1. Introduction
Upper-limb prostheses are extremely helpful devices for people with amputations. By using surface electromyography (sEMG), so-called myoelectric prostheses can move actively. There are already various sEMG electrodes available on the market, and numerous research articles have focused on pattern recognition to enable movement at a high level of dexterity [
1,
2]. However, due to unsatisfactory robustness against various interferences [
3], most of these systems have not yet been applied for daily use. The amputee must choose between a basic system with low functionality and a system that enables high dexterity but requires more cognitive effort [
4]. If the system is unstable or requires excessive cognitive effort, amputees might switch to passive use or even reject the prostheses entirely, which leads to detrimental effects on the contralateral limb due to overuse [
5]. Robust control is highly important to result in great acceptance of the myoelectric prosthesis [
6]. The level of dexterity must be balanced such that the prosthesis can become part of the patient’s body [
7].
Most research into prostheses and exoskeleton control has concentrated on state-of-the-art sEMG electrodes, which need a conductive connection to the skin [
8,
9,
10]. These electrodes have various limitations due to the effects of skin perspiration, hair and fat content in the skin, and may cause skin irritations. Amongst others, these limitations lead to reduced stability of the sEMG electrodes. Hence, Cho et al. [
11] and Radmand et al. [
6] suggested using force myography as an alternative to sEMG. Connan et al. [
3] proposed multimodal sensing to increase the robustness and reliability of myocontrol. They combined force- and electromyography in a device for myocontrol. Fougner et al. [
12] combined EMG and accelerometers for hand motion classification.
Rather than alternatives or multimodal sensing approaches, this paper presents a capacitive EMG sensor that overcomes the limitations associated with state-of-the-art conductive sEMG electrodes. Capacitive sensors, also known as insulated electrodes, are insulated against direct contact to the skin. The electrolytic layer that is required in conductive electrodes is formed by an electrolytic gel or by a sweat film. In capacitive electrodes, this layer is replaced with a dielectric film. Due to the electrode’s design, no ohmic contact exists, so only displacement currents occur. Furthermore, no polarization effects arise, as there is no net charge flow from skin to electrode. Another advantage is that these sensors can be applied without skin preparation. A bio-compatible sensor design is straightforward, as only the insulating layer is in contact with the skin. Prance [
13] compared wet, dry and active insulated electrodes for biopotential sensing. She pointed out the problem of movement artifacts, which occur with all electrodes.
Unlike for conductive sEMG sensors, little research has been published on capacitive biopotential sensors. Since electrocardiography (ECG) has higher signal amplitudes than EMG, most insulated biosignal sensors have been developed for ECG [
14,
15,
16]. The idea of measuring biopotentials with capacitive sensors is old, but continues to offer great potential. As early as 1967, Richardson et al. [
17] patented an electrocardiography and bioelectric capacitive electrode using a transistor circuit for impedance conversion. In recent research, several systems have been suggested for capacitive biopotential sensing. These sensors have been developed for various applications, such as prosthesis control and exoskeletons, and use different techniques to deal with bias currents at the input. Various methods have been proposed for referencing to the body to reduce the subject’s common-mode voltage. Shielding is also dealt with, as it plays an important role in insulated biopotential electrodes.
Insulated electrodes allow biopotentials to be measured even through thin cloth or across a small air gap. Lim et al. [
16] presented a system for measuring the ECG signals of a subject who is sitting on a chair and wearing clothes. This system requires a large ground plate integrated into a seat surface of the chair. Clearly, integration of such a large ground plate is impossible in other applications such as upper-limb prostheses. Lee et al. [
14] also developed a capacitive sensor for measuring ECG signals through cloth, where the active electrode was designed to be thin and flexible. They implemented an active shield to reduce stray capacities. A capacitive driven-right-leg electrode with maximized surface area on the chest belt was used to reduce the subject’s common-mode voltage. In their work, motion artifacts distorted the signals; these artifacts limit biopotential measurement. As long as the signal is in the operating range of the analog circuit, the signal peaks can be identified after bandpass filtering. Further, Spinelli et al. [
15] and Ueno et al. [
18] measured biopotentials through cloth. The former [
15] provided a detailed description of the input circuit and the latter [
18] showed ECG and EMG measurement through cloth. Insulated sensors have even been used in underwater applications [
19]. Oehler [
20] developed a capacitive ECG sensor and a capacitive electroencephalography (EEG) helmet in his dissertation. Loong et al. [
21] compared a variety of textile materials used as insulators for biopotential sensing. However, the sensing area itself was not textile but a rigid copper plate.
