Electromyography:

Processing

D. Gordon E. Robertson, PhD, FCSB

Biomechanics Laboratory,
School of Human Kinetics,
University of Ottawa, Ottawa, Canada
Types of Signal Processing
Raw (with or without band-pass filtering)
Full-wave rectified (absolute value)
Averaged or root-mean-square (RMS)
Linear envelope
Ensemble-averaged
Integrated EMG (iEMG)
Frequency or power spectrum (Fourier)
Fatigue analysis (sequential Fourier)
Amplitude probability distribution function (APDF)
and CAPDF
Conduction velocity
Wavelet transform

Biomechanics Laboratory, University of Ottawa 2

 Raw EMG
wide frequency spectrum (20-500 Hz)
most complete information
needs 1000 Hz or greater sampling rates
requires large memory storage
difficult to determine levels of contraction
bursts of activity and onset times may be
determined from this signal
best for examining problems with recording
following slides show some errors that can be
detected from the raw signal
Biomechanics Laboratory, University of Ottawa 3
Errors when Recording EMGs

clean signal

ECG crosstalk
heart rate
detected

Biomechanics Laboratory, University of Ottawa 4

 ECG Crosstalk

ECG crosstalk occurs when recording

near the heart (ECG has higher voltages
then EMG)
EEG crosstalk when near scalp (rare)
difficult to resolve
use right side of body (away from heart)
move electrodes as far away from heart as
possible
signal averaging (average many trials)
indwelling electrodes

Biomechanics Laboratory, University of Ottawa 5

 Muscle Crosstalk

one muscles EMG is picked up by another

muscles electrodes
can be reduced by careful electrode
positioning or double differential amplifier
can be determined by cross-correlation
difficult to distinguish crosstalk from
synergistic contractions, however, biarticular
muscles have extra bursts of activity
compared to monoarticular muscles (thus
crosstalk is not a problem)
Biomechanics Laboratory, University of Ottawa 6
Errors when Recording EMGs

line (AC) interference

60 Hz
noise

DC-offset or DC-bias

baseline
not at
zero volts
Biomechanics Laboratory, University of Ottawa 7
 Solutions

To interference (line AC and radio frequency RF etc.)

Keep away from fluorescent lighting
Keep away from large electrical devices and power cords
(especially leads and cabling)
Use room lined with grounded conductive material
Keep leads short and braided (vs. radio)
Use preamplified electrodes (signal is stronger)
Use extremely narrow notch filter in post processing (e.g.,
59.5-60.5 Hz)
For DC-offsets (telemetry systems often have DC-offsets)
Use a good ground electrode over electrically neutral area
Use high-pass filter (510 Hz) to remove in post-processing

Biomechanics Laboratory, University of Ottawa 8

Errors when Recording EMGs

movement artifact

electrodes
were struck

amplifier saturation (+/0.5 V)

clipped at
+/0.5 V

Biomechanics Laboratory, University of Ottawa 9

 Solutions

To movement artifacts
Affix leads to subject (tape, wrap, webbing)
Prevent electrodes from being struck (use lateral muscles)
Avoid rapid motions
Use strong high-pass filter in post-processing
Amplifier saturation
Test with maximal contractions before recording
Reduce gain if peaks and valleys top out or bottom out
Use larger range A/D converter (+/10 V vs. +/5 V)

Biomechanics Laboratory, University of Ottawa 10

 Full-wave Rectified EMG

same as taking the absolute value

of the raw signal
mainly used as an intermediate
step before another process (e.g.,
averaging, linear envelope, and
integration)
can be done electronically in real-
time
Biomechanics Laboratory, University of Ottawa 11
 Sample EMGs
raw EMG (band-passed filtered, 20-500 Hz)

full-wave rectified

Biomechanics Laboratory, University of Ottawa 12

 Averaged EMG

simple to compute
can be done in real-time
averaged EMG is a moving average of a
full-wave rectified EMG
must select an appropriate window
width that changes with sampling rate
easy for determining levels of contraction

Biomechanics Laboratory, University of Ottawa 13

 Sample Averaged EMG
raw EMG (1010 Hz sampling rate)

averaged EMG (moving average, 51 points)

Biomechanics Laboratory, University of Ottawa 14

 Linear Envelope EMG
requires two-step process: full-wave rectification
followed by low-pass filter (4-10 Hz cutoff)
can be done electronically (but adds a delay)
reduces frequency content of EMG and thus lowers
sampling rates (e.g., 100 Hz) and memory storage
easy to interpret and to detect onset of activity
can be ensemble-averaged to obtain patterns
difficult to detect artifacts
useful as a control (myoelectric) signal
Biomechanics Laboratory, University of Ottawa 15
 Sample LE-EMG
raw (band-passed filtered) EMG

linear envelope EMG (cutoff 4 Hz)

can
have a
time lag

Biomechanics Laboratory, University of Ottawa 16

 Ensemble-Averaged EMG
usually applied to cyclic activities and linear
envelope EMGs
requires method for identifying start of a cycle
or start and end of an activity
foot switches or force platforms can be used for gait
studies
microswitches, optoelectric, or electromagnetic
sensors for other activities
can also use a threshold detector of a kinematic or
EMG channel
each cycle of activity must be time normalized
Biomechanics Laboratory, University of Ottawa 17
 Ensemble-Averaged EMG contd
amplitude normalization is often done
to maximal voluntary contraction (MVC)
to submaximal isometric contraction
to EMG of a functional activity (e.g., holding a load)
mean and standard deviations for each proportion of
cycle are computed
mean and s.d. or 95% confidence interval may be
presented to show representative contraction during
activity cycle
easier to make comparisons among subjects
grand ensemble-averages (average of averages) for
comparisons among several experimental conditions
Biomechanics Laboratory, University of Ottawa 18
Ensemble-Averages from Squat Lift

mean +/ S.D.

