Need an example

[nicolas-f]

An example of generating a signal, running fft and retrieve parameters would be welcome.

I try to make a unit test, but I need help to retrieve the expected value.

Would expect to find such help http://fr.mathworks.com/help/matlab/ref/fft.html

        // Make 50 Hz signal
        final int sampleRate = 44100;
        final int signalFrequency = 50;
        float[] signal = new float[sampleRate];
        for(int s = 0; s < sampleRate; s++) {
            double t = s * (1 / (double)sampleRate);
            signal[s] = (float)Math.sin(2 * Math.PI * signalFrequency  * t);
        }


        // Execute FFT
        FloatFFT_1D floatFFT_1D = new FloatFFT_1D(sampleRate);
        floatFFT_1D.realForward(signal);

        // Recover expected 50 Hz frequency with value 1
       ...

thank you

[nicolas-f]

Ok, found it.

        // Make 50 Hz signal
        final int sampleRate = 44100;
        final int signalFrequency = 50;
        float[] signal = new float[sampleRate];
        for(int s = 0; s < sampleRate; s++) {
            double t = s * (1 / (double)sampleRate);
            signal[s] = (float)Math.sin(2 * Math.PI * signalFrequency * t);
        }

        // Execute FFT
        FloatFFT_1D floatFFT_1D = new FloatFFT_1D(sampleRate);
        floatFFT_1D.realForward(signal);

        // Extract Real part
        float localMax = Float.MIN_VALUE;
        int maxValueFreq = -1;
        float[] result = new float[signal.length / 2];
        for(int s = 0; s < result.length; s++) {
            //result[s] = Math.abs(signal[2*s]);
            float re = signal[s * 2];
            float im = signal[s * 2 + 1];
            result[s] = (float) Math.sqrt(re * re + im * im) / result.length;
            if(result[s] > localMax) {
                maxValueFreq = s;
            }
            localMax = Math.max(localMax, result[s]);
        }

        // 50 Hz signal is graduated to 1
        assertEquals(1, localMax, 1e-12);
        // Expected 1 value on 50 Hz band
        assertEquals(signalFrequency, maxValueFreq);

Some documentation are still welcome. But I close this issue.

[brunoccc]

Thanks for the example, it helped me a lot.
On a side node, I think that the code above does not deal properly with the fact that results[1] isn't the imaginary part of the first item, but actually the real part of the item at index n/2, as explained in the documentation:

 a[2*k] = Re[k], 0<=k<n/2 
 a[2*k+1] = Im[k], 0<k<n/2 
 a[1] = Re[n/2]

Of course, the code works well because the example signal has just a frequency component at 50Hz, but if you plan to adapt the code above to real world applications, you may need some extra step.

[UQuark]

Could you please briefly explain me how does it store values?
a[1] is real part of the item n/2 but where does it store imaginary part of the first item?

[wendykierp]

The imaginary part of the first item is always 0, so there is no need to store it.