IEEE TRANSACTIONS ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, VOL. ASSP-35, NO.

8, AUGUST 1987 1223

The compromise for achieving thisefficiency is the dependency Taking the products of (15a)’ and (15b)’, and (15c)’ and (15d)’,
of the programs length on the length of the transform, i.e., for we get
longer transforms longer programs are generated.
plexity order of the program length is
But the com-
N loglog N,and the propor- (ai + u.:
- U: - 1 ) - ~ ( u O U I - ~ 2 < ) 0 (16)’
tionality constant is small, resulting in the program lengths shown and
in TableIV. These results areof real practical significance and can
be immediately applied to create digital signal processing software (ai U : - + - 1) +
~ ( u O U-~~ 2 < ) 0, (17)’
of very high time efficiency. Since the cost of memory chips is respectively. The condition (14) of the correspondence follows if
going down rapidly, the concept of variable lengthof programs will we multiply (16)’ and (17)’. Therefore, the multipliersof the pro-
become important for efficiency even if they occupy more space posed structure are also real when Z;’ is taken as the delay oper-
than programs of fixed size do. ator.
A program generation technique for the radix-2 FFT that gen-
erates a program of length 0( N ) with no array references is re- REFERENCES
ported in Takaoka [SI. D. M. Goodman, “A design technique for circularly symmetric low-
ACKNOWLEDGMENT pass filters,” IEEE Trans. Acoust., Speech, Signal Processing, vol.
ASSP-26, pp. 290-304, Aug. 1978.
The authors would like to thank an anonymous referee for val-
uable comments.
REFERENCES
I. J. Good,“TheinteractionalgorithmandpracticalFourieranaly-
sis,’’ J . Roy. Statist. SOC.,ser. B, vol. 20, pp. 361-372, 1958.
C. S. Burrus and P. W. Eschenbacher, “In-place in-order prime factor Passive Source Localization Employing Intersecting
FFT algorithm,” IEEE Trans. Acoust., Speech, Signal Processing, vol.
ASSP-29, pp. 806-817, Aug. 1981. Spherical Surfaces from Time-of-Arrival Differences
H. W. Johnson and C. S. Burrus, “On the structure of efficient DFT
algorithms,” IEEETrans.Acoust.,Speech, Signal Processing, vol. H. C. SCHAU AND A.Z.ROBINSON
ASSP-33, pp. 248-254, Feb. 1985.
H. W. Johnson, “The design of DFT algorithms,” Ph.D. dissertation,
Dep. Elec. Eng., Rice Univ., Houston, TX, Apr. 1982. Abstruct-Problems associated with the use of intersecting hyper-
L. R. Morris, “Automatic generation of time efficient signal process- boloids for passive source localization from time-of-arrival difference
ing software,” IEEE Trans. Acoust., Speech, Signal Processing, vol. signals are discussed. A closed-form solution for source location is pre-
ASSP-25, pp. 74-79, Feb. 1977. sented given time-of-arrival difference measurements when the dis-
B . J. McKenzie and T. Takaoka, “A control structure for a variable tance from the source to any arbitrary reference is unknown.
number of nested loops,” The Computer Journal, vol. 26, no. 3, pp.
282-283,1983.
G. H. Hardy and E. M. Wright, An Introduction to the Theory ofNum- I. INTRODUCTION
bers. London,England:OxfordUniversityPress,1954.
T. Takaoka, “A program generation technique for the radix-2 FFT,” The problemof passive source localization using time-of-arrival
Dep. Inform. Sci., Ibaraki Univ., Tech. Rep., 1986. difference signals from multiple sensors is an active research area
in acoustics and radar. Application techniques include the use of
sensors along a line or surface array where beamforming and wave-
front curvature ranging may be employed [1]-[4], and the use of
single sensors where interference effects such as Lloyd’s mirror
may be employed [SI. In many applications, the problem may be
A Note on “Realization of First-OrderTwo- made more tractableby reduction to two dimensions, e.g., azimuth
Dimensional All-Pass Digital Filters” and range or range and depth [11-[5].
The most common approach for passive source localization is to
M. SUDHAKARAREDDY, S . C. DUTTAROY, use time delays between pairs of sensors to define curves of con-
AND S. N. HAZRA stant time difference which are hyperboloids of revolution (of two
sheets). Since it is known thata minimum of four sensors areusu-
In the above correspondence,’ the stability constraints (15), used ally required to locate a source, the intersection of three sets of
to satisfy thecondition(14),assume Zi,i = 1, 2, as the delay hyperboloids defines the source location. This solution is typically
operators 111. With the commonly used symbol Z;‘ for the delay found numerically since closed-formed solutions to simultaneous
operator, the stability constraints (15) of the paper become hyperbolic algebraic equations are difficult to find. In many appli-
cations the problem is constrained to a plane where the intersection
ao-al-a~+l>O (15a)’ of two hyperboloids in two variables is sufficient [3], [5]. In some
ao-a~+a2-1<0 (15b)‘ applications, the source can be assumed to lie ata distance which
is great compared to the sensor spacingso that the directionof the
ao+a~-U2-1<0 (1%)’ source is given by the hyperboloid asymptote [l], [2]. In two di-
mensions, this reduces to the intersection of two lines; whereas in
a,+al+a2+1>0. (15d)’ three dimensions, the source is found by the intersection of three
sets of cones (asymptote lines rotated about the axis of each pair
of sensors). The linearized case is not computationally difficult,
Manuscript received December 30, 1986; revised February 16, 1987.
The authors are with the Department of Electrical Engineering, Indian
Institute of Technology, New Delhi 110016, India. Manuscript received October 22, 1985; revised January 29, 1987.
IEEE Log Number 8715125. The authors are with the Naval Research Laboratory, Underwater Sound
‘M. S. Reddy, S. C. Dutta Roy, and S . N. Hazra, IEEE Trans. Acoust., Reference Detachment, Orlando, FL 328564337,
Speech, Signal Processing, vol. ASSP-34, pp. 1011-1013, Aug. 1986. IEEE Log Number 8715128.

