A Pseudodifferential Amplifier for Bioelectric Events With DC-Offset ...
300
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
A Pseudodifferential Amplifier for Bioelectric Events With DC-Offset Compensation Using Two-Wired Amplifying Electrodes Thomas Degen, Member, IEEE, and Heinz Jäckel, Member, IEEE
Abstract—Most wired active electrodes reported so far have a gain of one and require at least three wires. This leads to stiff cables, large connectors and additional noise for the amplifier. The theoretical advantages of amplifying the signal on the electrodes right from the source has often been described, however, rarely implemented. This is because a difference in the gain of the electrodes due to component tolerances strongly limits the achievable common mode rejection ratio (CMRR). In this paper, we introduce an amplifier for bioelectric events where the major part of the amplification (40 dB) is achieved on the electrodes to minimize pick-up noise. The electrodes require only two wires of which one can be used for shielding, thus enabling smaller connecters and smoother cables. Saturation of the electrodes is prevented by a dc-offset cancelation scheme with an active range of 250 mV . This error feedback simultaneously allows to measure the low frequency components down to dc. This enables the measurement of slow varying signals, e.g., the change of alertness or the depolarization before an epileptic seizure normally not visible in a standard electroencephalogram (EEG). The amplifier stage provides the necessary supply current for the electrodes and generates the error signal for the feedback loop. The amplifier generates a pseudodifferential signal where the amplified bioelectric event is present on one lead, but the common mode signal is present on both leads. Based on the pseudodifferential signal we were able to develop a new method to compensate for a difference in the gain of the active electrodes which is purely software based. The amplifier system is then characterized and the input referred noise as well as the CMRR are measured. For the prototype circuit the CMRR evaluated to 78 dB (without the driven-right-leg circuit). The applicability of the system is further demonstrated by the recording of an ECG. Index Terms—Amplifying electrodes, bioelectric recordings, common mode rejection, dc-offset compensation, instrumentation amplifier.
I. INTRODUCTION
A
CTIVE electrodes are far less used than it could be expected given their relatively long history. A very early device was reported in 1968 [1] and since then several publications have described the development [2], [3] as well as the intrinsic advantages of active electrodes, e.g., the reduction of power line interference [4]. Most of the active electrodes described need
Manuscript received June 18, 2004; revised April 24,2005Asterisk indicates corresponding author. T. Degen is with the Department of Information Technology and Electrical Engeering, Swiss Federal Institute of Technology, Zurich 8001, Switzerland (e-mail: [email protected]) H. Jäckel is with the Department of Information Technology and Electrical Engeering, Swiss Federal Institute of Technology, Zurich 8001, Switzerland. Digital Object Identifier 10.1109/TBME.2005.862531
at least three wires to be operated and have a gain of only one. Thus, the improved signal quality is paid for by stiff wires, large connectors and added noise. In [5], a two-wired buffer electrode has been reported, yet the electrode has the restriction of requiring the input potential to be the lowest in the circuit. There is no method provided as to how this requirement might be met in the case of bioelectric recordings. On the other hand very few devices reported have a gain greater than one although this would reduce the input referred noise. One problem is that a high gain on the electrodes has the potential to limit the maximal allowable dc-offset. Therefore, only small gains were implemented [6], [7] or at least a reduced dc gain was realized by adding a filter circuit on the active electrodes [8]. This filter normally requires an additional wire to adapt the time constant. Having a large gain on the electrodes would be beneficial for three reasons. First, using the same opamp with a higher gain reducestheinputreferrednoiseasconfirmedbynoisemeasurements (see Section IV, Fig. 9). Second, the noise requirements of the following amplifier stage can be lowered accordingly, which allows toreducethepowerconsumption.Third,comparedtousingbuffer electrodes the total number of opamps can be reduced because a large part of the gain is implemented on the electrodes. Yet all amplifying electrodes suffer from the possibility of a reduced common mode rejection ratio (CMRR). This is because any variation in gain, due to the component tolerances, limits the maximal achievable CMRR in multistage amplifiers [9]. Therefore, a difference in gain has to be compensated for. This is possible by using coupled electrodes [7], by manual trimming the gain of each electrode [6] or of each differential amplifier [10]. The latter can also be done automatically both for coupled and noncoupled electrodes [11]. All these methods require at least four wires per electrode. In this paper, we introduce a new method to compensate for the difference in gain. The method is relaying on a new pseudodifferential amplifier which allows to compensate for the difference in gain similar to [11], but implemented in software. This allows to reduce the complexity of the analog circuit. It is an often overlooked fact, that most amplifiers for bioelectric events eliminate clinically relevant data by removing the low-frequency components of the measured signal. Examples of clinically relevant data normally not measured by reported amplifiers are the change of alertness that can be measured by EEG [12] or the depolarization prior to an epileptic seizure [13] or the sleep pattern of preterm infants [14]. This is because the amplifiers need to suppress the dc-offset present at the skin-electrode
0018-9294/$20.00 © 2006 IEEE
DEGEN AND JÄCKEL: PSEUDODIFFERENTIAL AMPLIFIER FOR BIOELECTRIC EVENTS WITH DC-OFFSET COMPENSATION
Fig. 1. Principle of a two-wired amplifying active electrode biased by a constant current source I . The active part of the electrode consists of an operational amplifier, two resistors to define the gain and a diode used for level shifting purposes. The mass symbol refers to a common reference point, e.g., the negative pole of the battery.
interface. In our system, the error signal matches the low-frequency parts of the bioelectric signal which can, therefore, be measured down to dc although with a smaller resolution. We have designed a new pseudodifferential amplifier based on two-wired amplifying electrodes with a gain of 40 dB. The saturation of the electrodes is prevented by a dc-offset compensation based on an error-feedback loop. The difference in gain is compensated for in software. The circuit employs only offthe-shelf components which leads to a low-cost and yet highly customizable circuit. The description of the amplifier is split in three parts. In Section II, we will discuss the characteristics of amplifying active electrode and and the limits within they operate. In Section III, we then present the corresponding amplifier including the error feedback loop necessary for the electrodes to stay within the limits found beforehand. Finally, in Section III-B, we introduce a new method to compensate for the difference in gain implemented in software. II. AMPLIFYING TWO-WIRED ACTIVE ELECTRODES The general principle of two-wired active electrodes is to use a constant current source with the electrode as load and having the electrodes vary the voltage drop in relation to the bioelectric signal. A first electrode based on this principle was presented in [5] having a gain of one. We use the same principle, but use a slightly modified noninverting amplifier on the electrode with a gain of 40 dB. The full circuit is depicted in Fig. 1. Instead of a voltage supply the electrode receives a constant , part of which is the supply current of the opercurrent ational amplifier (opamp) and part of which flows through the into the amplifier output and the resistive network condiode nected to it. The reason a diode is placed between the positive supply of the electrode and the output of the opamp is to ensure that the output voltage of the opamp stays about 0.7 V below its positive supply voltage. This is necessary because common opamps can not drive their output voltage to either of the two supply voltages. Most of the current through the diode will then flow through the internal output transistor to the negative supply of the opamp. The remaining part of the current passes the gain setting resistors and . The negative supply is connected
301
Fig. 2. Characteristic voltage gain of a amplifying electrode measured for = 0. The opamp is a LMC7111, the three different supply currents and V diode is a BAS16, R and R have a value of 39 , respectively, 3.9 k .
which is typically held around 500 to a reference voltage mV above ground and which will play an important role in the dc-offset compensation scheme. The output voltage of the electrode is expressed as (1) This equation is true as long as the current source delivers enough current to supply the active electrode without exceeding the current sink capacity of the opamp. Another condition is that the supply voltage of the opamp ( ) is maintained within its specified limits (2) is the minimal supply voltage of the opamp. The value the opamp has to depends on the amount of supply current sink (see Fig. 2) and can be estimated by a measurement to 2.6 V for a reference current of 1 mA. is limited by the current The maximal output voltage of the subsequent circuit. source and the supply voltage If stands for the minimal voltage drop over the current source the maximal output voltage is given by (3) In our prototype circuit, we used two current regulator diodes J503 from Vishay in parallel. For sourcing 1 mA they require a voltage drop of about 1.2 V. The supply voltage being 5 V we have a maximal output voltage of 3.8 V. The voltage swing of the reference voltage is necessary to compensate for the offset between different electrodes as we will see later in Section III-A. Its amplitude corresponds to the value of the dc-offset which can be compensated for. Satisfying these conditions will allow the active electrode to function properly and to amplify the input signal by 40 dB as defined by (1). The input range for which the electrode operates in the linear region is depicted in Fig. 2. The input range of the electrodes is primarily given by the supply voltage divided by their gain corresponds to 0 V. For a maximal output expressed in (1), voltage of 3.8 V and a reference current of 1 mA the linear input range is about 12.5 mV starting at 20 mV.
302
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
This input range may appear to be too small to accommodate the power line interference. This is not the case, the DRL-loop of the reference electrode will adapt the reference voltage shifting the input range in order to follow the power line interference. Later in Section III-B, Fig. 5 we will see that the common mode gain of the amplifier is approximately one (at 50 Hz). This means that common mode voltage appearing at the input is hardly amplified by the reference electrode. The common mode input range (for 50 Hz) is, therefore, limited by the output swing . of the reference electrode which is Initially, when the power to the circuit is switched on, the opamp is turned off and will not sink any current. The current source will then increase the voltage over its load (the active electrode) until the desired current can flow. The input voltage corresponds to the body potential whose dc value is controlled by the driven-right-leg circuit (DRL). Fig. 2 shows a measurement of the output voltage in relation of 0 V. The to the input voltage for a reference voltage opamp is an LMC7111 and the maximal supply voltage in the prototype circuit is about 3.7 V. The important specifications of the LMC7111 are; a minimal guaranteed supply voltage of 2.7 between 13 V, rail-to-rail input and output, a supply current , an input resistance of over 1 and a bias current and 15 of about 60 pA. The dotted line in Fig. 2 corresponds to the regression line calculated by the measurement equipment between 25 mV and 35 mV for a supply current of 1 mA. The linear gain of the electrode is given by the gradient of the latter, for this particular electrode it resolves to 100.7. The difference between this value and the gain calculated in (1) is due to component tolerances, the implication of the latter will be discussed in detail in Section III-B. over the diode corresponds approximately The voltage drop -axis, i.e., to the intersection of the regression line with the 555 mV. For a supply current of 1 mA there is a sharp drop at . This is because when the current via the gain setting resistors reaches the maximal possible value of about 1 mA the voltage of the negative input of the opamp cannot rise any higher. Thus, the opamp will short-circuit the diode in the attempt to raise the output voltage. The curve then becomes flat . and stays at the value By increasing the supply current to 2 mA the gain remains constant over the full measurement range and the voltage drop increases slightly. On the other side, the minover the diode necessary to sink the imal supply voltage of the opamp current increases by about 400 mV. This observation remains true if we again augment the supply current by 1 mA. The next characteristic of interest is the differential output defined by the variation of the output voltage resistance in relation to a variation of the supply current. Its value can be over the diode directly extracted from the voltage drop (4) where is the thermal voltage and yields about is the reverse 25 mV for an ambient temperature of 300 .
