b r o a d b a n d noises occurring at the electrode, membrane and the amplifier. We are currently exploring these and related applications.

References

BANKMAN,I. N., and THAKOR,N. V. (1990): 'Noise reduction in biological step signals: application to saccadic EOG,' Med. Biol. Eng. Comput., 28, pp. 544-549 BIN, Z., and YI-SHENG,Z. (1990): 'An adaptive potential filter and its application to the EOG.' Proc. 12th Ann. Int. Conf. o f l E E E Engineering in Medicine Biology Society, Philadelphia, Pennsylvania, pp. 784-785 HARRIS, R. W., CHABRIES,D. M., and BISHOP,F. A. (1986): 'A variable step (VS) adaptive filter algorithm,' I E E E Trans., ASSP-34, pp. 309-316 INCHmGOLO,P., and SPANIO,M. (1985): 'On the identification and analysis of saccadic eye movements--a quantitative study of the processing procedures,' I E E E Trans., BME-32, pp. 683-695 KWONG, R. H., and JOHNSON,E. W. (1992): 'A variable step size LMS algorithm,' I E E E Trans., ASSP-40, pp. 1633-1640 MIKHAEL,W. B., Wu, F. H., KAZOVSKY,L. G., KANG,G. S., and FRANSEN, L. J. (1986): 'Adaptive filters with individual adaptation of parameters,' I E E E Trans., CAS-33, pp. 677-686 SAKMANN, B., and NEHER, E. (Eds.) (1983): 'Single-channel recording' (Plenum Press, New York) WtDROW, B., GLOVER, J. R., Jr., McCooL, J. M., KAUNITZ, J., WILLIAMS,C. S., HEARN,R. H., ZEIDLER,J. R., DUNG, E., Jr., and GOODLIN, R. C. (1975): 'Adaptive noise cancelling: principles and applications,' Proc. I E E E , 63, pp. 1692-1716 WIDROW, B., and STEARNS, S. D. (1985): 'Adaptive signal processing' (Prentice Hall, Englewood Cliffs, New Jersey)

1 Introduction IN THE past decade or so medical evidence has persuaded people of the benefits of regular exercise in promoting physical fitness and reducing the risks of high blood pressure, cardiovascular disease and heart failure (BJORNTORP, 1982; BLAIR e t aL, 1984; MORRIS et aL, 1980; PAFFENBARGER and HYOE, 1980). This has led to a marked increase in the number of people participating in leisure sporting activities, amateur athletics and serious training programmes. Individuals who have previously led a sedentary life style face the danger of overstress when beginning an exercise Correspondence should be addressed to Dr. M. J. Burke. First received 1 February and in final form 7 September 1993

9 IFMBE: 1994 678

6 Appendix

6.1 Adaptive L M S filter The conventional AF receives two sets of sampled data inputs (time index denoted as i); a primary input s i = di + hi, where d i is the desired underlying signal and n= is an additive uncorrelated noise; and a reference input X i that is correlated with the noise component of the primary signal. The filter weights are adapted so as to minimise the MSE between the primary input and the filter output yi = x~r~W~ = Bff~X~. The method of steepest descent is employed to iteratively minimise the MSE, defined as ei = E[e2], where e i = s= - y~ is the filter error at time i, and E['] indicates the expected value. The filter weights are adjusted in the descending direction of the gradient of the error surface. The filter weights W at iteration i + 1 are updated according to W,+ 1 =

(25)

W , - - tz - -

aw,

An approximation, that the expected value of error e~ can be substituted by the instantaneous value of the squared error e2, results in the least-mean-square algorithm: c3ei/OWi = OE[e2]/O W~ ~- c3e2/c3Wi = - - 2 e , X i

(26)

Therefore, the filter weight adaptation in the LMS algorithm is carried out as follows: Wi+l = Wi + 2#e, Xi

(27)

The parameter/~ governs the rate of convergence and also strongly influences the stability and the steady-state misadjustment error. Detailed analysis of the performance of the LMS algorithm is presented elsewhere (WIt3ROW et aL, 1975; WIDROW and STEARN, 1985).

programme, as do professional athletes when undergoing rigorous training (American College of Sports Medicine, 1978; SHEPARD, 1978; SMITH and GILLIGAN, 1983). It has been well established that an individual's heart rate during exercise is closely correlated to the degree of physical exertion placed on their b o d y and can be used as an indicator of stress (BORG and LINOERHOLM, 1967; MAXFIELD and BROUHA, 1963; MOREHOUSE, 1972). M a n y portable battery-operated heart rate monitors have been marketed to allow exercising individuals to keep track of their heart rate and maintain it within safe limits without interrupting their exercise. However, studies of a number of these instruments have shown them to be extremely inaccurate and unreliable (BURKE and WHELAN, 1987; THIVIERGE and LEGER, 1986).

