Automatic indexing of individual cycles in continuous sinusoidal data

Share icon

Mon, 05/14/2018 - 01:37 pm

I am revisiting a problem I had to shelve last July. Our lab has an instrument that acquires sinusoidal position data from an actuator at a rate of 50 samples/s, with an oscillation frequency of 1 Hz. I would like to index the minima on either side of each cycle (hundreds of thousands) by row location, essentially giving me the "start" and "stop" rows of the cycle. This is time-prohibitive to do manually using GetInfo cursors, but it's how I've been doing it. The intent is to use those start and stop row values to generate graphical analysis of an arbitrary cycle or subset of cycles out of the whole 100k+ set.

To complicate matters, with much staring at previous data sets, I am aware there is a ca. 11.5 microsecond/cycle error in the position signal coming from our instrument (permanent firmware issues), so the start and stop times of the nth cycle are not where you'd expect them to be based on those of the first cycle. I need to create a loop routine that accounts for this indeterminacy, and I have come up with the following strategy, but am having trouble coding it:

1) Make wave to store (Cycle#, StartRow, StopRow)
2) Set dimension labels according to step 1
3) Define data source wave
4) Define input range variables to use in WaveStats as [Start, Stop], these will need to be incremented by 50 rows after each loop iteration
5) Loop:
-Find V_minRowLoc between first and second cycles using WaveStats, with initial range values determined by manually placed cursors, call this point P
-Assign P as first StopRow value
-First StartRow value = (First StopRow - 50)
-Increment WaveStats range values to encompass next minimum, using (P-25) for Start and (P+25) for Stop

This seems like it might be overly complicated, and perhaps a more elegant solution exists. I do know I eventually require the precision of WaveStats, rather than FindPeak. Below is what I have so far, based on a gracious response from Andy Hegedus when I attempted to tackle this problem last time:

#pragma TextEncoding = "UTF-8"
#pragma rtGlobals=3     // Use modern global access method and strict wave access.

Function AutoIndex()
 
    //Create Wave to store results
    Make/O /N=(0,3) Limits
 
    setdimlabel 1,0,Cycle,Limits
    setdimlabel 1,1,StartRow,Limits
    setdimlabel 1,2,StopRow,Limits
 
    //Define sources of waves
    Wave Position
 
    //define the range of inputs
    SVAR CycleList
    Variable cycle, maxcycle
    Variable start, stop
    maxcycle = itemsinlist(CycleList)// allow for indeterminent number of cycles
 
    For(cycle=0;cycle<maxcycle;cycle+=1)
        WaveStats/R=[start,stop] Position
        Limits[cycle][%Cycle]= {cycle+1}
        Limits[cycle][%StopRow] = {V_minRowLoc}
        Limits[cycle][%StartRow] = {(StopRow-50)}
        //more code required

    Endfor
end
If the signal is sinusoidal, at least locally, maybe you could fit a sine (perhaps sequentially over suitable subranges centered about the peak) to get the start/end of cycle positions? Each fit should give you good starting guesses for coefficients for the next.

May 14, 2018 at 05:28 pm - Permalink

Another scheme if the data has fairly good precision (i.e. not very much noise).

Take a derivative and then Findlevels /edge=2 derivativeWave 0

The resulting `W_FindLevels (or your own wave name if you use the /Dest flag) will give you all the maxima positions.

Andy

May 14, 2018 at 05:54 pm - Permalink

Thank you all for your helpful comments. I have revised my procedure as follows, and it at least compiles without any errors:

#pragma TextEncoding = "UTF-8"
#pragma rtGlobals=3     // Use modern global access method and strict wave access.

