You are on page 1of 8

Home Search Collections Journals About Contact us My IOPscience

An ultra-low-noise electrical-potential probe for human-body scanning

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2000 Meas. Sci. Technol. 11 291

(http://iopscience.iop.org/0957-0233/11/3/318)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 153.19.58.31
This content was downloaded on 12/11/2015 at 20:14

Please note that terms and conditions apply.


Meas. Sci. Technol. 11 (2000) 291–297. Printed in the UK PII: S0957-0233(00)09206-7

An ultra-low-noise electrical-potential
probe for human-body scanning
R J Prance, A Debray, T D Clark, H Prance, M Nock, C J Harland
and A J Clippingdale
Physical Electronics Group, School of Engineering, University of Sussex, Brighton,
Sussex BN1 9QT, UK
E-mail: t.d.clark@sussex.ac.uk and r.j.prance@sussex.ac.uk

Received 2 November 1999, in final form and accepted for publication 13 January 2000

Abstract. In this paper we describe a new very-low-noise, high-input-impedance probe


developed to make non-contact measurements of electrical potentials generated by currents
flowing in the human body. With a noise level of 2 µV Hz−1/2 at 1 Hz, down to
0.1 µV Hz−1/2 at 1 kHz, and an operational bandwidth from 0.01 Hz to 100 KHz, this probe
would seem well suited to the detection of a wide range of electrical activity in the body.

Keywords: electrical field strength, electrical potential, charge (Coulomb) meter,


human-body imager, monitoring human-body electrical activity

1. Introduction the sensor and the skin. Clearly, the preparation of an array of
in-contact BPSM sensors is time consuming. Furthermore,
In previous work [1–3] we described the use of a high-input- since it is difficult to keep the conducting-paste interfaces
impedance electrical potential probe to make non-contact entirely separate in an array of sensors, cross coupling
electrical measurements of the human body. We introduced between neighbouring sensors (through leakage currents)
this probe precisely to overcome limitations associated with is common. These problems are eliminated completely if
conventional electrocardiographic (ECG) sensors. There displacement-current electrical-potential probes are used. By
were two key features in its design. First, it operated on way of illustration, we have already published an example
displacement, not real charge, current. This eliminated of non-invasive electrical-potential-probe heart imaging,
the need to make direct electrical contact with the body. utilizing an array of 25 elements (plus a separate, independent
For comparison, conventional ECG sensors always need to array of eight elements when required) [1].
maintain this good electrical contact with the body. In its
In this earlier work [1–4] we were able to engineer
practical embodiment [1, 2] the probe was positioned off the
very substantial improvements in the input impedance and
body at a distance of 3 mm, with a body-to-probe capacitance
noise performance of an electrometer-grade operational
of 7 pF. Second, the high input impedance of the probe (a few
amplifier (an OPA 111) by using this in a stable buffer-
times 1012  at 1 Hz) ensured that the surface of the body, as
amplifier configuration with precisely unity gain. The
the source, was not loaded down.
Surface potentials due to heart currents are spatio- basis of these improvements was a combination of well-
temporal in nature. In conventional ECG work a single sensor known guarding techniques and novel feedback circuits.
(plus one or more reference electrodes) is used to record In particular, we were able to make the effective (i.e.
temporal information which changes with its location on the electronically manipulated) input impedance as high as
surface of the body. Clearly, in this electrical technique the 1016  [3], corresponding to an input capacitance close to
functioning of the heart can best be analysed by recording 10−17 F. In this [4], and other, previous designs it was always
all of the relevant potential information on this surface, i.e. necessary to stabilize the operating point of the operational
by constructing a body-surface-potential map (BSPM). In amplifier at zero frequency by providing an internal source of
principle, this entails making a time record of the potential dc bias current within the probe. We were able to achieve this
at each point on the surface. In practice, of course, we are without making any compromises in the input impedance of
constrained to accept finite spatial resolution. Even so, to the probe within its operating bandwidth (typically 0.01 Hz
obtain a satisfactory resolution (say 1 cm to a few centimetres) to 50 kHz). A block diagram of the probe, including the dc
means that we must make use of a large number of separate biasing circuit, is shown in figure 1. With this configuration
sensor probes. If these are conventional in nature (as is it is apparent that the probe is acting as a voltmeter. A typical
the case for BSPM), each sensor must make low-resistance measured noise-voltage spectral plot for this type of probe is
contact with the skin. This low resistance is achieved by using shown in figure 2 for two values [2] (approximately 0.2 and
a high-conductivity electrolytic paste as the interface between 1 pF) of source–probe capacitance.

