3. AMPLIFICATION OF GASTRIC ELECTRICAL ACTIVITY
3.1. AMPLIFIER REQUIREMENTS.
Although gastric electrical activity, which manifests itself as biovoltages, has been recorded for many decades, no special attention has been devoted to the problem of its proper amplification. The preferred approach was empirical modification of cut-off frequencies of all-purpose multichannel amplification units for the needs of gastric electrical signals (11, 17, 45). Because of the fact that this approach usually provides results, the problem has not been addressed until recently.
3.1.1. FLEXIBLE FREQUENCY RANGE.
The results from the previous section clearly indicate that the more similar cutaneous recordings are to the signals recorded invasively, the better the chances are for the useful clinical application of EGG. A well designed amplifier might record cutaneous LDB or even SDB gastric electrical signals comparable to the internal ones. The fact that today the EGG is recorded only with the a LDB configuration using standard EKG electrodes with a large active surface area, does not necessarily mean that this is the only possible EGG waveform. Signals with different waveforms have, in general, frequency spectra with different bandwidths even if their periods are the same (42). Therefore, if different electrode configurations or/and active surface areas are to be used, different low-pass and high-pass cut-off frequencies of the bandpass characteristic of the amplifier should be available. Whatever the waveform of the GEA signal is, its period of repetition is about 3 cpm, but variations from about 2 cpm (bradygastria) to 9 cpm (severe tachygastria) can be expected. Internal SDB signals can contain higher harmonics up to 4-5 Hz (11, 42), while the information from EGG signals is concentrated primarily in the fundamental (17, 40, 42, 45).
3.1.2. WIDE RANGE OF GAINS.
In the next section special attention is devoted to the importance of the EGG amplitude. Here it should be mentioned, that the GEA amplitude is also dependent on the interelectrode distance and electrode active surface area. Cutaneous signals become weaker as the distance between the stomach and the abdominal electrodes becomes greater (e.g. in obese patients), when the distance between the electrodes in one bipolar pair is less, or when the recording electrodes have smaller active surface area. A peak-to-peak amplitude range between 0.01 and 0.5 mV can be expected for EGG. The amplitude range for invasively recorded GEA signals is 0.1-10 mV (9, 12). This implies that a wide range of gains is necessary.
3.1.3. DIFFERENTIAL INPUT.
It was pointed out earlier that monopolar recordings of gastric electrical activity are rarely used, because the monopolar GEA signal is superimposed upon a higher amplitude EKG, respiration and motion artifacts, etc., all of which would be amplified. Therefore, differential input is preferred when amplifying GEA. Fortunately, high-quality instrumentation amplifiers are now available to meet these requirements. These instrumentation amplifiers are usually DC coupled, which means that the signal is not high-pass filtered before the next amplification stage. This contradicts the need to get as much gain as possible as soon as possible and thus eliminate the negative impact of the noises in the next amplifier stages (46, 47). A major problem for a high gain of the differential input stage is the polarization voltages of the electrodes (48, 49, 50). Although Ag-AgCl electrodes are deemed to be non-polarizable (48), it is worthwhile to reexamine their behavior in the the infralow frequency range. Huhta and Webster (50) as well as Metting van Rijn et al. (49) studied the interference currents through the patient's body in bioelectric measurements. They pointed out that there are three types of interference currents to be considered:
1. Interference currents through the body;
2. Interference currents into the amplifier;
3. Interference currents into the measurement cables.
The interference currents through the body are caused by the capacitances between the patient, the power lines and earth, and have a typical peak-to-peak value of 0.5 mA (49). A portion of these currents can flow through the neutral electrode thus causing a potential difference between the average potential of the body and the amplifier common: the common-mode voltage.
When doing an isolated bioelectric measurement, the capacitances between the isolated ground and mains and the isolated and non-isolated grounds are the major sources of the interference currents into the amplifier. These capacitances should be made as small as possible, to reduce the negative impact of these currents, portion of which flows through the neutral electrode contributing to the common-mode voltage. The interference currents into the cables have a typical value of 10 nA peak-to-peak and flow through the recording electrodes and the electrode-skin impedance associated with them (49). This can potentially produce a relatively large differential voltage and saturate the input stage. The problem can be addressed by connecting a simple passive RC network at each input with the input impedance of each of the operational amplifiers serving as R. This would provide the low cut-off frequency of the whole system converting it to an AC amplifier. However, in case of accidental saturation of the input stage (for example due to motion artifacts) there is not a low-resistance path available for discharge of these capacitors except through the patient. The discharge can take minutes after removing the overload conditions (53). Van Heuningen et al. (53) suggested a frequency dependent negative feedback as a possible solution for this problem. They pointed out that the suggested circuit can handle time constants up to 1000 s. In this design, the differential amplifier of the standard instrumentation circuit used at the input is fed back by an integrator.