Function IndexCycles()
 
    //Create Wave to store results
    Make/O /N=(0,3) Limits
 
    setdimlabel 1,0,Cycle,Limits
    setdimlabel 1,1,StartRow,Limits
    setdimlabel 1,2,StopRow,Limits
 
    //Define sources of waves
    Wave Position
 
    //define the range of inputs
    Variable cycle
    Variable start, stop
    Variable a=pcsr(A),b=pcsr(B)
    Variable V_minRowLoc
 
    For(cycle=0;cycle<10;cycle+=1)//max cycle count has to be entered manually
        If(cycle>0)
            a=V_minRowLoc+35
            b=V_minRowLoc+65
        endIf
            WaveStats/R=[start,stop] Position
            Limits[cycle][%Cycle]= {cycle+1} //make it 1 based
            Limits[cycle][%StopRow] = {V_minRowLoc} // include all increase more than one
            Limits[cycle][%StartRow] = {V_minRowLoc-50} // include all increase more than one
           
       
    Endfor
    Edit Limits
end


I'm sure I'll show my newbie colors with this question, but how do I execute the .ipf I've just complied from the command window, without copying and pasting all the code?

Per Thomas Braun's request, I have attached a sample experiment with 10 cycles.
Sinusoidal_data_10cycles.pxp

May 15, 2018 at 12:09 pm - Permalink

Here is what I did with your problem:

1. Calculate a derivative: Differentiate/METH=1 Position/X=t_sec/D=Position_DIF//History label from menu item differentiate

2. Find where it crosses zero findLevels/P /Edge=2 Position_dif 0 //the edge flag says only consider crossings that are decreasing (maxima)

Igor replies: V_Flag= 1; V_LevelsFound= 10; I use point scaling so now you have a wave with 10 values on the maxima.

If you want the were it crosses zero as the marker in the cycle, you can take the second derivative and do the find levels at 0.

In the picture I created some additional waves to hold the Y position.

Andy
Example Method.png

May 15, 2018 at 01:01 pm - Permalink

You can get all the indices where a certain threshold value is crossed. Using some heuristics one can write something like:
make/o/n=881 w_max = (position[p] > 1197) ? p : nan

Now, w_max stores all indices where the sin wave is greater than the heuristic threshold value, 1197.

If you want you can zapnans the wave like so:
wavetransform/o zapnans w_max

Now, w_max will store only the indices without the nan.

Since the number of points of position and t_sec are the same, you can write the time at which these extrema occur in another wave, like so:
make/o/n=881 w_max_t = (position[p] > 1197) ? t_sec[p] : nan; wavetransform/o zapnans w_max_t;

And there really are many variations on the theme.

best,
_sk

May 16, 2018 at 02:37 am - Permalink

Thank you all for the input. Andy, I tried your method of differentiating and then using FindLevels on my most recent data set, which comprises 1,000,000 cycles. I Set /EDGE=1 so I could obtain the minima zero crossings. Unexpectedly, the operation returned V_LevelsFound = 1,000,010. I decided to check this by performing FindLevels on the position data (not the derivative) with /EDGE=2 and a level equal to half the amplitude and obtained V_LevelsFound = 1,000,000 as expected. I'm not sure why there is a discrepancy there.

The following updated function works well for #cycles up to several hundred thousand, but analysis of the derived point values for StartRow and StopRow show the data appear to be off by 2 whole cycles at the end of the execution: I counted backwards from the last cycle visually, then placed cursors according the StartRow and StopRow for that cycle. I added the move cursor commands so I could watch the cursors step through the entire data set as sort of a pseudo progress bar, to make sure they weren't "hopping over" a cycle by mistake, and they are always somewhere around half amplitude for each cycle step.

Two additional observations:
1) Often in the latter cycles, the difference between StartRow and StopRow is not 50, as is supposed to be according to my code. What could be causing that discrepancy? I thought maybe it was due to the type of numerical data being output, which according to WaveType is 64-bit floating point, but I am stumped.
2) A visual inspection of the graph at latter cycles shows V_minRowLoc is incorrect, and off by a couple points. How is WaveStats missing this?

#pragma TextEncoding = "UTF-8"
#pragma rtGlobals=3     // Use modern global access method and strict wave access.