0957-0233/00/030291+07$30.00 © 2000 IOP Publishing Ltd 291


R J Prance et al

the√stabilization network. Hence, it should be proportional to


1/ R, which it is over most of the resistance range displayed
in figure 3. For example, at R = 1011  the noise has reduced
to 5.5 µV Hz−1/2 . Unfortunately, resistors with resistances >
1011 , with both small tolerances (i.e. a few per cent) and
stability, do not exist. Thus, these extremely large resistances
often vary by up to 50% from sample to sample; furthermore,
these resistors are not temperature stable. However, we have
made a spot noise measurement on the OPA 111-based system
using a (quoted) 1012  resistor of the glass-encapsulated,
carbon-film type, this being the highest-resistance resistor
we have been able to obtain commercially. The data √ point,
shown in figure 3, lies close to, but not on, the 1/ R line,
Figure 1. A block diagram of the earlier version of the
electrical-potential probe [3] based on the OPA 111, including the which we assume to be due to the uncertainty in the value of
input-bias current circuit. the resistance. Clearly, from the trend shown by the first four
data points in figure√ 3 the noise level at 1 Hz appears to be
proportional to 1/ R, as one would expect. On this basis
Working with probes designed especially for heart
the minimum noise level would be achieved if we could set
imaging [1, 2] (i.e. with input electrode areas for the probe
R → ∞.
electronics of about 3 cm2 and a probe-input capacitance
By definition, an R = ∞ input resistance would make
close to 1 pF) we were able to establish a noise performance
the input impedance of the amplifier purely capacitive. Now,
in the very low audio range (0.1–100 Hz) which was adequate
in principle the FET at the front end of an ideal operational
for detecting heart signals. This noise (referred to the
amplifier has a purely capacitive input impedance. This
input) was 30 µV Hz−1/2 at 1 Hz. For most routine heart
implies that, provided that we can dispense with the input
diagnostics this is sufficient, providing, as it does, a 30:1 bias network, we should be able to construct a charge
signal-to-noise ratio for peak heart signals [5] (around 1 mV). amplifier (coulombmeter) which responds to arbitrarily low
However, the electrical potentials generated by the body can frequencies (i.e. down to dc), rather than a voltmeter with
vary widely [5, 6], with the generally accepted minimum a built in low-frequency roll off. In practice, however,
(≈10 µV) arising from foetal heart action [5]. Hence, there will always be a finite on-chip leakage current which
there is a clear need for a non-contact sensor with a noise will act to prevent this situation from being realized. In
level close to 1 µV Hz−1/2 at 1 Hz. By taking advantage general, in the absence of an input bias network, the gate
of our previous experience, we have been able to create of the FET will charge via the leakage-current path and
the basic circuit building block for a new series of non- eventually cause the amplifier to saturate (usually after only
invasive electrical sensors. Using currently available, state- a few seconds). Previous attempts at producing high-input-
of-the-art, electronic components we have developed a new impedance amplifiers [1–4] have, to a large extent, relied
sensor amplifier with extremely accurate guarding. This on the use of precise guarding techniques to minimize the
has both a low input capacitance (≈1 pF) and a low noise effects of these leakage currents and so reduce the restrictions
level (2 µV Hz−1/2 at 1 Hz). Compared with previous imposed by the presence of an input bias network. Recently,
configurations, it has the added advantages of a simpler however, a new generation of operational amplifiers which
design, with fewer components, and with all problems of extends the capabilities of guarding techniques with the
stability and drift eliminated. We should also add that, just provision of on-chip guarding facilities has become available.
like in the previous work reported in the literature [1], we To the best of our knowledge, the best example is the Burr-
have not come across any disadvantages in using electrical- Brown INA 116 dual-input, instrumentation amplifier. We
potential probes for human-body scanning. These probes are have incorporated this amplifier into a planar configured
truly non-invasive, drawing no real current, their coupling probe circuit which has been designed by us to extend the on-
efficiency can be essentially perfect and they have a very chip guarding to the external input electrode structure. This
much lower noise level than do conventional ECG sensors. design has proved to be very successful and we are now able
to operate the INA 116-based probe as an unconditionally
2. Ultra-low-noise probes stable charge amplifier for long periods of time (≈24 h)
without any provision for input bias current. This is in
2.1. Design and construction stark contrast to the best of the previous generation of
electrometer amplifiers such as the OPA 111 which, in our
The electrical-potential probe, such as that depicted in figure experience, cannot be stabilized in this way. Indeed, for
1, is clearly a significant advance over conventional ECG the latter amplifier, stable operation in the absence of a
sensors. Even so, there is a quite obvious need for better bias network is conditional upon maintaining very precise
noise performance. As we have discovered, the noise level supply voltages, even for relatively short time periods ('1 h).
in the probe can be reduced by increasing the resistance R of Furthermore, the INA 116 amplifier has a quoted open-circuit
the input-bias-stabilization network, as shown in figure 1. In input impedance of 1015  in parallel with about 1 pF and a
figure 3 we show our measured output noise voltage at 1 Hz quoted spot-noise level at this frequency of 2 µV Hz−1/2 .
for the OPA 111-based probe as a function of R. We have Our choice of the INA 116 for all of the coulombmeter-
identified this noise as being derived from the current noise in configuration work described in this paper was based on its

