In present biomedical research, recording systems with a high number of measuring channels are often required. Examples of multichannel recordings are brain mapping: EEG measurements with 20 channels (Duffy, 1982), 28 channels (Gruzelier et al., 1988) or 36 channels (Duff, 1980) and body surface mapping: ECG measurements with 64 channels (Reek et al., 1984; Spekhorst et al., 1990). Over the last years, large progress has been made in the development of the digital part of recording systems with respect to the processing of the large amount of data gathered. The quality of the analog front end, however, has remained much the same and in many cases does not stand up to its task. The proper measurement and amplification of bioelectric signals present some specific problems. General amplifier design techniques are well developed and extensively described in literature. However, in designing a biomedical amplifier much attention should be paid to the special character of a bioelectric measurement. The fact that electrodes are used as signal transducers in bioelectric recordings has important consequences for the required performance of the amplifier. Because of its impedance, offset voltage and noise contribution, the electrode-skin interface may cause various problems, especially in multichannel recordings. Finally, the goal to make the amplifier applicable to many different bioelectric (multichannel) measurement situations received extra consideration.
The electrode-skin interface has a complex impedance between 1 k and 1 M at 50 Hz (Almasi and Schmitt, 1970; Grimnes, 1983; Geddes, 1972; Rosell et al., 1988). The electrode-skin impedance depends on many factors like skin condition and preparation, and large variations in electrode-skin impedance can be found. High electrode-skin impedances and/or large differences between the electrode-skin impedances may cause the measurement system to be susceptible to several forms of interference. The theory was extensively treated in a previous study (Metting van Rijn et al., in press). To ensure correct operation with typical electrode-skin impedances and typical interference sources, an amplifier should meet the following demands:
The electrode-skin interface forms a galvanic half-cell (Geddes, 1972). The exact
half-cell voltage of a electrode-skin interface depends mainly on the condition of the
skin and electrodes. To the signal measured between the electrodes, the difference between
the half-cell voltages of the electrode-skin interfaces must be added. This voltage
usually varies slowly during a recording, resulting in a low frequent noise signal. An
amplifier should suppress these relatively large (several tens of millivolts),
low-frequency input signals in order to prevent saturation. A cut-off frequency of
approximately 0.15 Hz for the high-pass filtering is commonly used for this objective. The
signal magnitude that minimally can be measured is defined in principle by the thermal
noise of the electrode-skin impedance (Et = 1.3 x 10-3 Re micro;Vrms in a
bandwidth of 0.1 - 100 Hz). In practice however, noise levels this low are almost never
obtained due to variations in the half-cell voltage of the electrode-skin interface. In
order to be able to decide about clinical death, an equivalent input noise level lower
than 2 micro;Vp-p in a bandwidth of 0.1 - 30 Hz is mandatory for clinical EEG amplifiers
(Cooper et al., 1969; Silverman et al. 1969). Consequently, the target value for the
equivalent input noise voltage in this study is chosen to be 0.5 micro;Vrms in a bandwidth
0.1 - 100 Hz . An equivalent input noise current less than 5 pArms in a bandwidth of 0.1 -
100 Hz is demanded to ensure that only a small amount of noise is added with typical
electrode-skin impedances (total equivalent input noise voltage is 0.6 micro;Vrms in a
bandwidth of 0.1 - 100 Hz with electrode-skin impedances of 50 k ).
In many multichannel recordings the signal differences between each of the measuring electrodes and a common reference voltage is measured (monopolar recording). This reference voltage is sometimes derived from one electrode but often the average voltage from two (EEG) or three electrodes (ECG) is used (Clark, 1978; Geddes and Baker, 1975). These measuring configurations should be considered when designing a multichannel amplifier. Small dimensions are important for several reasons. It is obvious that small dimensions will enable the amplifier module to be used in all kinds of portable and telemetry measurements but a small size of the amplifier also reduces the capacitance between the amplifier and its environment in an isolated amplifier set-up. Reduction of this capacitance lowers the common mode voltage, which voltage is one of the main causes for interference (Metting van Rijn et al., in press).
