Functions to locate and measure the positive peaks and valleys in noisy data sets.
A set of fast customizable functions for locating and measuring the peaks in noisy timeseries signals. Adjustable parameters allow discrimination of "real" signal peaks from noise and background. Determines the position, height, and width of each peak by leastsquares curvefitting. It can find and count over 10,000 peaks per second, and find and measure 1800 peaks per second, in very large signals. Includes two interactive versions, one with mousecontrolled sliders and one with keyboard control, for adjusting the peak finding criteria in realtime. Selfcontained demos show how it works. See http://terpconnect.umd.edu/~toh/spectrum/PeakFindingandMeasurement.htm for details.
4.0  Added several new functions including measurepeaks, autofindpeaks, and autopeaks plus test scripts for each. 

3.3  Bug fix 

3.2  Included version 7.7 of iPeak. 

3.1  Bug fix. Data vectors x and y can be either column or row vectors. 

3.0  Bug fixes and speed improvements. 

2.0  This version contains several small corrections, additions, and bug fixes. 

1.19  This version includes updated versions of several functions. 

1.18  This version includes updated versions of findpeaksfit.m and findpeaksb.m. 

1.17  This version includes findpeaksL.m for Lorentzian peaks, and findpeaksGSS.m and findpeaksLSS that compute the 1% start and end positions. 

1.16  This version includes new commandline functions findpeaksplot.m, findpeaksnr.m, and peakstats.m. plus version 5.3 of iPeak. 

1.14  Version 5 has faster jumps between peaks with spacebar and tab keys; other small bug fixes. 

1.13  This version includes some additional demos plus version 4 of the interactive iPeak function. 

1.11  Typo correction. 

1.10  This zip file includes version 3.9 of the interactive iPeak function. 

1.8  This version includes the findvalleys function and version 3.81 of the interactive iPeak function. 

1.5  Bug fix 

1.4  Includes the version 3.2 update of iPeak and version 4 of findpeaks.m, as well as the older 2008 version with sliders. 

1.3  The rtslid.m function in this version has been


1.1  Fixes a bug in the FindPeakSlidersG script that caused a malfunction in the individual peak viewing mode when the xaxis data series did not start at zero. 

enhancements 

Change to file details 

Change to file details (title, description, requirements, ...) 

Change to file details (title, description, requirements, ...),


Enhancements: Version 1.2 is a faster version of the original findpeaks function. 

Bug fixes and enhancements (Version 1.1) 

