Physiological Measurement

Related content
- Current approaches to analogue
A wide-band AC-coupled current source for instrumentation design in electrical
impedance tomography
electrical impedance tomography K G Boone and D S Holder

- A comparison of modified Howland circuits

as current generators with current mirror
To cite this article: R Bragos et al 1994 Physiol. Meas. 15 A91 type circuits
P Bertemes-Filho, B H Brown and A J
Wilson

- Current injection electrodes for electrical

impedance tomography
View the article online for updates and enhancements. S Armstrong and D Jennings

Recent citations
- Hanen Nouri et al

- A Parametric EIT System Spice Simulation

with Phantom Equivalent Circuits
Christos Dimas et al

- DSP-based current source for electrical

impedance tomography
Gary J Saulnier et al

This content was downloaded from IP address 37.14.32.212 on 17/03/2021 at 20:38

Physiol. Meas. 15 (1994) A91-A99. Printed in the UK

A wide-band Ac-coupled current source for electrical

impedance tomography

Ramon Brag&, Javier Rose11 and Pere Riu

Divisi6 d'lnsmmentaci6 i Bioenginyeria, Deparrament d'Enginyeria Elecubnica. Universitat
Polithcnica de Catalunya, Barcelona, Spain

Abstract. A current source suitable for application in electrical impedance tomography (UT)
is described. The fim stage of the commercially available current-feedback amplifier AD844
constitutes a current-conveyor implementation and allows the consmction of wide-bandwidth
current sources, thus avoiding the mismatching and temperahire-induced problem that arise in
discrete realizations. The lack in gain accuracy of this circuit is avercome by the inclusion of
ill input buffer in an operational amplifier (op amp) feedback loop. Saturation problems that
appear when placing a oc-blocking capacitor between the source and the electrode are solved by
a DC feedback that maintains oc voltage at the output near to 0 V without reducing the output
impedance of the source. Two Ac-coupled current sources, in both inverting and non-inverting
configurations, are described and their possible applications to EX are listed.

1. Introduction

Multifrequency BIT needs high-quality controlled current sources, able to maintain 0.1%
accuracy in the output current over a wide bandwidth (10 kHZ-1 MHz), even with large
variations of the electrode contact impedances.
In the last four years, some implementations based on current sensing at the supply
terminals of an amplifier have been published (Lidgey et nl 1990, h u n g et nl 1990). The
amplifier is usually an open-loop AB-class buffer or an op amp-based voltage follower.
In both cases, current mirrors are used to copy the supply currents from the amplifier and
drive them to the load, as shown in figure l(u). That structure constitutes an implementation
of a second-generation current conveyor (CCII), a powerful building block for current-mode
design whose symbol and hybrid parameters are shown in figure l(b) and whose principles
and applications have been widely described (Toumazou et nl 1990%b).
The advantages of these circuits are inherent stability, high output impedance and
wide-band operation. The drawbacks arise from the non-idealities of current mirrors and
amplifiers.
The main problem of the mirror implementation is that arrays of complementary
transistors are not easily available. Non-complementary transistors produce harmonic
distortion and matched discrete transistors result in a high temperature coefficient, a long
warm-up time and a laborious device substitution during repair. On the other hand, AB-class
open-loop amplifiers do not allow accurate setting of output current, while op amp-based
amplifiers provide both output and quiescent currents to the mirrors, therefore reducing the
output impedance of the current source and increasing its current noise.

2. The AD844 cnrrent-feedback amplifier

The current-feedback amplifier (CFA) AD844 (Analog Devices), whose internal structure is
shown in figure 2 (dotted zone), has accessible an internal high-impedance node (2)for

0967-3334/94/SAO091+09$19.50@ 1994 IOP Publishing Ltd A9 1

A92 R Bra& ef a1

Figure 1. (a) A mnt-sensingbased current souce. (b) A current-conveyor symbol and

hybrid parameten.

external frequency compensation. This feature permits its use as a current conveyor. In
fact, some papers have referred to it as a ‘commercially available current-conveyor’ (Bruun
and Haxtausen 1991, Lee and Tsao 1992).

