| T O P I C R E V I E W |
| mikee927 |
Posted - 10/30/2013 : 06:36:04 AM
Origin Ver. and Service Release (Select Help-->About Origin):9.0.0 SRI 64 bit
Operating System: Windows 8
I believe my data can be fit by the convolution of two functions (a gauss and possibly lognormal distribution). However, when I try to deconvolute my data with a gaussian distribution (both added below) I only receive one data point (for linear deconvolution), and what looks like random signal for circular deconvolution. What am I doing wrong? Both data sets have the same "X" values and the same number of points.
Note: When I use the convolute (a gaussian and lognormal) I do receive a function close to my data.
Thank you for any help you can provide,
Michael Eller
My Data:
0.32101 0.00741
14.59114 0.01398
28.85117 0.04752
43.10111 0.16084
57.34097 0.32593
71.57073 0.5701
85.79041 0.87638
100 1
114.1995 0.84255
128.38891 0.64125
142.56824 0.49696
156.73747 0.37429
170.89662 0.24733
185.04568 0.15223
199.18465 0.08994
213.31353 0.04238
227.43233 0.02365
241.54103 0.01579
255.63965 0.01035
269.72818 0.00824
283.80662 0.00617
297.87497 0.00464
311.93323 0.00445
325.9814 0.00358
340.01949 0.00311
354.04749 0.00307
368.0654 0.00261
382.07322 0.00239
396.07095 0.0025
410.05859 0.00222
424.03615 0.00239
438.00362 0.00245
451.96099 0.00213
465.90828 0.00211
479.84549 0.00216
493.7726 0.00211
507.68962 0.00199
521.59656 0.00185
535.49341 0.00213
549.38017 0.00197
563.25684 0.00157
577.12342 0.00166
590.97991 0.0016
604.82632 0.00163
618.66263 0.00148
632.48886 0.00148
646.305 0.00155
660.11105 0.00145
673.90702 0.00146
687.69289 0.0014
701.46868 0.00145
715.23438 0.00141
728.98998 0.00136
Gauss Function:
0.32101 0.00566
14.59114 0.0127
28.85117 0.04176
43.10111 0.12945
57.34097 0.31906
71.57073 0.60114
85.79041 0.85818
100 0.92663
114.1995 0.75694
128.38891 0.46854
142.56824 0.22086
156.73747 0.08072
170.89662 0.02468
185.04568 0.00837
199.18465 0.00486
213.31353 0.00429
227.43233 0.00423
241.54103 0.00422
255.63965 0.00422
269.72818 0.00422
283.80662 0.00422
297.87497 0.00422
311.93323 0.00422
325.9814 0.00422
340.01949 0.00422
354.04749 0.00422
368.0654 0.00422
382.07322 0.00422
396.07095 0.00422
410.05859 0.00422
424.03615 0.00422
438.00362 0.00422
451.96099 0.00422
465.90828 0.00422
479.84549 0.00422
493.7726 0.00422
507.68962 0.00422
521.59656 0.00422
535.49341 0.00422
549.38017 0.00422
563.25684 0.00422
577.12342 0.00422
590.97991 0.00422
604.82632 0.00422
618.66263 0.00422
632.48886 0.00422
646.305 0.00422
660.11105 0.00422
673.90702 0.00422
687.69289 0.00422
701.46868 0.00422
715.23438 0.00422
728.98998 0.00422
My linear deconvolution:
0 0.4449
My circular deconvolution:
0 -21.48392
14.01286 12.25111
28.02573 4.38697
42.03859 -19.34746
56.05146 22.11939
70.06432 -8.5506
84.07719 -12.01457
98.09005 23.70207
112.10292 -12.22833
126.11578 -8.66763
140.12865 23.61465
154.14151 -22.39345
168.15438 9.9502
182.16724 4.85218
196.18011 -11.81425
210.19297 6.08467
224.20584 10.71826
238.2187 -30.478
252.23157 42.26196
266.24443 -37.9766
280.2573 17.35395
294.27016 11.20502
308.28303 -34.58256
322.29589 41.70449
336.30876 -29.43226
350.32162 4.29272
364.33448 20.85559
378.34735 -33.51709
392.36021 27.92228
406.37308 -7.82563
420.38594 -15.77093
434.39881 30.50226
448.41167 -29.0914
462.42454 12.86104
476.4374 9.27107
490.45027 -25.72578
504.46313 28.26966
518.476 -16.13486
532.48886 -3.90061
546.50173 21.22433
560.51459 -27.0299
574.52746 18.76146
588.54032 -1.23046
602.55319 -16.19865
616.56605 24.56494
630.57892 -19.94557
644.59178 5.2421
658.60465 11.54447
672.61751 -21.71185
686.63038 20.33371
700.64324 -8.59049
714.6561 -7.28208
728.66897 19.30436 |
| 5 L A T E S T R E P L I E S (Newest First) |
| Sam Fang |
Posted - 03/31/2014 : 02:33:52 AM
Thanks for your suggestion.
Tool in the menu Signal Processing: Deconvolution is used to deconvolve a signal which is exactly a convolution of two signals. However most spectrum data are the convolution result with noise. And if we use this tool, the result may not make sense. And we should use deconvolution in fitting then, e.g. the tutorial "Fitting with Convolution of Two Functions".
To avoid misleading, we will remove it from menu as well as the page that you pointed out:
http://www.originlab.com/index.aspx?go=Products/Origin/DataAnalysis/SignalProcessing/Convolution&pid=72
If you have any question, please let us know.
Thanks.
Sam
OriginLab Technical Services |
| juredemsar |
Posted - 03/26/2014 : 1:02:26 PM
Hi Penn,
all I was trying to do is to follow the procedure here: http://www.originlab.com/index.aspx?go=Products/Origin/DataAnalysis/SignalProcessing/Convolution&pid=72
. I used all options (circular, linear, normalize on/off, wrap on/off), the length of the response function was the same or much smaller than that of the signal. As I said I just cannot get anything sensible out, so I gave up and used "Fitting with Convolution of Two Functions" instead. Still, if the deconvolution is not functioning it should not be offered.
Cheers, Jure |
| Penn |
Posted - 03/26/2014 : 03:07:06 AM
Hi Jure,
In our algorithm, the size of the deconvolution result is calculated by (input size - response size + 1). So, even using the analytical function, if the size of response is the same as it, the linear deconvolution result size is also 1. As you can perform linear convolution to see the size of the result, it is calculated by (input size + response size - 1).
Here I will explain more "pure" convolution. It means that, to get back the signal before response by linear deconvolution, the signal after response should be exactly the result of the convolution between the signal before response and the response.
Penn |
| juredemsar |
Posted - 03/25/2014 : 4:43:56 PM
Hi,
I am having the same problem as Michael. And this cannot be due to noisy data. When trying to find out where the problem is, I generated the data from an analytical function. When trying to deconvolute with simple Gaussian I get garbage as Michael...
Cheers, Jure |
| Penn |
Posted - 11/01/2013 : 02:39:03 AM
Hi Michael Eller,
The result may be from the noised signals, but not "pure" convolution of two digital signals, so I am afraid the deconvolution tool is not able to be used directly onto the noised signals. And for such case, deconvolution is not always reliable as the result can be very sensitive to any noise present in the data.
There are already two tutorials for such case, and you can refer to Fitting with Convlution and Fitting with Convolution of Two Functions.
If you still have problem on handling it, please provide the function of the other signal (possibly lognormal distribution) and parameters, then we can have a try.
Penn |
|
|