Various ways of applying insulated biopotential sensors have been described in the literature. They can be used, for example, for ECG, EEG and EMG. Here, we present sensors developed for capacitive EMG measurement. They can be employed in basic systems which allow simple movements of a prosthesis, in an array enabling a high dexterity level, or even in combination with other measuring methods. We focused on an application for a standard upper-limb prostheses available on the market and provided by Otto Bock Healthcare GmbH. This standard prosthesis uses EMG sensors at two muscle groups. By co-contraction, the movement mode of this prosthesis can be switched; this is achieved by simultaneous short, strong contraction of both muscle groups. Clearly, this sensor can be adjusted to other applications.
Our insulated sensor is highly robust and designed for clinical application. Various sensor assemblies were investigated for optimal signal coupling, using flexible materials to achieve good adaption to the anatomy of the human forearm. The resulting measurement system is compact, but well thought out. It aims for ease of use, long-term application, low-noise performance and stability and usability in a real-world environment. We presented a previous prototype of our measurement set-up [
22] and its digital signal processing path [
23].
First, we present the fundamentals of the conductive and capacitive sensing principles. We then describe various different sensor assemblies and the sensor electronics of our capacitive EMG measurement system. This is followed by a presentation of the transfer functions of the measurement set-up. We then compare the results from experiments using the different sensor assemblies and describe the shielding set-ups.
2. Comparison of the Capacitive and Conductive Measurement Principles
To explain the challenges associated with capacitive EMG measurement, we briefly compare the capacitive and conductive measurement principles in this section.
Figure 1 shows the principle of the capacitive and conductive measurement set-up with the coupled amplitudes. The transmission of the action potential to the human skin via the electrical network in the human tissue was covered by Roland et al. [
24].
The input amplifier stage is realized by an instrumentation amplifier (INA) with high common-mode rejection ratio (CMRR) for interference reduction. The main difference in terms of INA between conductive and capacitive measurement is that the direct current (DC) operating point is defined by the reference electrode in the former case and by the bias resistor RB in the latter case.
A stable DC operating point within the operating range of the amplifier is essential to signal acquisition. The measurement signal is transmitted via alternating current (AC) transmission behavior.
For a stable DC operating point, capacitive measurement requires a bias resistor RB through which the bias current can flow. The capacitive measurement principle forms a highpass, which attenuates the low-frequency components. Sensing of biopotentials is sensitive to low-frequency interferences, which are already attenuated due to the highpass resulting from the measurement principle.
The signal is not only attenuated in the low-frequency components, but across the entire frequency spectrum. The smaller signal amplitude at insulated electrodes results from the impedances at the voltage divider, which is formed by
CC and the impedance of
CP parallel to
RB. The adjustment of these impedances is explained in
Section 3.1.1.
Conductive measurement requires no bias resistor, because the bias current path is established via the conductive connection to the skin. Therefore, the high input impedance of the amplifier can be exploited. The DC potential of the signal lines are defined by the reference electrode, which also has a conductive connection to the skin.
Despite the smaller amplitude, the capacitively coupled EMG signal can be measured with high-precision electronics.
5. Experimental
The same electronics were used in all measurements except for the components to be compared. The common-mode shielding set-up from
Figure 7 was used for these comparisons. The comb- and highpass-filtered EMG signal was applied to the DAC and measured by means of a Handyscope HS3 oscilloscope [
66].
In our set-up, the frequency independent gains were:
26 at the INA;
8 at the C internal OpAmps;
16 at the ADC (accumulation of 64 samples and division by 4);
0.75 at the digital comb filter; and
0.25 at the DAC (right-shift by two to be within the 12-bit value range of the DAC).
Note that the signal was damped in EMG frequency range by filtering and capacitive coupling. With this frequency independent amplification, this damping behavior was compensated. The frequency independent and frequency dependent gains depended on the sensor assembly used and they ranged from 342 to 1437 at the total signal chain at the peak frequency (see
Section 6.2.3).
