Tweet
Hello everyone,
I have noticed lately that there are a number of members of
our community who use matlab to filter their position data. In the
past I have used matlab, specifically the cheby1 option to perform
these operations, with good results. However, a journal reviewer has
recently raised a question regarding my methods and it caused me
to think more closely about matlab's filtering.
Enclosed is a sample .m file that can be cut and pasted into
matlab. The data is from an led placed on the leg segment during
lifting and is sampled at 100 Hz (therefore a Nyquist frequency of
50). Three different cheby1 filters are defined with cut off
frequencies of 2 Hz, 2Hz and and seemingly low 0.17 Hz. Now for
those of you who use matlab, you may know that the cheby1 option also
requires you to state the dB of ripple in the passband. This has
been set to either 0.5 (the recommended starting value if you are
unsure) and 0.0001.
Let me define...
Filter 1. Cut off = 2Hz, 0.50 dB ripple in passband.
Filter 2. Cut off = 2Hz, .00001 dB ripple in the passband.
Filter 3. Cut off= 0.17 Hz, 0.00001 dB ripple in the passband.
When you run the matlab file you will see the raw and smoothed
data for each of these filters, and also the Bode plots. When I
qualitatively look at the "performanace" of the three filters I judge
Filter 3 to be superior (cutoff frequency 0.17, .00001 dB
ripple). Filter 1 shifts the data, while Filter 2 does not
sufficiently smooth the data.
The Bode plots show that the actual cut off frequency for Filter 1
and Filter 3 are approximately 2-3 Hz, and Filter 2 is about 25 Hz.
I am familiar with the basics of signal processing but am hoping
someone can shed a bit of light on this curious finding. If you are
reporting the use of this filter in a manuscript is it appropriate to
state the cutoff filter from the bode plot, or the software
specifications. This is the dilema I am in. I have reported a
cutoff frequency of 0.17 Hz, and it obviously raised some concern,
but the bode plot shows that this isn't really the case.
In any event, thank you for reading this and I hope you find it
interesting. It seems like many people are using this software
without any mention of this behaviour. I would like to know more.
Thanks in advance.
With regards,
David
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> >
% 1. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
[b,a]=cheby1(2,.5, 2/50)
% 2. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
[d,c]=cheby1(2,.00001, 2/50)
% 3. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
[f,e]=cheby1(2,.00001, 0.17/50)
x=[
.114822
.1146299
.1150839
.114484
.1146179
.114633
.1146589
.1146399
.1146719
.1147149
.1149649
.113958
.1139909
.114163
.1140569
.113802
.1131449
.1134449
.1137949
.1136789
.1129589
.11354
.1131579
.113164
.1126179
.1110199
.1106
.110174
.1096169
.1091869
.108957
.1089249
.1095219
.108583
.1082939
.1078519
.1073229
.107018
.1075059
.10689
.1074089
.1071169
.10728
.107021
.1064789
.107816
.1076579
.107882
.1083399
.1079879
.108153
.1087429
.109526
.1097259
.109468
.1093499
.1093799
.110306
.1103769
.1104199
.1104369
.110917
.1114089
.1115589
.1117549
.1115039
.1109289
.1122789
.1111689
.111987
.110328
.1110179
.111331
.111481
.1112219
.1115759
.111
.1110299
.1106539
.1104049
.10986
.1091429
.107982
.1080579
.1071539
.1073219
.108611
.1086179
.109554
.109718
.10966
.1096829
.109244
.1085269
.1079889
.107389
.107582
.108075
.109264
.109883
.109613
.1106999
.1107949
.1099179
.1088979
.1087539
.108042
.1077499
.107137
.107271
.106835
.106981
.106967
.1071399
.106816
.1063669
.1066699
.106968
.107321
.1072699
.107872
.10819
.107855
.1082789
.1088759
.109231
.108415
.109483
.10975
.110206
.110648
.111267
.1114979
.111676
.11165
.112693
.111905
.112773
.113094
.1136559
.1136419
.1136189
.113591
.113832
.114431
.114577
.114526
.1146339
.1144779
.1150589
.116052
.1160139
.116605
.1164209
.1174309
.1174869
.117512
.1177629