Medical & Biological Engineering & Computing

November 1994

7~

If there is an error in measuring the cardiac cycle time of AT, then an erroneous value of the heart rate results, which is given by

6O

~

50

R)t = 60/(T + AT) beats m i n - i

e--~.~ 30 N o 20

The error in the heart rate is then AR~ = R~ - R n

1

60 0

40

60

- T+ AT

80 120 160 200 240 hear rate, beats min ~

Fig. 1 A ~ w a b ~ error m measurement ~ cardiac ~ c ~ t ~ e ~ r •

(2)

60.AT

T -

min-~ (3)

T 2 + T.ATbeats

If the magnitude of AT is small compared with T, this is reduced to

beat m m - t error in heart rate

IARul = ~60AT T - beats m i n - t Accurate measurements of the heart rate in exercising subjects are best made from the electrocardiogram (ECG) signal. Conventionally, monitoring the ECG requires electrodes to be correctly placed on and attached to an individual's body using a coupling paste, and it is generally carried out by a trained technician. However, it is possible to record the ECG signal without the need for a coupling paste or skin preparation using dry conductive electrodes composed of a metal-impregnated carbon-based material (BERGEY et aL, 1971; K o and HYNECEK, 1974). These electrodes are strategically mounted on an elasticated belt worn by an exercising subject, which obviates the need for a trained technician. Sophisticated recording amplifiers and signal processing equipment can be used to accurately detect the peaks of the QRS complexes in the ECG signal from which the heart rate is measured. In the portable battery-operated instruments carried by athletes, however, size, weight and power consumption limitations rule out the use of sophisticated digital signal-processing circuitry, and a simpler approach must be adopted. This work describes the design and testing of a low-power amplifier and detector specifically intended for use with portable dry-electrode ECG-based heart rate monitors. It is an analogue circuit which uses commercial CMOS operational amplifiers, powered from a single 5 V supply, and provides a CMOS-compatible digital output signal.

(4)

Conversely, the allowable error in measuring the cardiac cycle time AT for a maximum desirable error A R n in the heart rate is T 2 .AR u 60. AR n

IATI--

60

= ~

s

(5)

For a maximum error of -I-i heat rain -~ in the heart rate, the allowable error in cardiac cycle time is

IATI = 6 0 / R ~ s

(6)

The value of this error is shown plotted as a function of the heart rate in Fig. 1. The principal sources of error in measuring the cardiac cycle time T are variation in the signal amplitude, modulation by interference signals and phase distortion of the signal profile (BERSON and P1PBERGER, 1966; BERSON and PIPBERGER,1967; TAYLER and VINCENT, 1983; WINTER and WEBSTER,1983; ZIPP and AHRENS, 1979). V R E2 .......... "A

approximationto the ECG signal upper threshold voltage

/i~/

input signal to the

/~__ Vr.

.

threshold detector VTL

.

.

.

.

Q/ :'

.

.

.

.

.

.

lower threshold voltage

I ' ' , \

:IS

oo,ut0svVli

:

t = 0 t r . Ton,

2 Requirements

from the

In clinical diagnosis involving the ECG signal, it is usually of the utmost importance that the profile of the signal is as faithfully preserved as possible en route from the electrodes to the recorder output. When monitoring the heart rate, however, the aim is to obtain an output logic signal from the amplifier which accurately represents the subject's pulse, usually using the peak of the QRS complex as a reference point. In this case, the requirements of the amplifier are not as stringent, and considerable distortion of the signal can be tolerated.

threshold detector

t

...........