3.1.4. HIGH INPUT IMPEDANCE.
The electrode-skin interface has a complex impedance between 1K and 1M at frequencies up to 50-60 Hz (48, 49, 54). It increases in the infralow (0.01 - 1 Hz) frequency range. This impedance depends on the skin condition and its preparation, the equivalent impedance of the electrodes, the fat volume underneath the skin layer and the distance between the source of the electrical field and the electrodes. An increase of the electrode-skin impedance is to be expected when recording in the infralow frequency range, because the capacitive component would be great. A high input impedance would prevent the formation of voltage dividers between the electrode-skin impedance and the input of the amplifier. That is why as a general rule the input impedance of a biomedical amplifier should be as high as possible (46, 47).
3.1.5. GALVANIC ISOLATION.
Galvanic isolation between input and output is a must for any electrophysiological measurements in vivo. The primary reason for this is patient safety (53, 49). Leakage currents from the mains which flow into the body through the amplification system can cause fibrillation of the heart (50, 53, 54). Currents even as low as 10 mA could be dangerous. For all equipment in contact with the patient, not only must the case be electrically grounded, but connections to the patient must have very low current flowing through them (54). In galvanically isolated systems, there is a danger of undesired connections beween the isolated and non-isolated grounds. Therefore, the general requirement is to separate physically the two grounds as much as possible (50, 50). The absence of leakage currents also diminishes the unwanted line voltage input to the amplifier through the electrodes (53).
3.1.6. HIGH COMMON-MODE REJECTION RATIO (CMRR).
The signals s1(t) and s2(t) in each of the active inputs of an instrumentation amplifier are entirely differential at the moment T if s1(T) = s2(T) [3.2] providing the supply voltages of the amplifier are symmetrical. Modern differential amplifiers can easily provide common-mode rejection ratios (CMRR) higher than 100 dB. However, this feature alone does not ensure high CMRR for the whole amplification system. There are two possibilities for common-mode signals to go through despite the high CMRR of the input stage: because of the limited common-mode rejection ratio (CMRR) of the stage itself, and due to a partial transformation of the common-mode signal into a differential at the input. While the first factor is rather technological, the conversion of the input common-mode signal into differential can be controlled. This conversion is due to the fact that ideal symmetry at the two active inputs of the differential amplifier cannot be provided.
The more inputs the amplification system has the worse the symmetry problem gets, because it becomes more difficult to match them all. Since the differential input impedances of the instrumentation amplifiers can be matched quite precisely, the greatest unbalance in the equivalent input impedance comes from the electrode-skin contact and the cables. High input impedance, skin conditioning, galvanic isolation, appropriate cable shielding and driving the shields to the isolated ground or to the common-mode signal can increase CMRR by reducing the differences in the equivalent input impedances thus providing better input symmetry (50, 51, 53). In some cases, driving the body to the common-mode voltage gives very good results (50, 51, 55).
3.1.7. LOW NOISE.
There are three major sources of noise in amplification systems for electrophysiological measurements: (a) noise of the operational amplifiers Eoa; (b) thermal noise from the electrodes, cables and passive components Et; (c) noise introduced by the power supply Ep. These noises are nearly random in nature and determine the ultimate lower limit of signal-handling capability of the amplifier (46).
Whatever power supply is used, it introduces additional noise in the amplification system (46, 49, 50). Modern isolation amplifiers usually have built-in isolated power supply, i.e. they are internally powered. Isolation amplifiers with a modest cost would have relatively big ripple current and voltage associated with the isolated power supply. A typical example is Burr-Brown ISO 107 which is recommended for biomedical applications, but has voltage ripples from the power supply in the range of 60 mV (peak-to-peak). This discussion clearly indicates that the power supply is the major source of noise in amplification systems with galvanic isolation. There are two ways of addressing this problem:
(a) filtering the high-frequency ripple according to the recommendations of the manufacturer;
(b) synchronizing the ripples so that they become common-mode signal before the next stage of the amplifier.
While the first option is obvious, the second option implies that two synchronized isolation amplifiers (one for each input) would be required for each channel and the instrumentation amplifier would be in the non-isolated part of the circuit. This would almost double the cost of the device.
3.1.8. GOOD OFFSET COMPENSATION.
Although most of the high-quality instrumentation and operational amplifiers have laser-trimmed offset, in an amplification system with multiple stages and built-in active filters some offset voltage is to be expected. In the modern systems the offset is automatically compensated to provide better output range for the amplified signals.