current of the diode. With this equation and with the relation we find (5) For 1 mA we obtain a value close to 25.6 . The output resistance of the opamp does not contribute to this value because does not depend on the latter as indicated by (1). of the opamp Until now we considered the supply current to be constant. In reality, the supply current increases in an relatively linear way with the supply voltage. The maximal variation of the supply voltage in our circuit will be less than 1 V and the corresponding variance of the supply current corresponding to . This will lead to the same variance the datasheet reads 2.9 which in the case of corin the diode current responds to a variance of 3‰ which is so small that it will be neglected throughout this paper. We have now described a two-wired active electrode with a gain of 40 dB. The amplification allows the input referred noise to be reduced when compared with active buffer electrodes (see Fig. 9). The noise requirements of the following stage are reduced accordingly to the gain. The low wire-count will allow the reduction of the cable stiffness and the connector size when compared with other amplifying electrodes. To use the amplifying electrode we must now define a corresponding amplifier. III. AMPLIFIER STAGE The most important requirement of the subsequently described amplifier is to ensure that the electrodes do not saturate. This is the stumbling block which until now prohibited the use of amplifying electrodes with a large gain. To avoid the saturation of the electrodes we will adapt the reference voltage of the electrodes which is simultaneously the lower (negative) supply voltage of the opamp. We can condense the function of the amplifier stage into three main tasks: 1) provide the necessary supply current for the electrodes; in order to 2) generate the adequate reference voltage compensate for any dc-offset between electrodes; 3) reduce any common mode voltage by feeding it back to the body via a DRL circuit which is simultaneously determining the dc-component of the input voltage of the reference electrode. These three functions are achieved in the above mentioned order by a current source and two different feedback loops. Fig. 3 depicts the amplifier stage. The first feedback loop (“DRL loop”) is realized around a reference electrode and the DRL electrode. The purpose of the first loop is to define the correct bias voltage for the reference of electrode. In detail, this means that if the output voltage a current will flow the reference electrode is higher than and the opamp which drives the DRL into the resistor electrode has to lower its output voltage to allow this current . This lowers the input voltage to flow likewise through of all electrodes connected to the DRL electrode via the human body. Yet lowering the input voltage of the reference electrode while the reference voltage stays fixed at 500 mV will
DEGEN AND JÄCKEL: PSEUDODIFFERENTIAL AMPLIFIER FOR BIOELECTRIC EVENTS WITH DC-OFFSET COMPENSATION
303
is altered by for the th electrode. The reference voltage is the loop as long as the dc value of the output voltage . The guiding principle is the same as for different from the first feedback loop, except that this time it is the reference is fixed (by voltage which is adapted while the input voltage the first feedback loop). For the loop to be stable we need to invert the output of the corresponding integrator. For this purpose we added a voltage inverter with the following transfer function: (7)
Fig. 3. Amplifier stage for providing the necessary bias voltages and currents based on a current source and two different feedback loops. The first feedback loop (“DRL loop”) is built around the reference electrode and the DRL electrode. The second loop (“error feedback loop”) is built around each signal electrode of which only the electrode “i” is drawn. The amplified output signal for each channel is V V . All opamps are of the LT1490 type. The ground symbol stands for any reference, e.g., the battery ground of the amplifier.
0
also lower the output voltage which closes the loop. The was chosen in such a way that it was right in the value of also middle of the maximal and minimal possible value of highlighted in Fig. 2 (6) The minimal and maximal value of were specified in (2) is 3.55 V. The task of the and (3). The resulting value for first feedback loop can be resumed as to maintain the voltage between the ground of the amplifier system and the surface potential of the body at a constant level, and simultaneously reduce is any common mode signal at the input. As a consequence hold around 525 mV, the exact value results from (1). The corner frequency of the feedback loop is defined by . To guarantee the stability of the first loop and we chose the values resulting in a open loop phase margin of 30 [15]. Smaller values will increase the CMRR but may lead to oscillation or instability within the feedback loop. The second loop (“error feedback loop”) is built individually around each signal electrode. The purpose of the second loop is to implement a high-pass behavior for each signal electrode based on an error feedback. We will examine the loop in detail
The inverter will also convert the range of which is 0 to to the desired reference voltage swing by its gain of . As we will see later in (10), this gain will also reduce the size of the feedback capacitor for a required time constant. the mid Furthermore, by choosing an adequate value for point of the output range can be shifted to . As a conseafter power-up will be quence the initial reference voltage . close to Finally, the output range limitation of the inverter will encannot be substantially higher than . This sure that the is important because otherwise if built-in protection diode of the opamp on the reference electrode would start to conduct and forestall the functioning of the first feedback loop. , and The values used were . The second loop forces the mean value of to be equal to plus-minus a residual offset given by (8) is the input offset voltage of the opamp building the corresponds to the bias current at the negintegrator, ative input. In our prototype circuit, the value of the residual offset was about 40 mV. The accuracy of the dc-offset measurement is, thus, limited by the same offset but divided by the gain of the electrodes and yields about 0.4 mV. The function of the second feedback loop can be resumed as to implement a high-pass behavior and, thus, remove the dc-offset between the corresponding signal electrode and the reference electrode avoiding saturation of the amplifying electrodes. The time constant of the second feedback loop is much lower (at least five decades) than the time constant of the first feedback loop. This ensures the stability of the second feedback loop despite it being interlinked with the first one (see Section III-B, Fig. 5). The amplifier stage is of a pseudodifferential nature, the amplified and dc-offset free bioelectric signal is represented by . Yet the bioelectric signal is only visible on whereas the common mode signal is present on both signals. would result in large common mode inConsidering only terferences being visible. Fig. 4 illustrates the overall transfer function of the pseudodifferential amplifier and the electrodes. The transfer function can be expressed by the forward gain and the feedback gain as follows: (9)
304
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
A particularity of this amplifier is the fact, that the cutoff frequency of the high-pass depends on the gain of the active electrode. We can use the same amplifier for EEG by using a low-noise electrode with a gain of 60 dB and, therefore, a cutoff frequency of 0.15 Hz. A. DC-Offset Measurement
Fig. 4. Transfer function of the pseudodifferential amplifier stage with the electrodes. The upper corner frequency is fixed by the employed opamp (LMC7111), the overall gain is defined in (9) and the lower corner frequency depends on the feedback loop as expressed in (10).
f
According to the ANSI standard [17] a circuit for biosignals must be able to suppress a possible dc-offset of at least 300 mV between the inputs. Our circuit handles an offset between electrodes by adapting the reference voltage of the electrodes themselves. To analyze this feature more closely we will assume of the th signal electrode and of the a dc-offset between reference electrode. The second feedback loop will then adjust until the dc component of the signal the reference voltage . This leads to the following relation: is again (11) (12) For a large gain equals which corresponds to the offset between electrodes. This means that by measuring the reference voltage of an electrode it is possible to deduce the offset voltage of this electrode compared to the reference electrode. Although this dc-offset is composed of the physiological signal and the skin-electrode offset, the latter varies only very slowly [18]. Fig. 4 displays the frequency response of (12) which can also be expressed as
Fig. 5. CMRR of the amplifier stage is defined by the ratio of the differential of the input signal divided by the common mode gain of gain the input signal (the two solid lines). Both functions show the difference in gain of the corresponding single ended output signals. and are the differential and common mode gain of the output voltage of the reference electrode . Whereas and are the differential and common mode gain of the output of the signal electrode “i” (dotted lines). The dashed line “LMC 7111” represents the gain of the active electrodes without the second feedback loop. The higher frequencies are suppressed as a result of the transfer function of the opamp. The squares represent measured values of the CMRR for our prototype circuit. The value of CMRR was limited by the resistor tolerances as described in (18) to 30 dB.
G
(13)
G
G
V
G
G
G
corresponds to the gain of the electrodes as stated in (1). The bandwidth of is given by the GBW (gain-bandwidth product) of the employed opamp. corresponds to the transfer function of the integrator multiplied by a constant factor given by (7) The resulting function is a high-pass where the corner frequency is . Consequently we can express defined by the condition by the cutoff frequency
holds the same information but with an amplitude ten times larger.To our knowledge this is the first circuit reported which allows to measure a bioelectric signal with frequencies down to dc. A second consequence of (12) is that the limits within which the circuit can compensate a dc-offset are given directly by the and, thus, by the gain in the feedback voltage swing of loop given in (7). B. Amplifier Design and CMRR The main problem of amplifying electrodes is the possible reduction of CMRR due to a variance in gain between electrodes. and Let us assume that the reference electrode has a gain of the signal electrode ì’ has a gain of . If we now calculate the difference we obtain
(10)
(14)
Below the corner frequency the transfer function can be approxand above by . We used imated by and which results in a corner frequency of as required for recording an ECG. To restore the dc value after replacing the electrodes or to allow for fast recovery enabling after an artifact we placed a switch in parallel to to temporarily shift the corner frequency.
(15) Where the differential input and the common mode input are given by (16) (17)
DEGEN AND JÄCKEL: PSEUDODIFFERENTIAL AMPLIFIER FOR BIOELECTRIC EVENTS WITH DC-OFFSET COMPENSATION
305
This allows to directly evaluate the differential gain as well as the common mode gain and therefrom the CMRR as follows: (18) This is a very general relation which also allows to calculate the CMRR limitation based on the “potential divider effect.” Any couple of amplifying electrodes will introduce a severe CMRR limitation if their gain mismatch is not compensated for. As an example we estimate the CMRR of our electrodes. If we assume that we use resistors with a 1% variance the two gains may vary up to 2% and the worst case value for the CMRR is as low as 28 dB. Even if we consider that the CMRR will be improved by the DRL circuit [19] we must conclude, that 28 dB is much less than a standard amplifier for bioelectric events should offer. The other way around, a CMRR of 80 dB corresponds to a resistor tolerance of at least 0.025‰. Before we proceed to the description of the gain compensating method we would like to detail the complex function of the CMRR in the case of a pseudodifferential amplifier. Therefore, we simulated the different gain functions and measured the CMRR in the frequency domain between 5 Hz and 50 kHz. The results are shown in Fig. 5. The simulation and measurements correspond to the case of a noncompensated gain mismatch and do not describe the performance of the final amplifier system. Fig. 5 is an asymptotic representation of the different transfer functions defining the CMRR of the pseudodifferential amplifier based on two spice simulations, one for the differential gain and one for the common mode gain. The definitions of the different lines representing the decomposition of the overall CMRR are (19) (20) (21) The differential mode gain of the reference output voltage is very low because is forced to a constant value by corresponds to the transfer functhe first feedback loop. tion of the active electrodes (line LMC 7111) but is damped at low frequencies by the second feedback loop. Because the differential to differential gain already drawn in Fig. 4 superimposes . The two common mode gains and almost overlap, their difference is given by (18) and was measured for this particular pair of point of is defined by electrodes to be 30 dB. The ) the time constant of the first feedback loop ( which evaluates to about 50 Hz for our prototype circuit. of the second feedback loop they Below the time constant is diverge. Therefore, the differential common mode gain mostly 30 dB below the single common mode gains but follows below . To summarize we can express the CMRR at 50 Hz by (22) For our prototype we obtained a CMRR of 61.7 dB at 50 Hz (with the DRL) for a noncompensated gain mismatch.