. . . . . . -50 -25 0 25 50 75 100 125 150 175 200 225 250 time, ms a V E ............. ,

input signal VrL.VT" to the threshold detector

ECG signal contaminated with 50Hz mains interference

''~ , / u p ~p, _e~ t hr r e~volstageh o l d

",_ _ _

2.1 H e a r t rate accuracy In normal subjects, the heart rate generally lies well within the range 30-250 beatsmin-X. It is usually satisfactory to measure the heart rate to a resolution of l b e a t m i n - ~ with an accuracy of + l b e a t m i n - ~ . In exercising subjects, short-term variations in the heart rate can be important, and these are monitored by measuring the heart rate RH on a beat-to-beat basis from the cardiac cycle time T as R~ = 60/Tbeats min- 1 Medical & Biological Engineering & Computing

(1)

::~tTH', .... output pulse V | measured cardiac cycle time ' from the 1 error t : t truecardiaccy c e time iF~'p~ threshoId detector -50 -25 0 25 50 75 100 125 150 175 200 225 250 time, ms b I

Fig. 2 Simple threshold detection o f ECG signal." (a) in absence of mains interference, ( b ) in presence of mains interference

November 1994

679

+5V

C .J~,33t~1 'C,d ,o ~10% ..n.2.q'l ,;'a',33#F 1MuU ~ ' i 0 %

Rj

10MO 5%

0"57 lOOk~I R6 ~- I ~ 6 gF lgM~ 5% / 1M~ differential RTQ33k~ input signal C3L(~94pF from the electrodes

/ T

,'I

~

/

R.

C,,

R., Rio

68k~ 220k~2 _ ~

10"--~.,~ ~ 47k~,}lSOk.q I?

~I~F

0"57gF100k,

R,2

C~ FI~ I Rs

[

|

I

C~= :47nF

C7 ~ 6"8nF T

,A#~ R... ......

[

10M~

l

5%

I

7&~.s

"J'=4"C7nF

output logic signal

IC,,IC2: TLC27L4CN T,: 2N4117A

O m

Fig. 3 Schematic diagram of ECG amplifier~detector circuit," resistors have 1% tolerance and capacitors have 5% tolerance unless otherwise indicated 2.2 Signal amplitude variations In general, the amplitude of the ECG signal, and in particular of the QRS complex, varies little with the heart rate in a healthy individual. Most variation takes place between individuals. Measurements on several individuals under exercise conditions showed that an amplitude range of 0.5-25 mV should be catered for. In the designed amplifier, the variation in the output signal level was reduced to 6 dB using automatic gain control, which is discussed later. 2.3 Interference signal rejection The QRS complex of the ECG signal can be closely approximated by a triangular waveform as shown in Fig. 2a, which also illustrates simple threshold detection of this signal to obtain a rectangular pulse. The signal can be defined during its rising slope from Q to R as V(t) = (E/ToR)t

(7)

The signal reaches the threshold voltage for detection at a time twn such that Vr n = (E/TQR)trn

Fig. 2b illustrates the contaminating effect of an added interference signal such as mains hum. The result is a shift in the time at which the contaminated signal reaches the threshold voltage compared with the uncontaminated signal. If the interference signal has a peak voltage V~ and frequency fi, then the contaminated signal can be described prior to detection as (9)

where 49 is an arbitrary phase angle defining the alignment in time of the ECG signal and the interfering signal. This contaminated signal will reach the threshold voltage at a time t~.nsuch that Vrn = (E/rQR)t'rn + Vii sin (2zfi t~rHq - 49)

(10)

Equating eqns. 8 and 10 yields

(E/TQR)t'Ttt + Vii sin (2;tfi t~-tt + 49) = (E/To.R)tTH 680

Atrn = t'Tn -- trH = -(Vi/E)TQR sin (2n f/t'Tu + 49) (12) When the signals are aligned accordingly, this has a maximum value of AtMA x = ++_( V# E) To R

(13)

As the interfering signal may align with the desired signal to cause a positive error in detecting one QRS complex and a negative error in the next, as shown in Fig. 2b, then the maximum possible error in the cardiac cycle time ATMAx must be taken as ATMAx = 4-2(V#E)TQR

(14)

In order to limit the error in heart rate, the ratio of the magnitudes of the interference signal and the QRS complex must fulfil the condition _> ~

(15)

Vrn (8)

V'(t) = (E/ToR)t + V~ sin (2rcfd + 49)

From this the error in the time at which the threshold voltage is reached can be obtained as

(1 I)

