IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 40, NO.

5, MAY 1992

917

Capture Models for Mobile Packet Radio Networks

Abstruct- The probability qi of successful reception in a nonfading mobile radio channel with i contending mobiles transmitting to a central base station is studied for a number of different capture and spatial distribution models. It is shown that a generalized capture model can be used to estimate qi for a simplified example system which uses noncoherent frequency shift keying modulation. This model can be applied to other systems as well. An example of the use of the qis in the throughput evaluation of a finite population slotted ALOHA system is given. In most practical systems, the mobiles cannot get arbitrarily close to the base station. The effect of this constraint on qi is examined. Finally, the dependence of the capture probability for a test mobile on its distance from the base station is obtained.

11. PRELIMINARIES

In this section, a number of considerations which will be necessary in the ensuing analysis are discussed.
A. Propagation Model

In this study, we will use a simple model of flat terrestrial propagation [21], in which the normalized received signal power r can be written as
1 qT) =r1

I. INTRODUCTION

where T is the distance between the transmitter and the receiver. It is assumed that the system is contention-limited.
B. Capture Effect

HIS paper addresses the problem of contention in a communication system in which a number of mobile users transmit over a common channel to a central base station. A data packet collision occurs whenever two or more users transmit at about the same time. In classical analyses of random access protocols, it is assumed that all the packets involved in a collision are destroyed [1]-[6]. This is reasonable when the packets are received with nearly equal powers. However, the assumption is somewhat pessimistic in a mobile radio environment. Typically the signal received at the base station from a given terminal will be subject to attenuation (a function of the base-to-terminal distance) and possibly fading. When packets from different terminals collide, it may still be possible to successfully decode the packet with the strongest received signal strength [7]-[20]. We will examine the probability q; of successful packet reception (capture) when i users transmit simultaneously for a number of different captureispatial distribution models described in Section 11. The evaluation of q; is discussed in Section 111. This is followed by the application of a certain capture model to predict qi in a system using noncoherent frequency shift keying modulation. The effect of using spatial distributions in which the mobiles cannot be arbitrarily close to the base station is considered. An example of the use of the results in evaluating the throughput of a finite population slotted ALOHA system is given. The probability that a packet transmitted by a test mobile at distance rt from the base station will capture the receiver is evaluated in Section VII.
Paper approved by the Editor for Mobile Communications of the IEEE Communications Society. Manuscript received August 24, 1989; revised December 1, 1989. This work was supported in part by NSERC under Grant OGP0001731. This paper was presented in part at the 1990 IEEE International Conference on Communications, Atlanta, CA. The authors are with the Department of Electrical Engineering, University of British Columbia, Vancouver, BC V6T 124, Canada. IEEE Log Number 9108002.

Since the mobile terminals will generally be at different distances from the base station, their respective received signal powers will not be the same. In the event of a collision between the data packets, it may still be possible for the receiver to successfully decode the packet with the highest signal strength. It is assumed that the received signal powers are more or less constant over a packet duration. The performance improvement in various systems which results from the ability of the receiver to capture a signal has been studied in [7], [9]-[20]. Two capture conditions for a packet from mobile t , t E { 1 , 2 , . . . , i } , have been commonly assumed, namely,

rt > cr,.
and

j = 1 . 2. . . . ,2 . j

#t

(2)

rt > c

2 r,
J=1
J#t

(3)

where I is the number of on-air packets, rk, k = 1.2, . . . I is the received power of mobile k , and c is the capture ratio. The value of c will depend on the modulation and coding schemes used. Condition (2) results in a somewhat optimistic model. However, its use [7], [9], [12], [14] often leads to simpler analytic derivations. Condition (3) has been used in [lo], [11J, ~31. We also introduce a new capture condition in which successful decoding of a test packet is a probabilistic function of the test signal-to-interference ratio 2 defined by

( 4 )
where 2 r J .For a given value of 2,the probability that a test packet from mobile t is successfully

rrL

zg=l,Jzt

0090-6778/92$03.00 0 1992 IEEE

91x

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 40, NO. 5 , MAY l Y Y 2

decoded is

A . Capture Model I

z<l

(5)

