A simulation model of the surface EMG signal for analysis of muscle

Computers in Biology and Medicine 36 (2006) 601–618www.intl.elsevierhealth.com/journals/cobm

A simulation model of the surface EMG signal for analysis ofmuscle activity during the gait cycle

W. Wang∗, A. De. Stefano, R. Allen

Institute of Sound and Vibration Research, University of Southampton, Southampton, UK

Received 27 July 2004; received in revised form 28 February 2005; accepted 14 April 2005

Abstract

This work describes a model able to synthetize the surface EMG (electromyography) signal acquired from tibialisanterior and gastrocnemious medialis muscles during walking of asymptomatic adult subjects. The model assumesa muscle structure where the volume conductor is represented by multiple layers of anisotropic media. This modeloriginates from analysis of the single fiber action potential characterized by the conduction velocity. The surfaceEMG of voluntary contraction is calculated by gathering motor unit action potentials estimated by the summationof all activities of muscle fibers assumed to have a uniformly parallel distribution. The parameters related to thegait cycle, such as onset and cessation timings of muscle activation, amplitude of muscle contraction, periods andsequences of motor units’ recruitment, are included in the model presented. In addition, the relative positions of theelectrodes during gait can also be specified in order to adapt the simulation to the different acquisition settings.� 2005 Elsevier Ltd. All rights reserved.

Keywords: Simulation model; Surface EMG; Motor unit action potential; Motor unit recruitment; Gait cycle

1. Introduction

The surface EMG (electromyography) signal represents the characteristics of muscle function andprovides information about muscle activities. The analysis of this signal provides diagnostic informationand can be used as an aid for choosing the most appropriate methods of treatment for muscle dysfunction.

New methods or algorithms for the analysis of muscle activity need to be assessed by comparing actualand known values of parameters with the values obtained using the new methods. Therefore, a simulation

∗ Corresponding author. Tel.: +44 23 8059 4932; fax: +44 23 8059 3190.E-mail address: [email protected] (W. Wang).

0010-4825/$ - see front matter � 2005 Elsevier Ltd. All rights reserved.doi:10.1016/j.compbiomed.2005.04.002

http://www.intl.elsevierhealth.com/journals/cobm

mailto:[email protected]

602 W. Wang et al. / Computers in Biology and Medicine 36 (2006) 601–618

model of surface EMG during the gait cycle is a valuable aid in order to assess novel techniques. This isthe problem addressed in this work.

A wide range of models has been presented over recent decades by many research groups [1–13]. Someof the models aimed to simulate the single fiber action potential [1], or voluntary and elicited musclecontractions [2–4]. The characteristics of the surface EMG signal detected by the electrodes depend onanatomical, physiological and experimental system parameters. Other simulation models [5–7] are usedto test the influence of various parameters on the surface EMG signals, such as the effect of an anisotropichomogeneous medium, the interpretation of the influence of internal bone [8], and illustration of the effectof the detection system [9] such as the influence of the electrode shape and size [10,11].

Despite the large number of publications on the subject of surface EMG simulation, only few researchershave attempted to explore the relationship between the surface EMG and the phases of the gait cycle,and to model the surface EMG signal based upon those different phases. In this paper, we describea novel method to produce a model of the surface EMG signal during the gait cycle, which covers thecharacteristics of the surface EMG, such as the fiber distribution, motor unit type, location and recruitment,tissue anisotropy, electrode configuration and phases of the gait cycle.

2. Simulation model

2.1. Physiological basis of the model

The muscles are composed of many muscle fibers, which are united into functional units called motorunits. The motor unit is the basic level of nervous system of the muscle. The motor unit has many branchesand innervates a great deal of muscle fibers. Various lengths and diameters of the branching nerve fibersthat spread to each muscle fiber cause the time that the nerve action potentials arrive at the endplate tovary. As a result, the activations of muscle fibers within a given motor unit are asynchronous. Gatheringthe different temporal muscle fiber action potentials comes into being motor unit action potential. Motorunit recruitment territories are imagined as small, overlapping circles on a cross-section of muscle. Whena muscle is activated, motor unit recruitment abides by the size principle. The smaller motor units arerecruited at the beginning of the muscle contraction occurrence, and subsequently, the larger motor unitsare activated, which play an important role in increasing the surface EMG signals. In general, the rangeof firing rate of human muscle fibers is between 8 and 50 Hz [14]. Accompanying the increase of musclecontraction, the firing rate of motor units changes from the lower to the higher value. Both the additionalmotor unit recruitment and the increase in firing rate result in stronger surface EMG signals.