The Handyscope HS3 [
66] was connected to the PC, and the Multi Channel software [
67] from TiePie engineering was used to display the signals. The voltage range was set to ±4 V, the sampling frequency was set to 10 kHz, and the time per division was 1 s. Thus, we stored 10 s time snippets. The measurement ran for 10 s, but only data starting from 2 s were used in the evaluation, as the test subject was starting the contraction within the first 2 s. Therefore, distortion of the onset of the muscle contraction due to signal cancellation at low forces [
68,
69,
70] and possible movement artifacts due to the relative movement of the muscle tissue to the EMG sensor were not included in the evaluation.
For the measurements, the EMG sensors were placed at the left and right human forearm. The distance between epicondylus lateralis and the ulna distal end was measured. At one third of this distance from the epicondylus lateralis, the sensor center was placed. The sensor was positioned along the longitudinal axis of the muscle, and a cuff was wrapped around the forearm for sensor fixation. It was crucial to place the sensor at exactly the same position in all measurements, as otherwise sensor positioning would have had a greater influence on the EMG signal power than the parameters to be compared (e.g., sensor assembly or shielding). The subjects performed a maximum voluntary static extension in the wrist joint by pulling back the hand against the restraint of the joint (see Jiralerspong et al. [
71]). It was also highly important to apply the cuff with the same amount of pressure in each measurement.
Figure 12 shows the measurement set-up (
Figure 12a) and the PCB print (
Figure 12b).
5.1. Normalization
The EMG power varied for different subjects, which had to be considered in the comparison. Hence, the sum of the root mean square (RMS) per person and per arm was normalized to 1 to compare sensor performance and not subject EMG power. For each subject, each RMS value was divided by the sum of the RMS values over all sensors according to:
where
are the measured RMS values of Sensor 1–6.
5.2. Comparison of Sensor Assemblies
For optimal signal coupling at the interface between electrode and skin, various multilayer-sensor constructs were compared. The sensor descriptions can be found in
Section 3.1.1. The sensor assembly was changed between measurements while the circuit board with the measurement electronics remained the same.
5.2.1. Measurement Procedure
For this study, ten able-bodied subjects aged between 23 and 62 were selected. The subjects were asked to perform maximum voluntary muscle contraction of the musculus extensor digitorum. Three measurements were conducted per subject, arm and sensor, resulting in 18 measurements per forearm. Each measurement had a duration of 10 s, the contraction onset in the first 2 s was discarded and the remaining 8 s EMG of maximum voluntary wrist extension was used in the evaluation (see
Section 5). There was a resting period of 30 s between the measurements and after every third measurement there was a longer resting period of 2 min, which was used to apply the next sensor to the forearm. The order of the sensors was randomized. Although the subjects were not blinded to the application of the different sensor assemblies, they were naive to the purpose of the experiment.
5.2.2. Data Evaluation
The data evaluation was done in Matlab
® [
72]. Eight seconds of actual EMG contraction were selected from the measurement files and the DC offset was subtracted. The three measurements of the same sensor, person and arm were stacked in an array. This stacking was performed for all measurements. One RMS value was calculated for each stack of three measurements.
The VRMSnormalized for the left and the right arm were stacked into one array each, treating left and right arm as separate subjects. The mean and the standard deviation were calculated for each sensor based on the data of all subjects.
The resulting RMS values per sensor were compared statistically to determine which sensor coupled the highest EMG RMS. The RMS values for each sensor were tested for normal distribution by means of the Chi-Square Goodness of Fit Test [
73]. To run multi-sample tests for equal variances, we employed the Bartlett Test [
74]. Finally, we used pairwise comparison to test whether the sensor means were from the same group. The
t-test [
75] for equal means was applied with the according options. The significance levels were set to 5%.
The sensor noise level caused by the sensor assembly was negligible and deviations were caused by interferences from the environment rather than by the sensor assembly. Hence, we did not calculate the signal-to-noise ratios when comparing the sensors, as it would not have helped with answering the question of which sensor has the optimal EMG signal coupling.