Fig. 6. Two ways of interfacing the amplifier by (a) converting all four single ended signals separately or (b) using three differential ADCs to convert them in a pseudodifferential way. This enables simple software based algorithms to compensate for differences in gain and, thus, improving the CMRR.
This is lower than the requirement stated in [18], namely; (without the DRL). This is probably the main reason why amplifying electrodes are not used often in EEG and ECG. The difference in gain of the amplifying electrodes has to be compensated for. This can be done by trimming the gain of the electrodes [6] or by coupling the electrodes [7]. Even coupled electrodes have a limited CMRR due to the component tolerances which can be compensated for by trimming the gain of the differential stage [10]. This trimming can also be applied to noncoupled electrodes. In [11], we describe an autonomous compensation scheme for coupled and noncoupled electrodes. All these schemes could be implemented with the two-wired amplifying electrodes presented in this paper. But there is a even more elegant way to compensate for the gain mismatch which is only possible due to the pseudodifferential structure of the amplifier we developed in Section III. The trimming implemented in [11] is deferred to the software while the analog part is kept small and low-power. The new approach consists in shifting the differentiation normally performed in the amplifier to the software. For this purpose we need a fully differential interface circuit like the two examples shown in Fig. 6. The differential approach requires one more analog-to-digital channel for the reference electrode. The total number of channels is equal to the number of signal electrodes plus one. They can either be supplied by one analog-to-digital converter (ADC) and a multiplexer combined with a sample and hold circuit, or by using several ADCs in parallel. In each case, the bioelectric is then calculated by the following: signal (23) (24) and stand for the gain of the reference electrode where and the th signal electrode, respectively, which has first to be obtained by software. and is to measure a common mode A way to obtain signal, e.g., the 50-Hz power line interference as demonstrated in [11], or by calculating the discrete fourier transform of the
306
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
sampled values and use the values representing the 50-Hz interference. Unlike the method described in [11] this method does not necessitate a phase shift measurement because the two 50-Hz common mode signals are in phase. By implementing (24) based on the measurement of the 50-Hz interference, the latter can be completely removed as shown in Section IV. Any remaining 50-Hz component of the output signal is part of the differential bioelectric signal itself. An example is shown in Figs. 12 and 13. The CMRR is improved to the value of (25) The general value for the improved CMRR can not be given, because the improvement depends also on the value of the measured common mode signal itself. We can say that the CMRR increases as much as necessary to ensure, that the measured common mode signal is minimized on the reading. We can tough give an upper limit to which this schema can function. In the optimal case (the common mode signal is as large as the input range of the ADC), the upper limit can be given by (26) Where “ ” corresponds to the resolution of the ADC. Contrary to a 50-Hz notch filter the method will reduce all common modes signals without altering the differential mode signal, i.e., the bioelectric signal. The difference between the two circuits in Fig. 6 is the required input range of the ADC. In Fig. 6(a), the input range must whereas in Fig. 6(b) the input range can be at least be as low as . As a consequence the resolution for the same number of bits is higher in Fig. 6(b). As an example, the and proAD7731 can have an input range as low as vides three differential ADC in one IC. At 400 Hz the AD7731 . The upper limit of the has a resolution of 14.5 bit or 1.73 CMRR improvement according to (26) would evaluate to 87 dB (in addition to at least 28 dB CMRR of the electrodes). C. Digital Control Loop Regardless of which AD-conversion scheme is chosen, the second feedback loop responsible for the dc-offset compensation can be implemented optionally by a digital controlled loop [20] as shown in Fig. 7. The digital loop functions in a similar way to the analog one. is generated by the digital-to-analog The feedback voltage converter (DAC) while the digital filter [finite impulse response filter (FIR)] generates the necessary codes for the DAC, based on the differential input of the ADC. The opamp stage is needed for sourcing the current and adapting the output range of the DAC to the desired reference voltage range. Equation (7) is still valid and defines the function of the opamp stage. The main differences between the analog and the digital implementation are as follows: 1) the digital loop requires no capacitor; 2) the digital loop introduces discrete steps in the output waveform (quantization noise).
Fig. 7. Second feedback loop built as a digital control loop by replacing the integrator of Fig. 3 with a digital filter (FIR) and a DAC. The feedback loop is again shown for the electrode “i.” There is no need for a capacitor.
Between these steps the signal does not show the typical superimposed exponential variation which is otherwise characteristic of high-pass filtered signals. The absence of a capacitor in the digital control loop enables to control the time behavior of the circuit in a much more flexible and accurate way. The time constants between different channels match perfectly, a switch to temporarily change the time constant of the high-pass filter is not required any more. Instead adaptive filter techniques may be used resulting in shorter settling times. Likewise the part count decreases as the DAC replaces an opamp together with some passive components as well as the ADC necessary to convert the dc-offset. The part count is even lower than in the system reported in [8]. The fully digital circuit also enables a seamless integration of the amplifier circuit. No large capacitors or large resistors are needed which both are not easily implemented in a standard IC-design process. the Because of the discrete nature of the output voltage dc-offset of the bioelectric signal cannot be completely nulled. There remains a residual offset which can be calculated in relation to the resolution of the DAC (27) denotes the number of bits of the DAC. For a supply voltage of 5 V and a DAC with 12 bits this results in a residual offset . In other words, each time the digital code voltage of will of the DAC is changed by the FIR filter, the signal display a discrete step (quantization noise) with an amplitude corresponding to a multiple of the residual offset calculated in (27). If not handled appropriately for, this leads to an oscillating with a mean value of zero behavior of the output voltage volt. This behavior is exemplarily shown in Section IV, Fig. 14. Because the exact timing of these steps is known a priori by the software the latter can be removed before displaying or storing the signal. A better solution is to implement two thresholds which differ by the value of the residual offset calculated in (27). As long as the signal stays between the thresholds a change of the DAC will be inhibited.
DEGEN AND JÄCKEL: PSEUDODIFFERENTIAL AMPLIFIER FOR BIOELECTRIC EVENTS WITH DC-OFFSET COMPENSATION
Fig. 8. Shielded amplifying two-wired active electrode mounted onto a button in order to attach different types of disposable electrodes.
Fig. 9. Noise spectrum of the amplifying two-wired electrode and the total system (no additional gain) was measured with a vector signal analyzer (Stanford SR785). For comparison we added the noise spectrum of a two-wired buffer electrode (gain = 1) using the same operational amplifier. Different sampling rates were used above and below 100 Hz. f denotes the higher bandwidth of the amplifier system (200 Hz).
IV. RESULTS The fabrication of the prototype electrode was relatively straight-forward. We mounted the components on a printed circuit board which was then soldered onto the back of a snap-button cut from a pair of trousers allowing us to use different kinds of disposable electrodes. Fig. 8 shows a disposable pregelled EEG electrode clipped to an active electrode. The prototype electrodes are left unprotected to ease changing of the soldered components. First we measured the input and output referred noise of the amplifying electrodes as well as of the total amplifier. The measurement is shown in Fig. 9. The noise of the amplifying electrode is mainly due to the opamp used, which for this implementation is an LMC7111. The total input referred noise of the electrode corresponds to the integrated noise spectrum over the desired bandwidth. For a bandwidth of 1 Hz–1 kHz the computed integral results in 4.7 . If we consider the total amplifier then the noise is mainly due to two sources, the active electrode and the noise generated by the error feedback loop. This is because the noise of and the error feedback loop affect the reference voltage is, therefore, amplified by the active electrode. The total noise of the system in the bandwidth of 1 Hz-1 kHz computes to 7.4 . This is low enough for our test application. Currently, we are developing a EEG amplifier based on the herein proposed noise. method with less than 1
307
Fig. 10. A sample ECG recording made with the circuit shown in Fig. 3, the recording was made with disposable pregelled electrodes from Conmed. The gain of the active electrode was about 100, the values of the resistors were R = 47 k , R = 9:1 k , R = 4:7 k , R = 10 M , C = 10 F, = 220 k and C = 2:2 nF. The ADC were two ADA400A from R Tektronix.
For comparison we added the noise spectrum of the same active electrode with a gain of 1. (“buffer electrode” with and ). This confirms, that using a amplifying electrode with a gain of 40 dB lowers the noise when compared to the same active electrode configured as a buffer electrode. This improvement may be considered to be small, but it comes without an increased supply and it allows to reduce the noise requirements of the following stage. The noise reduction becomes the more important the higher the gain of the active electrode is. To our knowledge 40 dB corresponds to the second highest gain of any reported input stage for a bioelectric amplifier [8]. We then evaluated our amplifier by measuring a three lead ECG. The system was composed of a DRL electrode together with the active reference electrode and one signal electrode as shown in Fig. 3. We believe that an ECG is best suited to evaluate the quality of a bioelectric signal due to its distinctive shape. The ECG signal was recorded without any skin preparation by using disposable self-adhering pregelled electrodes as shown in Fig. 8. measured Fig. 10 shows a recorded ECG signal with a differential ADC as well as the dc-offset voltage between the electrodes. The ECG in Fig. 10 shows a flat baseline and very little 50-Hz interference and this with an overall gain of only 100. The offset voltage of the ECG should be zero as a result of the error feedback loop; the 40 mV visible in the figure correspond to the value calculated in (8). The signal amplitude of an ECG is around 1 mV (times 100), therefore, we chose a differential ADC with an input range of 200 mV. The low frequency parts ). The of the signal was measured simultaneously ( total dc-offset between the electrodes is about 12 mV, it is probably mainly due to the polarization at the skin-electrode interface. The next measurement in Fig. 11 shows the two signals of a and pseudodifferential configuration corresponding to circuit Fig. 6(b). These are the two components of the pseudodifferential signal normally not accessible in standard bioelectric measurements.
308
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
Fig. 11. Pseudodifferential ECG recording made with the configuration shown in Fig. 6(b). The recording was made with the same equipment as before. These two waveforms are normally not available with common amplifiers. The distinct noise visible on both waveforms are the common mode signals which will be removed by calculating the difference similar to the signal shown in Fig. 10. For V of 20 mV. better visualization we added an artificial offset to V
0
0
Fig. 13. Digitally calculated difference of the two pseudodifferential signals of Fig. 11. The enlargement shows no 50-Hz interference, the discrete fourier transform in Fig. 12(d) reveals an amplitude of 2.85 V of the 50-Hz component. This component is probably due to the bioelectric signal itself. The high frequency spikes are probably due to the absence of any anti-aliasing filter.
only present on one lead; . Two factors are helpful in this situation, firstly, the 50-Hz common mode signal is usually much stronger than the 50-Hz component of the bioelectric signal and secondly, the two 50-Hz common mode signals are in phase. Thus, by taking the difference between the two unipolar signals the common mode signal will be strongly reduced. This would result in a signal similar to the signal measured in Fig. 10 with the corresponding frequency decompensation shown in Fig. 12(c). The achieved CMRR corresponds to the value defined by (22). That there is a 50-Hz component of the ECG signal which is not in phase with the common mode signal can be seen by the fact that the difference of the two 50-Hz components of the unipolar signal and the 50-Hz component of ). the 50-Hz signal do not perfectly match ( Now we can go further and obtain a better reduction of the 50-Hz component by using the following formula (only valid for this specific measurement): (28)
Fig. 12. Frequency decomposition of the different signals involved. (a) and (b) Two pseudodifferential signals shown in Fig. 11. (c) and (d) Digitally obtained differential signals. (c) Corresponds to the simple difference of the two signals shown in Fig. 11 and (d) corresponds to the digitally improved signal shown in Fig. 13. The value written in italic corresponds to the absolute value of the 50-Hz component.
