Ton van den Bogert wrote:

>It is not too hard to determine the transfer function in the

frequency

>domain of such a filter.

After sampling, the discrete-time moving average system (Example 2.2,

p. 17 in Oppenhem, Schafer "Discrete-Time signal processing", Prentice

Hall, 1989):

y[n]=(1/(M1+M2+1))(x[n+M1]+x[n+M1-1]+...+x[n]+x[n-1]+...+x[n-M2]),

where M1 and M2 are numbers of points around x[n]. M1 is the number of

samples received after x[n] - noncausal processing, can be implemented

only off-line, M2 is number of samples received before x[n]. If you

apply moving average such that the 10th point y[10] is the weighted sum

of x[10], x[9],...,x[1] then M1=0, M2=9. Anyway, it's the total number

of averaged points that is important for determining the magnitude of

the frequency response.

Magnitude of frequency response of this discrete-time system (Example

2.15, p. 44, ibid):

H(Wdiscr)=abs( 1/(M1+M2+1)*sin(W discr*(M1+M2+1)/2)/sin(W discr/2)

),

or

H(W discr)=abs ( 1/10*sin(5*Wdiscr)/sin(0.5*Wdiscr) ).

>From a graph of H(Wdiscr) (quick look using Matlab for Wdiscr=0 to pi,

step 0.01):

H(wdiscr)=0.707 for wdiscr=0.28.

A corresponding continuous-time system has equivalent (effective)

frequency response for Wcont=Wdiscr/Tsamp, and

Fcutoff=Wdiscr/2piTsamp.

If sampling at 100Hz:

Tsamp=0.01s, W discr=0.28 => Wcont=28 rad/s, Fcutoff=4.46Hz,

i.e., smoothing with 100ms window in this case is equivalent to

low-pass filtering with 4.46Hz cutoff freq.

Examples for different #points averaged, and for a couple of sample

freqs are given below. After Wdiscr is found and Fcutoff/Window is

calculated for one Fsamp the rest is found by multiplying/dividing by

the ratio F'samp/Fsamp.

#points Wdiscr Fcutoff Hz Fcutoff Hz Fcutoff Hz

averaged window ms window ms window ms

(Fsamp=100Hz) (Fsamp=500Hz) (Fsamp=1000Hz)

3 0.976 15.5 HZ, 30 ms 77.5 Hz, 6 ms 155 Hz, 3 ms

5 0.565 9 Hz, 50 ms 45 Hz, 10 ms 90 Hz, 5 ms

10 0.28 4.5 Hz, 100ms 22.5 Hz, 20ms 45 Hz, 10 ms

20 0.139 2.2 Hz , 200ms 11 Hz, 40ms 22 Hz, 20 ms

Mark Shapiro

***********************

Mark B. Shapiro, Ph.D

School of Kinesiology (M/C 194)

University of Illinois at Chicago

901 West Roosevelt Road

Chicago, Illinois 60608

tel: (312) 942-5414

e-mail: mshapi2@uic.edu

fax (312) 355-2305

***********************

---------------------------------------------------------------

To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl

For information and archives: http://isb.ri.ccf.org/biomch-l

---------------------------------------------------------------

>It is not too hard to determine the transfer function in the

frequency

>domain of such a filter.

After sampling, the discrete-time moving average system (Example 2.2,

p. 17 in Oppenhem, Schafer "Discrete-Time signal processing", Prentice

Hall, 1989):

y[n]=(1/(M1+M2+1))(x[n+M1]+x[n+M1-1]+...+x[n]+x[n-1]+...+x[n-M2]),

where M1 and M2 are numbers of points around x[n]. M1 is the number of

samples received after x[n] - noncausal processing, can be implemented

only off-line, M2 is number of samples received before x[n]. If you

apply moving average such that the 10th point y[10] is the weighted sum

of x[10], x[9],...,x[1] then M1=0, M2=9. Anyway, it's the total number

of averaged points that is important for determining the magnitude of

the frequency response.

Magnitude of frequency response of this discrete-time system (Example

2.15, p. 44, ibid):

H(Wdiscr)=abs( 1/(M1+M2+1)*sin(W discr*(M1+M2+1)/2)/sin(W discr/2)

),

or

H(W discr)=abs ( 1/10*sin(5*Wdiscr)/sin(0.5*Wdiscr) ).

>From a graph of H(Wdiscr) (quick look using Matlab for Wdiscr=0 to pi,

step 0.01):

H(wdiscr)=0.707 for wdiscr=0.28.

A corresponding continuous-time system has equivalent (effective)

frequency response for Wcont=Wdiscr/Tsamp, and

Fcutoff=Wdiscr/2piTsamp.

If sampling at 100Hz:

Tsamp=0.01s, W discr=0.28 => Wcont=28 rad/s, Fcutoff=4.46Hz,

i.e., smoothing with 100ms window in this case is equivalent to

low-pass filtering with 4.46Hz cutoff freq.

Examples for different #points averaged, and for a couple of sample

freqs are given below. After Wdiscr is found and Fcutoff/Window is

calculated for one Fsamp the rest is found by multiplying/dividing by

the ratio F'samp/Fsamp.

#points Wdiscr Fcutoff Hz Fcutoff Hz Fcutoff Hz

averaged window ms window ms window ms

(Fsamp=100Hz) (Fsamp=500Hz) (Fsamp=1000Hz)

3 0.976 15.5 HZ, 30 ms 77.5 Hz, 6 ms 155 Hz, 3 ms

5 0.565 9 Hz, 50 ms 45 Hz, 10 ms 90 Hz, 5 ms

10 0.28 4.5 Hz, 100ms 22.5 Hz, 20ms 45 Hz, 10 ms

20 0.139 2.2 Hz , 200ms 11 Hz, 40ms 22 Hz, 20 ms

Mark Shapiro

***********************

Mark B. Shapiro, Ph.D

School of Kinesiology (M/C 194)

University of Illinois at Chicago

901 West Roosevelt Road

Chicago, Illinois 60608

tel: (312) 942-5414

e-mail: mshapi2@uic.edu

fax (312) 355-2305

***********************

---------------------------------------------------------------

To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl

For information and archives: http://isb.ri.ccf.org/biomch-l

---------------------------------------------------------------