Minimum power consumption is important because it allows small sized batteries to be used as power supply. This is of course essential if the amplifier module is used in ambulant recordings but can present some important advantages in stationary measurement systems as well. In a system which is powered by the mains supply by the use of an isolation transformer, most of the capacitance across the isolation barrier is usually caused by the power supply (Metting van Rijn et al., in press). A small battery-powered amplifier can be designed to have very small capacitance to the environment, resulting in a low common mode voltage. Another advantage is the greatly improved safety because there are no high voltages present in the amplifier cabinet. Finally, batteries deliver a very "clean" supply voltage which is essential for low-noise operation (Motchenbacher and Fitchen, 1972). Mains supply voltages are often polluted with a lot of high-frequency interference (spikes).
Before concentrating on multichannel amplifiers it is instructive to review the almost universally used instrumentation amplifier with three operational amplifiers, pictured in Fig. 1A (Tobey et al.,1971). The advantages of the circuit are:
Fig. 1 : (a) Instrumentation amplifier with three operational amplifiers. Biomedical amplifiers are usually based upon this design. The voltage at point Pav is equal to the common mode voltage. (b) Multichannel instrumentation amplifier with operational amplifiers. The circuit is an extension of the three operational amplifier circuit of Fig. 1A and its favorable properties are preserved.
The three operational amplifier circuit, being widely available as a single monolithic
or hybrid IC, is an attractive choice as a building block for a multichannel amplifier
(Smit et al., 1987) when it is used as an independent instrumentation amplifier with its
own pair of measuring electrodes (bipolar recording). However, when the potential
difference between a number of electrodes and a common reference electrode is to be
measured (monopolar recording, for example clinical ECG and EEG measurements) the three
operational amplifier circuit is not optimal. It is tempting to interconnect several
amplifier inputs to obtain a common reference input. However, this method introduces
differences in common mode input impedance between the reference input and the other (not
interconnected) inputs resulting in an increase of interference due to the "potential
divider effect" (Metting van Rijn et al., in press). The inclusion of a buffer
amplifier at the reference input may prevent the degradation of common mode input
impedance but will add extra noise to the input signals, which are not amplified at this
point. A multichannel amplifier set-up without these drawbacks is shown in Fig. 1B. It is
basically an extension of the three operational amplifier configuration and all the
favorable characteristics mentioned above are preserved (O'Brien et al., 1983). Each input
signal is amplified with respect to the voltage at point Pav, which voltage is equal to
the average of the input voltages (= common mode voltage). Note that all amplifier inputs
are equal. For practical biomedical use the circuit is completed with appropriate
high-pass and low-pass filtering, guarding circuits and a driven right leg circuit. 4.
Biomedical amplifiers with discrete input stages In the present study, the goal was to
develop a miniature multichannel analog front end with optimal noise properties and
minimum power consumption. The specifications were based on the considerations given in
Section 2 and are listed in the first column of Table 1. The design had to be suitable to
be manufactured with thick-film production techniques which offers higher reliability and
far smaller dimensions than printed circuit board production techniques (Sergent, 1981).
Table I: Amplifier module specifications.
|Parameter||Target||Design Fig. 3§|
|Equivalent input noise voltage (µVrms, 0.1 - 100 Hz)||< 0.5||0.35|
|Equivalent input noise current (pArms, 0.1 - 100 Hz)||< 5||4|
|Bandwidth +0,-3 dB (Hz)||0.16-100||0.16-100|
|Differential mode DC input range (mV)||> 150||190|
|Differential mode AC input range (mVp-p)||> 20||25|
|Common mode input range (Vp-p)||> 2||2.5|
|Input bias current (nA per input)||< 50||23 +/- 5|
|Gain: 500 500 - differential mode input impedance (Mohm at 50 Hz)||> 10||> 15|
|Common mode input impedance : with unshielded input leads (Mohm at 50 Hz)||> 100||250|
|Common mode input impedance : with guarded input leads (M ohm at 50 Hz) (cable capacitance is 330 pF)||> 100||200|
|Common Mode Rejection Ratio (dB at 50 Hz)||> 80||91 +/- 5|
|power consumption (mW per channel)||< 1||0.6|
The amplifier design shown in Fig. 1B seemed attractive because of its simplicity. However, with this circuit it proved impossible to achieve the desired combination of low noise and low power consumption. It can be calculated (see Appendix A1 and A2) that it is possible in theory to surpass our specifications (Table 1) with a power consumption 10 - 100 times lower than offered by currently available low-noise operational amplifiers. The relatively high power consumption of operational amplifiers is mainly due to an extended bandwidth which however is superfluous for most physiological applications. When programmable operational amplifiers are used, they can be adjusted for a small bandwidth and low power consumption but none of these devices has an acceptable noise level . Therefore a multichannel instrumentation amplifier with a discrete input stage was developed. An input stage based on discrete components can offer some important advantages: the best discrete transistors are superior to the input transistors in IC operational amplifiers (Nelson, 1980) and the increased design freedom allows the input stage to be optimized for biomedical applications. In the next section a one-channel version of the amplifier is described. In addition, a new concept for expanding a one channel instrumentation amplifier to more channels was developed. The resulting multichannel design that was used in the thick-film modules is described in section 6.
