Sei sulla pagina 1di 4

Practical considerations for the use of a Howland

current source for neuro-stimulation


Philippe Pouliquen Jacob Vogelstein Ralph Etienne-Cummings
Electrical and Computer Engineering Applied Physics Laboratory Electrical and Computer Engineering
The Johns Hopkins University The Johns Hopkins University The Johns Hopkins University
Baltimore, MD 21218 Laurel, MD 20723 Baltimore, MD 21218
Email: philippe@jhu.edu Email: Jacob.Vogelstein@jhuapl.edu Email: retienne@jhu.edu

AbstractWe analyze the Howland current source circuit device. However, as shown here, creating even a simple reli-
topology in detail with regards to non-ideal circuit component able constant-current stimulator requires non-trivial engineer-
properties. This circuit is one of the few which can be built ing efforts and a judicious selection of electronic components.
with a small number of discrete components, offers very good
performance due to the use of an operational amplifier, and In the following sections, we present a formal analysis of our
supplies true symmetric bi-directional currents, and is hence well design with the hope that this treatment will allow others to
suited for use in the multi-channel electrical stimulation of nerve more rapidly and effectively assemble their own experimental
fibers and neurons. instrumentation.
I. I NTRODUCTION II. H OWLAND CURRENT SOURCE
Electrical stimulation has been used as an investigative tool The Howland current source [8] is an operational-amplifier
in neuroscience for over two hundred years, starting with circuit topology which effectively forms a linear differential
the pioneering work of Luigi Galvani in 1791 [1]. Since voltage-to-current converter. The basic diagram for this circuit
that time, countless devices have been designed to provide is reproduced in Fig. 1.
electrical stimuli via a host of different mechanisms, but
it is only recently that we have come to a comprehensive R10 R2
Vin- q
understanding of the physical and physiological processes in- I I
volved [2]. Perhaps because so many approaches have proven V-
Q
Q q
Q
viable, the neuroscience community has not standardized on  Vo
a single bench-top stimulator design, and most neuroscience V+ 
I q I

researchers construct their own home-grown solution for Vin+
each experiment. Although this is frequently a cost-effective R1 R20
L
solution, it results in redundant efforts across research labo- O I
ratories because the designs are not typically disseminated in A ?L
publications. D
For our application, we wanted a simple design that could
be easily tiled to accommodate large multielectrode arrays
such as the Cyberkinetics Microelectrode Array (Cyberkinetics Fig. 1. Howland current source.
Neurotechnology Systems, Foxborough, MA) or the Floating
Microelectrode Array (Micro Probe, Gaithersburg, MD), and The advantage of this circuit over other current source
supply biphasic currents as large as 1mA with a resolution topologies is that it does not employ additional transistors or
on the order of 1nA. Because none of the existing published switches, and when used with a dual-rail power supply, can
stimulator designs we reviewed [3], [4], [5], [6], [7] met all produce both positive and negative currents without cross-
of our specifications, we recently developed stimulator cir- over distortion or transconductance mismatches. The main
cuit utilizing one small microcontroller, one digital-to-analog disadvantage is that the effective source conductance is low
converter and a Howland current source per channel. The only if the resistors are well matched.
microcontroller stores the stimulus waveform and the event The basic operation of the circuit for ideal components is
times for the stimulus train, which can be at a fixed interval straightforward. Assuming that R1 = R10 and R2 = R20 , and
or an arbitrary pattern (with a capacity to store on the order setting the current flowing into and out of each node to be
of 1000 events). The total printed circuit board area required equal, the load current IL can be determined as follows:
is on the order of 1.5 square inches.
By basing our design on a standard Howland current source, Vin- V- V - Vo V + Vo V V+
we assumed that we could rapidly assemble a functional = and = in+ IL (1)
R1 R2 R2 R1