5.2.3. Fatigue Evaluation
To evaluate influence due to fatigue in the sensor comparison measurements, five able-bodied subjects aged between 26 and 52 were measured at the left and right forearm, resulting in 18 measurements at ten forearms. For the details of contraction and sensor placement, see
Section 5. Sensor 5 was used for all measurements. There was a resting period of 30 s after each sensor and 2 min after every third measurement to resemble the sensor comparison measurements. In the sensor comparison measurements, the mean of three measurements was calculated per sensor. Correspondingly, the mean of three subsequent measurements was calculated in the fatigue evaluation. These values were normalized to eliminate inter subject deviations and, finally, the mean
VRMSnormalized of the five subjects was calculated, resulting in six mean
VRMSnormalized values. To evaluate the fatigue, the change of the
VRMSnormalized over the maximum voluntary contractions was evaluated. The subjects were naive to the purpose of the experiment.
5.3. Shielding (Active Shield and Common-Mode Shield)
Shielding is a key element of insulated biosignal sensing electronics. Active shielding and common-mode shielding were compared in terms of EMG signal coupling and noise performance. The electronics used in the measurements differed only in the shielding (
Figure 6 and
Figure 7).
5.3.1. Shielding Measurement Procedure
For the comparison of the shielding circuits, four 26 year-old able-bodied subjects were measured at the left and at the right forearm. Three measurements per arm and subject were conducted at maximum voluntary contraction and one measurement was performed with the muscle relaxed to determine the noise. Note that the instrumentation amplifier input pins were not connected to VREF for this noise measurement to incorporate the shielding behavior at the input stage. These signals were measured with the active shield and the common-mode shield electronics. The copper sensor (Sensor 1) was used for these measurements. In the shielding measurements, the subjects were blinded to which electronics they were wearing and they were naive to the purpose of the experiment.
5.3.2. Shielding-Data Evaluation
Matlab
® [
72] was used for signal evaluation. The
VRMS of the coupled signal and the noise
VRMS were calculated. The
VRMS was normalized for each forearm to eliminate inter-subject deviations, as described in
Section 5.2.2.
The common-mode shielding and the active shielding group were compared in terms of their EMG
VRMS and their noise
VRMS. Chi-Square Goodness of Fit Test [
73] and Bartlett Test [
74] were used to test the groups for normal distribution and equal variances. The two-sample
t-test [
75] with the null hypothesis that the two groups are from populations with equal means was used with a significance level of 5%.
5.4. Proportionality to Force Level
To show the
VRMS of the EMG signal at different force levels, Sensor 5 was applied above the flexor carpi radialis. Five able-bodied subjects aged between 26 and 52 were measured at the left and right forearm, which were treated as individual subjects in the data evaluation. A digital body scale was placed between thumb and the other digits and it was squeezed at the respective force level. The force levels were set in 10% steps of the MVC. The 5 s measurement with the oscilloscope [
66] was started as soon as the respective force level was reached, so the onset of the muscle contraction was not included. Each subject performed the measurements once at each forearm and at each force level. The measurements were conducted in the order: 50%, 40%, 60%, 30%, 70%, 20%, 80%, 10%, 90%, 0%, and 100%. Resting periods of 40 s were added in between the measurements to avoid fatigue from influencing the measurements. The
VRMS was normalized to eliminate inter-subject deviations.
5.5. Wearing Comfort
Five subjects answered a survey after the sensor comparison measurements, which comprised questions about the feel of the sensor when applying the measurement setup, the feel at the beginning and at the end of the measurements. It was asked if the subjects were sweating under the sensor or if pressure marks were obtained. The subjects were asked to rank the different sensor materials according to their preference for long-term use. In the survey, the sensors were categorized into copper, textile and flex sensors.
6. Results
6.1. Measured EMG Signal in the Time- and Frequency-Domains
Figure 13a shows a typical capacitively measured EMG signal with its corresponding amplitude spectrum (
Figure 13b). The flex sensor (Sensor 5) and the electronics and software described above were used in this measurement.
6.2. Theoretical Analysis of the Measurement System
The entire sensor set-up was examined analytically and theoretical amplitudes were calculated for comparison with real-world measurements.
6.2.1. Input Stage
In
Figure 14, the calculated transfer functions of the input stage are plotted for the six different sensors. The equivalent circuit diagram and the transfer function are described in
Section 4.2.