292
An ultra-low-noise probe

Figure 2. The noise-voltage spectral density of the OPA 111-based probe of figure 1 with input-electrode arrangements giving
source–probe coupling capacitances of approximately 0.2 and 1 pF [4].

Figure 3. A plot of the probe-output noise voltage at 1 Hz versus the bias-stabilization-network resistance R for the OPA 111-based system.
The 1012  data point was taken using a glass-encapsulated, carbon-film, resistor.

noise performance, high input impedance, on-chip guarding, The quality of the fabrication of the chip is such that the
superior (unconditional) stability and extended frequency effects of low-frequency fluctuations and drift (thermally or
response. A block diagram of the INA 116, configured as a otherwise induced) are almost exactly balanced out between
charge (coulombmeter) amplifier, is shown in figure 4, with the inputs. This makes the INA 116 a very suitable amplifier
the signal applied to the non-inverting input and the inverting for our purposes.
input grounded. It can be seen here that, although guarding The coulombmeter (INA 116-based) probe was
is applied to both inputs, the inverting input is treated as a constructed using modern surface-mounting techniques.
dummy (i.e. grounded). However, this inverting amplifier, This allowed us to extend and maintain the level of on-
on-chip with its non-inverting dual, plays an essential role. chip guarding through to the input-electrode structure. In

293
R J Prance et al

Figure 4. The block diagram of the INA 116, configured as a charge amplifier, with the signal applied to the non-inverting input and the
inverting input grounded.