The gait cycle is a main parameter of this simulation model. When the body attempts to move forward,one limb performs the support function while the other one moves itself to the new support site. Then,the two limbs exchange their roles. Walking is composed of a repetitious sequence of limb functionsboth moving the body forward and maintaining body stability. A single sequence of these functionsaccomplished by one limb is defined as the gait cycle [15]. There is no specific onset or cessation pointin a gait cycle. In our model, the initial floor contact with heel is selected as the onset of the gait cyclebecause the moment of initial contact is easier to identify than other events. In other words, a completegait cycle begins when the heel of one foot contacts the ground and ends when it contacts the groundagain. Each gait cycle is made up of two periods, stance phase and swing phase. Stance phase describesthe period during which the foot is on the ground. Correspondingly, the swing phase presents the period

W. Wang et al. / Computers in Biology and Medicine 36 (2006) 601–618 603

Fig. 1. The multiple-layer model comprising muscle, fat and skin layer.

during which the foot is in the air and the limb moves forward. Stance phase starts at the initial contactand is divided into four intervals, loading response, mid-stance, terminal stance and pre-swing. Swingphase begins when the foot is lifted off the ground and is separated into three intervals, initial swing,mid-swing and terminal swing. A gait cycle and its phases are shown in Fig. 2.

2.2. Structure of the model

Limb muscles, such as the tibialis anterior and gastrocnemius medialis, are treated as the target musclesbecause the purpose of this work is to simulate the surface EMG signal during the gait cycle. As most ofthe recent models for limb muscles [5], these muscles are described as a cylindrical volume conductorcomposed of anisotropic media which are representative of skin, fat and muscle tissue. The multiple-layerstructure is shown in Fig. 1.

2.3. Method to generate the surface EMG signal from the model

The model is based on the single fiber action potential generation, which depends on the fiber conductionvelocity. The motor unit action potential is then estimated by summation of all fiber activities assumingthat they have a uniformly parallel distribution in each motor unit. The surface EMG of muscle voluntarycontraction is calculated by grouping the motor unit action potentials on the basis of their different sizes,positions, recruitment periods and sequences. The surface EMG during a gait cycle is simulated by fixingthe values of motor unit action potential trains in accordance with features of the gait phases, such asactivation timing and intensity of muscle contraction. In addition, the characteristics of the acquisitionsystem, such as location and configuration of the electrodes, sampling frequency and signal-to-noise ratio(SNR) are also taken into account.

The effect of non-homogeneous tissues on the surface EMG signal is considered by using the boundaryconditions. These state that normal components of the current density must be continuous at a boundarybetween two conducting media.

The expression of boundary condition can be written as

�1En1 = �2En2, (1)


where En1, En2 are the normal components of the electric field in the first and second media; �1, �2 arethe first and second media conductivities [8,16,17].

Let us now analyze the individual elements of the simulation model.

2.3.1. Single fiber action potential simulationThe depolarization of the muscle fiber produces the action potential, and this charge in transmembrane

potential travels along the muscle fiber at a velocity in the range from 1 to 5 ms−1, with amplitude ofapproximately 100 mV [18]. The single fiber action potential is simulated by using Dimitrov’s assumption.Dimitrov et al. [10] indicated that a muscle fiber can be described as a polarized shell. The assumption oftwo stacks of double-layer disks can be used to simulate the fiber excitation, thus, the single fiber actionpotential Vf(x, y, z) at an observation point can be expressed as [10]

Vf(x, y, z) = − �i

4��m

∫s

ds

∫ ∞

−∞�ei(z)

�z

�(1/r)

�zdz, (2)

ei(z) = 96z3e−z − 90, (3)

where ei(z) is the intracellular potential of the fiber; z is the axial direction, in mm [19]; s is the fibersection; �i is the intracellular conductivity; �m is the muscle conductivity,

�m = √�y�z, (4)

where �y is the muscle radial conductivity, �z is the muscle axial conductivity and r is the distancebetween the fiber sections to the observation point.