.........................................................

AD844

Rt

...........................

Fiyre 2. An equivalent schematic diagram of the AD844 CFA and ifs connection as a vccs,

Acting as a CFA, the voltage applied to the high-impedance non-inverting input is

copied to the low-impedance inverting input while the output current of this node is
 Wide-bandAc-coupled current souxe for EIT A93

replicated by the mirrors so as to flow in the transimpedance ( Z , = RI I[ C,). The voltage

developed across Zt is buffered by the unity-gain voltage follower. The high value of Z,
(RI = 3 MQ, Ct = 4.5 pF), provides a high open-loop voltage gain to the amplifier. Closing
the loop through a resistor network results in a very low current at the inverting input, thus
generating a virtual high-impedance node. Therefore, the CFA has an external behaviour
similar to the voltageoriented op amp.
The amplifier can also be considered as a current conveyor followed by an output buffer.
Then, the transimpedance of the CFA (Z,) becomes the output impedance of the conveyor,
and the output impedance of the voltage-controlled current source (vccs) when connected
as in figure 2.
The use of this integrated circuit overcomes the problems due to asymmetry of mirrors
and thermal stability, but errors due to gain accuracy of the first stage and its non-zero output
resistance Rob are not overcome. The resistor R,which determines the transconductance of
the current source is connected to the output of the buffer, thus being placed in series with
Rob and modifying the nominal transconductance (specified Rob is typically 50 and 65 a
maximum). We could minimize the influence of Rob variations by increasing the value for
q and R, but according to the manufacturer the voltage at the non-inverting input must be
kept below +I V for best results. Therefore. R cannot be arbitrarily increased.
These gain errors can be calibrated in single sources, but they would induce errors in
multiple-source systems and common-mode problems in floating-load sources.
Another problem that arises when applying any current source to EIT is the charging
of the Dc-blocking capacitor because of residual DC currents. In the circuit of figure 2,
current mirrors are driven to saturation if this capacitor is placed between the Z output and
the load. Traditional solutions based on placing a resistor at the output node of the current
source reduce its high output impedance. In addition, the resulting DC voltage presents slow
variations that affect the measurements’ repeatability.

3. Circuit description

The circuit shown in figure 3 overcomes both problems as follows.

(i) The gain accuracy of the amplifier is guaranteed by the inclusion of the input buffer
in the feedback loop of a wideband op amp (LT1220, AD829, EL2044,. . .). In addition,
the output resistance of the resulting amplifier is virtually zero and adds no error to the
intended transconductance. A 30% variation in Rob results in a 5% variation in the output
current when there is no feedback loop and R = 200 a,and in a 4 x variation when
the loop is closed.
(ii) The DC output voltage before the capacitor is kept near to 0 V by using a DC feedback
loop, which measures the actual DC voltage at the Z node (by using the output buffer of the
CFA) and applies a correction to the input. The use of the output buffer makes unnecessary
additional circuitry at the Z node, which would reduce its impedance. The output buffer
can also be used to drive a guard in order to reduce the effect of stray capacitances in the
WE and wiring between the current source and the current injection multiplexer.
Figure 3 shows two possible configurations, inverting and non-inverting, depending
on the phase shift between q and Io. Some differences can be expected between the
behaviour of the two configurations. The non-inverting version presents a high impedance
to the voltage source, while the inverting version presents an input impedance equal to R .
Therefore in the inverting case, the transconductance is affected by the output resistance of
A94 R Bra& et a1

Vcf

.............

:......................................................... :

VCC

AD844 i

.................................... :

Figure 3. An AC-Coupled current s o m (a) non-inverting;(b) inverting.

