You are on page 1of 6

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 49, NO.

1, FEBRUARY 2000 37

Minimization of the Optical Smoke Detector False


Alarm Probability by Optimizing its Frequency
Characteristic
Željko J. Aleksić

Abstract—The exclusive use of the time domain analysis in the area contamination of optical components, temperature changes
design tuning to lower the probability of false alarms is caused by and corresponding variations of optical component parameters,
the output signal binary nature prevailing in the optical smoke de- the aging of the components, moisture, etc.
tector designs. However, the characteristic of a standard detector
circuit is fairly linear up to the output comparator. On the basis False alarms caused by the sources from the first group are
of the generalized functional block scheme, for this linear depen- suppressed by the application of various pulse timing schemes
dence the Laplace transform is derived in this paper. The transfer [6], [7], by modulation [3], integration, i.e., averaging of the
function obtained is optimized with respect to false alarm proba- output pulses during several detection periods [8], etc.
bility minimization, for the second and the fourth order cases. Ex- Detector sensitivity dependence on the slowly varying para-
ploiting the optimization results presented, the improvement of the
detector design to obtain lower probability of false alarm reduces sitic influences is often suppressed by compensation techniques
to the calculation of element parameters from the coefficients of the [9] including a calibration as a common practice. The regular
optimized transfer function. If the requests for detector time re- service procedure assumes the cleaning of optical components
sponse behavior are different from those adopted here, this paper to cope with their contamination. Another method for attenu-
presents a systematic set of design steps that have to be performed ating slowly varying factors relies upon blocking the dc compo-
to obtain the optimal performance.
nent in the optical receiver output signal, and monitoring only
Index Terms—Circuit optimization, detectors, monitoring, reli- its dynamic component [10].
ability, safety, transfer functions.
It is evident that techniques mentioned above are intended to
stabilize the gain, the dc operating point, or to limit the detector
I. INTRODUCTION sensitivity passband. However, there is no any eligible analysis
in the current literature directed toward the evaluation and ap-
A BSORPTION index [1]–[4] is a fire parameter used in
the threshold sensitivity tests of the optical smoke detec-
tors. To avoid false alarms, caused by fluctuations of the normal
propriate optimization of the transfer function.
It has to be noticed that some novel reports [11]–[14] disclose
great efforts to reach the optimal performance in an alternative
ambient smoke and dust concentrations, or by small deteriora-
way. They are based on another approach relying upon the bene-
tion of the detector characteristics, the minimal value of the ab-
fits of the modern digital technology. Owing to the implementa-
sorption index threshold level is prescribed by the standards. On
tion of the low power analog-to-digital (A/D) converters and mi-
the other hand, detectors are classified according to the maximal
crocontrollers in the “smart” detectors that often monitor several
attained values of the fire parameters initiating the alarm in the
fire parameters, the algorithms developed by the neuro fuzzy
fire sensitivity tests.
logic are used to detect fire. Fire detection is based on the time
Lower false alarm probability is usually achieved, among
dependence of the signals that are processed by these algorithms
others, by the detector dynamic characteristics limitation.
[11]. In some designs the decision task is transferred to the host
Acceptable limits depend on the possible values of the ab-
unit, whereas detectors, with their output A/D converters, act as
sorption index rate of change. From the analysis and the data
linear meters. Nevertheless, these novel techniques do not mini-
on the spatial and time distribution characteristics of the fire
mize the importance of the evaluation of the optimum character-
parameters [3], [5], it stems out that 1 s and 1 h are the rough
istic. In the “smart” fire detectors monitoring several fire param-
limits of the equivalent time constant. Hence, the amplitude
eters, this characteristic has to be used for the filter processing
characteristic of the detector sensitivity should be of a bandpass
the smoke sensor output. Also, it can be used in the smoke de-
type, with the orders of magnitude of the edge frequencies of
tector realizations for one of the filters extracting specific fea-
1/3600 and 1 rad/s.
tures from the input signal [11]. In the low cost analog smoke
The sources of detector malfunctioning can be grouped into
detectors, the optimal characteristic is the only solution for min-
two separate classes. Stochastic impulse interferences, having a
imal false alarm probability.
wide power spectrum, belong to the first group. The items of the
The transfer characteristic of the typical smoke detector cir-
other group are slowly varying parasitic influences like an active
cuit is approximately linear, except for the output comparator
(which is omitted in some realizations). Hence, assuming the
Manuscript received March 15, 1997; revised November 12, 1998. absorption index at the input, and the electrical signal at the
The author is with the Department of Electronics, Faculty of Electrical Engi-
neering, University of Belgrade, Belgrade, Yugoslavia. output of this linear detector part, the transfer function can be de-
Publisher Item Identifier S 0018-9456(00)02857-6. rived for their ratio. The magnitude characteristic of this transfer
0018–9456/00$10.00 © 2000 IEEE
38 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 49, NO. 1, FEBRUARY 2000