978-1-4244-2879-3/08/$25.00 2008 IEEE 33


Since V+ = V- , The right-hand term encapsulates the source resistance, and
V Vin- by comparing it with (2), we find that the equivalent source
IL = in+ resistance is:
R1
This equation shows that the transconductance of the circuit R1 R10 R1
RS =
is related only to the value of R1 , and hence the designer R10 R1 tol1
is free to choose a value of R2 to satisfy other design
constraints. (It may seem at first glance that a small value where tol1 is the expected tolerance of the input resistors
of R2 is desirable, but in fact, the selection of R2 will be a (e.g. 1% for typical metal-film resistors).
tradeoff between offset, discussed in Sec. V, and output swing, Similarly, if all components are ideal and R1 = R10 but
discussed in Sec. VI.) R2 6= R20 , then
The following sections will analyze the circuits behavior
with respect to the non-ideal properties of the component parts. Vin- V- V- V o V + Vo V V+
= and = in+ IL
III. R ESISTOR MISMATCH R1 R2 R20 R1

Again using V+ = V- = VL ,
VL
q "
Vin+ R2 #
R20 Vin- R20 R2
 
 L IL = VL
IS RS O I R1 R1 R20
H
? A ?L
q D
The equivalent source resistance is therefore:

Fig. 2. Current source model. R1 R20 R1


RS = 0
R2 R2 tol2
A generic current source can be modeled as an ideal current
source IS in parallel with a resistor RS , as shown in Fig. 2. In where tol2 is the expected tolerance of the feedback resis-
order for the load current IL to be as close as possible to the tors.
source current IS , the source resistance must be as large as In summary, the performance of the current source in
possible. Otherwise, a portion of the source current will be lost terms of its output resistance is directly related to the resistor
to the source resistance according to the following equation: tolerance (or matching if the resistors are hand-selected) and
the value of the input resistors.
V
IL = IS L (2)
RS
IV. A MPLIFIER INPUT BIAS CURRENT
In the case of the Howland current source, if the circuit com-
ponents are ideal and the input resistors and feedback resistors In the previous analyses, we assumed that the amplifier input
are perfectly matched, then the load current is independent of bias currents Ib- and Ib+ were zero. If we include the input
the load voltage VL , and the equivalent source resistance is bias current, (1) becomes
infinite.
However, assuming as before that all components are ideal Vin- V- V- V o
Ib- =
and R2 = R20 but R1 6= R10 , then R1 R2
Vin- V- V - Vo V + Vo V V+ and
0 = and = in+ IL
R1 R2 R2 R1 V + Vo V V+
= in+ Ib+ IL
Since V+ = V- = VL , R2 R1

Vin+ R
" #
R10 Vin-
1  0
Letting V+ = V- ,

R1 R1
IL = V L R R0
R0 1 1 1
V Vin-
IL = in+

The left-hand term of the equation indicates that there will Ib+ Ib-
R1
be a mismatch between the transconductance of the circuit
relative to the two input control signals Vin+ and Vin- . We As can be seen from the above equation, only the amplifier
shall show in another section that there is yet another source input bias current difference (commonly called the input offset
of asymmetry for the two input signals, and therefore it is current Ios = Ib+ Ib- ) affects the load current. Hence, only
advisable to apply AC signals to only one of the inputs, and operational amplifiers with relatively low Ios are suited for
keep the other at a fixed bias. use in a Howland current source configuration.

34
V. A MPLIFIER INPUT OFFSET In general, the smaller R2 is the greater the compliance
In the previous analyses, we assumed that V+ V- = 0. In limit of the current source. However, note also that for normal
the presence of an amplifier input offset however, this becomes passive loads (positive load resistance), the load voltage VL
V+ V- = Voff . Assuming that R1 = R10 and R2 = R20 as will be of the same sign as the load current IL , which in turn
before, we have is proportional to Vin+ Vin- . Hence, by using Vin+ to control
the current we obtain a greater compliance limit than by using
Vin- V- V- V o V + Vo V V+
= and = in+ IL Vin- .
R1 R2 R2 R1
Substituting V+ = VL and V- = VL Voff , VII. A MPLIFIER SLEW RATE
Vin+ Vin-
   