0096-3518/87/0800-1223$01.00 O 1987 IEEE

1224 IEEE TRANSACTIONS
ACOUSTICS, ON SPEECH, AND SIGNAL PROCESSING, VOL.ASSP-35, NO. 8, AUGUST 1987

thereby reducing the numerical complexity of the solution consid- source squared, R,‘, yields
erably.
This correspondence is concerned with the general problem of
R,’ = x: + y : + Z;
passive localization o f an acoustic source where the application is = -(X: + y: + z:) + 0: + 2 x 4 , + 2y4Ys + 2242,
three dimensional by nature; i.e., two-dimensional simplifications
may not be appropriate. = -R: + 0: + 2x4Xs + 2y4Ys + 2242,. (6)
One of the basic problems with numerically, solving the simul- Substituting into the first three equations in (4) yields
taneous hyperbolic equations is the hyperboloid itself. The point
of intersection of two hyperboloids can move considerably for a R; - (di4 + D4)’ + 0: - R: = 2xjx, + 2yiy, + 2ziz, - 2x4x,
relatively small change in eccentricity of one of the hyperboloids. - 2y4Ys - 2Z4Zs, i = 1 , 3 .
This is not true for intersecting spheres when the radius of one of
the spheres is changed. This is due to the fact that hyperboloids (7)
definesurfaces of constantdistance difference betweenpairs of
Without loss of generality, we may take sensor 4 to be the origin.
sensors, while spheres define surfaces of constant distance from a
This makes (x4, y4, 2 4 ) = (0, 0 , 0 ) , R4 = 0. Substituting into (7)
single sensor. Measurement errors can, in some cases, make the
yields
existence ofa solution using intersecting hyperboloids difficult.
With this in mind. and because of the numerical difficulties asso-
ciated withintersectinghyperboloids ina general three-dimen-
sional geometry, we attempt to recast the problem into one which
employs spheres.
11. THEORY-ALTERNATE DERIVATION EMPLOYING SPHERICAL (8)
SURFACES after recalling that now R,’ = 02 from (6). Symbolically, (8) can
For four sensors, one can define thedifference in distance from be written
source to sensor i and source to sensor j , d, in terms of the (un-
A - 2Rsd = 2 3 n x (9)
known) absolute distancesDi from the source to each of the sensors
as follows:
d..
V
= D. - D. i, j = 1, 4

du
l

=
J

d.. = - d .
ti - t j ) = cAt,
where
R: - d:4

(1)
A =‘ [ R;
R; - dd;4
i]’

In (9) the unknown appears in

X I Y1 ZI

d = [I:]’ =[:I’
d24 x

both sides of the equation, the source

zd‘
where data must have the self-consistency shown below, c is the
speed of sound, and At is the time-of-arrival (TOA) difference. position x on the right-hand side, and source radiusR, on the left-
hand side. If one assumes for the moment that the source radius R,
d12 = d13 - d23 = d14 - d24 were known, (9) has the solution
dl3dl2 + di3 = dl4 - d34 X = +3n-’(A - 2R,d) (10)
d23 = d13 - dl2 = d24 - d34 provided the matrix of sensor positions 317. is not singular (sensors
d14 = d12 + d24 = + d34 do not provide redundant information). Solution (10) still involves
the as-yet unknown source radiusR,. To solve this problem, recall
d24 = dl4 - dl2 = d23 + dg R, = (xT.) 1 /2 . (11)
d34 = dl4 - dl3 = d24 - d23. (2)
Substituting (10) into (1 1) yields after expansion
Consider the definition of the distance from source to sensor;
this yields four equations R:[4 - 4 d T ( 3 n - 1 ) T 3 n - 1 d +
] R,[2dT(3n-1)T317.-1A