(a)

(b)

Ms
Fig. 1. (a) Generalized block scheme of the optical smoke detector; ( ) is the Laplace transform of the absorption index and T (s) is the transfer function of
the whole circuit. (b) Combined principal scheme of the optical chamber for transmission and reflective type detector.

function determines the detector smoke sensitivity frequency the optical chamber for both transmission and the reflective type
response, i.e., the sensitivity threshold has the reciprocal fre- detector is shown in Fig. 1(b). If is the optical flux intensity
quency dependence. The passband gain depends on the desired at the output of the collimated light beam source, and is the
absorption index threshold and on the output threshold level, optical flux intensity at the light receiver input of the transmis-
i.e., the comparator reference signal value. The evaluation and sion type smoke detector, in the absence of the smoke .
optimization of the transfer function of the detector (assuming The chamber static characteristic is a function of the smoke
its linear part) are discussed in the following sections. absorption index

II. GENERALIZED BLOCK SCHEME AND THE TRANSFER


FUNCTION OF THE OPTICAL SMOKE DETECTOR
(1)
Essential parts of the detector generalized block scheme are
a light source, optical chamber with smoke entrance capability where is the length of the light beam where the smoke enters.
and a light receiver, in a cascade connection. Besides these op- If is the optical flux intensity at the light receiver input of
toelectronic blocks, it is necessary to include a filter for the the reflective type smoke detector, in the absence of the smoke
transfer function frequency shaping. The filter can be connected , where is the constant
serially after the light receiver block, and/or between the light depending on the light receiver position in the chamber of the
receiver output and the light source control input, to form the reflective type detector. In this case the chamber static charac-
feedback loop. The later topology has not been considered yet in teristic is
the accessible literature. Transfer functions of these filter blocks
are designated as and , respectively. Generalized
block scheme consisting of the aforementioned blocks is shown
(2)
in Fig. 1(a).
The transfer functions (gain constants) of the light source From (1) and (2), the expressions can be evaluated for and
and light receiver are determined by the slope of their corre- , respectively, as functions of and , and then, in
sponding static characteristics in the quiescent operating points. the usual way [15], the Laplace transform of the light receiver
Transfer functions and of the two-input block repre- input signal. For both cases it can be written in the form
senting optical characteristics of the chamber where the smoke
enters are evaluated in a similar way. The principal scheme of (3)
ALEKSIĆ: MINIMIZATION OF OPTICAL SMOKE DETECTOR FALSE ALARM PROBABILITY 39

where , and are transforms of the cor- For the second order bandpass function with one finite zero
responding signals. Of course, the expressions for the optical it can be shown that
chamber block transfer functions (gain constants) and
for the transmission type detector differ from the respective ex-
pressions for the reflective type. (8)
With block transfer functions evaluated as above, overall
transfer function, , is given as
where is the pole product of the pole pair having the quality
factor .
(4) For the fourth order function with a pair of double poles and
two finite real zeros, the integration yields
For the transmission type detector it is

(9)
(5)
and for the reflective type
where
product of the poles;
quality factor;
and transfer function zeros.
If the pole pairs are not equal, the expression is more complex,
(6) and it can be written in the form

Assuming in (5) that has positive value, product


has to be negative to give a negative value for the
loop gain constant.
The factor in the parentheses in (6) is approximately zero, (10)
because for the reflective type detectors it is usually
. Thus, for this kind of detectors the loop
gain is zero, and therefore where and are pole products of the corresponding pole
pairs, and and are their quality factors. Coefficients ,
, , and , as functions of , , , , , and are
(7)

(11)
The above expressions determine overall transfer functions,
i.e., frequency characteristics of the detector sensitivity of the
analyzed circuits. where