There is a straight relation between space (z) and time (t) on the basis of the conduction velocity (v),z = vt . The distribution of the intracellular potential ei(z) along the fiber can be expressed by the time t ,

ei(t) = 96(vt)3e−vt − 90. (5)

2.3.2. Motor unit action potential simulationThe motor unit action potential is calculated by summation of all of the single fiber action potentials

in time domain. The parameters, such as the length of the fibers, position of the fiber endplate, number ofthe muscle fibers and recruitment territories of the motor unit are basic variables for computing the motorunit action potential. All parameter values of muscle fiber and its distribution used to estimate the motorunit action potential are given in Table 1. Each muscle fiber action potential at the observation point canbe calculated by Eq. (2) on the basis of the various values of parameters. Our model assumes that allfibers in a muscle have a constant diameter and run uniformly parallel to the central axis of muscle andskin surface. The number of muscle fibers contained within each motor unit may be enumerated by thevalue equal to the area of motor unit divided by the area of muscle fiber. The motor unit action potentialcan be expressed as

Vmu =Nf∑i=1

Vfi , (6)

where Vfi is each single fiber action potential contained in a given motor unit and Nf is the number ofmuscle fibers.


Table 1The parameters of the simulation model

Parameters Value

Muscle axial conductivity �z = 0.328 S m−1 [26]Muscle radial conductivity �y = 0.063 S m−1 [26]Intracellular conductivity �I = 1.010 S m−1 [26]Skin conductivity �s = 4.55 × 10−4 S m−1 [27]Fat conductivity �f = 0.0379 S m−1 [27]Conduction velocity v = 4 m s−1 [14]Muscle radius R Tibialis anterior: Rt = 20 mm

Gastrocnemius medialis: Rg = 30 mm [15]Thickness of fat and skin hfat = 3 mm, hskin = 1 mmFiber diameter d Tibialis anterior: dt = 57 �m

Gastrocnemius medialis: dg = 54 �m [15]Muscle fiber length Mean value: LM = 100 mm; [20]

Length variation: Gaussian distribution: SD = 1 mm,Mean = 0;

Fiberendplate position Gaussian distribution: [2]Mean = 0, SD = 1 mm, range = ±3 mm

Fiber position Uniform distribution:range ±1 motor unit radius

Motor unit size Circle shape, Poisson distribution: [14]• Tibialis anterior range 5–10 mm, lambda = 6 mm;• Gastrocnemius medialis range 5–10 mm, lambda = 8 mm.

Motor unit recruitment sequence From small size (diameter = 5 mm) to large size(diameter = 10 mm)

Motor unit position Uniform distribution:range ±1 muscle radius

Motor unit firing rates Range from 8 to 90 Hz. [20]At the beginning of maximum muscle contraction,range 70–90 Hz.Steady voluntary contraction:

• Tibialis anterior Poisson distribution, range 8–40 Hz, lambda = 12 Hz;• Gastrocnemius medialis Poisson distribution, range 8–40 Hz, lambda = 12 Hz.

Motor unit starting time Uniform distribution:range from 0 to 1/(mean firing rate)

Gait cycle period Period = 1.03 ± 0.08 s [18][15,25]

Muscle activation timing Mean value: on-time =58%, off-time =9%, stronger• Tibialis anterior contraction time =95%; time variation: Gaussian

distribution, mean =0, SD = ±5%.Mean value: on-time =9%, off-time =50%, strongercontraction time =38%; time variation: Gaussian

•Gastrocnemius medialis distribution, mean =0, SD = ±5%.

The motor unit recruitment territories are assumed as small and overlapping circles. The territories ofthe motor units have cross-sectional diameters of 5–10 mm, with 10–25 units overlapping with each other[14,20,21]. The diameters of motor units are assumed to be a Poisson distribution [2]. The distribution ofmotor units in the section of tibialis anterior muscle is shown in Fig. 6(a).


2.3.3. Motor unit action potential trains and voluntary contraction simulationThe different types of muscle activities are caused by the various actions of the motor units contained

within the muscle. The surface EMG signal is determined by motor unit action potential trains thatare constituted by an asynchronous series of action potentials that vary in amplitude and duration. Theactivities of the motor units can generally be described by parameters such as size, position, starting timeand firing rate as well as sequence of activation. For a given muscle, each motor unit action potentialtrain can be estimated by firing rate, start timing and the motor unit action potential that is calculated bysummation of all single fiber action potentials within this motor unit. In this model, the range of the firingrate between 8 and 90 Hz is divided into two parts, high firing rate and low firing rate. High firing rateoccurs at the beginning of maximum muscle contraction, which has a uniform distribution between 70and 90 Hz. Low firing rate happens at the muscle steady voluntary contraction, which is described by thePoisson distribution with the range from 8 to 40 Hz, with a mean value equal to 12 Hz. The order of motorunit activation is decided by their sizes and is described by a general criterion that the motor units havinga small size are activated earlier than the large motor units. Therefore, the motor unit that has a smallerdiameter may be recruited first; the motor unit with a larger diameter may be activated subsequently.The start time of motor unit activation can be computed by the parameters both mean of firing rate andnumber of motor units contained in a given muscle. The start time (tst) of each motor unit activation canbe expressed as the equation

tst = n�tint = n1/fr

Nmu, (7)

where n=1, 2, . . . , Nmu; Nmu is the number of motor units within a given muscle; fr is the mean of firingrate, in this case, default value 1/fr = 83 ms.