There was already experience gathered with a previously developed one-channel instrumentation amplifier (Hamstra et al., 1984), Starting from that design an amplifier which met more severe specifications was developed. The circuit is shown in Fig. 2 :
Fig. 2 : Simplified schematic of an instrumentation amplifier for biomedical
applications. The combination of low-noise and very low power consumption is achieved by
the use of discrete transistors in the input stage.
The input stage is based on the well known current balance amplifier circuit (Wong and
Ott, 1976; Graeme, 1977) with current sources T3 and T4 replacing the common emitter
resistor of the input transistors. The high output conductance of these current sources
provides a very high common mode input impedance . The bias of T3 is fixed, T4 is part of
the feedback circuit. In the second stage operational amplifiers are used to keep the
number of parts low. The programmable operational amplifiers used (LM4250, National
Semiconductor) can be set for very low power consumption while their rather high noise is
not troublesome because of the amplification in the first amplifier stage. Operational
amplifier "amp A" produces a single ended output. Operational amplifiers
"amp B" and "amp C" regulate the feedback signal. The design employs
the "current feedback" technique which offers high common mode rejection ratio,
without the need for very precise resistor matching. Operational amplifier "amp
C" is used as an integrator to provide maximum feedback for DC input voltages. DC
suppression without the need for large capacitors is accomplished with this circuit
(capacitors larger than 100 nF are difficult to apply with thick-film production
The measured performance characteristics of the design of Fig. 2 are listed in Table 1, second column. When high quality resistors (with a low excess-noise level, see Appendix A2) are used, the equivalent input noise voltage is determined by shot-noise currents of the four transistors in the input stage only. With the use of high quality transistors (LM194, National Semiconductor) a compromise between low noise, low power consumption, DC input range and differential mode input impedance was obtained superior to an input stage made with operational amplifiers. A detailed description of how to find the best balance between the various specifications is given in Appendix A1.
When the circuit described in Section 5 is to be used in a multichannel amplifier for monopolar recordings it is again not advisable to interconnect the inputs of a number of independent instrumentation amplifiers. The situation is comparable with the one in Section 3; the amplifier circuit itself must be adapted for multichannel use. We developed an unconventional method to expand the design to more channels. In our multichannel version, none of the good specifications of the one channel design is compromised. The way the extension to more channels was accomplished is shown for n=2 in Fig. 3 (n = number of channels).
Fig. 3 : Simplified schematic of a multichannel instrumentation amplifier with discrete transistors in the input stage. The circuit is based on the one-channel design of Fig. 2. The amplifier is equipped with Guarding and Driven Right Leg circuits and an additional CM-sense input stage.
The operation of the circuit is easily understood when the behavior of the circuit is examined with the same (common mode) signal applied to all inputs. In this case the (AC) currents in all sections are equal and there is no voltage across the resistors Rg which interconnect the sections. Because there are no potential differences between the inputs of the operational amplifiers "amp A", all of the operational amplifiers will have a zero output signal.
When differential mode input signals (signals at the measuring inputs with respect to the reference input) are added to the common mode signal, the voltage at point Pav will be equal to the average input signal (= common mode input signal). The coupling point Pav in Fig. 3 is similar in this aspect to the resistor coupling point Pav in the design with operational amplifiers, see Fig. 1B. However, the differential mode input signals produce differences in the currents between the sections. These differences in current produce potential differences between the inputs of the operational amplifiers "amp A" and consequently each operational amplifier will have an output voltage proportional to the voltage between the reference electrode (input E0) and the corresponding measuring electrode (input E1 or E2).
