希尔伯特频谱算法Hilbert-Huang spectral analysis（matlab代码）

前段时间磕盐接触到了希尔伯特频谱，它是一种信号分解方法，1998年提出来的，主旨是把复杂信号分解为简单信号的加权和，就像傅里叶变换小波变换一样，但是他和傅里叶变换等方法的区别是他是纯粹时间域的分解，但是每个子信号却可以表示不同的频率成分，于是可以得到像小波变换那样的时频平面，但是这个方法明显比小波分解冷门的多，而且在我的实验结果里确实远远远远弱于小波分解，不过也算是自己辛苦几天看论文和写代码的成果，特此记录，如果后面有人用到这个，可以快速入门

文献链接：The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis

伪代码，我读论文总结出来的，主要就是经验模式分解EMD和希尔伯特变换两步

伪代码的latex 代码

\begin{algorithm}
	\caption{Hilbert-Huang spectral analysis \cite{Branch1998}}
	\label{hhsa}
	\begin{algorithmic}[1] 
		\Require The original signal vector $\boldsymbol x$.
		\Ensure  The Hilbert-Huang Spectrum, i.e. an energy-time-frequency distribution of $\boldsymbol x$. 
		\Function{EMD}{$\boldsymbol x, SegLen, ResidueThreshold, SD_T$}		
		\State $\boldsymbol{IMF} \gets \boldsymbol 0$
		\State $i \gets 0$
		\State $ N \gets length(\boldsymbol{x}) / SegLen$
		\State $residue \gets \infty$
		\While {$residue > ResidueThreshold$}
		\State $i \gets i+1$	
		\State $\boldsymbol{x_i} \gets \boldsymbol{x}-\sum_i \boldsymbol{IMF}$
		\State $SD \gets \infty$
		\While{$SD > SD_T$}	
		\For {$j=1 \to j=N$}	
		\State $\boldsymbol{seg_{ij}} \gets \boldsymbol{x_i}[(1+(j-1)*SegLen) : (j*SegLen)]$
		\State $[LocalMax_{ij}, IndMax_{ij}]\gets max(\boldsymbol{seg_{ij}})$
		\State $IndMax_{ij}\gets IndMax_{ij}+(j-1)*SegLen$
		\State $[LocalMin_{ij}, IndMin_{ij}]\gets min(\boldsymbol{seg_{ij}})$
		\State $IndMin_{ij}\gets IndMin_{ij}+(j-1)*SegLen$
		\EndFor
		\State $\boldsymbol{UpperEnv_i} \gets spline(\boldsymbol{IndMax_i},\boldsymbol{LocalMax_i},1:length(\boldsymbol{x_i}))$
		\State $\boldsymbol{LowerEnv_i} \gets spline(\boldsymbol{IndMin_i},\boldsymbol{LocalMin_i},1:length(\boldsymbol{x_i}))$
		\State $\boldsymbol{LocalMeanApprox_i} \gets (\boldsymbol{UpperEnv_i} + \boldsymbol{LowerEnv_i})/2$
		\State $\boldsymbol {x_i} \gets \boldsymbol {x_i} - \boldsymbol{LocalMeanApprox_i} $
		
		\State $SD\gets \sum[\frac{(\boldsymbol{xTemp}-\boldsymbol{x_i})^2}{\boldsymbol{xTemp}^2}]$
		\EndWhile
		\State $\boldsymbol{IMF_i} \gets \boldsymbol x_i$
		\State $residue \gets mean(\boldsymbol x-\sum_i \boldsymbol{IMF})$
		\EndWhile
		\State \Return{$\boldsymbol{IMF}$}
		\EndFunction
		\State
		\Function{HT}{$\boldsymbol{IMF},fs$}
		\State $\boldsymbol{zt} \gets hilbert(\boldsymbol{IMF})$
		\State $\boldsymbol{Energy} \gets (imag(\boldsymbol{zt}))^2+(real(\boldsymbol{zt}))^2$
		\State $\boldsymbol{Phase} \gets arctan\frac{imag(\boldsymbol{zt})}{real(\boldsymbol{zt})+eps}$
		\State $\boldsymbol{frequency} \gets \boldsymbol{Phase}/(2*\pi*fs)$
		\State \Return{$\boldsymbol{Energy,frequency}$} 
		\EndFunction
	\end{algorithmic}
\end{algorithm}