 Wide-bandAC-coupled current source for E m A95
the voltage source. On the other hand, the inverting configuration has its inputs and output
at a voltage near to 0 V and will therefore be more robust to the slew rate limitations of the
amplifiers and to the influence of input capacitances of the op amp. These characteristics
will cause some differences between the frequency response of the two configurations.
Depending on the op amp used in the AC feedback loop, a slightly peaking frequency
response could be expected in the non-inverting case (for low load values).
The DC feedback is applied to a different node depending on the configuration of the
current source,.and has consequently an inverting or non-inverting gain. In the non-inverting
mode, the feedback is applied to resistor R and, therefore, the output of the DC feedback
has to be a virtual ground for the A c signal. Then, the op amp should be chosen to have a
very low closed-loop output resistance in the whole bandwidth.
The DC-feedback circuit must have a very high DC gain and a negligible gain at the
frequencies of interest. Their passive components are chosen in order to solve the trade-off
between response time (high bandwidth) and low gain (< 60 dB) at the lower frequency to be
injected to the load (10 kHz) (low bandwidth). The resulting system presents unconditional
stability and a theoretical zero steady-state error in its step response.
The main purpose of the circuits in figure 3 is to act as a vccs in a multifrequency
EIT system. Therefore, if the two frequencies are separately generated, they must be added
before being applied to the VCCS. If a classical op amp-based adder is used, the output
resistance of the voltage source has no influence on the transconductance value for the
inverting configuration.
In addition, for the non-inverting case, the DC feedback can be applied as a third input
to the adder, thus allowing us to connect R to earth. This additional circuitry will affect
the frequency response with a supplementary phase shift but, given that this shift will not
depend on the value of the load, it can be calibrated out.

4. Circuit performance

The following results have been obtained using an LT1220 as a buffer and a E O 8 1 in the DC
feedback loop. An adder based on an EL2044 has also been included in both configurations
in order to incorporate its effects in the circuit performance. The following parameters have
been measured distortion, noise, frequency reponse, output impedance, voltage and current
operation margins and DC feedback performance.
In current-sensing-based vccs. the main cause of distortion is the mismatch between
source and sink current mirrors, because each one copies approximately a half period of
the reference current to the output. Therefore, if their gains are not equal, the reconstructed
output current will present harmonic distortion. The advantage of using the AD844 is
that npn and pnp transistors have similar cment gains (they are built using a bipolar
complementary (BC) process) and temperature matching is ensured by the integration. The
residual mismatch together with the non-linearities of buffers and amplifiers yield the
following total harmonic distortion (m)results (second and third harmonic), expressed
in decibels with respect to a 1 mA carrier: maximum, -47 dB over 5 kQ at 1 MHz;
minimum, -61 dE3 over 100 Q at 100 kHz. Similar results have been obtained for both
configurations.
When using coherent demodulation, the noise bandwidth is reduced to that of the output
filter. The system is thus relatively robust to the noise. Nevertheless, given the number of
parts in the measurement chain, the noise of each part should be minimized. The noise output
current has been measured over 50 ki2 and with grounded inputs. For the inverting source,
A96 R Bragds et a1