We can confirm the pseudodifferential nature of the configuration by the fact, that the biopotential signal is present only on the signal electrode, yet the common mode interference is present on both signals (the visible “noise”). Both signals still show all the common mode noise, e.g., the 50-Hz power line interference is clearly visible in the enlargement. The higher frequency contributions visible are mostly induced by the oscilloscope (used for visualization) standing close to the subject. The frequency decomposition obtained by Fourier transformation of the two signals can be seen in the corresponding clipping; namely Fig. 12(a) and (b). As we can see, the 50-Hz components of the two pseudodifferential signals are not the same. This is mainly due to the difference in the gain of the two electrodes but also to the fact that the bioelectric signal itself has a 50-Hz component which is
This is simply the implementation of (25). In Fig. 13, we can see the corresponding waveform and Fig. 12(d) shows the corresponding frequency decomposition. The 50-Hz component is which is most probably no common mode inabout 2.85 terference but the bioelectric signal itself. This is supported by the fact, that the phase between the 50-Hz component of the difference signal and the 50-Hz component of the reference signal is 88.7 which is nearly orthogonal. It is impossible to give an exact value for the CMRR, because we can only roughly estimate the common mode part and the difference mode part of the were indeed only 50-Hz component. If the remaining 2.85 due to the differential signal the CMRR at 50 Hz would be infibeing in phase nite. If we estimate that the part of the 2.85 were the remaining common mode interferwith the 943 ence, then the improvement of the CMRR can be calculated by (29) The total CMRR (including the 30 dB of the electrodes) would then evaluate to about 78 dB and the final CMRR (including the DRL) would be equivalent to about 128 dB for 50 Hz.
DEGEN AND JÄCKEL: PSEUDODIFFERENTIAL AMPLIFIER FOR BIOELECTRIC EVENTS WITH DC-OFFSET COMPENSATION
Fig. 14. Sample ECG recording obtained with a digital control loop as shown in Fig. 7. The oscillating is due to the discrete steps of the DAC. The amplitude corresponds to the residual offset calculated in (27) which was artificially increased to 24 mV for better visualization. The frequency can be calculated similarly to (10). The mean value of the recording is 0 V. The oscillation can be suppressed by adding thresholds to the FIR filter or by software.
The trace of the digitally calculated ECG shows no 50 Hz common mode interference as shown in the corresponding frequency spectrum in Fig. 12(d). There is visibly more noise than in the fully differential recording shown in Fig. 10. One reason is that by using two different ADCs their input noise will add up. The other reason may be that for this prototype circuit there was no anti-aliasing filter and, therefore, no reduction of high frequency noise. The different recordings were taken on different days and the electrodes were not placed at exactly the same location. This last paragraph is to give some results regarding the optional use of a digital control loop. Fig. 10 shows an ECG recorded with a digital control loop as shown in Fig. 7. The purpose of this recording is to demonstrate the oscillation (quantization noise) due to the discrete steps of the DAC resulting in jumps of the size calculated in (27). The gray line shows the recording when the software would inhibit any ADC change while the mean value of the bioelectric signal lays within a higher and a lower threshold. The two thresholds are indicated by the semitransparent region behind the recordings. The advantage of the digital control loop is the absence of any capacitor. There is also no superimposed hyperbolical relaxation curve like in high-pass filtered signals. This enables for example a precise measurement of the S-T elevation even shortly after a spike has temporarily driven the amplifier into saturation. Using a digital control loop is the ultimate step in transferring the signal treatment from the analog world to the digital world. Digital filters have the advantage of a perfectly linear phase response allowing to keep the original shape of the bioelectric signal. V. CONCLUSION In this paper, we presented a new amplifying two-wired elecand trode with a gain of 40 dB, an input resistance of over 1 a output resistance of about 25 as well as the corresponding pseudodifferential bioelectric amplifier. The amplifier section provides the supply current for the electrodes as well as the necessary dc-offset compensation by adapting the reference voltage of the amplifying electrodes. The entire gain of the amplifier is
309
concentrated at the first stage enabling a low noise and low part count design. In addition, the circuit allows to simultaneously measure the bioelectric signal as well as the low-frequency components of the same signal, although with two different resolutions. This enables to measure clinically relevant data normally not available with other reported amplifiers. We then analyzed the common mode rejection ratio (CMRR) of the amplifier. Because of its pseudodifferential structure the difference in gain of the amplifying electrodes can, for a first time, be compensated for by the digital signal processing. This overcomes the lack of an easy to implement method to compensate for a difference in gain of noncoupled amplifying electrodes. Unlike a 50-Hz notch filter the improvement the CMRR reduces common mode interferences at all frequencies of interest. with a The circuit is able to compensate up to supply voltage of 5 V and a power consumption of 7.5 mW per channel. The supply voltage can be reduced in future by using operational amplifiers working with a supply of 1.8 V or less. A prototype for an EEG recorder is currently under development with 3.7-V power supply and 1.8 mW power consumption per channel. To our knowledge this is the first published amplifier for biosignals with this specific combination of features, namely, high gain in the input stage, truly dc compliant and a digital-only implemented improvement of the CMRR based on gain adaption. In addition, the amplifier uses two-wired electrodes that reduce the stiffness of the connecting wires and the size of the connectors, has a very low part-count, is highly customizable and offers the possibility to digitally control all filter parameters. REFERENCES [1] P. C. Richardson, F. K. Coombs, and R. M. Adams, “Some new electrode techniques for long-term physiologic monitoring,” Aerosp. Med., pp. 745–750, 1968. [2] W. H. Ko, “Active electrodes for EEG and evoked potential,” in Proc. 20th Annu. Int. Conf. IEEE Engineering in Medicine and Biology Society, vol. 4, H. K. Chang and Y. T. Zhang, Eds., 1998, pp. 2221–2224. [3] S. Nishimura, Y. Tomita, and T. Horiuchi, “Clinical application of an active electrode using an operational amplifier,” IEEE Trans. Biomed. Eng., vol. 39, no. 10, pp. 1096–1099, Oct. 1992. [4] M. Fernandez and R. Pallas-Areny, “A simple active electrode for power line interference reduction in high resolution biopotential measurements,” in Proc. 18th Annu. Int. Conf. IEEE Engineering in Medicine and Biology Society, vol. 1, H. Boom, C. Robinson, W. Rutten, M. Neuman, and H. Wijkstra, Eds., 1997, pp. 97–98. [5] F. Z. Padmadinata, J. J. Veerhoek, G. J. A. van Dijk, and J. H. Huijsing, “Microelectronic skin electrode,” Sensors Actuators B (Chemical), vol. B1, no. 1, pp. 491–494, Jan. 1990. [6] W. J. R. Dunseath and E. F. Kelly, “Multichannel pc-based data-acquisition system for high-resolution EEG,” IEEE Trans. Biomed. Eng., vol. 42, no. 12, pp. 1212–1217, Dec. 1995. [7] E. S. Valchinov and N. E. Pallikarakis, “An active electrode for biopotential recording from small localized bio-sources,” Biomed. Eng. Online, vol. 3, no. 25, Jul. 2004. [8] A. C. MettingVanRijn, A. P. Kuiper, T. E. Dankers, and C. A. Grimbergen, “Low-cost active electrode improves the resolution in biopotential recordings,” in Proc. 18th Annu. Int. Conf. IEEE Engineering in Medicine and Biology Society, vol. 1, 1997, pp. 101–102. [9] R. Pallas-Areny and J. G. Webster, “Common mode rejection ratio for cascaded differential amplifier stages,” IEEE Trans. Instrum. Meas., vol. 40, no. 4, pp. 677–681, Aug. 1991. [10] , “Common mode rejection ratio in differential amplifiers,” IEEE Trans. Instrum. Meas., vol. 40, no. 4, pp. 669–676, Aug. 1991.
310
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
[11] T. Degen and H. Jäckel, “Enhancing interference rejection of amplifying electrodes by automated gain adaption,” IEEE Trans. Biomed. Eng., vol. 51, no. 11, pp. 2031–2039, Nov. 2004, to be published. [12] S. Zschocke, Klinische Elektroencephalographie. Berlin, Germany: Springer, 2002, pp. 18–19. [13] A. Ikeda et al., “Focal ictal direct current shifts in human epilepsy as studied by subdural and scalp recording,” Brain, vol. 5, no. 122, pp. 827–838, 1999. [14] S. Vanhatalo et al., “DC-EEG discloses prominent, very slow activity patterns during sleep in preterm infants,” Clin. Neurophysiol., vol. 113, no. 11, pp. 1822–1825, Nov. 2002. [15] B. B. Winter and J. G. Webster, “Driven-right-leg circuit design,” IEEE Trans. Biomed. Eng., vol. BME-30, no. 1, pp. 62–66, Jan. 1983. [16] A. C. MettingVanRijn, A. Peper, and C. A. Grimberger, “Amplifiers for bioelectric events: A design with a minimal number of parts,” Med. Biol. Eng. Comput., vol. 32, no. 3, pp. 305–310, May 1994. [17] Ambulatory Electrocardiographs, American National Standard ANSI/AAMI EC38:1998, 1999. [18] A. C. Metting-van Rijn, A. Peper, and C. A. Grimbergen, “High-quality recording of bioelectric events. Part 2. Low-noise, low-power multichannel amplifier design,” Med. Biol. Eng. Comput., vol. 29, no. 4, pp. 433–440, Jul. 1991. , “High-quality recording of bioelectric events. Part 1. Interference [19] reduction, theory and practice,” Med. Biol. Eng. Comput., vol. 28, no. 5, pp. 389–397, Sep. 1990. [20] M. Verbeke, S. Gur, A. Ron, and R. Iraqi, “Manual or automatic nulling DC offset for physiological DC amplifier,” Med. Eng. Phys., vol. 16, pp. 171–171, Mar. 1994.
Thomas Degen (S’99–M’03) received the diploma in physical electronics in 1995 from the University of Neuchâtel, Neuchâtel, Switzerland. For the next two years he worked on industrial projects at the Interdisciplinary Institute of Integrated Systems at the Engineering School of Biel-Bienne. In 1997, he took part in an exchange program, doing research at the Indian Institute of Science, Bangalore, India. In 1998, he joined the Swiss Federal Institute of Technology, Zurich, Switzerland, where he is currently working toward the Ph.D. degree. His research interests include biomedical instrumentation, low-voltage analog circuit design, and wireless transmission.