In voluntary contraction, the activities of motor units are considered as random with no synchronizationprocess and therefore the starting times of motor unit activation are not identical. The relation of motorunit action potential trains (Vmuts) and single motor unit action potential train (Vmuti) can be illustratedby the array format:

Vmuts =

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

Vmut1(tst1, fr1, Vmu1)

Vmut2(tst2, fr2, Vmu2)

···

Vmuti(tsti , fri , Vmui)

···

Vmutn(tstn, frn, Vmun)

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

, (8)

where Vmui can be obtained by Eq. (6); fri is the firing rate that obeys the Poisson distribution. Forinstance, the firing rate of motor unit within the muscle, tibialis anterior, has a Poisson distribution witha range from 8 to 40 Hz and mean value 12 Hz when it is in the steady voluntary contraction. As a result,the simulated surface EMG signal during voluntary contraction is the summation of all motor unit actionpotential trains in the time domain.


Fig. 2. The timing of muscle activation and location of muscle maximum contraction during the different gait phases. The onsetand cessation points of the muscle activities are described by the percentage of the gait cycle. For the tibialis anterior muscle, theonset and cessation points of muscle activities are 58% and 9% of a gait cycle; a large force contraction occurs at approximately95% of a gait cycle. For the gastrocnemius medialis muscle, the onset and cessation points of muscle activities are 9% and 50%of a gait cycle, a maximum force contraction occurs at approximately 38% of a gait cycle. The white boxes at the beginning andthe end of each bar represent the standard deviation; value is ±5%.

2.3.4. Surface EMG simulation during the gait cycleDuring a gait cycle muscle activities such as voluntary contraction and relaxation are alternately

brought into being by activating or discharging the motor units and modulating the firing rates. In thissimulation model, increasing the firing rates of the motor units is the main mechanism to increase themuscle activity at the starting time of muscle contraction; the muscle force will continue to increaseby recruiting additional motor units during steady contraction. At the same time, the firing rates have arelated drop so that the muscle force increases smoothly [22]. At the start of muscle contraction, the orderof the motor unit’s activation is based upon the sequence of the motor units’ size [23]. The firing rates arein the frequency range from 70 to 90 Hz at the beginning of the maximum muscle contraction [20,24].During steady contraction, the firing rates descend to the frequency range from 8 to 40 Hz, with a meanvalue of 12 Hz, and additional motor units participate to the muscle contraction to produce an increase inactivity [20]. In addition, all motor units within the muscle are recruited simultaneously in order to obtainthe maximum contraction. In the model presented here approximately 400 motor units, each containing612 muscle fibers, are synchronously activated to produce the maximum voluntary contraction. Fig. 2illustrates the timing of muscle activation and intensity of muscle contraction during the different gaitphases [15,25].

Table 1 lists the values of the parameters used in our simulation model.

3. Simulation results

3.1. Simulated single fiber and motor unit action potentials

The simulated single fiber action potential of the tibialis anterior muscle is shown in Fig. 3. The fiberdiameter and length are assumed respectively to be 57 �m and 100 mm. The fiber endplate is assumedto be in the origin of the axis (coordinates (0, 0, 0)). The position of the electrode is described by the


Fig. 3. Single fiber action potential of the tibialis anterior muscle. The amplitude values shown in y-axis are normalized to themaximum value of amplitude.

Fig. 4. Motor unit action potential of the tibialis anterior muscle. The amplitude values shown in y-axis are normalized to themaximum value of amplitude.

coordinates (0, Rm + hf + hs, 10 mm), where Rm = 20 mm, hf = 3 mm, hs = 1 mm; the center of themuscle is the origin of the coordinate system (0, 0, 0). The motor unit action potential can be simulatedby summation of all of the single fiber action potentials in the time domain. The simulated motor unitaction potential of the tibialis anterior muscle is shown in Fig. 4. In this case, the motor unit includes 612muscle fibers, which run parallel to the surface skin. The endplates of muscle fibers in the axial directionis assumed to have Gaussian distribution in the range ±3 SD with zero mean value and SD = 1 mm [2].The fiber endplates in the radial direction are a uniform distribution in the range of ±1 motor unit radius.The fiber length variation has a Guassian distribution with zero mean value and SD = 1 mm.