The design is a multichannel instrumentation amplifier for monopolar measurements: the amplifier has one inverting (reference) input and a number of non-inverting inputs. All the important characteristics of the multichannel instrumentation amplifier with operational amplifiers (Fig. 1B) - no amplification of common mode signals in the input stage, amplification of differential mode signals in the input stage and equal common mode input impedances for all inputs - are also present in the design depicted in Fig. 3.
The measured characteristics of the multichannel design are the same as the characteristics of the one-channel amplifier with discrete input stage (see Table 1). When compared with n independent one-channel amplifiers (Fig. 2), the multichannel design of Fig. 3 offers a considerable reduction in the number of parts while the power consumption is reduced by approx. 40 %. Facilities are incorporated for operation with less than n channels. The circuit described above does not function properly in case of one or more badly functioning electrodes because the corresponding input sections influence the voltage at point Pav. Therefore, an extra input section was added (CM sense section, see Fig. 3) which provides the possibility to use the amplifier in two ways (notation: n is the number of channels in a module, x is the number of channels that is used in the recording):
It was experienced that some points need special attention if the one-channel design
(Fig. 2) is to be expanded into a multichannel design (Fig. 3).
In a one channel instrumentation amplifier the two input transistors can be chosen as a monolithic pair which ensures good matching (high common mode rejection ratio) and good tracking of the two transistors with temperature (low drift). The multichannel design (Fig. 3) needs a n+1 matched transistor array (n is the number of channels in a module) which is currently not available in the quality range needed. However, the matching between different LM194 pairs proved good enough to ensure a common mode rejection ratio of at least 86 dB (approx. 10 dB less than possible with a monolithic matched pair). The use of the thick-film technique helps to reduce temperature differences between the different transistor pairs because the ceramic substrate, on which all parts are mounted, has very good heat conducting properties. In addition, all the transistors are situated close together in the thick-film layout (see Fig. 4). Moreover, the offset voltages produced by temperature gradients are cancelled by the amplifiers DC suppression circuit. An extra buffer (operational amplifier "amp R" in Fig. 3) is included in the reference section to prevent differences in bias-currents to occur in the individual sections and to prevent loading of the reference section by a large number of interconnected operational amplifier inputs. Otherwise, the imbalance introduced in the circuit would decrease the common mode rejection ratio. The extra buffer ensures the equality of common mode currents in all sections while the noise contribution is insignificant because (differential mode) signals are already amplified more than 20 dB at the point where the buffer is incorporated in the circuit.
A thick-film module (60x40x7mm) has been developed which contains an 8 channel version
of the amplifier described above, together with a multiplexer (see Fig. 4). The noise
spectrum of one of the prototype modules is given in Fig. 5. It should be noted that the
1/f corner frequency is ca. 1 Hz, which is considerable lower than offered by currently
available low noise operational amplifiers (the industry-standard OP27 (PMI) low noise
operational amplifier has a 1/f corner frequency of 2.7 Hz). Because of the exceptionally
low 1/f noise of the input transistors (LM194, corner frequency < 0.1 Hz), the 1/f
corner frequency of the modules is determined by excess noise of the resistors (see
Appendix A2). The use of very high quality thin-film chip resistors (Reinhardt Microtech
AG, model URA) proved to be essential in achieving low noise.
Two or four modules can be coupled to make a 16 or 32 channel system with one multiplexed output. More modules can be coupled but the internal multiplexer can only handle up to 32 channels. CM-sense and reference inputs can be used in different configurations to obtain alternative reference choices. For example: with two modules, a 16 channel ECG front end with standard Wilson Central Terminal (WCT) reference (Wilson et al., 1934) can be formed. Guarding and driven right leg circuits in each module ensure that the mains interference is lower than the noise level of the amplifier in almost any situation. A module (8 channels with a multiplexer, 40 mW) operates for 48 hours on two standard 9 V radio-batteries (110 mAh each). Several modules have been manufactured by a thick-film production firm. The modules will become commercially available as soon as the current evaluation program has been completed.
This work was supported by the Technology Foundation (STW).