R1 + R 2
IL = Voff When a large step input is applied to Vin- or Vin+ , the rate
R1 R1 R2
at which the load current adjusts to its new level is dependent
In this expression, it can be seen that the amplifier input dV
on the amplifiers output slew rate dto . However, resistors R1
offset produces an error term which is dependent on the and R2 form an attenuator structure, so that the slew rate at
effective resistance of R1 in parallel with R2 . This is the main the load is given by
reason why R2 cannot be made arbitrarily small (e.g. zero),
since the error due to the offset would then become arbitrarily  
dVL R1 dVo
large. = (3)
Note that the amplifier input offset in the above equations dt R 1 + R2 dt
encapsulates two components, the random offset Vr-off that is
observed from device to device, as well as the deterministic As in the case of the amplifier compliance, a small value of
offset Vd-off due to the finite gain of the amplifier. Vd-off is R2 is desirable to maximize the current source slew rate. A
related to the amplifier output as similar analysis of the feedback factor [9] for the bandwidth
shows that making R2 small also maximizes the bandwidth.

Vo = Ad Vin+ Vin- An additional side-effect of the resistor attenuator structure
is that if a step input is applied to the circuit, it has an
where Ad is the amplifiers differential gain. Hence, instantaneous effect on VL if applied to Vin+ but not if it
Vo is applied to Vin- . This is shown schematically in Fig. 3.
Vd-off =
Ad
VI. A MPLIFIER OUTPUT SWING
C  C   
The limited output voltage swing of the operational am- C  C   
plifier results in a compliance voltage (the magnitude of the C  C  C C
maximum voltage the current source will go to). From (1), we C  C  C C
have (a) (b)
 
1 1 Vo V Vo V
VL + = + in- = + in+ IL
R1 R2 R2 R1 R2 R1
Fig. 3. Effect of input step on output current. (a) Step applied to Vin- . (b)
If the output of the operational amplifier is limited to Vmax Step applied to Vin+ with R1 = R2 .
and Vmin , then the load voltage will be limited to
When a square wave is applied to Vin- , the output VL will
R1 Vmax + R2 Vin-
VLmax = appear trapezoidal as in Fig. 3(a), according to (3). However,
R1 + R 2 when a square wave is applied to Vin+ , the output VL will
R1 Vmax + R2 Vin+ R1 R2 IL be a truncated trapezoid as in Fig. 3(b). The magnitude of the
= truncation is proportional to R2 /(R1 + R2 ), as the resistor
R1 + R2
attenuator structure appears in reverse relative to Vin+ . After
R1 Vmin + R2 Vin- the initial jump, the subsequent slew rate limited section will
VLmin = still follow (3), so that the overall settling time will be shorter
R1 + R 2
than if the same step input had been applied to Vin- .
R1 Vmin + R2 Vin+ R1 R2 IL
= This asymmetry in behavior relative to the two inputs makes
R1 + R 2 it difficult to accurately compute the delivered charge when
Since the amplifier inputs are not guaranteed to be equal using differential input signals. Although the problem can be
once the amplifier output has reached the limits of its swing, mitigated by limiting the input slew rate such that the amplifier
only the equations in Vin+ give the true relation between VL can keep up, in many cases it may be preferable to apply only
and Vo . However, the equations in Vin- are still useful in single-ended signals to either Vin- or Vin+ and keep the other
determining when compliance will be reached. input fixed.

35
VIII. S AMPLE RESULTS
We evaluated several operational amplifiers for use in a
stimulator design. The amplifiers were initially selected for
their availability in a common package (8-SOIC), and their
ability to operate with a +5/ 5V power supply (for com-
patibility with large commercial CMOS switch arrays such as
the ADG2128).
The amplifiers were all tested in the Howland current source
configuration with R1 = RL = 4.99k and R2 = 1.00k
(1% metal film resistors). The load resistance was chosen to
enable the observation of each amplifiers behavior at its com-
pliance limit as well as its capacity to supply large currents.
In practice, electrode impedances may be much higher, but
the stimulation currents will also tend to be correspondingly
lower. The experimental results are summarized in Table I. Fig. 4. Example printed circuit board section for two channels.