6.2.2. Analog Bandpass and Digital Signal Processing
The transfer functions of the analog bandpass from
Section 4.3 and the digital signal processing are plotted in
Figure 15.
6.2.3. Entire Measurement Set-Up
The transfer function of the entire measurement system (for plot see
Figure 16a) was applied to the EMG input amplitude spectrum as defined in
Figure 9 to obtain the results illustrated in
Figure 16b. We computed the RMS values of the calculated amplitude spectra and normalized their sum to 1 for comparison to the real-world measurements (
Section 6.3).
6.3. Comparison of Sensor Assemblies
The normalized mean RMS values and the standard deviations of the real-world measurements as well as the theoretical calculated RMS values are plotted in
Figure 17. Since the Chi-Square Goodness of Fit Test [
73] showed normal distribution, and the Bartlett Test [
74] did not show equal variances, the two sample
t-test [
75] for equal means without the assumption of equal variances was applied. This
t-test showed that the data resulted from different groups and that only Sensors 5 and 6, i.e., the sensor with the thin Platilon
® foil and the conductive sensor, fell within the same group; at a significance level of 5%, the
p-value was 0.3. For the other group comparisons, the largest
p-value was relatively small with
; all other sensors therefore led to different coupled amplitudes.
Sensors 5 and 6 showed the highest coupled amplitudes. Sensor 6 does not have a dielectric covering the sensor area, while Sensor 5 has a very thin foil in front of the sensor area. At Sensor 5, the ratio of the coupling capacity to the parasitic capacity led to the high coupled amplitude. Sensor 4 is also a flex sensor, but—due to its capacity ratio—it exhibited the lowest amplitudes. Sensor 1 has the same dielectric covering the sensor area as Sensor 4, but the ratio of the dielectrics causes a higher signal amplitude. Sensors 2 and 3 have the same parasitic capacity, but Sensor 2 showed higher amplitudes because it has a higher coupling capacity.
6.3.1. Fatigue Evaluation
The mean
VRMSnormalized did not decrease over the 18 measurements (see
Figure 18). To resemble the sensor comparison measurements, the mean value of three subsequent measurements was calculated. No fatigue occurred in the sensor comparison measurements.
6.4. Shielding (Active Shield and Common-Mode Shield)
The normalized
VRMS of the EMG signal and the noise values for both active and common-mode shielding are plotted in
Figure 19. It can be seen that the amplitudes of the coupled EMG signal for active shielding (0.5056
VRMS) and that for common-mode shielding (0.4944
VRMS) are almost identical. The slight difference might be due to higher noise and lower parasitic capacity, but it is not statistically significant.
The mean of the normalized noise for active shielding (0.5845 VRMS) is slightly higher than that for common-mode shielding (0.4155 VRMS). The active shielding circuit feeds back both input signals via the voltage follower. These OpAmps introduce differential-mode noise, which is coupled directly to the input signal lines and further amplified.
The statistical evaluation shows that the values of all four bars are normally distributed. The active shield and the common-mode shield have equal variances in the comparison of the EMG signals and the noise. The null hypothesis of the two-sample t-test is that the data are from populations with equal means. For the EMG signal VRMS, the null hypothesis is not rejected (p-value: 0.0816). For the noise VRMS, the null hypothesis is rejected (p-value: ), and therefore the noise is higher for active than for common-mode shielding at a significance level of 5%.
The EMG signal VRMS values of the two set-ups do not differ significantly, but the noise of the active shield set-up is higher, resulting in a lower SNR than for common-mode shielding. Furthermore, in the case of interferences, such as movement artifacts, the common-mode shield has a higher CMRR. For these reasons, the common-mode shielding was implemented in the final circuit of our low-noise high-stability sensor.
6.5. Proportionality to Force Level
The
VRMSnormalized is increasing with increasing muscle force level (see
Figure 20). Muscle force level correlates with measured EMG
VRMS. The resulting curve corresponds to force–EMG relations, as shown in the literature [
76,
77].