At a maximum heart rate of 250 beats min-1, the error ATMAx must be less than 1 ms as given by eqn. 6 if the error in heart rate is to be less than 4- 1 beat m i n - t. If a typical value of T~R is taken as 30 ms, then this yields ]E/V~[ > 60 or a signal-to-interference ratio of 36 dB. In exercising subjects, interference signals arise from several sources including motion-induced muscle noise and electrode contact artefacts, electromagnetically induced mains hum and low-frequency variations in the QRS complex amplitude due to respiration. The frequency spectrum of the ECG signal generally contains components from DC up to 200 Hz. Most of the spectral energy contained in the QRS complex, however, lies in the range 5-35 Hz (BRYDON, 1976; GOLDEN et aL, 1973; THAKOR et al., 1984). The most troublesome interference signals caused by bodily movement generally have maximum energy in the range 0.5-3.0Hz (BRYDON, 1976: THAKOR et al., 1984). These figures suggest that a low cut-offpoint in the amplifier frequency response of 5 Hz, with a steep roll-off slope, would reject most of this interference while retaining the principal components of the QRS complex.

Medical & Biological Engineering & Computing

November 1994

intended range of operation

I iiiiiitl

2"0

\] I Ili;ii

l ltl!ll

heart rate,

beatsmio'

1.8. 1.6" 1.4

C:ZC ~ "5 .-~ 9

-: tOO

i III1[I _,, 150

1.2

i l I1[11

1-0 0-8"

--

~

= 200

0"6-

250

0-4 9 0"20"0 0"1

1

10

100

output signal level (log. scale), mV

Fig. 4 Curves illustratiny operation of A GC Mains hum interferes with the ECG signal in both common and differential modes. Common mode interference can be effectively suppressed by matching the input impedance and gain on inverting and non-inverting sides of the amplifier (McCLELLAN, 1981; PALLAS-ARNEY, 1986; WINTER and WEBSTER, 1983). Differential mode hum appears as a differential input signal to the amplifier and is usually suppressed by the use of notch filters. The characteristics of these filters are often very sensitive to component tolerances and frequently require fine-tuning which may drift with time. Obtaining sufficient attenuation of the interfering signal in the narrow bandwidth concerned with an acceptable phase response is not an easy task and can require more stages of circuitry than the remainder of the amplifier, which increases power dissipation. As the predominant spectral content of the QRS complex extends only to 35 Hz, a simple alternative means of suppressing differential mains hum is to limit the high cut-off frequency of the amplifier to around 30 Hz and to use a steep roll-off slope. 2.4 Phase distortion Non-linearity in the phase characteristic of an amplifier causes distortion of the profile of a signal passing through it, which will vary depending on the spectral content of the signal and the non-linearity of the phase characteristic. This often leads to serious problems in clinical diagnosis (BERSON a n d PIPBERGER, 1966; BERSON and PIPBERGER, 1967; TAYLER and VINCENT, 1983). In the ECG signal, however, the nature of the QRS complex is largely independent of the heart rate in healthy individuals so that its spectral content remains relatively constant as the heart rate changes. This means that, although the QRS complex profile is altered at the amplifier output, the distortion does not change with heart rate. Considerable distortion of the QRS complex can be tolerated provided that a fiducial point is present that can be detected accurately and used to measure the heart rate. This reduces the need for phase compensation of the filter stages in the amplifier.

stage comprising RI-R,2, C1-C 3 and IC 1 pins 5-7, 8-10 and 12-14 provides a balanced high input impedance of 10 MD, a gain of 80, a minimum common-mode rejection ratio of 56 dB and 40 dB attenuation of motion-induced electrode artefact. This stage has a low cut-off frequency of 5 Hz. The input stage is followed by a second-order Sallen-Key high-pass filter consisting of R~3-Rts, C4, Cs and [C, pins 1-3. This stage also has a cut-off frequency of 5 Hz and combines with the input stage to provide a third-order high-pass Butterworth response with a roll-off slope of 18 dB octThis is followed by second-order Sallen-Key low-pass filters comprising R16-RI9 , C6-C9, IC 2 pins 5-7 and 12-14, positioned each side of the AGC stage consisting of Rzo-Rzr C1o-C13, D x, T 1 and ICz pins 8-10. These filters both have a cut-off frequency of 30 Hz and combine with the high-frequency response of the AGC stage to provide a fifth-order low-pass Butterworth response with a roll-off slope of 30 dB oct- 1. This provides an attenuation of 26 dB at 50 Hz to differential mains interference. The AGC stage provides an output signal level varying from 0-75-1.5Vpk for an input signal ranging from 0.5-25mVpk, which gives a reduction in amplitude variation of almost 30 dB. The output signal is negativegoing with respect to a DC bias voltage of 2"5 V. The circuit is configured to use an N-channel JFET, T1, as a voltage-controlled resistance in the feedback loop of up-amp, IC2 pins 8-10, to accomplish the necessary gain variation. The JFET is biased by the network R2o and C~o to maintain the drain-source voltage as close to zero as possible. Linearisation is provided by negative feedback via C l l , R22 and R23 to maximise the useful range of the JFET. The peak detector formed by D~, Rz,~ and C~3 provides the control voltage at the gate of the JFET. Resistor R2, must be selected to suit the characteristics of the particular JFET. The final stage of the circuit is a modified Schmidt-type threshold detector comprising R25-R29 , C14 , Cls and IC 2 pins 1-3, which acts as a comparator with hysteresis and detects the negative-going peaks of the amplified signal. It has a threshold voltage of 2-07 V set at half the value of the minimum signal peak. Once the input signal reaches this threshold, the output voltage at IC2 pin 1 goes to the positive supply rail. The circuit has a steady-state hysteresis of 0-2 V provided by the positive feedback via Rzv to IC 2 pin 3. However, a much larger amount of AC hysteresis is provided by the network R29 and Cls, which causes the duration of the output pulse of the comparator to be prolonged beyond that of the input signal, providing a monostable multivibrator action. The duration of the output pulse is governed by the time constant C~sRzg, which must be short enough to allow the circuit to return to its steady-state