Add or update the screenshot


replace missing function in ZIP file 

Add or update the screenshot 
Inspired: IDpeak: Peak Identifier Function, Demo functions for peak detection and fitting.
ZHENG WANG (view profile)
Very good code. Saved me!
Appreciate!
Tom O'Haver (view profile)
Black Hawk, the line number you mentioned don't match up with the current version 5, so you probably have an older version.
One restriction of the current version is that x and y must be column vectors (size 1 x n) not row vectors. I'll fix that in the next version.
You can always find the latest versions of all my functions and scripts at https://terpconnect.umd.edu/~toh/spectrum/functions.html
Tom O'Haver (view profile)
Black Hawk, sorry I didn't see this earlier; evidently I am no longer getting email notifications and I don't go on the Matlab site very often. I'll check the errors you got. Make sure you have the latest version  some debugging has already been done. Do the demos in the internal help file not work?
Black Hawk (view profile)
It appears like findpeaksb and findpeaksb3 have never been tested or debugged. That is extremely disappointing. There are just too many errors, and I tried debugging findpeaksb3 without success. Here are some basic examples for findpeaksb3:
 Line 164: why do you put "MeasuredError(peak)=Error" if Error is a 1x2 row vector? Or was it done on purpose..? That obviously gives the error message "In an assignment A(I) = B, the number of elements in B and I must be the same." So the line has to be modified to something like "MeasuredError(peak)=Error(1)", or "=Error(2)" depending on what you want.
 Lines 121, 236 and 151: why do you put "signal=[XXX',YYY']"? This makes a row vector of the independent variable concatenated with the dependent variable. Why didn't you simply choose "signal=[XXX,YYY]" so that signal is automatically a twocolumn matrix, 1st column=independent variable and 2nd column=dependent variable. That would also avoid the huge confusion when calling the peakfit function in which line 382 "if datasize(1)<datasize(2),signal=signal';end
" is not very useful...
 The cases when length(signal)=0 or 1 are not considered at all, yielding errors when these cases are encountered
Sorry for being very direct but I think the scripts should have been carefully tested and debugged prior to submission. It is very frustrating to spend time debugging from A to Z without success a code that one hasn't written.
arnold (view profile)
I posted the problem with val2ind and r2015b with your iPeak contribution but I guess it applies here as well.
Error: File: findpeaksb.m Line: 2168 Column: 29
Function with duplicate name "val2ind" cannot be defined.
there is a function like this in one of the matlab toolboxes and I think this might call this problem?
C:\Program Files\MATLAB\R2015b\toolbox\phased\phased
Tom O'Haver (view profile)
Oops, I meant to say that the area/width ratio is 1.06447...
Tom O'Haver (view profile)
There is a typo there on my part; the number should be 1.06447.... Here's why: The simplest expression for a Gaussian peak is y=exp(x^2). This has a fullwidth at half maximum of 2 * sqrt(ln(2)=1.66561 and the area under the curve is sqrt(pi)= 1.77245. So the width/height ratio is 1.06447.... The peak area is proportional to both peak height and width. Therefore the area is 1.06447 * peak height * peak width. Hint: You can use Wolfram Alpha (http://www.wolframalpha.com) to do all the algebra for stuff like that.
CC (view profile)
Dear Tom,
Sorry about this really basic question, but could you please explain to me the reasoning behind calculating the area of the peak using 1.0646 * peak height * peak width? I'm not sure where the 1.0646 value comes from.
Many thanks
Rahul Mohan (view profile)
Tom O'Haver (view profile)
Iman, you can use File > Import Data to import data files in various format, including three sound formats (.au, .snd, and .wav). These is also a Data Acquisition Toolbox that I would assume has additional data import capabilities.
iman (view profile)
How can i import the signal from the file? such as sound file or RF signal
Tom O'Haver (view profile)
Dan,
As you requested, I have created variants of findpeaks that compute the peak start and end positions. See
http://terpconnect.umd.edu/~toh/spectrum/PeakFindingandMeasurement.htm#PeakStartAndEnd
I'll include these in the next update of fileexchange entry.
Tom
Dan (view profile)
Tom, it works very well for me. Thank you!
The start and the end position of the peaks would also help me… the definition of 1% of the peak amplitude would be great.
Tom O'Haver (view profile)
Dieter, I would be glad to add that if you could define quantitatively what you mean by peak start and end  for example, when the amplitude of the signal reaches 1% of the peak amplitude. Or would some other definition be more appropriate for your work?
Dieter K. (view profile)
It is really good and easy to use. Only one thing I miss: The start and end position of the peaks would be really helpful.
Tom O'Haver (view profile)
Andre, iPeak reports only the position, height, and width of peak parameterized as Gaussians, not the parabolic results of polyfit. But you can generate the Gaussian model from height.*exp(((xposition)./(0.6005615.*width)) .^2) and plot it over the raw data. iPeak does this automatically for the peaks detected in the upper window.