we have obtained 290 PA Hz-'/' at 10 kHz, 100 pA Hz-'/' at 100 kHz,50 PA Hz-'/'
at 500 kHz and 50 pA Hz-'12 at 1 MHz. For the non-inverting source, we have obtained
410 pA &-'Iat' 10 kHz, 140 pA Hz-'' at 100 kHz, 80 pA &-'/2 at 500 kHz and
80 PA HZ-'/'at 1 MHz.
As stated above, the magnitude and phase variations that are independent of the load
can be calibrated out. Those variations that are load dependent can be characterized by the
output impedance. However, the differences between inverting and non-inverting frequency
responses will induce common-mode problems in floating output sources built with two
single sources as we will discuss in section 5. The measured frequency responses are shown
in figure 4. The maximum values for the magnitude variation and the corresponding phase
shift from 10 kHz to 1 M H z are as follows: for the non-inverting circuit and RL = 50 Q;
1.1%14.4'; RL = 1 kc, 1.1%18.4'; RL = 5 kQ; 4.3%123". For the inverting circuit and
RL = 50 Q; 0.1%16.3"; RL = 1 kQ; 0.1%[9.6"; RL = 5 k@ 3.5%123".There is an initial
phase shift of 4" (6") at 1MHz for the non-inverting (inverting) source due to the amplifiers,
which can be calibrated and does not depend on the value for RL. The high magnitude
variation in the non-inverting source is due to its peaking response and will probably be
cancelled when adding more stray capacitances at the output i n the whole EIT system.
According to the equivalent schematic supplied by Analog Devices, the output
impedance of the AD844 should be the transimpedance of the CFA. Therefore, it can
theoretically be modelled by the parallel connection of a resistor and a capacitor. The
method used for the measurement of this impedance is explained by Webster (1990), and
is based on the load commutation between two values. The inverse of the real part of the
measured admittance is assigned to the resistor value and its imaginary part is assigned to
the capacitor susceptance. When IO = 150 $A and RL is switched between 50 Q and 50 kQ
we have measured for the non-inverting source; CO= 10.7 pF and RO = 7.6 MQ at 10 kHz,
6.9 MQ at 100 kHz,2.7 MS2 at 500 kHz and 0.7 MQ at 1 MHz.; for the inverting source:
CO= 11.5 pF and Ro = 5.5 MR at 10 kHz, 4.1 MR at 100 IcHz,0.8 M R at 500 kHz and
0.3 MQ at 1 MHz. The real part of the admittance does not remain constant with frequency
and therefore it can not be modelled with the usual resistor-capacitor circuit.
The extreme operation margins are determined by the saturation of either the first stage
or the current mirrors of the AD844, depending on the values for R and RL. The measured
values are: IO,, N 10 mA; V, N 8 V (V, = f 1 5 V)

measured an output DC current of -

Without any DC feedbatk and without adjusting the offset of the amplifiers, we have
5 pA for both circuits in figure 3. This current
drives the output to saturation if a DC-blocking capacitor is placed in series with the output.

-
With the DC feedback connected, the measured DC voltage before the capacitor reduces to
1 mV.

5. Applications

The circuit in figure 3 can be directly used as a single earth-referred current source in both
inverting and non-inverting configurations; also, afloating-load current source could be built
by placing the load between the outputs of two single sources, one of each configuration
(Haslett and Rao 1979), as shown in figure 5. Both sources have the same reference current
and no adjustments of the transconductance determining resistors are needed. If there were
any difference between the two output currents, it would flow through the output impedance
of the single sources, thus inducing a large common-mode voltage that would be transferred
to the load, and affecting the voltage detecting amplifier. Even if both currents are equal at
 Wide-bandAC-coupled current source for Err A97

NormalizedtransmnduaanCa magninxle (dB)

-2

-41
-6

-*U
-10
-12 3 4 5

Normalized out~utcurrent
6 loslfl 7 - 1 2 j , ,
3 4
, ,
5
,

Normalizedoutput current
,
6
~y
loglf1
7

1.01
I

1.01 {
1.00

0.99
0.98
3 4 5 6 7
loglfl
Transconductance phase (degrees) Transmnductance phase (degrees)
1 0

-10

-20

-30

- 40
-50
3 4 5
Ioglf)
1 kn 0 5kP
(*)
Figure 4. The frequency responses for RL = 50 S l , 1 kS2 and 5 kS2: ( a ) non-inverting Current
source: (b) inverting current source.

low frequencies, the mismatch between the frequency responses of the two single sources
will produce a common mode degeneration at high frequencies.
When connecting as in figure 5, the interaction between the two DC feedback circuits
drives the outputs to saturation. This is because they modify both reference and output
currents, thus allowing a positive feedback path in the whole circuit. One solution, tested in
a discrete component version of the circuit, is to apply the DC feedback outputs to the low-
A98 R Bragds et a1