.118055
.117677
.1184279
.1183589
.11821
.11862
.1179919
.1184359
.1180969
.1180649
.1186349
.117835
.1173539
.1181339
.1173499
.117621
.1172829
.116401
.116607
.1158619
.115679
.1154159
.114747
.114725
.1141289
.113963
.113366
.1127609
.112387
.1115309
.111383
.111377
.1106019
.111322
.1106179
.111151
.1111709
.1109649
.1109839
.1100829
.110316
.1104929
.1101799
.110759
.1096799
.109819
.110105
.1099179
.1103349
.110036
.1101929
.110306
.1098399
.109647
.10948
.1091369
.109263
.108676
.109716
.108036
.1082919
.107257
.1067829
.1057149
.1048489
.104102
.102871
.103007
.10215
.101691
.1018259
.101549
.10065
.1005159
9.96640E-02
9.84770E-02
9.79070E-02
9.80570E-02
9.70340E-02
9.67660E-02
9.65890E-02
9.55700E-02
9.52850E-02
9.48550E-02
9.54420E-02
9.48640E-02
9.44030E-02
9.41390E-02
9.35640E-02
9.31330E-02
9.22800E-02
9.22880E-02
9.21110E-02
9.10890E-02
9.15140E-02
9.10670E-02
9.08170E-02
9.06580E-02
9.00690E-02
8.99380E-02
9.00860E-02
8.93470E-02
8.95000E-02
8.94770E-02
8.93440E-02
8.93320E-02
9.03580E-02
8.90460E-02
8.97630E-02
8.99150E-02
9.07620E-02
9.01880E-02
9.04950E-02
9.07830E-02
9.05000E-02
9.06760E-02
9.06470E-02
9.19690E-02
9.13770E-02
9.12410E-02
9.13900E-02
9.18170E-02
9.18170E-02
9.19770E-02
9.28490E-02
9.29950E-02
9.31220E-02
9.32750E-02
9.35840E-02
9.32900E-02
9.41360E-02
9.42760E-02
9.44330E-02
9.44220E-02
9.50050E-02
9.51570E-02
9.55790E-02
9.58830E-02
9.61510E-02
9.58670E-02
9.68810E-02
9.63190E-02
9.65940E-02
9.70540E-02
9.71700E-02
9.74630E-02
9.68810E-02
9.74630E-02
9.78900E-02
9.78880E-02
9.74340E-02
9.71450E-02
9.77410E-02
9.71370E-02
9.68510E-02
9.72660E-02
9.69720E-02
9.65300E-02
9.70810E-02
9.75370E-02
9.65030E-02
9.73720E-02
9.73530E-02
9.76450E-02
9.74640E-02
9.78890E-02
9.77360E-02
9.81580E-02
9.83230E-02
9.80200E-02
9.78550E-02
9.74130E-02
9.78660E-02
9.72630E-02
9.68410E-02
9.75320E-02
9.78260E-02
9.86840E-02
9.82480E-02
9.89760E-02
9.85050E-02
9.86680E-02
9.87960E-02
9.90950E-02
9.89600E-02
9.91150E-02
9.88080E-02
9.93650E-02
9.97790E-02
9.99430E-02
.10048
.10048
.1012369
.101222
.1016279
.1019449
.102497
.102457
.1031839
.1036409
.104497
.1043379
.1038919
.1041729
.105076
.1044609
.1054
.1056709
.106178
.106508
.107526
.1067759
.1066149
.107342
.107202
.1073539
.1067539
.106903
.107494
.1073529
.1067009
.107578
.1067289
.107588
.1079149
.108027
.1087599
.1084209
.1082929
.108693
.108302
.1087279
.110177
.109564
.1095829
.1097429
.1095869
.1105889
.1114649
.111596
.112199
.1126179
];
t= 1.00000E-02: 1.00000E-02: 4.0;
% 1. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
% [b,a]=cheby1(2,.5, 2/50)
% 2. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
% [d,c]=cheby1(2,.00001, 2/50)
% 3. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
% [e,f]=cheby1(2,.00001,.0035)
xba=filtfilt(b,a,x)
plot (t,x,t,xba)
title('[b,a]=cheby1(2,.5, 2/50)')
figure
freqz(b,a)
title('[b,a]=cheby1(2,.5, 2/50)')
figure
xdc=filtfilt(d,c,x)
plot(t,x,t,xdc)
title('[d,c]=cheby1(2,.00001, 2/50)')
figure
freqz(d,c)
title('[d,c]=cheby1(2,.00001, 2/50)')
figure
xfe=filtfilt(f,e,x)
plot(t,x,t,xfe)
title('[e,f]=cheby1(2,.00001, 0.17/50)')
figure
freqz(f,e)
title('[e,f]=cheby1(2,.00001, 0.17/50)')
Department of Rehabilitation Sciences
University of East London
Romford Road
London E15 4LZ
Phone 0181-590-7000 (4025)
d.m.hooper@uel.ac.uk
-------------------------------------------------------------------
To unsubscribe send UNSUBSCRIBE BIOMCH-L to LISTSERV@nic.surfnet.nl
For more information: http://www.kin.ucalgary.ca/isb/biomch-l.html
-------------------------------------------------------------------
I have noticed lately that there are a number of members of
our community who use matlab to filter their position data. In the
past I have used matlab, specifically the cheby1 option to perform
these operations, with good results. However, a journal reviewer has
recently raised a question regarding my methods and it caused me
to think more closely about matlab's filtering.