3.2. INPUT STAGE OF THE ISOLATED GEA AMPLIFIER.
Many of the above stated requirements for the GEA amplifier are to be met at the input of the amplification system. In fact, the input stage of a differential amplifier determines the CMRR and the input impedance of the whole system (46, 47, 53). In this study the solution for the input stage proposed by Van Heuningen et al. (53) was utilized for the gastric electrical signals. The frequency-dependent negative feedback was implemented on the differential operational amplifier of the monolithic instrumentation amplifier after the galvanic isolation (Fig. 3.1). Two synchronized isolation amplifiers were connected as buffers in the isolated part. Their common power supply was used to supply the two input FET operational amplifiers, which had a gain of 10. A quad (four-in-package) operational amplifier OPA404 was used, so one of the two amplifiers left was used to drive the shielding of the cables to the isolated ground. Appropriate shielding of the active inputs was also provided to ensure better symmetry.
Figure 3.1. Simplified diagram of the input stage of the amplifier.
A possibility for easy on-line change of the cut-off frequencies as well as computer control via software accessible registers should be provided if the requirements for a flexible frequency range are to be met. It is important to use active filters with monotonic frequency and phase characteristics (for example Butterworth filters) in order to avoid the negative impact of non-linear phase responses. Ripples in the phase characteristic of the amplifier can lead to misinterpretation of the recorded time shifts between different GEA channels. An example of a digitally controlled 4-pole low-pass filter which follows the input stage is shown on Fig.3.2.
Figure 3.2. Digitally controlled four-pole low-pass filter.
3.3. PROGRAMMABLE GAIN AND OFFSET COMPENSATION.
Computer controlled gain is relatively easy to realize with the new integrated circuit technology. The problem is how to cover wider range of gains. This is usually achieved by cascading several programmable operational amplifiers (56). Automatic offset compensation is still a problem in any computerized amplification system (46). The problem can be solved using an analog or digital compensation circuit. When doing an analog compensation, the input is short-circuited and the measured offset is stored in a holding capacitor. The stored value of the offset is later applied during the recording to the positive input of the programmable amplifier instead of grounding it. When using digital compensation, the offset is assessed similarly, but it is then converted to its digital value and returned back to analog by a digital-to-analog converter (DAC). The DAC output is applied to the positive input of the programmable amplifier. Both approaches have advantages and disadvantages. The major draw back of the analog approach is the leakage current of the holding capacitor, which prevents the offset voltage to be held unchanged indefinitely. However, this can be overcome with an appropriate compensation circuit.
The holding problem is avoided in the digital alternative, but the finite resolution of the digital-to-analog converter often makes the compensation voltage different from the real offset. To achieve good offset compensation in the microvolt range using the digital approach at least a 16-bit DAC would be required. In this study the analog approach is preferred because of its simplicity and modest cost. Good shielding around the active inputs of the programmable amplifier used for offset compensation is provided.
3.4. BLOCK AND CIRCUIT DIAGRAMS OF THE GEA AMPLIFIER.
The block diagram of the isolated GEA amplifier (Fig 3.3) contains several major blocks:
(a) isolated input stage with a circuit for driving the shielding;
(b) instrumentation amplifier with computer controlled frequency dependent negative feedback (high-pass filter);
(c) computer controlled low-pass filter;
(d) cascaded programmable amplifiers;
(e) automatic offset compensation.
The isolated EGG amplifier utilizes two non-inverting FET amplifiers with a gain of 10 (U1a, U1b) on each of the two inputs. The amplifier outputs are then applied to a differential amplifier made up of two unity gain ISO 107s (U2a & b) and a monolithic instrumentation amplifier (INA 101) with a gain of 10 (U3), thus providing an overall gain of 100. U1c drives all shielding to isolated common. A single-pole high pass filter (U4) implements the idea of frequency dependent negative feedback and provides filtering of the input signal with a cutoff frequency of 0.015, 0.029, 0.5, or 1.6 Hz selected by the appropriate setting of the analog switches (U20). The output of the INA 101 is applied to a variable four-pole low-pass filter (U5) which filters the signal with a cutoff frequency of either 0.1, 0.5, 2, or 3.4 Hz selected by the appropriate setting of the analog switches (U6-9). The filtered signal is then applied to a programmable gain block (U11 & 12) with a gain varying from +1 to +8000 by powers of 2 up to 8 (PGA203) and powers of 10 up to 1000 (PGA202). The value of the low and high-pass cutoff frequencies, as well as the gain, are determined by bits loaded into U18 and U19 by a software accessible register. The auto-offset group is made up of an integrator (U13b) and a leakage current compensator (U13a). This block is toggled active and inactive by the switch S1.
Figure 3.3. Block diagram of an isolated GEA amplifier. The shadow area is the isolated section.