Fig. 5. The influence of the motor unit distribution and features of recruitment on its potential. (a) Six motor unit distributionwithin the muscle. The positions are (7, 15, 0), (0, 10, 0), (−8, 5, 0), (0, 0, 0), (−7, −7, 0) and (5, −10, 0); the diameters are 6,8, 9, 6, 7 and 10 mm; firing rates are 8, 10, 13, 17, 20 and 25 Hz; the starting times are 0, 15, 30, 35, 50 and 65 ms. (b) Motorunit action potential trains based upon the different parameters of motor units.

3.2. Simulated surface EMG for voluntary muscle contraction

The surface EMG represents the activation of the motor units. Different combinations of active motorunits determine the various muscle activities such as reflex and voluntary contractions. The shape andpower range of the surface EMG are influenced by the parameters of recruitment and distribution ofthe motor units contained within the muscle. The effects of the motor units’ parameters on their actionpotentials are illustrated in Fig. 5(a) and (b). Fig. 5(b) shows six motor unit action potential trains oflength 1000 ms each, which are related to the distributions of the six motor units given in Fig. 5(a) havingdifferent sizes, positions, starting-times of contraction, and firing rates.

Motor unit action potential trains and the distribution of their positions and firing rates within themuscle in voluntary contraction are shown in Fig. 6(a) and (b). The number of motor units is 400. Thediameters of the motor units is assumed to be a Poisson distribution with a range from 5 to 10 mm, lambdais 6 mm. The position of the motor units is considered to be a uniform distribution with a range of between±1 muscle radius. The number of motor units and the size of a given muscle determine the overlapping.The motor unit firing rates of the tibialis anterior in the voluntary contraction are assumed as a Poissondistribution with the range from 8 to 40 Hz, mean value is 12 Hz. The small motor units are recruitedbefore large ones. The interval of recruitment is a uniform distribution within the value 1/(mean firingrate).

Fig. 7 shows the simulated surface EMG signals of the tibialis anterior muscle in voluntary contractionand its power spectral density. The surface EMG is computed by summation of 400 motor unit actionpotential trains in the time domain. The frequency content of the simulated surface EMG signals is from8 to 150 Hz, mainly concentrated in the range of 10–30 Hz.


Fig. 6. The motor unit action potential trains of the tibialis anterior muscle. The amplitude values shown in the y-axis (b) arenormalized to the maximum value of amplitude. (a) Motor unit distribution in the tibialis anterior muscle. (b) Motor unit actionpotential trains.

Fig. 7. The simulation EMG signals of the tibialis anterior muscle in voluntary contraction and corresponding power spectraldensity (400 motor units). The amplitude values shown in the y-axis are normalized to the maximum value of amplitude,respectively.

3.3. Simulated surface EMG signals during gait cycle

The simulated surface EMG signals, assuming collection from the tibialis anterior and gastrocnemiusmedialis muscles during the gait cycle, are illustrated in Fig. 8. The simulated surface signals are normal-ized by 100% gait cycle, which is defined by two consecutive heel-down signals. In this simulation, thetiming of muscle activation is described by the percentage of the gait cycle, the tibialis anterior muscle isactivated between from 58% to 9%, the stronger contraction occurs at approximately 95%; the gastroc-nemius medialis is activated between from 9% to 50%, the stronger contraction occurs at approximately38%; for the each gait cycle, there is ±5% standard deviation at the timing of muscle activation. Theintensity of the muscle contraction is simulated by the summation of the different motor unit action po-


Fig. 8. The simulated surface EMG signals of the tibialis anterior and gastrocnemius medialis muscles during the gait cycle,including four cycles. SNR=20 dB. The amplitude values shown in the y-axis are normalized to the maximum value of amplitude,respectively.

tential trains, which is calculated by changing the number and firing rate of the motor unit recruitment.The details are explained earlier in the method of model generation.

The simulated surface EMG signal during one gait cycle and its characteristics are shown in Fig. 9 (tib-ialis anterior) and Fig. 10 (gastrocnemius medialis). In clinical experiments, the surface EMG signals arealso detected from the two lower limb muscles, tibialis anterior and gastrocnemius medialis, during walk-ing. The signals are recorded by the BIOPAC system (MP100 acquisition system unit, AcknowledgeTM

software and surface EMG electrodes) and sampled with a sampling rate equal to 1000 Hz. A band-pass filter (lower and upper cutoff frequencies are 8 and 500 Hz, respectively) is also implemented. Thesurface EMG signals of tibialis anterior and gastrocnemius medialis muscles collected by the clinicalexperiment and their features, such as the timing and intensity of muscle activation and power spectraldensity, are shown in Fig. 11. Comparison of the real surface EMG signals collected from the tibialisanterior and gastrocnemius medialis muscles in the gait experiment with simulated surface EMG signalson the features of the timing of muscle activation and power spectral density of the signals, is presented inTable 2, and illustrates that the simulated signals can reproduce surface EMG signals due to various motorunit action potential trains determined by the anatomical conditions.