(x; - x,)’ + (yi - y,? + (zi- z,f = D;, i = 1, 4 (3) + 2AT(3n-’)*%-Id] - [AT(3 n - I ) ‘317.-’A] = 0. ( 12)

where ( x I , y zll , ) is the position of sensor 1 (similarly for sensors This equation involvesonly the unknown sourceradius R,, all other
2 and 3), (x,, y,, z,) is the source position, D l is the distance of quantities are known from thesensor locations (m),measured time
serrsor 1 from the source(similarly for sensors 2 and 3). Expanding delay differences ( d ) or both, ( A ) . The quantities in brackets are
(3) and substituting the values for D l , D2. D3 given by (1) yields seen to represent scalar results of inner products of the individual

x’ - 2xix, + x: + y’ - 2yjy, + ys” + Z’ +

- 2 2 ; ~ ~zf = (did + D 4 )2 , i = 1, 3
X i - 2X4Xs + X 3 + y: - 2y4ys + y: + - 2z4zs + zf = 02. (4)
In the above set, the direct solution to equation set ( 1 ) hasbeen vectors and matrices. Equation (12) has solution
employed, i.e., Dl = d I 4 D4, etc.+
alternative
some
cases,
an
In least-squares solution to (1) may -b G c
be employed more effectively (depending on the specific applica- R, = ; R, 2 0, (13)
2a
tion)
where
Dl = (dl2+ dl3 + 2d14 + 4 4 + d34)/4 + 0 4 , a = [ 4 - 4dr(3n-’)*3n-’d],
4 = (-dl2 + d23 + dl4 + 2d24 + & ) / 4 + 0 4 , +
b = [ 2 d T ( 3 n - 1 ) T 3 n - ’ A 2AT(9l-’)‘317.-Id],
0 3 = (-dl3 - d23 + dl4 + d24 + 2d34)/4 + 0 4 . (5) c = - [AT(3n-1)T9T-1A].
Solving the fourth equation in ( 4 ) for the distance from origin to Thus, (10) and (13) representa two-step solution to the passive
IEEE TRANSACTIONS ON ACOUSTICS, SPEECH, SIGNAL
AND PROCESSING, VOL. ASSP-35, NO. 8, AUGUST 1987 1225

source localization problem. Equation (13) is first solved to find the discussion presented. Each of the subjects is application de-
the source radius, this radius is substituted into (10) to find source pendent, particularly the N sensor case ( N > 4) which requires a
position. In some instances, (13) will permit two physical solutions pseudoinverse in (10). Since detailed analysis for this case can only
( R , 2 0). In these cases, two source locations will be found whichbe carried forth for a specific sensor number and position, its de-
will reproduce the measured data. In these situations, the two lo- velopment is more appropriately considered for each application
cations are usually far enough apart that the correct solution be can separately.
discerned by other physical reasoning such as one solution lying
outside the domain of interest. REFERENCES
111. CONCLUSION [l] G. C. Carter,“Timedelayestimation for passivesonarsignalpro-
cessing,” IEEE Trans. Acoust., Speech,Signal Processing, vol. ASSP-
The difficulties of passive localization of sound sources through 29, pp. 463-469, 1981.
intersecting hyperboloids has been discussed.To circumvent these [2] A. G. Piersol, “Time delay estimation using phase data,” IEEE Trans.
difficulties, an alternate method which employs intersecting spheres Acoust., Speech, Signal Processing, vol. ASSP-29, pp. 411-477, 1981.
has been presented. This formulation of the problem permits a sim- [3] T. N. Claytor, D. A. Greene, R. L. Randall, T. D. Ohe, and D. Ga-
ple closed-form solution for source location from the time’delay warecki, “Development of a passive acoustic imaging system,” Ar-
estimates and sensor locations. gonne Nat. Lab., ANL-83-102, Nov. 1983.
The specific solution presented herein is for the case of fourre- [4] L. Kastic, “Local steam transit time estimation in a boiling water re-
actor,” IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-
ceiving sensors. Implicit in the developmentis the assumption that 29, pp. 555-560, 1981.
a mechanism for measuring TOA differences exists. More general [5] R. F. Hudson, “A horizontal range vs. depth solution of sound source
aspects of this problem, such as methodology for measuring TOA position under general sound velocity conditions using the Lloyd’s mir-
differences, multiple sources, and situations where.more than four ror interferencepattern,” M.S. thesis,NavalPostgraduateSchool,
receiving sensors are employed, are capable of development from 1983.