35 ~

3oi

delay

251 2o~

,5i

3 Circuit design

A schematic diagram of the amplifier/detector is shown in Fig. 3. The primary objective was to design the simplest possible circuit using the minimum number of components to minimise size and power consumption. A low-power CMOS operational amplifier* was used. A differential input * Texas Instruments T L C 2 7 L 4 C N

Medical & Biological Engineering & Computing

input signat level 9 9 0"5 mV 5"0 mV = = 25-0 mV

10~

5i O:

9,

~ , 40

80

120

160

200

" ~ magnitude of delay ~ - relative to that at ~, . 250 beats min-' 240

heart rate, beats min"

Fig. 5 Curres illustrating signal delay introduced by ECG amplifier/detector November 1994

681

and recorded on a two-channel chart-recorderw running at a chart speed of 30 cm min-1. 100 paired readings were obtained from each subject under four conditions:

Table 1 Results of the exercise tests

test condition

sitting relaxed walking @3 k m h - I jogging @ 5 k m h -l running @10kmh -1

error range beats min-

+ 5 t o _+10 + 11 to 4-20 > _+20 4- 5 to 4- 10 -+ 11 to -+ 20 >-+20 _+5 to 4-10 4-11to4-20 >-+20 _+5 to _+10 4-11to4-20 >_+20

subject no. 1

2

3

4

5

6

0 0 0 0

0 0

0 0 0 1

0 0 0 1

4 0 0 0

1 0 0 0

0

0

0

0

1

0

0

0

0 0 0 0 * * *

0 0 0 0 0 0 0

1 0 1 5 1 0 9

1 3 7 9 * * *

0 0 0 0 6 0 1

0 0 0 0 5 0 0

* no reliable readings obtained due to poor quality of the conventional ECG signal between heart beats. The output from the detector is compatible with C M O S logic circuitry.

4 Performance evaluation

A lead II E C G signal with an amplitude ranging from 0-5-25 mV and a heart rate ranging from 30-250 beats min-1 was obtained from a test signal generator specially constructed to test the designed amplifier/detector. The amplifier proved to have a 3 dB bandwidth of 7-30-5 Hz with the intended out-of-band roll-off in the response. The rejection ratio for 50 Hz mains interference was 90 dB for c o m m o n - m o d e and 18 dB for differential-mode at a signal heart-rate of 100 beats m i n - 1. The operation of the A G C circuit is illustrated by the graph of output versus input signal level shown in Fig. 4 for selected heart rates. The performance is a little poorer than intended but may be improved by using a more complex AGC circuit with some increase in power consumption. The most important aspect of the circuit's performance is indicated in Fig. 5, which shows graphs of the delay measured between the peak of the QRS complex in the input signal and the leading edge of the output pulse from the threshold detector, plotted as a function of heart rate for several input signal amplitudes. The variation in this delay can be seen to be well within the limits allowed for a _ 1 beat min -1 error in measuring the heart rate, as indicated in Fig. 1. The performance of the circuit was tested under more practical conditions by assessing measurements of the beat-to-beat heart rate of six healthy male subjects, aged between 20 and 55 years, under several conditions of exercise. The subjects' upper clothing was removed to allow a safety harness to be worn on a motorised treadmill. Readings were obtained from a cardiotachometer (BURKE, 1990) connected to the amplifier/detector circuit under test used with an elasticated dry-electrode b e r t worn around the subject's chest. These readings were compared with readings obtained from a commercial rate meter, connected to a conventional E C G monitor*, picking up a lead II exercise E C G signal via flying leads from pasted metal electrodes. An analogue output signal representing the beat-to-beat heart rate was obtained from each monitor