Heinz Jäckel (M’82) received the Ph.D. degree in electrical engineering from the Swiss Federal Institute of Technology, Zürich, Switzerland, in 1979. From 1979 to 1993, he worked as a Research Staff Member and in managerial functions at the IBM research lab, Rüschlikon, Switzerland. In 1993, he returned to the Swiss Federal Institute of Technology, where he is now a full Professor with the Institute of Electronics and Head of the high-speed and analog electronics group. His research interests are design, characterization and technology of InP-based heterobipolar transistors, high-speed analog and digital circuitand IC-design, all-optical photonic devices for TB/s-data communication and CMOS IC-design for RF- and sensor-applications.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
A Pseudodifferential Amplifier for Bioelectric Events With DC-Offset Compensation Using Two-Wired Amplifying Electrodes Thomas Degen, Member, IEEE, and Heinz Jäckel, Member, IEEE
Abstract—Most wired active electrodes reported so far have a gain of one and require at least three wires. This leads to stiff cables, large connectors and additional noise for the amplifier. The theoretical advantages of amplifying the signal on the electrodes right from the source has often been described, however, rarely implemented. This is because a difference in the gain of the electrodes due to component tolerances strongly limits the achievable common mode rejection ratio (CMRR). In this paper, we introduce an amplifier for bioelectric events where the major part of the amplification (40 dB) is achieved on the electrodes to minimize pick-up noise. The electrodes require only two wires of which one can be used for shielding, thus enabling smaller connecters and smoother cables. Saturation of the electrodes is prevented by a dc-offset cancelation scheme with an active range of 250 mV . This error feedback simultaneously allows to measure the low frequency components down to dc. This enables the measurement of slow varying signals, e.g., the change of alertness or the depolarization before an epileptic seizure normally not visible in a standard electroencephalogram (EEG). The amplifier stage provides the necessary supply current for the electrodes and generates the error signal for the feedback loop. The amplifier generates a pseudodifferential signal where the amplified bioelectric event is present on one lead, but the common mode signal is present on both leads. Based on the pseudodifferential signal we were able to develop a new method to compensate for a difference in the gain of the active electrodes which is purely software based. The amplifier system is then characterized and the input referred noise as well as the CMRR are measured. For the prototype circuit the CMRR evaluated to 78 dB (without the driven-right-leg circuit). The applicability of the system is further demonstrated by the recording of an ECG. Index Terms—Amplifying electrodes, bioelectric recordings, common mode rejection, dc-offset compensation, instrumentation amplifier.
I. INTRODUCTION
A
CTIVE electrodes are far less used than it could be expected given their relatively long history. A very early device was reported in 1968 [1] and since then several publications have described the development [2], [3] as well as the intrinsic advantages of active electrodes, e.g., the reduction of power line interference [4]. Most of the active electrodes described need
Manuscript received June 18, 2004; revised April 24,2005Asterisk indicates corresponding author. T. Degen is with the Department of Information Technology and Electrical Engeering, Swiss Federal Institute of Technology, Zurich 8001, Switzerland (e-mail: [email protected]) H. Jäckel is with the Department of Information Technology and Electrical Engeering, Swiss Federal Institute of Technology, Zurich 8001, Switzerland. Digital Object Identifier 10.1109/TBME.2005.862531
at least three wires to be operated and have a gain of only one. Thus, the improved signal quality is paid for by stiff wires, large connectors and added noise. In [5], a two-wired buffer electrode has been reported, yet the electrode has the restriction of requiring the input potential to be the lowest in the circuit. There is no method provided as to how this requirement might be met in the case of bioelectric recordings. On the other hand very few devices reported have a gain greater than one although this would reduce the input referred noise. One problem is that a high gain on the electrodes has the potential to limit the maximal allowable dc-offset. Therefore, only small gains were implemented [6], [7] or at least a reduced dc gain was realized by adding a filter circuit on the active electrodes [8]. This filter normally requires an additional wire to adapt the time constant. Having a large gain on the electrodes would be beneficial for three reasons. First, using the same opamp with a higher gain reducestheinputreferrednoiseasconfirmedbynoisemeasurements (see Section IV, Fig. 9). Second, the noise requirements of the following amplifier stage can be lowered accordingly, which allows toreducethepowerconsumption.Third,comparedtousingbuffer electrodes the total number of opamps can be reduced because a large part of the gain is implemented on the electrodes. Yet all amplifying electrodes suffer from the possibility of a reduced common mode rejection ratio (CMRR). This is because any variation in gain, due to the component tolerances, limits the maximal achievable CMRR in multistage amplifiers [9]. Therefore, a difference in gain has to be compensated for. This is possible by using coupled electrodes [7], by manual trimming the gain of each electrode [6] or of each differential amplifier [10]. The latter can also be done automatically both for coupled and noncoupled electrodes [11]. All these methods require at least four wires per electrode. In this paper, we introduce a new method to compensate for the difference in gain. The method is relaying on a new pseudodifferential amplifier which allows to compensate for the difference in gain similar to [11], but implemented in software. This allows to reduce the complexity of the analog circuit. It is an often overlooked fact, that most amplifiers for bioelectric events eliminate clinically relevant data by removing the low-frequency components of the measured signal. Examples of clinically relevant data normally not measured by reported amplifiers are the change of alertness that can be measured by EEG [12] or the depolarization prior to an epileptic seizure [13] or the sleep pattern of preterm infants [14]. This is because the amplifiers need to suppress the dc-offset present at the skin-electrode
0018-9294/$20.00 © 2006 IEEE
DEGEN AND JÄCKEL: PSEUDODIFFERENTIAL AMPLIFIER FOR BIOELECTRIC EVENTS WITH DC-OFFSET COMPENSATION
Fig. 1. Principle of a two-wired amplifying active electrode biased by a constant current source I . The active part of the electrode consists of an operational amplifier, two resistors to define the gain and a diode used for level shifting purposes. The mass symbol refers to a common reference point, e.g., the negative pole of the battery.
interface. In our system, the error signal matches the low-frequency parts of the bioelectric signal which can, therefore, be measured down to dc although with a smaller resolution. We have designed a new pseudodifferential amplifier based on two-wired amplifying electrodes with a gain of 40 dB. The saturation of the electrodes is prevented by a dc-offset compensation based on an error-feedback loop. The difference in gain is compensated for in software. The circuit employs only offthe-shelf components which leads to a low-cost and yet highly customizable circuit. The description of the amplifier is split in three parts. In Section II, we will discuss the characteristics of amplifying active electrode and and the limits within they operate. In Section III, we then present the corresponding amplifier including the error feedback loop necessary for the electrodes to stay within the limits found beforehand. Finally, in Section III-B, we introduce a new method to compensate for the difference in gain implemented in software. II. AMPLIFYING TWO-WIRED ACTIVE ELECTRODES The general principle of two-wired active electrodes is to use a constant current source with the electrode as load and having the electrodes vary the voltage drop in relation to the bioelectric signal. A first electrode based on this principle was presented in [5] having a gain of one. We use the same principle, but use a slightly modified noninverting amplifier on the electrode with a gain of 40 dB. The full circuit is depicted in Fig. 1. Instead of a voltage supply the electrode receives a constant , part of which is the supply current of the opercurrent ational amplifier (opamp) and part of which flows through the into the amplifier output and the resistive network condiode nected to it. The reason a diode is placed between the positive supply of the electrode and the output of the opamp is to ensure that the output voltage of the opamp stays about 0.7 V below its positive supply voltage. This is necessary because common opamps can not drive their output voltage to either of the two supply voltages. Most of the current through the diode will then flow through the internal output transistor to the negative supply of the opamp. The remaining part of the current passes the gain setting resistors and . The negative supply is connected
301
Fig. 2. Characteristic voltage gain of a amplifying electrode measured for = 0. The opamp is a LMC7111, the three different supply currents and V diode is a BAS16, R and R have a value of 39 , respectively, 3.9 k .
which is typically held around 500 to a reference voltage mV above ground and which will play an important role in the dc-offset compensation scheme. The output voltage of the electrode is expressed as (1) This equation is true as long as the current source delivers enough current to supply the active electrode without exceeding the current sink capacity of the opamp. Another condition is that the supply voltage of the opamp ( ) is maintained within its specified limits (2) is the minimal supply voltage of the opamp. The value the opamp has to depends on the amount of supply current sink (see Fig. 2) and can be estimated by a measurement to 2.6 V for a reference current of 1 mA. is limited by the current The maximal output voltage of the subsequent circuit. source and the supply voltage If stands for the minimal voltage drop over the current source the maximal output voltage is given by (3) In our prototype circuit, we used two current regulator diodes J503 from Vishay in parallel. For sourcing 1 mA they require a voltage drop of about 1.2 V. The supply voltage being 5 V we have a maximal output voltage of 3.8 V. The voltage swing of the reference voltage is necessary to compensate for the offset between different electrodes as we will see later in Section III-A. Its amplitude corresponds to the value of the dc-offset which can be compensated for. Satisfying these conditions will allow the active electrode to function properly and to amplify the input signal by 40 dB as defined by (1). The input range for which the electrode operates in the linear region is depicted in Fig. 2. The input range of the electrodes is primarily given by the supply voltage divided by their gain corresponds to 0 V. For a maximal output expressed in (1), voltage of 3.8 V and a reference current of 1 mA the linear input range is about 12.5 mV starting at 20 mV.
302
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
This input range may appear to be too small to accommodate the power line interference. This is not the case, the DRL-loop of the reference electrode will adapt the reference voltage shifting the input range in order to follow the power line interference. Later in Section III-B, Fig. 5 we will see that the common mode gain of the amplifier is approximately one (at 50 Hz). This means that common mode voltage appearing at the input is hardly amplified by the reference electrode. The common mode input range (for 50 Hz) is, therefore, limited by the output swing . of the reference electrode which is Initially, when the power to the circuit is switched on, the opamp is turned off and will not sink any current. The current source will then increase the voltage over its load (the active electrode) until the desired current can flow. The input voltage corresponds to the body potential whose dc value is controlled by the driven-right-leg circuit (DRL). Fig. 2 shows a measurement of the output voltage in relation of 0 V. The to the input voltage for a reference voltage opamp is an LMC7111 and the maximal supply voltage in the prototype circuit is about 3.7 V. The important specifications of the LMC7111 are; a minimal guaranteed supply voltage of 2.7 between 13 V, rail-to-rail input and output, a supply current , an input resistance of over 1 and a bias current and 15 of about 60 pA. The dotted line in Fig. 2 corresponds to the regression line calculated by the measurement equipment between 25 mV and 35 mV for a supply current of 1 mA. The linear gain of the electrode is given by the gradient of the latter, for this particular electrode it resolves to 100.7. The difference between this value and the gain calculated in (1) is due to component tolerances, the implication of the latter will be discussed in detail in Section III-B. over the diode corresponds approximately The voltage drop -axis, i.e., to the intersection of the regression line with the 555 mV. For a supply current of 1 mA there is a sharp drop at . This is because when the current via the gain setting resistors reaches the maximal possible value of about 1 mA the voltage of the negative input of the opamp cannot rise any higher. Thus, the opamp will short-circuit the diode in the attempt to raise the output voltage. The curve then becomes flat . and stays at the value By increasing the supply current to 2 mA the gain remains constant over the full measurement range and the voltage drop increases slightly. On the other side, the minover the diode necessary to sink the imal supply voltage of the opamp current increases by about 400 mV. This observation remains true if we again augment the supply current by 1 mA. The next characteristic of interest is the differential output defined by the variation of the output voltage resistance in relation to a variation of the supply current. Its value can be over the diode directly extracted from the voltage drop (4) where is the thermal voltage and yields about is the reverse 25 mV for an ambient temperature of 300 .
