View Full Version : Resampling data for coherence and cross-correlations

Kearney, Robert
03-26-2001, 07:06 AM

In response to your questions you should:

(1) Resample all data to the sample sampling rate being sure to anti-alias
filter the data sampled at the higher sampling rate first. I do not think
there is any point in up-sampling since the resulting signal will have no
power at the higher frequencies. As a result you will simply get noisy
measures with no improvement of resolution. Depending on your application
You may also wish to rectify the EMG signals before decimating

(2) Decimating your data may change the power in two ways:

(i) By eliminating frequency components at frequencies about the
nyqist rate.
This is one reason to rectify EMG data before decimating
since the
rectification followed by anti-alias filtering will act as a

(ii) you may also get changes if you algorithm does not take into
account the
sampling interval. In computing the power you need to
explicitly account for
the sampling rate since there will be fewer samples at the
lower sampling

Hope this helps.

Rob Kearney
Robert E Kearney, Ph.D. Eng.
Professor & Chair, Department of Biomedical Engineering,
McGill University, 3775 University Street, Room 309
Montreal, Quebec, Canada H3A 2B4
Tel: (514) 398-6737; Fax: (514) 398-7461
E-mail: rob@bmed.mcgill.ca

Date: Thu, 22 Mar 2001 17:30:53 +1000
From: Justin Keogh
Subject: Resampling data for coherence and cross-correlations

0100,0100,0100Dear list,

We have collected some EMG (1000 Hz), accelerometry (100 Hz)
and video (50 Hz) data in a recent study. We seek to investigate
the relationships between these variables in the frequency
(coherence) and time (cross-correlation) domains.

Papers which have done EMG-tremor coherence (Halliday et al.
1999; Marsden et al. 1997; Vaillancourt et al. 2000) have all
sampled their different sources of data at the same rate. While we
would have liked to do that as well, that was not possible.

Therefore we need to resample our data so to obtain the same
number of data points, and to ensure that these data points were
measured at the same point of time.

So my questions are thus:

1. Is it better to downsample the data (eg EMG from 1000 to 100
Hz) or upsample tremor from 100 to 1000 Hz?

2. We have noticed a change in the power of the resampled data in
proportion to the change in signal sampling frequency. Is this a
problem for coherence and cross-correlation analyses?

Any other comments or papers to refer to would be appreciated.


Justin Keogh.


Halliday, D. M., Conway, B. A., Farmer, S. F., & Rosenberg, J. R. (1999).
independent contributions from motor-unit synchronization to human
physiological tremor. Journal of Neurophysiology, 82,

McAuley, J. H., Rothwell, J. C., & Marsden, C. D. (1997). Frequency peaks of
tremor, muscle vibration and electromyographic activity at 10 Hz and 40 Hz
during human finger muscle contraction may reflect rhythmicities of central
neural firing. Experimental Brain Research, 114, 525-541.

outVaillancourt, D. E., & Newell, K. M.
(2000). Amplitude changes in the 8-12, 20-25,

outand 40 Hz oscillations in finger tremor.
Clinical Neurophysiology, 111, 1792


Justin Keogh BHMS (Hons)
PhD Candidate
Griffith University, Gold Coast
School of Physiotherapy & Exercise Science
07 5552 8390 (W) 0419 714 921 (M)
07 5552 8674 (Fax)


To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l