4. Discussion

In this paper we proposed an approach for modelling the generation of the surface EMG signal duringthe gait cycle. Important features of the model include the multiple-layer volume conductor, motor unitaction potential trains determined by the anatomical conditions, electrode position, and characteristicsof gait phases, sequence and firing rate of the motor unit during the walking. The surface EMG signalsduring the gait cycle are simulated by modulating the parameters of the model extracted and determinedby the features of the real data detected from the gait experiment. We demonstrated that the motorunit action potential trains produced by this model can be changed according to the specific anatomical


Fig. 9. The simulated surface EMG signal of tibialis anterior muscle during one gait cycle and its characteristics. The amplitudevalues shown in the y-axis are normalized to the maximum value of amplitude, respectively. (a) The simulated surface EMG signalof the tibialis anterior muscle during one gait cycle and the timing of muscle activation. (b) The three-gray-level quantization ofthe instantaneous power of the simulated surface signal shows the localization of the signal peak and the on- and-off timing ofmuscle activation. (c) The power spectral density of the simulated surface EMG signal.


Fig. 10. The simulated surface EMG signal of gastrocnemius medialis muscle during one gait cycle and its characteristics. Theamplitude values shown in the y-axis are normalized to the maximum value of amplitude, respectively. (a) The simulated surfaceEMG signal of gastrocnemius medialis muscle during one gait cycle and the timing of muscle activation. (b) The three-gray-levelquantization of the instantaneous power of the simulated surface signal shows the localization of the signal peak and the on-and-off timing of muscle activation. (c) The power spectral density of the simulated surface EMG signal.


Fig. 11. The surface EMG signals during the one gait cycle and their characteristics, which are collected from the tibialis anteriorand gastrocnemius medialis muscles during the gait experiment. Sample frequency is 1000 Hz. A bandpass filter is used in thesignal acquisition; the low and high cutoff frequency is 8 and 500 Hz, respectively. The amplitude values shown in the y-axis arenormalized to the maximum value of amplitude, respectively. (a) The surface EMG signal of the tibialis anterior muscle duringone gait cycle and the timing of muscle activation. (b) The three-gray-level quantization of the instantaneous power of the surfacesignal of the tibialis anterior muscle shows the localization of the signal peak and the on- and-off timing of muscle activation.(c) The power spectral density of the surface EMG signal of the tibialis anterior muscle. (d) The surface EMG signal of thegastrocnemius medialis muscle during one gait cycle and the timing of muscle activation. (e) The three-gray-level quantizationof the instantaneous power of the surface signal of the gastrocnemius medialis muscle shows the localization of the signal peakand the on- and-off timing of muscle activation. (f) The power spectral density of the surface EMG signal of the gastrocnemiusmedialis muscle.


Fig. 11. (continued).

conditions and acquisition system parameters. The simulated surface EMG signals are achieved gatheringmotor unit action potential trains.

A simple approach to present this model is suitable for analysis of the surface EMG signals is tocompare the features of simulated surface signals with the parameters of the real signals in time domain.


Table 2The characteristics of the simulated surface EMG signals and the real surface EMG signals collected during the gait experiment

Muscles Timing of muscle activation Powerspectraldensity

Onset Offset Strongercontraction

Range(Hz)

Mean Std Mean Std Mean Std(%) (%) (%) (%) (%) (%)

Simulated surface EMG signals Tibialisanterior

60.5 2.1 6.5 2.4 93.1 2.0 Rangefrom 8to 50 Hz,focus on15–25 Hz

(SNR = 10 dB) Gastrocnemiusmedialis


Real surface EMG signals Tibialisanterior


Gastrocnemiusmedialis


For example, the onset and cessation time of muscle activation and the power of signals displaying thevarious activities of muscle contraction during a gait cycle. The simulated and real surface EMG signalsof tibialis anterior and gastrocnemius medialis muscles and their characteristics, such as the timing andintensity of muscle activation in time domain, are respectively given in Figs. 9–11. In contrast to the above-mentioned temporal method, the approach to transform the surface EMG into its frequency representationwhere the various frequency and amplitude can be used to describe the EMG is efficient to interpret thefeatures of the surface EMG. In the current work, we applied the power spectral density method to extractfrequency components and their power distribution from the surface EMG. The power spectral density ofthe simulated and real surface EMG signals of tibialis anterior and gastrocnemius medialis is respectivelyshown in Figs. 9–11.