practice, this hybrid guarding could be made so accurate probe (figure 4) over the range 1 mHz to 100 kHz. Here,
that, to an extremely high degree, leakage-current paths could it is clear that the response is flat down to 30 mHz, with a
be eliminated. In the actual embodiment a double-sided slight and irregular rise (<2 dB) from 30 to 1 mHz. That
printed-circuit-board layout was selected with the bottom the output remains stable over 15 h (>50 000 s) in figure 6
(reverse) face used as a guard plane while the top face was indicates that this response continues down to at least 20 µHz.
configured as a guarded input-electrode structure. As shown If required, the low-frequency rise below 30 mHz can be
in the layout of figure 5, with the INA 116 amplifier surface truncated by imposing a high-pass filter (3 dB point at 0.1 Hz)
mounted, guard tracks were provided either side of the input on the output of the probe (figure 7(b)). For comparison, we
electrode and maintained over the entire circuit board. This show in figure 8 the frequency response of the OPA 111-
planar design (top-surface tracks and reverse plane) provided based probe (figure 1) acting as a voltmeter [1–4] (with a
a remarkably high degree of guarding and shielding. The few times 1012  input impedance at 1 Hz). From the first-
design was implemented using a software package (IsoPro) order roll off in this response around 100 mHz, it is obvious
and cut using a Quick Circuit 5000 PCB milling machine that the input impedance of this probe contains a resistive
(T-Tech, Inc). The use of a milling machine ensured that component. This is in contrast to the purely capacitive input
the electrical-insulation properties of the circuit board were to the INA 116-based probe configuration of figure 4. Using
not compromised by wet chemical etching. Naively, without a potential-divider-based technique, with a 10 pF series input
this extremely accurate guarding, we would have expected to the probe, we found that its input capacitance was close to
the leakage to be that quoted by the manufacturer ('3 fA). 1 pF (corresponding to an input impedance >1012  at 1 Hz).
However, as we shall see, the actual leakage current is The improvement in noise performance of the INA 116-
negligible on the scale of femto-amperes. based system compared with previous probes is striking. All
noise measurements were carried out using a HP 3562A
3. The probe performance signal analyser in logarithmic resolution mode. A ×10 low-
voltage noise preamplifier was inserted between the probe
In order to demonstrate unambiguously that the INA and the signal analyser. This ensured that the probe noise
116-based probe was functioning as a charge amplifier was well above the noise floor of the HP 3562A. In figure 9
(coulombmeter) a dc voltage was applied to its (non- we show the noise response referred to the input (the noise
inverting) input through a 10 pF capacitor; the response with spectral density) between 1 Hz and 100 kHz. As can be seen,
time was then measured. In figure 6 we show a record the input noise voltage at 1 Hz is 2 µV Hz−1/2 , reducing
of this output taken over 15 h with a stabilized dc voltage to 0.2 µV Hz−1/2 at 100 Hz. At 1 Hz this constitutes a
of 10 V applied to the input. During this time the output factor of 15 improvement over the OPA 111-based voltmeter
from the source was monitored and remained within ±5 mV design, quite sufficient to make almost all of the electrical
at all times. As can be seen, with some fluctuations the activity known to occur in the human body accessible in
output remains that of the input over very long time scales. real time. We note that, although 1 Hz is clearly within
This is the result to be expected if the probe is acting as a the 1/f -noise regime in figure 9, the noise level is still only
coulombmeter, measuring charge rather than voltage, with 10 µV Hz−1/2 at 0.1 Hz (not shown). Quite obviously,
a concomitant frequency response flat down to dc. It is to make use of this noise performance the quasi-dc drift
apparent from figure 6 that this is the case. In figure 7(a) in the probe has to be very small. As we have already
we show the measured frequency response for the INA 116 pointed out, this certainly requires that the guarding on the

294
An ultra-low-noise probe

Figure 5. The printed-circuit-board layout of the INA 116-based probe, designed with IsoPro software and machined on a Quick Circuit
5000 PCB milling machine (copper tracks are shown shaded).

(a)
Figure 6. A plot of the long-time-scale (≈15 h) output-voltage
response of a coulombmeter configuration probe with 10 V of dc
offset applied to its (non-inverting) input.