6.6. Power Consumption
The current of the sensor was measured with 6 mA at a supply voltage of 3.7 V (=22.2 mW). This includes the
C, which requires 2.7 mA. At this measurement, the controller was set to 16 MHz clock frequency and the ADC, DAC, brown out detector and IO-pins were activated. The current of the controller includes also the integrated OpAmps for gain adaption, the non-volatile memory and the digital signal processing software. The BLE module would increase the current at 3.3 V by 1 mA in advertising mode and by 10 mA in send/receive mode [
78]. The send/receive mode is only used in the case of sensor configuration or data transfer to the PC.
6.7. Wearing Comfort
The survey showed that all sensors feel comfortable at the surface of the skin. The textile sensors (Sensors 2 and 3) have the highest wearing comfort, followed by the flex sensors (Sensors 4–6) and the copper sensor (Sensor 1) performed worst in the wearing comfort evaluation (
Table 5).
6.8. Comparison of Capacitive and Conductive Sensors
The presented capacitive EMG sensor was compared with a conductive EMG sensor, the 13E401 from Otto Bock [
79]. As described in
Section 2, the conductive measurement principle results in higher signal coupling. The 13E401 has high signal quality when there is some sweat, which forms the electrolyte between the EMG sensor and the skin. However, it takes some time to form that sweat film. The impedance of the stratum corneum decreases in the course of time [
80,
81], therefore gain level adaptions are necessary, if it is desired to exploit full operating range. When there is poor contact, e.g., due to a missing sweat film or high hairiness, the signal has lower quality. The same applies to conventional EMG sensors, such as the SX230 from Biometrics Ltd. [
82] or the Delsys Bagnoli [
83]. Delsys defines the SNR of their EMG equipment to be 65 dB. These sensors have protruding metal parts to establish the conductive connection to the skin, which can cause pressure marks.
The MyoWare
TM EMG sensor [
84] uses an electrolyte gel to establish the conductive connection to the skin. This sensor has high SNR straight away at application to the skin surface. However, gel sensors are not suitable for long-term application.
Although the conductive sensors have a high SNR at optimal skin contact, the presented capacitive sensor has a high SNR (40 dB at Sensor 1) as well and it is already given at application to the skin surface as it is independent of an electrolyte. The SNR would be even higher at sensor assemblies with better signal coupling (e.g., Sensor 5). Due to the flexible design, pressure marks can be avoided and especially the textile sensor feels comfortable to the skin (see
Section 6.7).
6.9. Alternative Reference Design: INA with DC Rejection Reference Design
The current sensor circuit was also built with DC rejection reference design (see
Section 3.2.2.3). In real-world applications, the circuit shows the desired filtering behavior directly at the first stage. However, when placing the sensor with asymmetric pressure, the signal at the INA and/or (depending on the interference) the feedback amplifier are/is saturated. The circuit is then unstable due to the feedback. Such instability is not acceptable in a real-world measurement system, so the INA with DC rejection reference design was not used in our final set-up. Further, fewer components are required when no DC rejection reference design is employed.
7. Discussion
7.1. Capacitive EMG Sensing Electrode
Six dry sensor set-ups were analyzed theoretically and measured to find the set-up with optimal signal coupling. Due to their flexible design, they adapt well to the forearm anatomy. The textile sensor in particular feels comfortable on the skin. A bio-compatible assembly is easy to achieve, since numerous bio-compatible insulating materials are available. This is, however, not the case for conductive materials. The textile and the copper sensors are connected to the electronics via coax-cables (UMCC), which have advantageous properties for signal transmission, but establishing a stable, conductive connection to the sensor area and the shielding is difficult. Mechanically, they do not exhibit the long-term stability of flex sensors, which are connected to a flat cable plug. To avoid short circuiting, insulating textile must not absorb sweat. Unlike textile or foil sensors, where sweat might travel along the UMCC cables into the sensors, flex sensors are insensitive to water. This problematic sensitivity to water could be addressed by sealing the cable inlets.
Due to the sensor design, the material forming the dielectric for the coupling capacity and also for the parasitic capacity must be flexible. For the coupling and parasitic capacities to remain constant, the material should further be incompressible and the dielectric constant should not change.
7.2. Body Reference
To shift the common-mode component of the measurement signal to within the operating range, we used a conductive reference in our set-up because it enables stable operation. Only one small reference is necessary for an entire array, and the reference is not as sensitive to lifting and changes in skin condition as in the case of a conductive sensor electrode. The reference can be placed at any part of the forearm where no movement is expected. A conductive reference is predominantly advantageous, and the advantage of the insulated measurement principle is maintained.
