Finding a maximum signal variation in MATLAB

6 次查看（过去 30 天）

Jimmy 2015-8-5

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/232495-finding-a-maximum-signal-variation-in-matlab

评论： Image Analyst 2016-5-22

Walk.mat

I have a signal which consist of one 1000000x1 array (data) and one 1000000x1 array (time). I'm having problem to detect the location where the signal have the largest variation with respect to the value of the array.

For example in this signal, I want to locate the variation of the signal from time 4 until 11. I found out that findpeaks command will detect all peaks but I want to detect only the highest peak(highest value from the array) and the data (eg: 5000 sampling) to the right and left from the highest peak.

Thank you in advance.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

采纳的回答

Star Strider 2015-8-5

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/232495-finding-a-maximum-signal-variation-in-matlab#answer_188353

If you are only finding one peak each time, I would use the max and min functions, each with two outputs, to get the absolute maximum and minimum. The second output is the index location of the maximum or minimum.

You can also use the Signal Processing Toolbox findpeaks function on the original signal to get the value and location of the highest peak, the use findpeaks again on the negative of the original signal to get that highest value and location.

39 个评论
显示 37更早的评论隐藏 37更早的评论

Star Strider 2015-8-6

在 MATLAB Online 中打开

Actually what you’re currently doing works, providing you use the hold function. I’m not sure what final result you want, because you haven’t described it. We’re doing this a bit at a time.

To combine all of them, do a simple horizontal concatenation:

time_vct = [left2_time  NaN  right2_time];
data_vct = [left2  NaN  right2];

The NaN is there to provide discontinuity, since I get the impression that you want to plot these separately.

If you want the entire sequence displayed at once, this works:

win_idx = (max_idx-2):(max_idx+2);
plot(time(win_idx), data(win_idx))

The entire code (in summary for continuity) now becomes:

data = [3 5 1 7 6 2 1];
time = [0 1 2 3 4 5 6];
[data_max, max_idx] = max(data);
time_at_max = time(max_idx);
left2  = data(max_idx-[2:-1:1]);
right2 = data(max_idx+[1:2]);
left2_time = time(max_idx-[2:-1:1]);
right2_time = time(max_idx+[1:2]);
time_vct = [left2_time  NaN  right2_time];
data_vct = [left2  NaN  right2];
figure(1)
plot(time_vct, data_vct, 'LineWidth',2)
grid
win_idx = (max_idx-2):(max_idx+2);
figure(2)
plot(time(win_idx), data(win_idx))
grid

Star Strider 2015-12-6

I would have done bandpass filtering to eliminate the d-c offset and any low-frequency baseline variations, as well as high-frequency noise. A Butterworth design would likely work for your data.

As I mentioned, I don’t have much recent experience with PCA, and without your data it would be difficult to give you a specific answer as to how PCA works with it.

You mentioned that your PSD records are column vectors (together comprising ‘X’), with your original data being time-domain data, and the data you’re presenting to the PCA being frequency-domain data, again as a column vector for each study subject. As I read the princomp documentation, you might have to transpose ‘X’ to use princomp correctly. The documentation states that the rows are observations (here, subjects either running or walking), and columns are variables (here, frequencies).

As I understand your study, your goal is to classify the FFT (or PSD) of each subject as either running or walking. The PCA would be appropriate for this, but you have to be certain you present your data correctly to the PCA, or your results will not be optimal.

It may be best for you to submit this as a new Question, and include the relevant parts of your code and at least a representative sample — if not all — of your data.

Save your data to a .mat file so we can simply load it and don’t have to write code to import and then parse a text or other file.

Jimmy 2016-5-22

在 MATLAB Online 中打开

Thank you for the reply. Regarding my signal processing technique, I am required to perform Haar wavelet transform. I'm having difficulties to use the scaling function and keep getting errors as I don't know what is the best scaling function to use for 1000000 data. Attached below is the coding for haar wavelet which I found from several references.

%assuming InpSig = data (which consists of 1000000x1 array)
Fs = 50000;
N = length(InpSig);                         %number of data points
freqRes = Fs/N;                             %frequency resolution
freqAxisLab = freqRes*[0:N-1];              %calculate the labels for the frequency axis
fc = centfrq('haar',1);                     %scaling function for haar is calculated based on central frequency and level of iteration
freqrange = [1 freqAxisLab];
scalerange = fc./(freqrange*(1/Fs)); 
scales = scalerange(end):0.05:scalerange(1);%scalerange is the initial value:steps of the scale:end value of the scale
inpSigWAV = cwt(InpSig,scales,'haar','plot');      %CWT calculation
cwt(InpSig, scales,'haar','plot');
figure
plot(freqAxisLab, inpSigWAV,'b')