amplifier be efficient so that, to an extremely high degree,


leakage-current paths (and accompanying fluctuations) can
be eliminated. It is also necessary that drift and low-
frequency noise due to temperature fluctuations in the
environment be minimized. Again, as we have discussed
above, the quality and uniformity of the on-chip processing
in the fabrication of the INA 116 instrumentation amplifier
allow us to null out (differentially compensate) almost all
of these fluctuation problems. With a combination of both
measures, fluctuations and drift can be kept to a minimum.
In figure 10 we show the dc drift, referred to the input,
over the settling period of the probe immediately after it (b)
has been turned on. After 40 min this is down to less than Figure 7. (a) The frequency response from 1 mHz to 100 kHz
2.5 mV min−1 . On much longer time scales, i.e. after settling, (full line) using the INA 116-based probe configuration of figure 4.
(b) The truncated response of the probe with a high-pass (0.1 Hz
this drift is very much lower, i.e. as shown in figure 11. It
roll off) filter on its output.
is apparent that this drift is typically less than 2.5 mV h−1 .
We note that the data of figures 11 and 7(a) imply that the
4. Conclusions
effective leakage current for the INA 116-based probe is less
than 10−18 A.
In this paper we have shown that it is possible to design and
As an illustration of the use of the INA 116-based system construct a high-impedance electric-potential probe, based
in non-invasive detection of human-body electrical activity, on the INA 116 instrumentation amplifier, with a noise
we show in figure 12 an ECG taken above the surface of the level of 2 µV Hz−1/2 at 1 Hz. This represents a very
chest of a student subject with, as previously [1, 2], the chest considerable improvement on that achieved (30 µV Hz−1/2
surface and the probe input electrode separated by 3 mm of at 1 Hz) using previous designs. At this noise level it should
rubber. prove possible to resolve even the smallest electrical signals

295
R J Prance et al

Figure 8. The frequency response of the OPA 111-based Figure 11. The quasi-dc drift curve for the INA 116-based probe
electrical-potential probe of figure 1 over the range 0.01 Hz to taken over 26 h; typically the drift is <2.5 mV h−1 .
1 MHz.

Figure 12. A time-domain plot (ECG) of the electrical potential


above the surface of the chest of a student subject using the INA
116-based probe with 3 mm of rubber insulation between this
surface and the input electrode of the probe.
Figure 9. The noise voltage spectral density (1 Hz to 100 kHz) of
the INA 116-based probe.
shown that, using the INA 116, the probe can be arranged
to have a purely capacitive input. This allows it to function
as a coulombmeter with a frequency response down to dc.
As an additional advantage, it is easy to provide an INA
116-based probe system with extremely efficient guarding
and shielding. This has allowed us to operate the INA 116
probe in a coulombmeter configuration with negligible input
leakage current. This, combined with the inherent on-chip
balancing of the input amplifier blocks of the INA 116,
means that drift and fluctuation effects can be kept very
small. The currently attained noise level can certainly be
reduced to 1 µV Hz−1/2 , without seriously compromising
performance, by implementing modest increases in coupling
(i.e. the input capacitance) between the body and the probe.
However, much more substantial improvements in noise
performance can be achieved by introducing gain into the
Figure 10. The quasi-dc settling curve for the INA 116-based probe. Preliminary measurements indicate that with a voltage
probe; the drift after 40 min is <2.5 mV min−1 . gain factor of ten, we can reach a noise level close to
200 nV Hz−1/2 at 1 Hz. We are now working on the feedback
generated by the body (for example, the foetal heartbeat networks [1–3] for this probe, which will allow us to produce
at around 10 µV peak signal in, say, an integrated RMS an electrometer with a stable output for use in future non-
noise of 6 µV from 2 to 100 Hz (figure 9)). We have also invasive cardiac-data-acquisition studies.

296
An ultra-low-noise probe

Acknowledgments [2] Clippindale A J 1993 The sensing of spatial electrical


potential PhD Thesis University of Sussex, p 27
We would like to thank the UK Engineering and Physical [3] Prance R J, Clark T D, Prance H and Nock M 1997
Electrometer arrays: sensing of spatio-temporal ELF fields
Sciences Research Council for its generous funding of this Proc. Marelec (London, 1997) p 3.4
work. [4] Prance R J, Clark T D, Prance H and Clippingdale A J 1998
Non-contact VLSI imaging using a scanning electric
potential microscope Meas. Sci. Technol. 9 1229–35
References [5] Katila T 1980 Instrumentation for biomedical applications
Proc. 3rd Int. Workshop on Biomagnetism (Berlin, 1980)
[1] Clippingdale A J, Prance R J, Clark T D and Watkins C 1994 pp 3–31
Ultrahigh impedance capacitively coupled heart imaging [6] Fisch B J 1991 Spehlmann’s EEG Prime (Amsterdam:
array Rev. Sci. Instrum. 65 269–70 Elsevier)

297

You might also like