In the model, the influence of the electrode size and shape is not considered; the position of electrode,which is modelled as point electrode, is introduced as this has more important effect for estimating thesurface EMG signals than electrodes’ size, because previous studies have reported only minor differencesin the recorded waveform when electrode size is changed, and these minor differences disappear alongwith the increasing distance between the electrode and muscle fiber [8,28]. In the current status, the


simulation model only produces the surface EMG during the normal adult gait cycle, but it has potentialin helping to assess the algorithms or methods that are used to analyze the surface EMG signals. Due toits existing modular parameters that can be modified and altered to suit the different requirements, thismodel is intended to assess the novel algorithms that are utilized to process pathological surface EMGsignals during abnormal walking.

5. Conclusion

This paper presents a method for simulation of the surface EMG signals of an asymptomatic subjectduring the gait cycle, which describes the anatomical conditions, features of locomotion and detectionsystem on the basis of the multiple-layer model. This model can provide researchers with an efficienttool to assess the implementation of new algorithms, such as estimating the onset and cessation timing ofmuscle activation during walking. The model could also have potential in combination with pathologicalinformation, such as the characteristics of the surface EMG signals obtained from children with cerebralpalsy, to investigate the effects of cerebral palsy on the surface EMG during walking and to produce asuitable simulation model to evaluate novel algorithms or methods that are utilized to analyze the surfaceEMG signals collected during gait of both asymptomatic and symptomatic subjects.

Acknowledgements

Wei Wang is supported by a scholarship of the Institute of Sound and Vibration Research and anOverseas Research Students Award, University of Southampton. We would like to thank Dr VictoriaYule,School of Health Professions and Rehabilitation Sciences, University of Southampton, for playing animportant role in the collection of surface EMG signals during walking.

References

[1] S.D. Nandedkar, E. Stalberg, Simulation of single fiber action potentials, Med. Biol. Eng. Comput. 21 (1983) 158–165.[2] J. Duchêne, J.Y. Hogrel, A model of EMG generation, IEEE Trans. Bio-med. Eng. 47 (2000) 192–201.[3] R. Merletti, L. Lo Conte, E. Avignone, P. Guglielminotti, Modeling of surface myoelectric signals—Part I: model

implementation, IEEE Trans. Bio-med. Eng. 46 (1999) 810–820.[4] R. Merletti, S.H. Roy, E. Kupa, S. Roatta, A. Granata, Modeling of the surface myoelectric signals—Part II: model-based

signal interpretation, IEEE Trans. Bio-med. Eng. 46 (1999) 821–829.[5] K. Roeleveld, J.H. Blok, D.F. Stegeman, A. Van Oosterom, Volume conduction models for surface EMG: confrontation

with measurements, J. Electromyography Kinesiol. (1997) 221–232.[6] D. Farina, R. Merletti, A novel approach for precise simulation of the EMG signal detected by surface electrodes, IEEE

Trans. Bio-med. Eng. 48 (2001) 637–647.[7] D. Farina, L. Mesin, S. Martina, R. Merletti, A surface EMG generation model with multilayer cylindrical description of

the volume conductor, IEEE Trans. Bio-med. Eng. 51 (2004) 415–426.[8] M.M. Lowery, N.S. Stoykov, A. Taflove, T.A. Kuiken, A multiple-layer, finite-element model of the surface EMG signal,

IEEE Trans. Bio-med. Eng. 49 (2002) 446–455.[9] D.F. Stegeman, J.H. Blok, H.J. Hermens, K. Roeleveld, Surface EMG models: properties and applications, J.

Electromyography Kinesiol. 10 (2000) 313–326.[10] G.V. Dimitrov, N.A. Dimitrova, Precise and fast calculation of the motor unit potentials detected by a point and rectangular

plate electrode, Med. Eng. Phys. 20 (1998) 374–381.[11] J.N. Helal, P. Bouissou, The spatial integration effect of surface electrode detecting myoelectric signal, IEEE Trans.

Bio-med. Eng. 39 (1992) 1161–1167.


[12] S.D. Nandedkar, E. Stalberg, D.S. Sanders, Simulation techniques in electromyography, IEEE Trans. Bio-med. Eng. 32(1985) 775–785.