III. TRANSFER FUNCTION OPTIMIZATION


A. Optimization Criterion
To optimize a transfer function of the detector, the statistics
of the input interferences have to be known. Generally, those
data are not available, since they depend on the surrounding
conditions in the operating ambient. However, it should be kept
(12)
in mind that the useful signal spectrum is in the extremely low
frequency range, so that every interference can be considered
as a Dirac pulse, when compared to the slowly varying input
signal. Hence, it is reasonable, and easier, to assume that the
power spectrum of the stochastic input interferences is constant
over the frequency range of interest. This reasoning relates to the
equivalent input generators representing electrical interferences
from power lines, too.
With constant power spectrum at the input, minimization
of the probability that stochastic output process attains the
(13)
threshold level is reduced to the minimization of the integral
of the squared amplitude characteristic, i.e., “noise equivalent (14)
bandwidth” (NEB) [16]. (15)
40 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 49, NO. 1, FEBRUARY 2000

where

(16)

(17)
(18)

The detector is activated if the output signal of the linear part


Fig. 2. Minimal NEB as a function of the zero for a second order transfer
rises over the comparator reference value. Besides the input function of the detector, for three different values of the dc gain, T (0). Time
signal level, i.e., smoke density, its rate of change affects the response constraints are not fulfilled above the dashed line. The data above
output signal, too. Transfer function coefficients have to be de- crosses designate respective values of T (0).
termined so that the detector threshold level is within the pre-
scribed limits for all realistic rates of changes of the absorption analog circuits are of the order of 10 rad/s. Due to this reason,
index. In other words, the following conditions have to be sat- in the optimization presented, the zero values have been fixed,
isfied: too.
1) When input signal, rising at any realistic rate of change, Obviously, the appropriate value of the passband gain is pro-
reaches the upper sensitivity threshold limit, the response portional to the comparator reference signal value. According to
at the output of the linear part has to exceed the com- (8)–(10), the value for NEB is proportional to the squared value
parator reference value. of this signal. To normalize the transfer function, the comparator
2) The response at the output of the linear part has to be reference signal was accepted to be unity. In this analysis it was
lower than the comparator reference value, if the input not necessary to specify the nature of the output signal, so
signal, rising at the maximal realistic rate of change, does has been left without dimension, that would obviously be either
not surpass the lower sensitivity threshold limit. V/(dB/m) or A/(dB/m). Using the criterion described, the opti-
The lower limit of the input threshold is taken to be mization was performed and the minimal NEB was calculated
dB/m in the present numerical calculations. For the upper by the implementation of the SIMPLEX algorithm [17].
limit 0.5 dB/m is selected, which corresponds to the class A
detectors in European standards [2]. B. Optimization Results
Based upon the data presented in the literature [3], [5], [8] To attain the constrained minimum of the NEB for the second
0.5/10 dB/m/s is taken as a maximal absorption index rate of order transfer function, and have been optimized, with the
change, and 0.5/3600 dB/m/s as a minimal. According to the assigned values for and . The selected values of
criterion adopted, condition 1 has to be fulfilled for these two were 0.01, 0.1, and 1. Resultant NEB as a function of the zero
extreme values of the absorption index rate of change. is shown in Fig. 2. It can be noticed that the minimal value for
Concerning condition 2, the step function is the worst pos- NEB is 7.79 rad/s for , irrespective of the selected
sible shape of the input signal, due to its infinite rate of change. value. At low values of the zero, the curve is horizontal up to its
Therefore, with the amplitude equal to the lower limit of the knee, and then rises abruptly, reaching soon the border (dashed
input threshold, it is accepted for the calculations presented. line) above which the time response constraints are not fulfilled.
NEB minimization under the time domain constraints Optimization is performed at the dots marked on the curves.
described above does not produce adequate attenuation of the In the optimization process, as mentioned above, the zeros are
slowly varying influences. In fact, because the useful frequency made to move toward the origin, if they are not fixed. Hence,
bandwidth is relatively wide, the value of the amplitude charac- in the case of the fourth order transfer function with two real
teristic below the passband is of negligible influence on NEB. zeros, it has been rational to optimize only the function with
Attempting to fulfill the criterion and time domain constraints, a double zero ( ). The optimization is performed for
the optimization algorithm rises the low frequency gain to an and . From the curves shown
unfavorable level. Thus, it is necessary to include an additional in Fig. 3, it can be seen that the increased order of the function
constraint for the attenuation of slowly varying influences. The does not reduce minimal NEB, that is still equal to 7.79 rad/s.
simplest way to do it is directly by accepting the value for In fact, the poles of the optimized transfer function are posi-
, as it has been done in this paper. tioned so that two of them (with the zeros) determine the selec-
With the stated constraints the optimization algorithm tends tivity at low frequencies, and other two the slope at high fre-
to move the transfer function zeros toward the origin, to sat- quencies. However, high frequency poles are real and mutually
isfy the condition. The resultant zero values can be unac- distant. Hence, the slope of the amplitude characteristic above
ceptably low for analog realizations. Because of the capacitor passband edge up to the undominant pole is the same as in the
time constants, the minimal values of the critical frequencies in previous (second order) case, resulting in the same NEB value.
ALEKSIĆ: MINIMIZATION OF OPTICAL SMOKE DETECTOR FALSE ALARM PROBABILITY 41