Y
=L Y1
CCll z """"
A*
CCll
4
X 1.3 X

impedance side of current mirrors, thus modifying the output current but not the reference
current. These nodes are not accessible in the AD844, hence the solution is not useful in this
case. With BC analogue design tools as described by Gallant (1992), an integrated version
of the floating-load, Ac-coupled current source could be built. If two resistors are placed in
parallel with the DC-blocking capacitors, the system remains stable. We can determine the
mismatch between output currents by measuring the common-mode voltage over a known
resistor pair. We have obtained 0.1% at 10 H z , 0.2% at 100 Wz, 0.5% at 500 lcHz
and 0.6% at 1 MHz. Depending on the value for load and common-mode impedances,
the ratio between common-mode and differential voltage may be excessive, even for these
low mismatches between sources. The common mode voltage could be reduced by some
additional circuitry.
Other possible applications in EIT of the circuits in figure 3 acting as a current conveyor
are as follows.
(i) A current-mode front end measured voltages are converted to currents in active
electrodes, thus allowing a single cable fransmission and avoiding large input voltage
excursions in switches. In addition, a voltage-driving structure can be built by applying
the voltage to the electrode through the input buffer of the current conveyor, so making the
output current at the Z node a replication of the electrode current.
(ii) A current-mode inshunentation amplifier (IA) (Toumazou et al 199Ob). If the
AD844 is used as a current conveyor in the implementation of the current-mode (IA), the
input feedback loop will reduce the common mode rejection ratio ( C m ) degeneration due
to the gain inaccuracy of the input stage.
(iii) Current outputs of D/A convertors in direct digital synthesizers could be directly
applied to the electrode through a current conveyor, thus avoiding the phase shifts in
successive I l V and V I I conversions.

6. Conclusions

The AD844 is a commercially available current conveyor, the use of which overcomes the
problems related to the symmetry of current mirrors and their thermal behaviour. It is
possible to build a current source suitable for EIT based on this device but its gain accuracy
is very limited. Furthermore, its DC output current will drive the output to saturation if a
Dc-blocking capacitor is used.
The circuits in figure 3 solve both problems by using feedback loops. The DC feedback
does not reduce the output impedance of the source.
 Wide-band AC-coupled current source for EIT A99

We have built an inverting and a non-inverting current source thus allowing the building
of a floating load current source.
The measurements show that EIT specifications (0.1%accuracy, Ro > 1 Ma)are nearly
accomplished between 10 lcHz and 500 kHz. The output capacitance is however greater
than the desirable 5 pF.
The circuit proposed can be used not only as a current injector but also as a precision
voltage-to-current convertor and as a current amplifier in analogue front-end architecture
and in signal conditioning applications.

Acknowledgment

This work was supported by grant PB89-0505 from the Spanish DGICYT.

References

BNUUE and Haxtlusen E 1991 Current conveyor based EMG amplifier with shutdown control Ebcfron Len. 27
21724
Gallant J 1992 Analog ASICS make your circuits leaner and meaner "24 53-52
Haslett J W and Roa M K N 1979 A high quality conholled current source IEEE Trans. Imfrwm Mew. lM-28
132-140
Lee J and Tsao H 1992. True RC integrators based on current conveyors with tunable time constants using active
ContIol and modified loop technique IEEE Tram. Insfrum. Mem. IM-41709-714
Leung H T L williams R J and Griffiths H 1990 A wide-band current s o w e for electrical impedance tomography
Pmc. Meeting on ~ ( C o p e n h n g e n1990)
, (Brussels: CEO pp 206-11
Lidgey I, Vere-hunt M md Toumawu C 1990 Developments in current driver circuitry Proc. Meeting on m
(Copenhngen, 1990) (Brussels: CEC) pp 183-90
Toumazou C, Lidgey F J and H i g h D G (ed) 1990a Analogue I C . Design: The Current-mode Approach (London:
Peregrinus)
Toumazou C, Lidgey F J, and Makris C A 1990b Extending volmge-mode op amps to current-mode performance
IEEProc. 137 116-130
Webster I G 1990 ElecfricalImpedance Tomography (Bristol: Institute of Physics) ch 4, pp 36-37