current of the diode. With this equation and with the relation we find (5) For 1 mA we obtain a value close to 25.6 . The output resistance of the opamp does not contribute to this value because does not depend on the latter as indicated by (1). of the opamp Until now we considered the supply current to be constant. In reality, the supply current increases in an relatively linear way with the supply voltage. The maximal variation of the supply voltage in our circuit will be less than 1 V and the corresponding variance of the supply current corresponding to . This will lead to the same variance the datasheet reads 2.9 which in the case of corin the diode current responds to a variance of 3‰ which is so small that it will be neglected throughout this paper. We have now described a two-wired active electrode with a gain of 40 dB. The amplification allows the input referred noise to be reduced when compared with active buffer electrodes (see Fig. 9). The noise requirements of the following stage are reduced accordingly to the gain. The low wire-count will allow the reduction of the cable stiffness and the connector size when compared with other amplifying electrodes. To use the amplifying electrode we must now define a corresponding amplifier. III. AMPLIFIER STAGE The most important requirement of the subsequently described amplifier is to ensure that the electrodes do not saturate. This is the stumbling block which until now prohibited the use of amplifying electrodes with a large gain. To avoid the saturation of the electrodes we will adapt the reference voltage of the electrodes which is simultaneously the lower (negative) supply voltage of the opamp. We can condense the function of the amplifier stage into three main tasks: 1) provide the necessary supply current for the electrodes; in order to 2) generate the adequate reference voltage compensate for any dc-offset between electrodes; 3) reduce any common mode voltage by feeding it back to the body via a DRL circuit which is simultaneously determining the dc-component of the input voltage of the reference electrode. These three functions are achieved in the above mentioned order by a current source and two different feedback loops. Fig. 3 depicts the amplifier stage. The first feedback loop (“DRL loop”) is realized around a reference electrode and the DRL electrode. The purpose of the first loop is to define the correct bias voltage for the reference of electrode. In detail, this means that if the output voltage a current will flow the reference electrode is higher than and the opamp which drives the DRL into the resistor electrode has to lower its output voltage to allow this current . This lowers the input voltage to flow likewise through of all electrodes connected to the DRL electrode via the human body. Yet lowering the input voltage of the reference electrode while the reference voltage stays fixed at 500 mV will
DEGEN AND JÄCKEL: PSEUDODIFFERENTIAL AMPLIFIER FOR BIOELECTRIC EVENTS WITH DC-OFFSET COMPENSATION
303
is altered by for the th electrode. The reference voltage is the loop as long as the dc value of the output voltage . The guiding principle is the same as for different from the first feedback loop, except that this time it is the reference is fixed (by voltage which is adapted while the input voltage the first feedback loop). For the loop to be stable we need to invert the output of the corresponding integrator. For this purpose we added a voltage inverter with the following transfer function: (7)
Fig. 3. Amplifier stage for providing the necessary bias voltages and currents based on a current source and two different feedback loops. The first feedback loop (“DRL loop”) is built around the reference electrode and the DRL electrode. The second loop (“error feedback loop”) is built around each signal electrode of which only the electrode “i” is drawn. The amplified output signal for each channel is V V . All opamps are of the LT1490 type. The ground symbol stands for any reference, e.g., the battery ground of the amplifier.
0
also lower the output voltage which closes the loop. The was chosen in such a way that it was right in the value of also middle of the maximal and minimal possible value of highlighted in Fig. 2 (6) The minimal and maximal value of were specified in (2) is 3.55 V. The task of the and (3). The resulting value for first feedback loop can be resumed as to maintain the voltage between the ground of the amplifier system and the surface potential of the body at a constant level, and simultaneously reduce is any common mode signal at the input. As a consequence hold around 525 mV, the exact value results from (1). The corner frequency of the feedback loop is defined by . To guarantee the stability of the first loop and we chose the values resulting in a open loop phase margin of 30 [15]. Smaller values will increase the CMRR but may lead to oscillation or instability within the feedback loop. The second loop (“error feedback loop”) is built individually around each signal electrode. The purpose of the second loop is to implement a high-pass behavior for each signal electrode based on an error feedback. We will examine the loop in detail
The inverter will also convert the range of which is 0 to to the desired reference voltage swing by its gain of . As we will see later in (10), this gain will also reduce the size of the feedback capacitor for a required time constant. the mid Furthermore, by choosing an adequate value for point of the output range can be shifted to . As a conseafter power-up will be quence the initial reference voltage . close to Finally, the output range limitation of the inverter will encannot be substantially higher than . This sure that the is important because otherwise if built-in protection diode of the opamp on the reference electrode would start to conduct and forestall the functioning of the first feedback loop. , and The values used were . The second loop forces the mean value of to be equal to plus-minus a residual offset given by (8) is the input offset voltage of the opamp building the corresponds to the bias current at the negintegrator, ative input. In our prototype circuit, the value of the residual offset was about 40 mV. The accuracy of the dc-offset measurement is, thus, limited by the same offset but divided by the gain of the electrodes and yields about 0.4 mV. The function of the second feedback loop can be resumed as to implement a high-pass behavior and, thus, remove the dc-offset between the corresponding signal electrode and the reference electrode avoiding saturation of the amplifying electrodes. The time constant of the second feedback loop is much lower (at least five decades) than the time constant of the first feedback loop. This ensures the stability of the second feedback loop despite it being interlinked with the first one (see Section III-B, Fig. 5). The amplifier stage is of a pseudodifferential nature, the amplified and dc-offset free bioelectric signal is represented by . Yet the bioelectric signal is only visible on whereas the common mode signal is present on both signals. would result in large common mode inConsidering only terferences being visible. Fig. 4 illustrates the overall transfer function of the pseudodifferential amplifier and the electrodes. The transfer function can be expressed by the forward gain and the feedback gain as follows: (9)
304
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
A particularity of this amplifier is the fact, that the cutoff frequency of the high-pass depends on the gain of the active electrode. We can use the same amplifier for EEG by using a low-noise electrode with a gain of 60 dB and, therefore, a cutoff frequency of 0.15 Hz. A. DC-Offset Measurement
Fig. 4. Transfer function of the pseudodifferential amplifier stage with the electrodes. The upper corner frequency is fixed by the employed opamp (LMC7111), the overall gain is defined in (9) and the lower corner frequency depends on the feedback loop as expressed in (10).
f
According to the ANSI standard [17] a circuit for biosignals must be able to suppress a possible dc-offset of at least 300 mV between the inputs. Our circuit handles an offset between electrodes by adapting the reference voltage of the electrodes themselves. To analyze this feature more closely we will assume of the th signal electrode and of the a dc-offset between reference electrode. The second feedback loop will then adjust until the dc component of the signal the reference voltage . This leads to the following relation: is again (11) (12) For a large gain equals which corresponds to the offset between electrodes. This means that by measuring the reference voltage of an electrode it is possible to deduce the offset voltage of this electrode compared to the reference electrode. Although this dc-offset is composed of the physiological signal and the skin-electrode offset, the latter varies only very slowly [18]. Fig. 4 displays the frequency response of (12) which can also be expressed as
Fig. 5. CMRR of the amplifier stage is defined by the ratio of the differential of the input signal divided by the common mode gain of gain the input signal (the two solid lines). Both functions show the difference in gain of the corresponding single ended output signals. and are the differential and common mode gain of the output voltage of the reference electrode . Whereas and are the differential and common mode gain of the output of the signal electrode “i” (dotted lines). The dashed line “LMC 7111” represents the gain of the active electrodes without the second feedback loop. The higher frequencies are suppressed as a result of the transfer function of the opamp. The squares represent measured values of the CMRR for our prototype circuit. The value of CMRR was limited by the resistor tolerances as described in (18) to 30 dB.
G
(13)
G
G
V
G
G
G
corresponds to the gain of the electrodes as stated in (1). The bandwidth of is given by the GBW (gain-bandwidth product) of the employed opamp. corresponds to the transfer function of the integrator multiplied by a constant factor given by (7) The resulting function is a high-pass where the corner frequency is . Consequently we can express defined by the condition by the cutoff frequency
holds the same information but with an amplitude ten times larger.To our knowledge this is the first circuit reported which allows to measure a bioelectric signal with frequencies down to dc. A second consequence of (12) is that the limits within which the circuit can compensate a dc-offset are given directly by the and, thus, by the gain in the feedback voltage swing of loop given in (7). B. Amplifier Design and CMRR The main problem of amplifying electrodes is the possible reduction of CMRR due to a variance in gain between electrodes. and Let us assume that the reference electrode has a gain of the signal electrode ì’ has a gain of . If we now calculate the difference we obtain
(10)
(14)
Below the corner frequency the transfer function can be approxand above by . We used imated by and which results in a corner frequency of as required for recording an ECG. To restore the dc value after replacing the electrodes or to allow for fast recovery enabling after an artifact we placed a switch in parallel to to temporarily shift the corner frequency.
(15) Where the differential input and the common mode input are given by (16) (17)
DEGEN AND JÄCKEL: PSEUDODIFFERENTIAL AMPLIFIER FOR BIOELECTRIC EVENTS WITH DC-OFFSET COMPENSATION
305
This allows to directly evaluate the differential gain as well as the common mode gain and therefrom the CMRR as follows: (18) This is a very general relation which also allows to calculate the CMRR limitation based on the “potential divider effect.” Any couple of amplifying electrodes will introduce a severe CMRR limitation if their gain mismatch is not compensated for. As an example we estimate the CMRR of our electrodes. If we assume that we use resistors with a 1% variance the two gains may vary up to 2% and the worst case value for the CMRR is as low as 28 dB. Even if we consider that the CMRR will be improved by the DRL circuit [19] we must conclude, that 28 dB is much less than a standard amplifier for bioelectric events should offer. The other way around, a CMRR of 80 dB corresponds to a resistor tolerance of at least 0.025‰. Before we proceed to the description of the gain compensating method we would like to detail the complex function of the CMRR in the case of a pseudodifferential amplifier. Therefore, we simulated the different gain functions and measured the CMRR in the frequency domain between 5 Hz and 50 kHz. The results are shown in Fig. 5. The simulation and measurements correspond to the case of a noncompensated gain mismatch and do not describe the performance of the final amplifier system. Fig. 5 is an asymptotic representation of the different transfer functions defining the CMRR of the pseudodifferential amplifier based on two spice simulations, one for the differential gain and one for the common mode gain. The definitions of the different lines representing the decomposition of the overall CMRR are (19) (20) (21) The differential mode gain of the reference output voltage is very low because is forced to a constant value by corresponds to the transfer functhe first feedback loop. tion of the active electrodes (line LMC 7111) but is damped at low frequencies by the second feedback loop. Because the differential to differential gain already drawn in Fig. 4 superimposes . The two common mode gains and almost overlap, their difference is given by (18) and was measured for this particular pair of point of is defined by electrodes to be 30 dB. The ) the time constant of the first feedback loop ( which evaluates to about 50 Hz for our prototype circuit. of the second feedback loop they Below the time constant is diverge. Therefore, the differential common mode gain mostly 30 dB below the single common mode gains but follows below . To summarize we can express the CMRR at 50 Hz by (22) For our prototype we obtained a CMRR of 61.7 dB at 50 Hz (with the DRL) for a noncompensated gain mismatch.
Fig. 6. Two ways of interfacing the amplifier by (a) converting all four single ended signals separately or (b) using three differential ADCs to convert them in a pseudodifferential way. This enables simple software based algorithms to compensate for differences in gain and, thus, improving the CMRR.