导言区需要包含的包

\usepackage{algorithm}
\usepackage{algorithmicx}
\usepackage{algpseudocode}  
\usepackage{amsmath}  % 用于算法伪代码中的数学公式
\renewcommand{\algorithmicrequire}{ \textbf{Input:}}     %Use Input in the format of Algorithm
\renewcommand{\algorithmicensure}{ \textbf{Output:}}    %UseOutput in the format of Algorithm

matlab代码后面加，被新冠病毒阻挡了去学校的脚步，代码在教研室电脑上，后面补上

结果图

代码：
这个代码计算了EEG信号的HHS，不能直接用，但是大致实现可以参考，写的很烂，，，大神随意看看就好

% EMD.m
% HHS hilbert huang spectrum  emd
function IMF = EMD(eeg)
IMF = zeros(size(eeg,1), 9);
for j=1:size(eeg,1)
    ch = eeg(j,:);
    x = ch;
    
    IMF = zeros(100, length(x)); % 每条IMF都和x等长
    seg_len = 4;
    iter_num = length(x)/seg_len;
    local_max = zeros(iter_num,1);local_min = zeros(iter_num,1);
    indmax = zeros(iter_num,1);indmin = zeros(iter_num,1);
    for s=1:100
        res = x - sum(IMF); % IMF第一行是IMF0,初始化为0,res残差
        mean(res)
        if(abs(mean(res))<.017) break; end
        xs = res;
        %                  figure,plot(1:200,res(1:200)),grid on;title('res')
        
        for q=1:100
            for i=1:iter_num
                seg = xs((1+(i-1)*seg_len):(i*seg_len)); % 对x分段
                [local_max(i),indma] = max(seg);[local_min(i),indmi] = min(seg); % 局部极值点
                indmax(i) = indma + (i-1)*seg_len; indmin(i) = indmi + (i-1)*seg_len;% 索引
            end
            % 插值计算上下包络
            up = spline(indmax,local_max,1:length(x));
            lo = spline(indmin,local_min,1:length(x));
            up1 = interp1(indmax,local_max,1:length(x),'pchip');
            lo1 = interp1(indmin,local_min,1:length(x),'pchip');
            %                     up2 = interp1(indmax,local_max,1:length(x),'nearest');
            %                     lo2 = interp1(indmin,local_min,1:length(x),'nearest');
            %                     up3 = interp1(indmax,local_max,1:length(x),'linear');
            %                     lo3 = interp1(indmin,local_min,1:length(x),'linear');
            %                     up4 = interp1(indmax,local_max,1:length(x),'nearest');
            %                     lo4 = interp1(indmin,local_min,1:length(x),'nearest');
            %                     for r=1:length(x)
            %                         if(up(r)<lo(r)) up(r)=lo(r); end
            %                     end
            %                     figure,
            %                     plot(1:200,lo(1:200),'b',1:200,xs(1:200),'k',1:200,up(1:200),'r')
            %                     set(gca,'XLim',[1 200]);
            
            %                     figure,
            %                     plot(1:200,lo1(1:200),'b',1:200,xs(1:200),'k',1:200,up1(1:200),'r')
            %                     figure,
            %                     plot(1:200,lo2(1:200),'b',1:200,xs(1:200),'k',1:200,up2(1:200),'r')
            %                     figure,
            %                     plot(1:200,lo3(1:200),'b',1:200,xs(1:200),'k',1:200,up3(1:200),'r')
            %                     figure,
            %                     plot(1:200,lo4(1:200),'b',1:200,xs(1:200),'k',1:200,up4(1:200),'r')
            
            %                     set(gca,'XLim',[1 200]);
            xtemp = xs;
            xs = xs - (up1+lo1)/2;
            SD = mean(((xtemp-xs).^2)./((xtemp).^2));
            if(SD<.3) break; end % 不满足则继续sifting，最终的x就是IMF % 满足则不再sifting
            
        end
        %
        IMF(s,:) = xs;
        figure,plot(1:200,xs(1:200),'k'),grid on;title('IMF')
        %                 res = x - sum(IMF);
        
        %                 set(gca,'GridLineStyle',':','GridColor','k','GridAlpha',.5);%虚线网格
        %                 if(j<=3)
        %                 subplot(3,1,j)
        %                 plot(1:256,xs(1:256),'k',1:256,up(1:256),'r',1:256,lo(1:256),'b')
        %                 axis([1 256 -2 2])
    end
    % 去除多余的行
    for i=1:100
        if(sum(IMF(i,:))==0)
            IMF(i:end,:) = [];
            break;
        end
    end
end