Fig. 3. Minimal NEB as a function of the zeros for the fourth order transfer
function of the detector, for five different values of the dc gain T (0). Fig. 4. Time responses for the fourth order transfer function of the detector;
T (0) = 0:1, s = s = 10 rad/s. Input signals for the responses shown
are: (a) linear function (0:5=10) 1 t 1 h(t) dB/m, (b) (0:5=3600) 1 t 1 h(t) dB/m,
Obviously, this result is a consequence of the time domain con- and (c) step function 0:05 1 h(t) dB/m.
straints in the optimization criterion, not allowing higher selec-
tivity above the passband.
paper presents a generalized block scheme and transfer func-
However, a comparison of the curves for the same value
tion derivation. After the detailed consideration of the detector
shows that the knee of the higher order function occurs at much
characteristics and behavior, the criterion for the false alarm
greater zero values. In other words, for the same zero value and
probability minimization is formulated, based upon the transfer
the same minimal NEB, a higher order function allows for much
function derived and the time domain performances prescribed
greater dc attenuation. Hence, it is especially attractive for the
by the standards. Using the established criterion, the parame-
design of the transmission type detectors. As a matter of fact,
ters of the second and the fourth order transfer functions are
due to the undominant high frequency pole of the fourth order
optimized. Optimization results disclose that the circuit selec-
case, it could be presumed that the third order function would
tivity enhancement at frequencies above the passband in order
provide the same results.
to lower the false alarm probability, as used by some manu-
Although the curves in Fig. 3 are similar to those in Fig. 2,
facturers, is unreasonable. Due to the specific time domain be-
they do not reach the border line. There is a pole pair at low
havior requirements, the increased selectivity above the pass-
frequencies in this case, and the rise of the NEB above the knee
band enhances the edge frequency and the NEB and, hence, the
of the curves is associated with the abrupt increase of the pole
false alarm probability. On the other hand, rising the order of the
factor, in an attempt to maintain the given value for and
transfer function allows for the higher attenuation of the slowly
the necessary passband gain.
varying parasitic influences. Optimization gave the same min-
Selecting the optimized function corresponding to a point on
imal NEB value of 7.79 rad/s for both considered functions, of
the horizontal part of the curve below the knee, introduces wider
different orders, normalized to have threshold levels 0.5 dB/m
margins in some of the time domain responses, with respect
and one at the input and output, respectively.
to the corresponding constraints. This can be seen in Fig. 4,
which presents all three time responses included in the opti- The optimization results presented in this paper can be easily
mization constraints, for the optimal fourth order function with implemented in the smoke detector design to yield the minimal
rad/s. Curve b, corresponding to susceptibility to false alarms. The initial of the design steps re-
the lowest absorption index rate of change, at 3600 s surpasses lates to the decision on the detector type, based upon the overall
the unity output threshold value for more than 50%. Similarly, design requirements. Depending on the type selected, attenu-
curve c, for the step function dB/m at the input, has ation below the passband and, consequently, the order of the
maximal value of 0.17, which is far less than the unity output overall circuit transfer function have to be determined. Based
threshold level. on the general block scheme and the transfer function derived
It has to be pointed out that the time domain constraints have in this paper, transfer function parameter values have to be de-
been formulated to include limit situations which are not very termined, with the help of Figs. 2 or 3. The calculation of the
likely to occur. This has been done to preserve a correct detector element values for the filter and other blocks is then easily per-
operation in all circumstances. In a typical situation [18] the formed, according to the particular block realizations.
time response margin is even wider, i.e., the output alarm level The presented results can be easily incorporated in the ex-
is reached at even smaller absorption index level than in Fig. 4 isting designs. In some realizations, the redesign reduces to the
(curves a and b). changes of the values of the time constants, to adjust the transfer
function according to the optimization results. Depending on the
particular circuit, addition of the dc blocking stage could be nec-
IV. CONCLUSION
essary.
Optical smoke detectors are basically linear circuits, except Detector performances and the stability of its characteristics
for the output comparator. For the linear part of the circuit the are strongly affected by the particular block realization. The de-
42 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 49, NO. 1, FEBRUARY 2000