TABLE I
S UMMARY OF EXPERIMENTAL RESULTS FOR SOME OPERATIONAL A prototype 4 channel stimulator has been partially as-
AMPLIFIERS .
sembled. Fig. 4 shows two of the four channels, with the
Op. Amp. Vmax Vmin Slew rate Howland current source operational amplifier on the left, DAC
(V) (V) (V/s) (lower) and CMOS switches (upper) in the middle, and the
AD8675 +4.95 -4.96 1.5
microcontroller on the right. Note that the Howland current
LT1218 +4.93 -4.94 0.06
source resistors have not been installed, as the high precision
OP177 +4.33 -4.18 n/a
resistors were not yet available at the time the photograph was
OP27 +4.27 -4.19 n/a
taken.
THS4221 non-linear n/a R EFERENCES
THS4225 non-linear n/a
[1] M. Piccolino, Luigi Galvani and animal electricity: two centuries after
TLC2201 +4.64 -4.62 n/a the foundation of electrophysiology, Trends in Neurosciences, vol. 20,
no. 10, pp. 443448, 1997.
[2] D. Merrill, M. Bikson, and J. Jefferys, Electrical stimulation of excitable
Of the amplifiers tested, two failed to supply an output tissue: design of efficacious and safe protocols, Journal of Neuroscience
current linearly related to the input (THS4221 and THS4225), Methods, vol. 141, no. 2, pp. 171198, 2005.
although the cause of the non-linearity was not further in- [3] J. Gwilliam and K. Horch, A charge-balanced pulse generator for nerve
stimulation applications, Journal of Neuroscience Methods, vol. 168,
vestigated. Three other amplifiers were not able to drive their no. 1, pp. 146150, 2008.
outputs completely from rail-to-rail (OP177, OP27, TLC2201) [4] C. Hofer, W. Mayr, H. Stohr, E. Unger, and H. Kern, A stimulator for
and hence were handicapped in terms of their compliance lim- functional activation of denervated muscles, Artifical Organs, vol. 26,
no. 3, pp. 276279, 2002.
its. Of the remaining two amplifiers, the AD8675 was finally [5] F. Jaw, S. Liu, T. Kuo, and C. Wang, Microcomputer-based pulse
selected because of its superior slew rate, even though the stimulator, Journal of Neuroscience Methods, vol. 62, no. 1-2, pp. 193
LT1218 exhibited a more graceful performance degradation 197, 1995.
[6] H. Wu, S. Young, and T. Kuo, A versatile multichannel direct-
when approaching its compliance limit. synthesized electrical stimulatorfor FES applications, IEEE Transactions
on Instrumentation and Measurement, vol. 51, no. 1, pp. 29, 2002.
IX. C ONCLUSION [7] J. Ross, D. Lovely, and P. Parker, Design of a PC controlled constant
current stimulator for evokedpotential studies, Proceedings of the 22nd
In summary, for best performance, a precision amplifier (low Annual International Conference of the IEEE Engineering in Medicine
offset voltage) should be used, with an input bias current offset and Biology Society, vol. 4, 2000.
better than the desired current resolution. Resistors should be [8] P. Horowitz and W. Hill, The Art of Electronics, 2nd ed. Cambridge
University Press, 1989.
matched as well as possible to maximize the equivalent source [9] A. Sedra and K. Smith, Microelectronic Circuits, 3rd ed. Saunders
resistance of the circuit, and R2 should be minimized but not College Publishing, 1991.
so much that the error due to the amplifiers offset becomes
too large. The control signal should be applied to only one
of the inputs, choosing Vin+ to maximize the compliance, or
Vin- to avoid distortions for step inputs.

36

Potrebbero piacerti anche