abscissa must
be normalized
to % cycle

ordinate may also

be normalized

Biomechanics Laboratory, University of Ottawa 19

 Integrated EMG (iEMG)
important for quantitative EMG relationships
(EMG vs. force, EMG vs. work)
best measure of the total muscular effort
useful for quantifying activity for ergonomic
research
various methods:
mathematical integration (area under absolute values of EMG
time series)
root-mean-square (RMS) times duration is similar but does
not require taking absolute values
electronically (see next page)
Biomechanics Laboratory, University of Ottawa 20
 Electronically Integrated EMG
always requires full-wave rectification
various methods:
simple time integration (eventually saturates amplifier)
integration and reset after a fixed time interval
integration and reset after a particular value is reached
cannot recognize artifacts, noise become
included
especially important to first remove DC-offsets
must compute amount of iEMG from amplitude
or differences between 2 amplitudes
Biomechanics Laboratory, University of Ottawa 21
 Sample Integrated EMG
raw (band-passed filtered) EMG

integrated EMG (over contraction)

notice read total
units are iEMG from
V.s curve (i.e.,
320 V.s)
Biomechanics Laboratory, University of Ottawa 22
 Other iEMGs
integrate after preset time (0.1 s)

add each
notice
peak to
units are
get total
V.s
IEMG
integrate after preset voltage (20 V.s)
multiply
number of
peaks by
20 V.s
Biomechanics Laboratory, University of Ottawa 23
 Frequency Spectrum
useful for determining onset of muscle fatigue
mean or median frequency of spectrum in
unfatigued muscle is usually between 5080 Hz
as fatigue progresses fast-twitch fibres drop out,
shifting frequency spectrum to left (lowering
mean and median frequencies)
mean frequency is less variable and therefore is
better than median
useful for detecting neural abnormalities

Biomechanics Laboratory, University of Ottawa 24

 Sample Power Spectrum
flexor digitorum longus (MVC)

gradual
increase to
>95% after
200 Hz

median
frequency
approx. 70 Hz

Biomechanics Laboratory, University of Ottawa 25

 Fatigue Analysis
essentially a series of frequency analyses
select duration of window (1 to 5 s)
can overlap intervals to increase resolution
usually normalized to percentage of initial
mean or median frequency
mean frequencies are less variable than
median
need to decide a threshold for when fatigue
occurs (i.e., fatigue has occurred when mean
or median frequency is below a threshold)
Biomechanics Laboratory, University of Ottawa 26
 Sample Fatigue Analysis
erector spinae over 60 seconds (50% overlap)

gradual
decline of
mean and
median
frequencies
medians
are more
variable
Biomechanics Laboratory, University of Ottawa 27
Amplitude Probability Distribution
Function (APDF & CAPDF)
developed by Hagberg & Jonsson for ergonomics
research (Ergonomics, 18:311-319)
EMG is amplitude normalized to %MVC then
sampled to compute frequencies of various
amplitudes, usually for long durations (hours)
Cumulative APDF is calculated to compute three
thresholds:
10%tile < 25% MVC for level of rest
50%tile < 1014% MVC for work load
90%tile < 5070% MVC for heavy contractions

Biomechanics Laboratory, University of Ottawa 28

 Sample APDF & CAPDF
neck flexor (only 5 minutes)

90%tile
=52%MVC

50%tile
=8%MVC
10%tile
=2%MVC

Biomechanics Laboratory, University of Ottawa 29

 Muscle Fibre Conduction
Velocity
requires two amplifiers and three electrodes
electrodes are arranged in a line over a known
distance (15 mm)
middle electrode connected as ground to both
amplifiers
divide distance between electrodes by time difference
between similar peaks (Andreassen & Arendt-
Neilsen, J Physiology, 319:561-71, 1987)

Biomechanics Laboratory, University of Ottawa 30

 Wavelet Analysis
decomposition of EMG time series into a time-
frequency space, to determine the dominant modes of
variability and their temporal changes
figure shows EMG signal and
its wavelet transform (SIMI)
used to de-noise EMG signals,
to detect fatigue and for feature
extraction

Biomechanics Laboratory, University of Ottawa 31

 Other Techniques
auto-correlation (correlate signal with itself
shifted in time, gives signal characteristics)
cross-correlation (correlate signal with
another EMG signal, tests for crosstalk)
zero-crossings (the more crossings the greater
the level of recruitment)
spike (peak) counting (number of spikes
above a threshold)
single motor unit detection
double differential amplifier (velocity of
propagation)
Biomechanics Laboratory, University of Ottawa 32