7.3. Analog Circuit
The bias resistor was chosen to be relatively low for stability, although—together with the parasitic capacity—it damps the signal. The damping effects due to parasitic capacities can be compensated by higher amplification.
7.4. Comparison of Sensor Assemblies
The measurements for the sensor comparison were conducted on healthy subjects and not on amputees; nevertheless, the results also extend to amputees. If the remaining muscle tissue is reduced, sensor area and inter-electrode distance and total gain can be adapted.
Differences between theory and real-world measurement may have resulted from the assumptions made for the calculation. The impedance of the stratum corneum was assumed to be an ohmic resistance, but its electrical properties vary depending on skin condition. Since the theoretical calculation is only an estimation of the insulated measurement, we consider the assumptions for the stratum corneum to be sufficient. The impedance in the measurements depends on the pressure applied to the sensor. Air entrapped between the sensor assembly layers may have led to differences in the capacities. Further, due to the sensor-making process, layer thickness may have differed from that given in the manufacturer data sheet.
Differences in the measurements may have resulted from slight variations in sensor positioning, although we strove for accurate placement. Further, the pressure of the cuff may have varied slightly between measurements. The subjects were asked to perform maximum voluntary contraction, so slight variations could not be excluded. These facts led to minor variations but the results are statistically sound. Although the sensors were placed above the musculus extensor digitorum, some of the EMG signal power might have resulted from other muscles due to crosstalk.
In terms of signal coupling, the conductive sensor (Sensor 6) had the highest amplitude in the real-world measurements. However, the t-test did not show a significant difference between the conductive sensor and Sensor 5, which has a thin foil as coupling dielectric. The capacitive sensor was more stable, as it is insensitive to skin conditions; adaption of the amplification for real-word application is therefore unnecessary. With increasing parasitic capacity, the common-mode rejection ratio increases, but the coupled signal amplitude is increasingly attenuated. By decreasing the parasitic capacity, the coupled signal amplitude can be increased. The capacity ratio can be set by selecting the appropriate dielectrics.
8. Conclusions
Insulated EMG measurement is associated with several challenges. Due to the capacitive measurement principle, the impedances at the coupling are much higher than in conductive measurement. Very small signals must be measured with precision electronics. Low-frequency movement artifacts pose a problem in biosignal measurement because they can have higher amplitudes than the EMG itself. Further, noise from various sources, such as the 50 Hz hum, disturbs the EMG signal.
Considering and addressing these challenges, we developed a highly stable insulated EMG measurement system. Our sensor is particularly suitable for the Otto Bock standard prosthesis in real-world applications. The set-up we have introduced has high signal quality (SNR = 40 dB at Sensor 1) and therefore allows accurate measurement. Using a sensor assembly with better signal coupling, such as Sensor 5, further increases the SNR. The prototype sensor is a compact two-layer PCB print with 27 mm × 46 mm and it requires only 22 mW. No feedback loops are included in our system to guarantee high stability. The flexible sensors avoid occurrence of pressure marks, and their design reduces movement artifacts. High wearing comfort is achieved, especially for the textile sensors, and skin irritations are avoided when using biocompatible insulating materials. The capacitive sensor shows greater stability because it is independent of skin condition, so no gain-level adaption is necessary.
Digital signal processing is essential to high-quality EMG measurement. We developed algorithms particularly for this system to achieve low power consumption and real-time capability for use in prosthesis.
As part of future work, we will improve the flexible sensors for higher mechanical stability. The textile sensors in particular are very promising due to their high wearing comfort, but they must be sealed to be waterproof; this is essential to long-term application, as otherwise sweat could cause short circuiting. The sensor can be washed when waterproof, which is a requirement in prosthesis applications.
Further, we will ask amputees to test our sensor set-up to identify potential challenges associated with its use in prostheses. The insulated EMG sensor can then be adapted to enable even more stable operation.
Possible future applications of this sensor include, for example, exoskeletons, sport physiology, medical diagnosis and analysis of human movement by means of EMG signals in real-world environments.