从技术上讲，这不是一个 MATLAB 式的论坛，但我可以为您更详细地解释这些步骤：假设您的输入信号是，它的 DFT 是。对于真实信号，您可以使用一侧的 DFT，因为当您查看其功率谱密度时，另一半将是多余的。（PSD）。$x[n]$$X(f)$

计算信号的 DFT 后，PSD 就是。也就是说，您需要将 DFT 结果的绝对幅度取平方。$|X(f)|^2$

您现在需要对 PSD 进行归一化，以便将其视为概率密度函数 (PDF)。因此，归一化的 PSD（让我们称之为）将简单地为：$PSD_n$

$$PSD_n(f) = \frac{PSD(f)}{\sum_{f=\frac{-fs}{2}}^{f = \frac{f_s}{2}} PSD(f)}$$

最后，您的谱熵将是：

$$E = -\sum_{f=\frac{-fs}{2}}^{f = \frac{f_s}{2}} PSD_n(f) log_2[PSD_n(f)]$$

我只是在这里

我的源代码：

    [x, Fs, nbits] = wavread('ederwander.wav'); 


    winSize = 2048;

    n_samples = length(x);


    %50% overlap or 0 to not use overlap
    OverlapStep = 50;

    if OverlapStep > 0

        Overlap = floor((OverlapStep*winSize) / 100); 
        nFrames=floor(n_samples/Overlap)-1; 
    else
        Overlap= winSize;
        nFrames=floor(n_samples/Overlap)-1;
    end

    Entropy = zeros(nFrames,1);

    k=1;
    inc=1;

    while ( (k+winSize-1) <= n_samples )

        FrameSignal = x(k:k+winSize-1);

        v = FrameSignal .* hann(length(FrameSignal));           

        N = length(v);

        Y=fft(v);

        % Compute the Power Spectrum
        sqrtPyy = ((sqrt(abs(Y).*abs(Y)) * 2 )/N);
        sqrtPyy = sqrtPyy(1:winSize/2);



       %Normalization
       d=sqrtPyy(:);
       d=d/sum(d+ 1e-12);

       %Entropy Calculation
       logd = log2(d + 1e-12);
       Entropy(inc) = -sum(d.*logd)/log2(length(d));


       k=k+Overlap;
       inc=inc+1;
end

此源代码从每个帧块进行光谱熵计算...