This is lower than the requirement stated in [18], namely; (without the DRL). This is probably the main reason why amplifying electrodes are not used often in EEG and ECG. The difference in gain of the amplifying electrodes has to be compensated for. This can be done by trimming the gain of the electrodes [6] or by coupling the electrodes [7]. Even coupled electrodes have a limited CMRR due to the component tolerances which can be compensated for by trimming the gain of the differential stage [10]. This trimming can also be applied to noncoupled electrodes. In [11], we describe an autonomous compensation scheme for coupled and noncoupled electrodes. All these schemes could be implemented with the two-wired amplifying electrodes presented in this paper. But there is a even more elegant way to compensate for the gain mismatch which is only possible due to the pseudodifferential structure of the amplifier we developed in Section III. The trimming implemented in [11] is deferred to the software while the analog part is kept small and low-power. The new approach consists in shifting the differentiation normally performed in the amplifier to the software. For this purpose we need a fully differential interface circuit like the two examples shown in Fig. 6. The differential approach requires one more analog-to-digital channel for the reference electrode. The total number of channels is equal to the number of signal electrodes plus one. They can either be supplied by one analog-to-digital converter (ADC) and a multiplexer combined with a sample and hold circuit, or by using several ADCs in parallel. In each case, the bioelectric is then calculated by the following: signal (23) (24) and stand for the gain of the reference electrode where and the th signal electrode, respectively, which has first to be obtained by software. and is to measure a common mode A way to obtain signal, e.g., the 50-Hz power line interference as demonstrated in [11], or by calculating the discrete fourier transform of the
306
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
sampled values and use the values representing the 50-Hz interference. Unlike the method described in [11] this method does not necessitate a phase shift measurement because the two 50-Hz common mode signals are in phase. By implementing (24) based on the measurement of the 50-Hz interference, the latter can be completely removed as shown in Section IV. Any remaining 50-Hz component of the output signal is part of the differential bioelectric signal itself. An example is shown in Figs. 12 and 13. The CMRR is improved to the value of (25) The general value for the improved CMRR can not be given, because the improvement depends also on the value of the measured common mode signal itself. We can say that the CMRR increases as much as necessary to ensure, that the measured common mode signal is minimized on the reading. We can tough give an upper limit to which this schema can function. In the optimal case (the common mode signal is as large as the input range of the ADC), the upper limit can be given by (26) Where “ ” corresponds to the resolution of the ADC. Contrary to a 50-Hz notch filter the method will reduce all common modes signals without altering the differential mode signal, i.e., the bioelectric signal. The difference between the two circuits in Fig. 6 is the required input range of the ADC. In Fig. 6(a), the input range must whereas in Fig. 6(b) the input range can be at least be as low as . As a consequence the resolution for the same number of bits is higher in Fig. 6(b). As an example, the and proAD7731 can have an input range as low as vides three differential ADC in one IC. At 400 Hz the AD7731 . The upper limit of the has a resolution of 14.5 bit or 1.73 CMRR improvement according to (26) would evaluate to 87 dB (in addition to at least 28 dB CMRR of the electrodes). C. Digital Control Loop Regardless of which AD-conversion scheme is chosen, the second feedback loop responsible for the dc-offset compensation can be implemented optionally by a digital controlled loop [20] as shown in Fig. 7. The digital loop functions in a similar way to the analog one. is generated by the digital-to-analog The feedback voltage converter (DAC) while the digital filter [finite impulse response filter (FIR)] generates the necessary codes for the DAC, based on the differential input of the ADC. The opamp stage is needed for sourcing the current and adapting the output range of the DAC to the desired reference voltage range. Equation (7) is still valid and defines the function of the opamp stage. The main differences between the analog and the digital implementation are as follows: 1) the digital loop requires no capacitor; 2) the digital loop introduces discrete steps in the output waveform (quantization noise).
Fig. 7. Second feedback loop built as a digital control loop by replacing the integrator of Fig. 3 with a digital filter (FIR) and a DAC. The feedback loop is again shown for the electrode “i.” There is no need for a capacitor.
Between these steps the signal does not show the typical superimposed exponential variation which is otherwise characteristic of high-pass filtered signals. The absence of a capacitor in the digital control loop enables to control the time behavior of the circuit in a much more flexible and accurate way. The time constants between different channels match perfectly, a switch to temporarily change the time constant of the high-pass filter is not required any more. Instead adaptive filter techniques may be used resulting in shorter settling times. Likewise the part count decreases as the DAC replaces an opamp together with some passive components as well as the ADC necessary to convert the dc-offset. The part count is even lower than in the system reported in [8]. The fully digital circuit also enables a seamless integration of the amplifier circuit. No large capacitors or large resistors are needed which both are not easily implemented in a standard IC-design process. the Because of the discrete nature of the output voltage dc-offset of the bioelectric signal cannot be completely nulled. There remains a residual offset which can be calculated in relation to the resolution of the DAC (27) denotes the number of bits of the DAC. For a supply voltage of 5 V and a DAC with 12 bits this results in a residual offset . In other words, each time the digital code voltage of will of the DAC is changed by the FIR filter, the signal display a discrete step (quantization noise) with an amplitude corresponding to a multiple of the residual offset calculated in (27). If not handled appropriately for, this leads to an oscillating with a mean value of zero behavior of the output voltage volt. This behavior is exemplarily shown in Section IV, Fig. 14. Because the exact timing of these steps is known a priori by the software the latter can be removed before displaying or storing the signal. A better solution is to implement two thresholds which differ by the value of the residual offset calculated in (27). As long as the signal stays between the thresholds a change of the DAC will be inhibited.
DEGEN AND JÄCKEL: PSEUDODIFFERENTIAL AMPLIFIER FOR BIOELECTRIC EVENTS WITH DC-OFFSET COMPENSATION
Fig. 8. Shielded amplifying two-wired active electrode mounted onto a button in order to attach different types of disposable electrodes.
Fig. 9. Noise spectrum of the amplifying two-wired electrode and the total system (no additional gain) was measured with a vector signal analyzer (Stanford SR785). For comparison we added the noise spectrum of a two-wired buffer electrode (gain = 1) using the same operational amplifier. Different sampling rates were used above and below 100 Hz. f denotes the higher bandwidth of the amplifier system (200 Hz).
IV. RESULTS The fabrication of the prototype electrode was relatively straight-forward. We mounted the components on a printed circuit board which was then soldered onto the back of a snap-button cut from a pair of trousers allowing us to use different kinds of disposable electrodes. Fig. 8 shows a disposable pregelled EEG electrode clipped to an active electrode. The prototype electrodes are left unprotected to ease changing of the soldered components. First we measured the input and output referred noise of the amplifying electrodes as well as of the total amplifier. The measurement is shown in Fig. 9. The noise of the amplifying electrode is mainly due to the opamp used, which for this implementation is an LMC7111. The total input referred noise of the electrode corresponds to the integrated noise spectrum over the desired bandwidth. For a bandwidth of 1 Hz–1 kHz the computed integral results in 4.7 . If we consider the total amplifier then the noise is mainly due to two sources, the active electrode and the noise generated by the error feedback loop. This is because the noise of and the error feedback loop affect the reference voltage is, therefore, amplified by the active electrode. The total noise of the system in the bandwidth of 1 Hz-1 kHz computes to 7.4 . This is low enough for our test application. Currently, we are developing a EEG amplifier based on the herein proposed noise. method with less than 1
307
Fig. 10. A sample ECG recording made with the circuit shown in Fig. 3, the recording was made with disposable pregelled electrodes from Conmed. The gain of the active electrode was about 100, the values of the resistors were R = 47 k , R = 9:1 k , R = 4:7 k , R = 10 M , C = 10 F, = 220 k and C = 2:2 nF. The ADC were two ADA400A from R Tektronix.
For comparison we added the noise spectrum of the same active electrode with a gain of 1. (“buffer electrode” with and ). This confirms, that using a amplifying electrode with a gain of 40 dB lowers the noise when compared to the same active electrode configured as a buffer electrode. This improvement may be considered to be small, but it comes without an increased supply and it allows to reduce the noise requirements of the following stage. The noise reduction becomes the more important the higher the gain of the active electrode is. To our knowledge 40 dB corresponds to the second highest gain of any reported input stage for a bioelectric amplifier [8]. We then evaluated our amplifier by measuring a three lead ECG. The system was composed of a DRL electrode together with the active reference electrode and one signal electrode as shown in Fig. 3. We believe that an ECG is best suited to evaluate the quality of a bioelectric signal due to its distinctive shape. The ECG signal was recorded without any skin preparation by using disposable self-adhering pregelled electrodes as shown in Fig. 8. measured Fig. 10 shows a recorded ECG signal with a differential ADC as well as the dc-offset voltage between the electrodes. The ECG in Fig. 10 shows a flat baseline and very little 50-Hz interference and this with an overall gain of only 100. The offset voltage of the ECG should be zero as a result of the error feedback loop; the 40 mV visible in the figure correspond to the value calculated in (8). The signal amplitude of an ECG is around 1 mV (times 100), therefore, we chose a differential ADC with an input range of 200 mV. The low frequency parts ). The of the signal was measured simultaneously ( total dc-offset between the electrodes is about 12 mV, it is probably mainly due to the polarization at the skin-electrode interface. The next measurement in Fig. 11 shows the two signals of a and pseudodifferential configuration corresponding to circuit Fig. 6(b). These are the two components of the pseudodifferential signal normally not accessible in standard bioelectric measurements.
308
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
Fig. 11. Pseudodifferential ECG recording made with the configuration shown in Fig. 6(b). The recording was made with the same equipment as before. These two waveforms are normally not available with common amplifiers. The distinct noise visible on both waveforms are the common mode signals which will be removed by calculating the difference similar to the signal shown in Fig. 10. For V of 20 mV. better visualization we added an artificial offset to V
0
0
Fig. 13. Digitally calculated difference of the two pseudodifferential signals of Fig. 11. The enlargement shows no 50-Hz interference, the discrete fourier transform in Fig. 12(d) reveals an amplitude of 2.85 V of the 50-Hz component. This component is probably due to the bioelectric signal itself. The high frequency spikes are probably due to the absence of any anti-aliasing filter.
only present on one lead; . Two factors are helpful in this situation, firstly, the 50-Hz common mode signal is usually much stronger than the 50-Hz component of the bioelectric signal and secondly, the two 50-Hz common mode signals are in phase. Thus, by taking the difference between the two unipolar signals the common mode signal will be strongly reduced. This would result in a signal similar to the signal measured in Fig. 10 with the corresponding frequency decompensation shown in Fig. 12(c). The achieved CMRR corresponds to the value defined by (22). That there is a 50-Hz component of the ECG signal which is not in phase with the common mode signal can be seen by the fact that the difference of the two 50-Hz components of the unipolar signal and the 50-Hz component of ). the 50-Hz signal do not perfectly match ( Now we can go further and obtain a better reduction of the 50-Hz component by using the following formula (only valid for this specific measurement): (28)
Fig. 12. Frequency decomposition of the different signals involved. (a) and (b) Two pseudodifferential signals shown in Fig. 11. (c) and (d) Digitally obtained differential signals. (c) Corresponds to the simple difference of the two signals shown in Fig. 11 and (d) corresponds to the digitally improved signal shown in Fig. 13. The value written in italic corresponds to the absolute value of the 50-Hz component.