[13] D.W. Stashuk, Simulation of electromyographic signals, J. Electromyography Kinesiol. 3 (1993) 157–173.[14] J.R. Cram, G.S. Kasman, J. Holtz, Introduction to Surface Electromyograph, Aspen Publishers, Gaithersburg, MD, 1998.[15] J. Perry, Gait Analysis: Normal and Pathological Function, SLACK, Thorofare, NJ, 1992.[16] L.D. Landau, E.M. Lifshitz, Electrodynamics of Continuous Media, Pergamon Press, Oxford, England, 1984.[17] H.C. Ohanian, Classical Electrodynamics, Allyn and Bacon, Newton, MA, 1988.[18] B.H. Brown, P.V. Lawford, R.H. Smallwood, D.R. Hose, D.C. Barber, Medical Physics and Biomedical Engineering,

Institute of Physics Publishing, London, 1999.[19] P. Rosenfalck, Intra- and extracellular potential fields of active nerve and muscle fibers, Acta Physiol. Scand. 321 (1969)

1–168.[20] A.J. McComas, Skeletal Muscle Form and Function, Human Kinetics, Leeds, 1996.[21] R.E. Burke, D.N. Nevine, P. Tsairis, F.E. Zajac, Physiological types and histochemical profiles in motor units of the cat

gastronemius, J. Physiol. (1973) 723–748[22] H. Broman, C.J. DeLuca, B. Mambrito, Motor unit recruitment and firing rates interaction in the control of human muscles,

Brain Res. 337 (1985) 311–319.[23] E. Henneman, G. Somjen, D.O. Carpenter, Functional significance of cell size in spinal motoneurones, J. Neurophysiol.

28 (1965) 560–580.[24] J. Tanji, M. Kato, Discharge of single motor units at voluntary contraction of abductor digiti minimi muscle in man, Brain

Res. 45 (1972) 590–593.[25] J.R. Gage, Gait Analysis in Cerebral Palsy, Mac Keith Press, London, 1991.[26] S. Andreassen, A. Rosenfalck, Relationship of intracellular and extracellular action potentials of skeletal muscle fibers,

Crit. Rev. Biomed. Eng. 6 (1981) 267–306.[27] C. Gabriel, S. Gabriel, E. Corthout, The dielectric properties of biological tissues: I literature survey, Phys. Med. Biol. 41

(1996) 2231–2249.[28] A.J. Fuglevand, D.A. Winter, A.E. Patla, D. Stashuk, Detection of motor unit action potentials with surface electrodes:

influence of electrode size and spacing, Biol. Cybernetics 67 (1992) 143–153.

Wei Wang received the B.E. degree in Biomedical Engineering and the B.A. degree in Economic Law from Tianjin Univer-sity, P.R. China, in 2001. She is currently working for her Ph.D. degree in Biomedical Engineering at the Institute of Soundand Vibration Research, University of Southampton, United Kingdom. Her main research interests include biomedical signalprocessing—modelling and analysis of the surface EMG during dynamic movements. Her recent research focuses on analysisof muscle activity during the gait cycle based upon simulation and measurement of the surface EMG signal.

Antonio De Stefano is Distance Learning Coordinator for the Master Training Packages in Biomedical Signal Processing atthe ISVR in the University of Southampton, UK. He received the Electronics Engineering 5 Years Degree with specializationin Biomedical Sciences and the ChEng qualification from the University Federico II in Naples (Italy), and PhD from the ISVRin the University of Southampton (UK). His research interests include wavelet-based noise reduction and image enhancement,biomedical signal processing, implementation of distance learning packages for biomedical subjects, techniques for EMGanalysis during walking of pathological children, and mechanical models for speech design. He has over 30 publications injournals and international conferences.

Robert Allen holds a Personal Chair in Biodynamics and Control at the Institute of Sound and Vibration Research (ISVR),University of Southampton, UK. Following training in the machine tool industry and a period as numerical control programmeengineer he studied at the University of Leeds and graduated with BSc (Hons) in Control Engineering and was awarded a PhDfor research on modelling the dynamic characteristics of neural receptors. He is currently Editor-in-Chief of the internationaljournal Medical Engineering and Physics (http://www.elsevier.co.uk/), a member of the Association of Institutions concernedwith Medical Engineering and a member of the Publications Committee (Institute of Physics and Engineering in Medicine). Heis a Fellow of the Institution of Electrical Engineers, the Institution of Mechanical Engineers and the Institute of Physics andEngineering in Medicine. He is also a Member of the Institute of Electrical & Electronics Engineers.

http://www.elsevier.co.uk/

Documents

A simulation model of the surface EMG signal for analysis of muscle