sitting relaxed on a stool. walking on the treadmill at a speed of 3 km h - 1 jogging on the treadmill at a speed of 5 km h - 1 running on the treadmill at a speed of 10 km h - t. Each set of readings was analysed, and the differences between readings obtained from the two monitors were classified into three ranges of errors for each condition of exercise, as presented in Table 1. The accuracy of the rate meter was limited to + 2~ but this was the only instrument available that would provide the required output signal. For this reason, errors of + 4 beats min -1 or less were not assessed as they could not be attributed definitely to the circuit under test. The reason for the four minor errors in the readings obtained from subject 5 when sitting relaxed is unknown. Most of the other errors can be attributed to motion-induced interference in the conventionally monitored E C G signal. Overall, the performance of the amplifier/detector under the exercise conditions assessed is very reasonable. Of the total of 2200 readings analysed, 2-6% contained errors of + 5 beats rain- i or more, 1-6% contained errors of more than + 10 beats m i n - 1 and 1.2% contained errors of more than + 2 0 b e a t s m i n -1. The circuit performed notably better on some subjects than on others, and there is r o o m for improvement in the suppression of motion-induced interference. 5 Conclusion

The amplifier/detector has been constructed on a single-sided printed circuit board using surface-mounted components and fits into a package, measuring 42 x 42 x 12 mm and weighing 20 g, which can be mounted on the electrode belt. The circuit draws a current of 120/~A from a 5 V supply giving a power consumption of only 0-6 mW. The output signal from the unit is CMOS-compatible and directly interfaces with a microprocessor over a 1 m length of thin screened lead. The circuit presented in this work is intended to be used in conjunction with dry electrodes for the purposes of interfacing with a portable cardiotachometer. It provides an effective low-power analogue alternative to more sophisticated digital signal-processing circuits, which are very power-consuming by comparison. With some modification, it could be adapted to suit clinical monitoring of the E C G for cardiac stress testing and other exercise-related measurements. References

American College of Sports Medicine (1978): 'Position statement on the recommended quantity and quality of exercise for developing and maintaining fitness in healthy adults,' Med. Sci. Sports Exer., 10, (3), pp. vii-x BERGEY, G. E., SQUIRES, R. D., and SIPPLE, W. C. (1971): 'Electrocardiogram recording with pasteless electrodes,' IEEE Trans., BME-18, pp. 206-211 BERSON, A. S., and PIPBERGER,H. V. (1966): 'The low-frequency response of electrocardiographs, a frequent source of recording errors,' Am. Heart J., 71, pp. 779-789 BEaSON, A. S., and PIPBEROER,H. V. (1967): 'Electrocardiographic distortions caused by inadequate high-frequency response of direct-writing electrocardiographs,' ibid., 74, pp. 208-218