signs with the improved resistance to the characteristics deterio- [11] M. Thuillard, “The development of algorithms for a smoke detector with
ration due to the parasitic influences (especially to the tempera- neuro-fuzzy logic,” Fuzzy Sets and Syst., Jan. 29, 1996.
[12] H. C. Muller and A. Fischer, “A robust fire detection algorithm for tem-
ture and the optical component contamination) deserve maximal perature and optical smoke density using fuzzy logic,” in Proc. IEEE
attention, too. 29th Annu. Int. Carnahan Conf. Security Technology, Sanderstead, U.K.,
Oct. 18–20, 1995.
[13] T. Jandeweith, “The use of an expert system to expand a multi sensor
ACKNOWLEDGMENT fire detector to a robust fire detector,” in Proc.IEEE 29th Annu. Int. Car-
nahan Conf. Security Technology, Sanderstead, U.K., Oct. 18–20, 1995.
The author would like to thank Prof. D. Rabrenović for his [14] P. Ryser and G. Pfister, “Optical fire and security technology: Sensor
extensive help and encouragement throughout the research pre- principles and detection intelligence ,” in TRANSDUCERS ’91 Int.
sented. Conf. Solid-State Sensors and Actuators, San Francisco, CA, USA,
June 24—27, 1991.
[15] J. Cadzow and H. Landingham, Signals, Systems, and Trans-
REFERENCES forms. Englewood Cliffs, NJ: Prentice-Hall, 1985.
[16] A. Papoulis, Probability, Random Variables, and Stochastic Pro-
[1] EN 54, “Components of automatic fire detection systems—Part 7: Point
type smoke detectors; Detectors using scattered light, transmitted light cesses. Tokyo: McGraw-Hill Kogakusha Ltd., 1965.
[17] J. Nelder and R. Mead, “A simplex method for function minimization,”
or ionization,” in CEN, Brussels, Belgium, July 1982.
[2] EN 54, “Components of automatic fire detection systems—Part 9: Fire Computer J., vol. 7, no. 3, 1965.
[18] R. Dobriner, “Smoke detectors coming on strong—Look for ICs,” Elec-
sensitivity test,” in CEN, Brussels, Belgium, July 1982.
[3] Principles of the accomplishing of the automatic fire signaling devices tron. Design, no. 22, Oct. 25, 1977.
and systems, Stroiizdat, Moscow, 1983.
[4] ANSI/UL 217, “Single and multiple station smoke detectors,” Under-
writers Laboratories Inc., Oct. 4, 1978.
[5] ANSI/NFPA 72E, “Automatic fire detectors,” NFPA, Quincy, USA,
June 14, 1984. Željko J. Aleksić was born in Yugoslavia in 1950. He
[6] H. Tanaka, “Optical smoke detector,” U.S. Patent 4 199,755,, Apr. 22, received the Eng. and M.S. degrees in electrical en-
1980. gineering from the Faculty of Electrical Engineering,
[7] H. Tagashira, “Optical smoke detector,” U.S. Patent 4 258 360,, Mar. 24, Belgrade, Yugoslavia, in 1974 and 1996, respectively.
1981. Since 1975, he has been working in the Depart-
[8] MS 9, “Automatic fire detectors,” Cerberus Ltd, Mannedorf, Sept. 83. ment of Electronics of the Faculty of Electrical Engi-
[9] J. Muggli, H. Guttinger, and Z. Horvath, “Optical smoke detector with neering, Belgrade. His research interests include low-
contamination detection circuitry,” U.S. Patent 4 555 634, Nov. 26, 1985. power detectors, embedded control, and test and mea-
[10] Y. Yamauchi and H. Homma, “Light decrease type smoke sensor,” U.S. surement instrumentation.
Patent 4 203 100, May 13, 1980.

You might also like