We first derive a general expression for qL for capture model 1. Here, it is possible to consider a more general received power to distance relationship of the form

where y(z) is a function which depends on the type of modulation and coding used. It might be noted that the use of spread-spectrum systems could allow successful decoding even for values of 2 less than unity. It can be seen that condition (3) is a special case of (5) with P ( S l z ) = u ( z - c) where U(.) is the unit step function. For convenience, we will refer to condition (2) as capture model 1. The conditions implied by ( 5 ) and (3) will be termed capture models 2 and 2.1, respectively. Model 2 will be used to examine the sensitivity of q2 to deviations from model 2.1.
C. Spatial Distribution of Mobiles

[ j can vary between 2 and

is 4. In practice, the power loss factor 5 depending on the terrain. Note from (10) and (2) that a test packet from mobile t is assumed to be successfully received if and only if In (l), the value of

Rj > QRt.
where
Q

1,2,...,*i, j

#t

(11)

c1l3. Thus, we have

qt = 2

1
00

f R t (7.t)

The probability of capture q2 will also depend on the spatial distribution of the mobiles around the base station. It is assumed that the locations of the mobiles are independent random variables. We consider two models defined by the following. 1) A uniform spatial traffic density G u ( r )over a unit radius circle centered at the base station, i.e.,

.Pr{R1,R2,...,Rt-1, Rt+l,...,R; > art}drt

=iffRtirt)

[[fR3(rj)drj]

drt.

(12)

GZL(T = )

{ +,
0,

O<r<l otherwise.

We now consider the evaluation of (12) for the uniform and bell-shaped spatial density functions. I ) Uniform Spatial Density: The distances RI R2,. . . , R i are independent, identically distributed random variables with common pdf as in (7). Substituting (7) in (12) yields

In (6), the traffic on the channel has been normalized to unity. The corresponding probability density function (pdf) of the distance R between a mobile and the base station is
fR(r) =

2r. 0 < T < l 0. otherwise.

(7)

2) A bell-shaped spatial traffic density function [ 131centered at the base station, i.e.,

which can be shown to be equal to $- for i = 2 , 3 , 4 , . . . Thus, when there are two or more contending packets, the probability of capture with model 1 is independent of the number of contending packets! 2) Bell-Shaped Spatial D e n s i t y : Here, the common pdf of { R I R2. , . . . ,R,} is given by (9). Using (9) in (12) we obtain

In this case, the corresponding pdf of the distance a mobile and the base station is

R between

These two density functions will be used to illustrate the dependence of q; on the spatial distribution of the mobiles.
111. ANALYSIS OF CAPTURE PROBABILITIES
AND

NUMERICAL RESULTS

In this section, we examine the capture probability qz for the different capture and spatial distribution models discussed in the preceding section. Clearly, qo = 0 and q1 = 1. So in the sequel, the expressions derived for yz will apply for i = 2,3,4;.. .

e - x L dx is the complementary error where erfc(y) 2 function. Equation (14) can be readily evaluated using a numerical integration routine. The truncation error E which results from setting the upper limit of the integral in (14) to b is discussed in Appendix A. It is shown in Appendix B Hence, the asymptotic value of qL that 1imt+= 4%= is the same as with the uniform spatial density model. This asymptotic value increases with [j and decreases with The lower bound derived in Appendix B also shows that the qz for the bell-shaped spatial density model is at least as big as that for the uniform spatial density model.
(8.

&

LAW AND LEUNG: MODELS FOR MOBILE PACKET RADIO NETWORKS

91Y

B. Capture Model 2

Using (20) and (21) in (15) yields

From (4) and (5), we can write

The pdf of 2 is given by

t i ( . )

/
2 . c

By using the fact [23, (2.124.1) that

fz.r,, (2. T 7 L dTIL )

(16)

we can reduce (22) to where f z , r , > ( z . T n )2 jrt,r,?(Yt.Tn) 1-l and J ( $ ) is the Jacobian of the coordinate transformation from 27 a ( & - J C I ) T to <[22]. q,=; si - 1 1) Uniform Spatial Density: The common pdf fr.u,,lform(T) of the received powers {rl, I72,. . . . r,}can be derived from (1) and (7) and is given by

~(2)

- _ _ _ +ne--

( "

(i - 1)

The asymptotic value of qi is given by The evaluation of 4%is complicated by the fact that it is difficult to obtain a closed-form expression for the pdf of [given by the (7 - 1)-fold convolution of fr,uI,lforII1(y)] when L is large. Our approach is to use a Monte-Carlo simulation to obtain values for q2 which will then be compared to the case of the bell spatial density function for which analytic results can be obtained. 2) Bell Spatial Density: Equations (1) and (9) can be used to obtain the common pdf fr,be1l(y) of the received powers {I'1.r2....,r,}. The result is 1 f r bCl1(y) = - ~ - ~ / ~ r - ~ ' 2 ~ 0. 7 , 2

r,
a + x

The expression for qa for capture model 2.1 can be obtained by letting a + 0 0 in (22). In this case, c1 = c 2 = c and the second integral yields
q z , m o d e l 2.1