Andre (view profile)
This works really well for me, thanks!
One quick question: I would like to overlay the fitted parabola onto the raw data that I used findpeaks on, for visualization purposes (and to show that the fits are reasonable).
I tried going into the function and doing polyval on the coefs that are returned by polyfit, but this doesn't produce anything sensible.
I would like the drawn parabola to have the same units as the original data. Perhaps I could recreate the parabola from the output of findpeaks (ie the width, height, mean, etc), but it seems that polyval and other parabola drawers parameterize the parabolas in other ways.
Any ideas on which way to go??? Thanks!
Andre
mitra devkota (view profile)
Dear MATLAB users, I have x and y as vectors, I plotted them and now I need to detect the peaks and also find the corresponding peak locations. How can I do it?
Thanks in advance,
Mitra
zaidan (view profile)
great code for beginner
mohammad (view profile)
thanks a lot Tom
this code is so useful
Tom O'Haver (view profile)
When looking at a signal for the first time, it is best to use iPeak.m, iSignal.m, or ipf.m to inspect the data interactively. All of these can be downloaded from the Matlab File Exchange or from http://terpconnect.umd.edu/~toh/spectrum/SignalProcessingTools.html
Use the "pan and zoom" capabilities of these programs to inspect the whole signal closely. If the signal has important features that cover a very wide amplitude range or are superimposed on a much larger baseline or background, then those features may not be visible when the entire signal is plotted normally. Try using a semilog(x,y) plot, or use iSignal's semilog mode (H key).
mohammad (view profile)
i have a .xls file, i saved data of that in a matrix like : r=xlsread(.xls)
now its needed to find peaks but this discrete data (r) is noisy data and using of 'findpeaks' command give me a lot of numbers of peaks (because of nois).
i think your code can help me. so, could you tell me how can i use this?
best regard
Tom O'Haver (view profile)
I am not exactly sure what you mean by "matrix (discrete) wave" .Please explain.
mohammad (view profile)
dear how can i use this for a matrix (discrete) wave?
Tom O'Haver (view profile)
See http://www.mathworks.com/matlabcentral/fileexchange/23850 for a version that includes iterative curve fitting to selected peak shapes.
Tom O'Haver (view profile)
See http://www.mathworks.com/matlabcentral/fileexchange/23850ipeakinteractivepeakdetectorversion12 for a version that uses keyboard control and that works in recent versions of Matlab.
Andre Guy Tranquille (view profile)
ZULIN (view profile)
Michael Coen (view profile)
And comment out line 372 of rtslid.m as well... :)
Michael Coen (view profile)
Ah, commenting out line 337 of rtslid.m:
"figure(params.fig);"
eliminates crashes for later versions of Matlab...
Michael Coen (view profile)
Very nifty idea but it crashes quickly in Matlab 2008b, 2009a, and 2009b. There are clearly some GUI incompatibilities.
Tom O'Haver (view profile)
Latest version just uploaded corrects a bug in the FindPeakSlidersG script that caused a malfunction when individual peaks are viewed if the xaxis data series does not start with zero.
Just what I needed! Thank you very much!
fast and reliable tool ... thanks for sharing!
Very useful interactive software! Thanks for sharing this.
Please gat back to me with your private phone number.
Regards,
Dr. Hugo.
This is a good algorithm but can it be changed to give peeks over a certain datum instead of the trough values.
Your algorithm is excellent.But I find a problem in it.
Your algorithm is limited only when the data are positive number.If all the data are negative and there is peak points.Your algorithm can't detect the peak points.
But I can translate the coordinate axis.
It is to make all the y data into positive number. Then using your algorithm we can find the peak points.
I think you can modify your algorithm and make it adapt to the situation that the data is all negative number.
If you can modify your algorithm,please send the message to my email.(lygwxr@yahoo.com.cn)
Thank your.
A student from BIT of China
You saved my MEng project with this one. I credited you thogh, don't worry.
Many thanks Tom,
Matt.
Perfect for scientific data processing. Good performance without being overly complicated. You just saved me a week of my time.
Would be greate if it will do deconvolution and gave the areas for deconvolved peaks.
In general  great tool!
K. Tampa is correct; extreme.m is the one to use if you want to find ALL the minima and maxima. But "Peak finding and measurement" is intended to distinguish real peaks from noise in experimentally measured signals and to measure their height, position, and width. Two completely different purposes.
Get EXTREME.M in this site and you'll find the ALL peaks (minima and maxima!) of
rand(10^6,1) in less than a second, 10^7 in about 10 sec, and from matrix too, so...
Version 1.2, dated August 24, 2006, is even faster than the previous version It can find and measure 1000 peaks in a 1,000,000 point signal in 13 seconds.
It is pretty good and easy to use. Yet some commercial software,such as origin, can do faster.