Enclosed is a sample .m file that can be cut and pasted into
matlab. The data is from an led placed on the leg segment during
lifting and is sampled at 100 Hz (therefore a Nyquist frequency of
50). Three different cheby1 filters are defined with cut off
frequencies of 2 Hz, 2Hz and and seemingly low 0.17 Hz. Now for
those of you who use matlab, you may know that the cheby1 option also
requires you to state the dB of ripple in the passband. This has
been set to either 0.5 (the recommended starting value if you are
unsure) and 0.0001.
Let me define...
Filter 1. Cut off = 2Hz, 0.50 dB ripple in passband.
Filter 2. Cut off = 2Hz, .00001 dB ripple in the passband.
Filter 3. Cut off= 0.17 Hz, 0.00001 dB ripple in the passband.
When you run the matlab file you will see the raw and smoothed
data for each of these filters, and also the Bode plots. When I
qualitatively look at the "performanace" of the three filters I judge
Filter 3 to be superior (cutoff frequency 0.17, .00001 dB
ripple). Filter 1 shifts the data, while Filter 2 does not
sufficiently smooth the data.
The Bode plots show that the actual cut off frequency for Filter 1
and Filter 3 are approximately 2-3 Hz, and Filter 2 is about 25 Hz.
I am familiar with the basics of signal processing but am hoping
someone can shed a bit of light on this curious finding. If you are
reporting the use of this filter in a manuscript is it appropriate to
state the cutoff filter from the bode plot, or the software
specifications. This is the dilema I am in. I have reported a
cutoff frequency of 0.17 Hz, and it obviously raised some concern,
but the bode plot shows that this isn't really the case.
In any event, thank you for reading this and I hope you find it
interesting. It seems like many people are using this software
without any mention of this behaviour. I would like to know more.
Thanks in advance.
With regards,
David
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> >
% 1. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
[b,a]=cheby1(2,.5, 2/50)
% 2. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
[d,c]=cheby1(2,.00001, 2/50)
% 3. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
[f,e]=cheby1(2,.00001, 0.17/50)
x=[
.114822
.1146299
.1150839
.114484
.1146179
.114633
.1146589
.1146399
.1146719
.1147149
.1149649
.113958
.1139909
.114163
.1140569
.113802
.1131449
.1134449
.1137949
.1136789
.1129589
.11354
.1131579
.113164
.1126179
.1110199
.1106
.110174
.1096169
.1091869
.108957
.1089249
.1095219
.108583
.1082939
.1078519
.1073229
.107018
.1075059
.10689
.1074089
.1071169
.10728
.107021
.1064789
.107816
.1076579
.107882
.1083399
.1079879
.108153
.1087429
.109526
.1097259
.109468
.1093499
.1093799
.110306
.1103769
.1104199
.1104369
.110917
.1114089
.1115589
.1117549
.1115039
.1109289
.1122789
.1111689
.111987
.110328
.1110179
.111331
.111481
.1112219
.1115759
.111
.1110299
.1106539
.1104049
.10986
.1091429
.107982
.1080579
.1071539
.1073219
.108611
.1086179
.109554
.109718
.10966
.1096829
.109244
.1085269
.1079889
.107389
.107582
.108075
.109264
.109883
.109613
.1106999
.1107949
.1099179
.1088979
.1087539
.108042
.1077499
.107137
.107271
.106835
.106981
.106967
.1071399
.106816
.1063669
.1066699
.106968
.107321
.1072699
.107872
.10819
.107855
.1082789
.1088759
.109231
.108415
.109483
.10975
.110206
.110648
.111267
.1114979
.111676
.11165
.112693
.111905
.112773
.113094
.1136559
.1136419
.1136189
.113591
.113832
.114431
.114577
.114526
.1146339
.1144779
.1150589
.116052
.1160139
.116605
.1164209
.1174309
.1174869
.117512
.1177629
.118055
.117677
.1184279