We can confirm the pseudodifferential nature of the configuration by the fact, that the biopotential signal is present only on the signal electrode, yet the common mode interference is present on both signals (the visible “noise”). Both signals still show all the common mode noise, e.g., the 50-Hz power line interference is clearly visible in the enlargement. The higher frequency contributions visible are mostly induced by the oscilloscope (used for visualization) standing close to the subject. The frequency decomposition obtained by Fourier transformation of the two signals can be seen in the corresponding clipping; namely Fig. 12(a) and (b). As we can see, the 50-Hz components of the two pseudodifferential signals are not the same. This is mainly due to the difference in the gain of the two electrodes but also to the fact that the bioelectric signal itself has a 50-Hz component which is
This is simply the implementation of (25). In Fig. 13, we can see the corresponding waveform and Fig. 12(d) shows the corresponding frequency decomposition. The 50-Hz component is which is most probably no common mode inabout 2.85 terference but the bioelectric signal itself. This is supported by the fact, that the phase between the 50-Hz component of the difference signal and the 50-Hz component of the reference signal is 88.7 which is nearly orthogonal. It is impossible to give an exact value for the CMRR, because we can only roughly estimate the common mode part and the difference mode part of the were indeed only 50-Hz component. If the remaining 2.85 due to the differential signal the CMRR at 50 Hz would be infibeing in phase nite. If we estimate that the part of the 2.85 were the remaining common mode interferwith the 943 ence, then the improvement of the CMRR can be calculated by (29) The total CMRR (including the 30 dB of the electrodes) would then evaluate to about 78 dB and the final CMRR (including the DRL) would be equivalent to about 128 dB for 50 Hz.
DEGEN AND JÄCKEL: PSEUDODIFFERENTIAL AMPLIFIER FOR BIOELECTRIC EVENTS WITH DC-OFFSET COMPENSATION
Fig. 14. Sample ECG recording obtained with a digital control loop as shown in Fig. 7. The oscillating is due to the discrete steps of the DAC. The amplitude corresponds to the residual offset calculated in (27) which was artificially increased to 24 mV for better visualization. The frequency can be calculated similarly to (10). The mean value of the recording is 0 V. The oscillation can be suppressed by adding thresholds to the FIR filter or by software.
The trace of the digitally calculated ECG shows no 50 Hz common mode interference as shown in the corresponding frequency spectrum in Fig. 12(d). There is visibly more noise than in the fully differential recording shown in Fig. 10. One reason is that by using two different ADCs their input noise will add up. The other reason may be that for this prototype circuit there was no anti-aliasing filter and, therefore, no reduction of high frequency noise. The different recordings were taken on different days and the electrodes were not placed at exactly the same location. This last paragraph is to give some results regarding the optional use of a digital control loop. Fig. 10 shows an ECG recorded with a digital control loop as shown in Fig. 7. The purpose of this recording is to demonstrate the oscillation (quantization noise) due to the discrete steps of the DAC resulting in jumps of the size calculated in (27). The gray line shows the recording when the software would inhibit any ADC change while the mean value of the bioelectric signal lays within a higher and a lower threshold. The two thresholds are indicated by the semitransparent region behind the recordings. The advantage of the digital control loop is the absence of any capacitor. There is also no superimposed hyperbolical relaxation curve like in high-pass filtered signals. This enables for example a precise measurement of the S-T elevation even shortly after a spike has temporarily driven the amplifier into saturation. Using a digital control loop is the ultimate step in transferring the signal treatment from the analog world to the digital world. Digital filters have the advantage of a perfectly linear phase response allowing to keep the original shape of the bioelectric signal. V. CONCLUSION In this paper, we presented a new amplifying two-wired elecand trode with a gain of 40 dB, an input resistance of over 1 a output resistance of about 25 as well as the corresponding pseudodifferential bioelectric amplifier. The amplifier section provides the supply current for the electrodes as well as the necessary dc-offset compensation by adapting the reference voltage of the amplifying electrodes. The entire gain of the amplifier is
309
concentrated at the first stage enabling a low noise and low part count design. In addition, the circuit allows to simultaneously measure the bioelectric signal as well as the low-frequency components of the same signal, although with two different resolutions. This enables to measure clinically relevant data normally not available with other reported amplifiers. We then analyzed the common mode rejection ratio (CMRR) of the amplifier. Because of its pseudodifferential structure the difference in gain of the amplifying electrodes can, for a first time, be compensated for by the digital signal processing. This overcomes the lack of an easy to implement method to compensate for a difference in gain of noncoupled amplifying electrodes. Unlike a 50-Hz notch filter the improvement the CMRR reduces common mode interferences at all frequencies of interest. with a The circuit is able to compensate up to supply voltage of 5 V and a power consumption of 7.5 mW per channel. The supply voltage can be reduced in future by using operational amplifiers working with a supply of 1.8 V or less. A prototype for an EEG recorder is currently under development with 3.7-V power supply and 1.8 mW power consumption per channel. To our knowledge this is the first published amplifier for biosignals with this specific combination of features, namely, high gain in the input stage, truly dc compliant and a digital-only implemented improvement of the CMRR based on gain adaption. In addition, the amplifier uses two-wired electrodes that reduce the stiffness of the connecting wires and the size of the connectors, has a very low part-count, is highly customizable and offers the possibility to digitally control all filter parameters. REFERENCES [1] P. C. Richardson, F. K. Coombs, and R. M. Adams, “Some new electrode techniques for long-term physiologic monitoring,” Aerosp. Med., pp. 745–750, 1968. [2] W. H. Ko, “Active electrodes for EEG and evoked potential,” in Proc. 20th Annu. Int. Conf. IEEE Engineering in Medicine and Biology Society, vol. 4, H. K. Chang and Y. T. Zhang, Eds., 1998, pp. 2221–2224. [3] S. Nishimura, Y. Tomita, and T. Horiuchi, “Clinical application of an active electrode using an operational amplifier,” IEEE Trans. Biomed. Eng., vol. 39, no. 10, pp. 1096–1099, Oct. 1992. [4] M. Fernandez and R. Pallas-Areny, “A simple active electrode for power line interference reduction in high resolution biopotential measurements,” in Proc. 18th Annu. Int. Conf. IEEE Engineering in Medicine and Biology Society, vol. 1, H. Boom, C. Robinson, W. Rutten, M. Neuman, and H. Wijkstra, Eds., 1997, pp. 97–98. [5] F. Z. Padmadinata, J. J. Veerhoek, G. J. A. van Dijk, and J. H. Huijsing, “Microelectronic skin electrode,” Sensors Actuators B (Chemical), vol. B1, no. 1, pp. 491–494, Jan. 1990. [6] W. J. R. Dunseath and E. F. Kelly, “Multichannel pc-based data-acquisition system for high-resolution EEG,” IEEE Trans. Biomed. Eng., vol. 42, no. 12, pp. 1212–1217, Dec. 1995. [7] E. S. Valchinov and N. E. Pallikarakis, “An active electrode for biopotential recording from small localized bio-sources,” Biomed. Eng. Online, vol. 3, no. 25, Jul. 2004. [8] A. C. MettingVanRijn, A. P. Kuiper, T. E. Dankers, and C. A. Grimbergen, “Low-cost active electrode improves the resolution in biopotential recordings,” in Proc. 18th Annu. Int. Conf. IEEE Engineering in Medicine and Biology Society, vol. 1, 1997, pp. 101–102. [9] R. Pallas-Areny and J. G. Webster, “Common mode rejection ratio for cascaded differential amplifier stages,” IEEE Trans. Instrum. Meas., vol. 40, no. 4, pp. 677–681, Aug. 1991. [10] , “Common mode rejection ratio in differential amplifiers,” IEEE Trans. Instrum. Meas., vol. 40, no. 4, pp. 669–676, Aug. 1991.
310
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 53, NO. 2, FEBRUARY 2006
[11] T. Degen and H. Jäckel, “Enhancing interference rejection of amplifying electrodes by automated gain adaption,” IEEE Trans. Biomed. Eng., vol. 51, no. 11, pp. 2031–2039, Nov. 2004, to be published. [12] S. Zschocke, Klinische Elektroencephalographie. Berlin, Germany: Springer, 2002, pp. 18–19. [13] A. Ikeda et al., “Focal ictal direct current shifts in human epilepsy as studied by subdural and scalp recording,” Brain, vol. 5, no. 122, pp. 827–838, 1999. [14] S. Vanhatalo et al., “DC-EEG discloses prominent, very slow activity patterns during sleep in preterm infants,” Clin. Neurophysiol., vol. 113, no. 11, pp. 1822–1825, Nov. 2002. [15] B. B. Winter and J. G. Webster, “Driven-right-leg circuit design,” IEEE Trans. Biomed. Eng., vol. BME-30, no. 1, pp. 62–66, Jan. 1983. [16] A. C. MettingVanRijn, A. Peper, and C. A. Grimberger, “Amplifiers for bioelectric events: A design with a minimal number of parts,” Med. Biol. Eng. Comput., vol. 32, no. 3, pp. 305–310, May 1994. [17] Ambulatory Electrocardiographs, American National Standard ANSI/AAMI EC38:1998, 1999. [18] A. C. Metting-van Rijn, A. Peper, and C. A. Grimbergen, “High-quality recording of bioelectric events. Part 2. Low-noise, low-power multichannel amplifier design,” Med. Biol. Eng. Comput., vol. 29, no. 4, pp. 433–440, Jul. 1991. , “High-quality recording of bioelectric events. Part 1. Interference [19] reduction, theory and practice,” Med. Biol. Eng. Comput., vol. 28, no. 5, pp. 389–397, Sep. 1990. [20] M. Verbeke, S. Gur, A. Ron, and R. Iraqi, “Manual or automatic nulling DC offset for physiological DC amplifier,” Med. Eng. Phys., vol. 16, pp. 171–171, Mar. 1994.
Thomas Degen (S’99–M’03) received the diploma in physical electronics in 1995 from the University of Neuchâtel, Neuchâtel, Switzerland. For the next two years he worked on industrial projects at the Interdisciplinary Institute of Integrated Systems at the Engineering School of Biel-Bienne. In 1997, he took part in an exchange program, doing research at the Indian Institute of Science, Bangalore, India. In 1998, he joined the Swiss Federal Institute of Technology, Zurich, Switzerland, where he is currently working toward the Ph.D. degree. His research interests include biomedical instrumentation, low-voltage analog circuit design, and wireless transmission.
Heinz Jäckel (M’82) received the Ph.D. degree in electrical engineering from the Swiss Federal Institute of Technology, Zürich, Switzerland, in 1979. From 1979 to 1993, he worked as a Research Staff Member and in managerial functions at the IBM research lab, Rüschlikon, Switzerland. In 1993, he returned to the Swiss Federal Institute of Technology, where he is now a full Professor with the Institute of Electronics and Head of the high-speed and analog electronics group. His research interests are design, characterization and technology of InP-based heterobipolar transistors, high-speed analog and digital circuitand IC-design, all-optical photonic devices for TB/s-data communication and CMOS IC-design for RF- and sensor-applications.
