs;        % length
    n = 0:l-1;              % normalized time
    w = (2*pi/fs)*pitchHz(:);  % reshape in column
    sine = sin((w.*n)+phi)';  % auto-expansion, use bsxfun for R2016a or prior
end

[Timo Kuchheuser]

Thank you!