Function IndexCycles()
 
    //Create Wave to store results
    Make/O /N=(0,5) Limits
 
    setdimlabel 1,0,Cycle,Limits
    setdimlabel 1,1,StartRow,Limits
    setdimlabel 1,2,StopRow,Limits
    setdimlabel 1,3,CsrAPos,Limits
    setdimlabel 1,4,CsrBPos,Limits
 
    //Define sources of waves
    Wave Pos
 
    //define the range of inputs
    Variable cycle
    Variable start=pcsr(A),stop=pcsr(B)
    Variable V_minRowLoc
 
    //max cycle count has to be entered manually, = (#cycles in data set - 1)
    For(cycle=0;cycle<999999;cycle+=1)
        If(cycle>0)
            start=V_minRowLoc+37
            stop=V_minRowLoc+63
            Cursor A Pos start
            Cursor B Pos stop
        endIf
        WaveStats/Q/R=[start,stop] Pos
        Limits[cycle][%Cycle] = {cycle+1}
        Limits[cycle][%StopRow] = {V_minRowLoc}
        Limits[cycle][%StartRow] = {V_minRowLoc-50}
        Limits[cycle][%CsrAPos] = {pcsr(A)}
        Limits[cycle][%CsrBPos] = {pcsr(B)}
    Endfor
   
    //last cycle does not obey If-endif conditions, enter limits manually
    WaveStats/R=[V_minRowLoc,(V_minRowLoc+50)] Pos
    Edit Limits.ld
    InsertPoints 0,1,Limits
   
end

May 17, 2018 at 01:14 pm - Permalink

After much trial and error, and consultation with the helpful engineers at Wavemetrics, I generated the following procedure to automatically index cycles. It uses a polynomial curve fit to the peak maxima, and calculates start and stop row locations based on that value. These results can now be used to create hysteresis loops, from simultaneously collected signals, of any number of cycles.

The procedure does throw a curve fit error at the end of execution: "Must contain at least as many points as there are parameters," or something to that effect, but the actual fit curve on the last cycle appears to be normal. Further, the row locations for the beginning and end of the cycle are correct.

#pragma TextEncoding = "UTF-8"
#pragma rtGlobals=3     // Use modern global access method and strict wave access.

    //Before executing program, perform FindLevels on data wave bounded by cursors, once for EDGE=1
    //and again for EDGE=2, store these positions in Lvls_L and Lvls_R, respectively.

    //Allocate #cycles
constant kNumCycle=1000000

Function IndexMax()

    //Create wave to store results
    Make/O/N=(kNumCycle,4) Limits
   
    SetDimLabel 1,0,Cycle,Limits
    SetDimLabel 1,1,StartRow,Limits
    SetDimLabel 1,2,StopRow,Limits
    SetDimLabel 1,3,Maxima,Limits
   
    //Define source wave
    Wave Pos,t_s
    Wave Lvls_L,Lvls_R
   
    //Define input variables, manually enter start and stop from cycle 0 of Lvls_L, Lvls_R
    Variable cycle
    Variable start=19194,stop=19210
    Variable Xcross
    Variable V_value
   
    For(cycle=0;cycle<kNumCycle;cycle+=1)
        If(cycle>0)
            start=(Lvls_L[cycle])
            stop=(Lvls_R[cycle])
        endIf
   
        CurveFit/Q/K={t_s[Lvls_L[cycle]]} poly_XOffset 3, Pos[Lvls_L[cycle],Lvls_R[cycle]] /X=t_s /D
        Xcross = ((-K1+2*K2*t_s[Lvls_L[cycle]])/(2*K2))
        FindValue/S=(Lvls_L[cycle]) /V=(Xcross) /T=0.015 t_s
        Limits[cycle][%Cycle] = {cycle+1}
        Limits[cycle][%StopRow] = {V_value+25}
        Limits[cycle][%StartRow] = {V_value-25}
        Limits[cycle][%Maxima] = {V_value}
    endFor
   
    Edit Limits.ld
    InsertPoints 0,1,Limits
   
end

May 24, 2018 at 12:04 pm - Permalink