1 Respironics Inc. t Devices Ltd., Model 4522 " Devices Ltd.

682

w Rikadenki R-50

Medical & Biological Engineering & Computing

November 1994

BJORNTORP,P. (1982): 'Hypertension and exercise,' Hypertens., 1, Suppl. III, pp. 56-59 BLAIR, S. N., GOODYEAR,N. N., GIBaONS, L. W., and COOPER, K. H. (1984): 'Physical fitness and incidence of hypertension in healthy normotensive men and women,' J. Am. Med. Assoc., 252, pp. 487-490 BORG, G., and LINDERHOLM,H. (1967): 'Perceived exertion and pulse rate during graded exercise in various age groups,' Acta. Med. Scand., 472, Suppl., pp. 194-206 BRYDON,J. (1976): 'Automatic monitoring of cardiac arrhythmias' in HILL, D. W., and WATSON, B. W. (Eds.): IEE Medical Electronics Monographs Nos. 18-22. (Peter Peregrinus Ltd., Stevenage) pp. 27-41 BURKE, M. J. (1990): 'A microcontroller based athletic cardiotachometer'. PhD Thesis, Trinity College, Dublin BURKE, M. J., and WHELAN, M. V. (1987): 'The accuracy and reliability of commercial heart-rate monitors,' Br. J. Sports Med., 21, pp. 29-32 GOLDEN, D. P., WOLTHUIS,R. A., and HOFFLER,C. W. (1973): 'A spectral analysis of the normal resting electrocardiogram,' IEEE Trans., 20, pp. 366-372 Ko, W. H., and HYNECEK,J. (1974): 'Dry electrodes and electrode amplifiers' in MILLER, A. C., and HARRISON, D. C. (Eds.) 'Biomedical electrode technology' (Academic Press, London) pp. 169-181 MAXEIELD,M. E., and BROUHA,L. (1963): 'Validity of heart rate as an indicator of cardiac strain,' J. Appl. Physiol., 18, pp. 1099-1104 MCCLELLAN, A. D. (1981): 'Extracellular amplifier with bootstrapped input stage results in high common-mode rejection,' Med. Biol. Eng. Comput., 19, pp. 657-658 MOREHOUSE, L. E. (1972): 'Exercise heart rate' in 'Laboratory

manual for physiology of exercise' (C. V. Mosby Co., London) pp. 63-74 MORRIS,J. N., EVER1TT,M. G., POLLARD,R., CHAVE,S. P. W., and SEMMENCE. A. M. (1980): 'Vigorous exercise in leisure-time: protection against coronary heart disease,' Lancet, 2, pp. 1207-1210 PAFFENBARGER, R. S., and HYDE, R. T. (1980): 'Exercise as protection against heart attack,' New Eng. J. Med., 302, pp. 1026-1027 PALLAS-ARNEY,R. (1986): 'On the reduction of interference due to common mode voltage in two-electrode biopotential amplifiers,' IEEE Trans., BME-33, pp. 1043-1046 SHEPARD, R. J. (1978): 'The physical working capacity of the athlete' in 'Human physiological work capacity' (Cambridge University Press, Cambridge) pp. 136-178 SMITH, E. L., and GILLIGAN, C. (1983): 'Physical activity prescription for the older adult,' Physician Sportsmed., 11, (8), pp. 91-101 TAYLER, D., and VINCENT, R. (1983): 'Signal distortion in the electrocardiogram due to inadequate phase response,' IEEE Trans., BME-30, pp. 352-356 THAKOR, N. V., WEBSTER, J. G., and TOMPKINS, W. J. (1984): 'Estimation of QRS complex power spectra for design of a QRS filter,' IEEE Trans., BME-31, pp. 702-706 THIVIERGE, M., and LINER, L. (1986): 'Validite des cardiotachometres,' L 'Entraineur, April-June, pp. 28-29 WINTER, B. B., and WEBSTER, J. G. (1983): 'Reduction of interference due to common mode voltage in biopotential amplifiers,' IEEE Trans., BME-30, pp. 58-62 ZIpP, P., and AHRENS,H. (1979): 'A model of bioelectrode motion artefact and reduction of artefact by amplifier input stage design,' J. Biomed. Eng., 1, pp. 273-276

1 Introduction

2 Frequency r e q u i r e m e n t s

IN THE bioelectric field there is a requirement to stimulate tissue and measure its response. In bioimpedance, especially electrical impedance tomography, there is a requirement for the accurate application of current and measurement of voltage; the latter with an amplitude error equal to or less than 0"1% (BROWN and SEAGER, 1987). A wide-band approach has been adopted to allow the possibility of tissue characterisation and detecting pathology by impedance spectroscopy (SINGH et al., 1979; GRIFFITHS, 1988).

Tissue distinguishability depends on the probe current frequency; 100 kHz for intracellular/extracellular structures in muscle, 2 5 0 k H z for liver, 5 0 0 k H z - 1 M H z for fat and > 3 M H z for blood, (GRIFFITHS, 1988; RIU I COSTA, 1991). For bioelectric work, a bandwidth of 10 kHz is sufficient for the fast moving signals observed in electromyography (DE LUCA, 1988). Errors in impedance measurement result from errors in both voltage and current. In four-point impedance measurement, the current source must have at least 5 M ~ output impedance and very low capacitance for the required accuracy. This is difficult to achieve for the required bandwidth. This design proposes a novel voltage source

First received 18 March and in final form 9 August 1993

9 IFMBE: 1994 Medical & Biological Engineering & Computing

November 1994

683