(18)

= - arctan
7i

2i

I ? ] is given by the ( 7 - 1)-fold The pdf of = convolution of f r h c l l ( ~ )and can be written as [13]

rn

with asymptotic value

2 lim qz.model 2.1 = -

(27)

Substituting (19) into (16), we obtain

C. Numerical Results

We now consider the special case of a piece-wise linear g ( z ) defined by

where e1 = c c2 = c and n 2 We will refer to this case as model 2(PL). The lowest value of a results from the assumption that P ( S l z ) = 0 for z < 1.

&.

i0:
y

+ n(z

z
-

c).

('1

I c1 I z 5 e2

(21)

2 c2

+&

A.

We now compare the q2 values for the capture and spatial distribution models discussed so far. In Fig. 1, q2 is plotted for capture model 1 with = 4, for both the uniform and bell spatial densities and different values of the capture ratio. It can be seen that the bell spatial density model yields a slightly higher probability of successful reception. The asymptotic value of is quite rapidly approached. Fig. 2 is a plot of q1 for capture model 2.1. The curves for the uniform spatial density model represent simulation results with 99% confidence intervals as shown. The bell spatial density curve is obtained from (26). As in the case of capture model 1, the bell spatial density model yields a slightly higher (I, value than the uniform spatial density model. From Fig. 2, the asymptotic 9%values for the two spatial density models appear to be the same, i.e.,

5.

920

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 40, NO. 5, MAY 1992

I ) ,

0.8

.0
a

2 z
0
w ,

0.6 -

e,
0
v?

E .P

..9

0.4 -

0.2

11

o ~ " " " " ' ~ " " " ' " ~ " ' "
10

20

" , I 30

0 1 ' '

'

10

20

30

Number of contending transmitters, i Fig. 1. Probability of successful reception versus the number of contending transmitters I for capture model 1, using = 4.

Number of contending transmitters, i Fig. 3. Probability of successful reception qr versus the number of contending transmitters I for capture model 2(PL) with the bell spatial density model.

1
Bell spatial density

__-

Uniform spatial density - - - - -

a system with a given P ( S l z ) curve. We use as example a simplified model of a system employing noncoherent frequency shift keying (NCFSK) modulation. In addition, the bell spatial density model will be used. For simplicity, we assume that the total interfering signal is Gaussian. This is reasonable when the number of interferers is large but will yield pessimistic results when the number is small [24]. Using this simplified model, the bit error rate (BER) for NCFSK modulation is given by (28) 2 A more accurate BER value can be obtained using the results in [24]. However, our main purpose here is to determine how closely capture model 2(PL) can estimate yz, given a P ( S l z ) curve. We will therefore use the simplified model and assume that bit errors are independent to conveniently obtain P ( S I z ) , the probability that a packet of length L contains no bit errors, as 1 pb = - e-+.

10

20

30

Number of contending transmitters, i Fig. 2. Probability of successful reception ql versus the number of contending transmitters t for capture model 2.1.

Fig. 3 shows the qz values for capture model 2(PL) with the bell spatial density model. Curves for different values of n and c are given. For given values of c and i , qz decreases with U. This follows from the fact that the pdf of the test signalto-interference ratio, f i ( z ) , is a monotonically decreasing function of z . However, there is little difference in qt as the slope a increases from 1 to infinity. This indicates that yz is not very sensitive to the exact shape of the g ( 2 ) curve, especially for large c, since fi(z) is small for large z . The corresponding qL curves for capture model 2(PL) with the uniform spatial density model were also obtained. For small values of L, they are somewhat lower than those in Fig. 3. The differences become negligible for large values of i .

P(S1.) = (1 - p # .

(29)

Substituting (29) and (20) in (15), we can obtain an expression for ql as shown in Appendix C. The resulting 4%expression can be evaluated numerically and is plotted as the solid curves in Fig. 4, for L = 100, 200. and 500. To apply model 2(PL), we first have to select appropriate values of c and a. The value of c is taken to be the value of z at which the P ( S I z )expression in (29) has value 0.5, namely
CUCFSI<

= -2111 2 1 - 0.51/L

[(

11

(30)

The value of U is chosen so that the piecewise linear P ( S Iz) line is tangent to (29) at c = C ~ C F S I < namely, ,

Iv.