.1183589
.11821
.11862
.1179919
.1184359
.1180969
.1180649
.1186349
.117835
.1173539
.1181339
.1173499
.117621
.1172829
.116401
.116607
.1158619
.115679
.1154159
.114747
.114725
.1141289
.113963
.113366
.1127609
.112387
.1115309
.111383
.111377
.1106019
.111322
.1106179
.111151
.1111709
.1109649
.1109839
.1100829
.110316
.1104929
.1101799
.110759
.1096799
.109819
.110105
.1099179
.1103349
.110036
.1101929
.110306
.1098399
.109647
.10948
.1091369
.109263
.108676
.109716
.108036
.1082919
.107257
.1067829
.1057149
.1048489
.104102
.102871
.103007
.10215
.101691
.1018259
.101549
.10065
.1005159
9.96640E-02
9.84770E-02
9.79070E-02
9.80570E-02
9.70340E-02
9.67660E-02
9.65890E-02
9.55700E-02
9.52850E-02
9.48550E-02
9.54420E-02
9.48640E-02
9.44030E-02
9.41390E-02
9.35640E-02
9.31330E-02
9.22800E-02
9.22880E-02
9.21110E-02
9.10890E-02
9.15140E-02
9.10670E-02
9.08170E-02
9.06580E-02
9.00690E-02
8.99380E-02
9.00860E-02
8.93470E-02
8.95000E-02
8.94770E-02
8.93440E-02
8.93320E-02
9.03580E-02
8.90460E-02
8.97630E-02
8.99150E-02
9.07620E-02
9.01880E-02
9.04950E-02
9.07830E-02
9.05000E-02
9.06760E-02
9.06470E-02
9.19690E-02
9.13770E-02
9.12410E-02
9.13900E-02
9.18170E-02
9.18170E-02
9.19770E-02
9.28490E-02
9.29950E-02
9.31220E-02
9.32750E-02
9.35840E-02
9.32900E-02
9.41360E-02
9.42760E-02
9.44330E-02
9.44220E-02
9.50050E-02
9.51570E-02
9.55790E-02
9.58830E-02
9.61510E-02
9.58670E-02
9.68810E-02
9.63190E-02
9.65940E-02
9.70540E-02
9.71700E-02
9.74630E-02
9.68810E-02
9.74630E-02
9.78900E-02
9.78880E-02
9.74340E-02
9.71450E-02
9.77410E-02
9.71370E-02
9.68510E-02
9.72660E-02
9.69720E-02
9.65300E-02
9.70810E-02
9.75370E-02
9.65030E-02
9.73720E-02
9.73530E-02
9.76450E-02
9.74640E-02
9.78890E-02
9.77360E-02
9.81580E-02
9.83230E-02
9.80200E-02
9.78550E-02
9.74130E-02
9.78660E-02
9.72630E-02
9.68410E-02
9.75320E-02
9.78260E-02
9.86840E-02
9.82480E-02
9.89760E-02
9.85050E-02
9.86680E-02
9.87960E-02
9.90950E-02
9.89600E-02
9.91150E-02
9.88080E-02
9.93650E-02
9.97790E-02
9.99430E-02
.10048
.10048
.1012369
.101222
.1016279
.1019449
.102497
.102457
.1031839
.1036409
.104497
.1043379
.1038919
.1041729
.105076
.1044609
.1054
.1056709
.106178
.106508
.107526
.1067759
.1066149
.107342
.107202
.1073539
.1067539
.106903
.107494
.1073529
.1067009
.107578
.1067289
.107588
.1079149
.108027
.1087599
.1084209
.1082929
.108693
.108302
.1087279
.110177
.109564
.1095829
.1097429
.1095869
.1105889
.1114649
.111596
.112199
.1126179
];
t= 1.00000E-02: 1.00000E-02: 4.0;
% 1. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
% [b,a]=cheby1(2,.5, 2/50)
% 2. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
% [d,c]=cheby1(2,.00001, 2/50)
% 3. Cutoff frequency perscribed as .035/50
% .00001 dB of Ripple in the passband
% [e,f]=cheby1(2,.00001,.0035)
xba=filtfilt(b,a,x)
plot (t,x,t,xba)
title('[b,a]=cheby1(2,.5, 2/50)')
figure
freqz(b,a)
title('[b,a]=cheby1(2,.5, 2/50)')
figure
xdc=filtfilt(d,c,x)
plot(t,x,t,xdc)
title('[d,c]=cheby1(2,.00001, 2/50)')
figure
freqz(d,c)
title('[d,c]=cheby1(2,.00001, 2/50)')
figure
xfe=filtfilt(f,e,x)
plot(t,x,t,xfe)
title('[e,f]=cheby1(2,.00001, 0.17/50)')
figure
freqz(f,e)
title('[e,f]=cheby1(2,.00001, 0.17/50)')
Department of Rehabilitation Sciences
University of East London
Romford Road
London E15 4LZ
Phone 0181-590-7000 (4025)
d.m.hooper@uel.ac.uk
-------------------------------------------------------------------
To unsubscribe send UNSUBSCRIBE BIOMCH-L to LISTSERV@nic.surfnet.nl
For more information: http://www.kin.ucalgary.ca/isb/biomch-l.html
-------------------------------------------------------------------