PROBABILITY O F SUCCESSFUL

RECEPTION FOR NCFSK SYSTEM

In this section, we examine how capture model 2(PL) can be used to estimate the probability of successful reception for

The qi values using c and a as given by (30) and (31) in (24) are plotted as dashed curves in Fig. 4. It can be seen that the

LAU AND LEUNG. MODELS FOR MOBILE PACKET RADIO NETWORKS

921

Piecewse linear approximation

--0.8

Bell spatial density, model 1 Uniform spalial density. model 1

Numerical inlegration of (50) -~

No capture

___

--LI

m * a c

0.6

8
L = 100
L-200

%
0.4

L = 500

0.2

ooo
10
20 30 0

Number of contending transmitters, Fig, 4.

Probability of successful reception (1, versus the number of contending transmitters I for a NCFSK system.

Fig. 5 . Throughput S versus p<, for capture model 1 with and = 10po.
I ,
Bell spalial density. model 2(PL), a = 1

c'

= 4, AV= 50,
I

results agree closely with the curves obtained from (50). This indicates once again that the exact shape of g ( z ) is of minor importance. Even though we treated the specific example of an NCFSK system, the technique can be readily extended to any system for which the P ( S 1 z ) curve can be derived or measured.
V. THROUGHPUT FOR A SLOTTED ALOHA SYSTEM

Bell spatial density. model 2(PL). a

1- - d -

Bell spatial density. m d e l 2 1 Uniform spatial density. model 2 1

-+-

In this section, we use the approach described in 121, 1111 to evaluate the throughput for a finite population slotted ALOHA system under the different capture and spatial distribution models of Section 111. Consider a slotted ALOHA system with a base station surrounded by N mobiles. At the end of a time slot, a mobile terminal can be in either the origination (0)-state or the retransmission (R)-state. An 0-state terminal transmits a packet in the next slot with probability p,. If the packet transmission is successful, the terminal returns to the 0-state; otherwise, it enters the R-state. In the R-state, the terminal attempts to retransmit the packet in the next slot with probability p , . The terminal stays in the R-state until the packet is successfully transmitted, at which time it returns to the 0-state. It is assumed that acknowledgments from the base station to the mobiles are received perfectly. Such a system can be described by a homogeneous Markov chain with N 1 states corresponding to the number of Rstate mobiles at the end of a time slot. Given p o , p , , and q l , the throughput S,defined as the average number of successful transmissions per time slot, can be evaluated 1111. The results are plotted in Figs. 5 and 6. In both figures, the following parameter values are used: = 50, I' = 4 and p r = lopo. Fig. 5 shows the throughput for capture model 1 as a function of po. It can be seen that the bell spatial density tends to have a higher throughput than the uniform spatial density near the peak of the S os p ( , curve, but the difference is relatively small. For comparison, the throughput for the classical slotted

IO-^

10-2

lo I

Po

Fig. 6. Throughput

S versus
c'

p u for capture models 2.1 and 2(PL) with

= 4. .Y = .XI., and I ) ! . = 10pu.

is also shown. The value of 5 ' for large p , is lowerbounded by q;o = 0.5. Fig. 6 compares the throughput curves for model 2.1 with the bell and uniform spatial densities. As in the case of model I , the bell spatial density gives a slightly higher throughput. Also shown are the curves for model 2(PL) with a = 1 and U = $. As would be expected from Fig. 3, there is very little difference between the curves for model 2.1 and model 2(PL) with n = 1. The value of S for large p , is lowerbounded by yso = 0.32. A comparison of Figs. 5 and 6 shows that systems with capture are not only capable of much higher throughput; the throughput also degrades gracefully under overload conditions.

VI. EFFECTOF PUNCTURlNG THE SPATlAL DENSITY FUNCTION

ALOHA system with no capture (i.e.,

? ( I

{ ;:otherwise )
x = 1.

922

IEEE TRANSACTIONS ON COMMUNICATIONS. VOL. 40, NO. 5. MAY 1992

The pdf of the distance R [see (7)] is changed to

2
(33)
9

p-01

0 2 -

0,

othcrwise.

The punctured bell-shaped density function is obtained by

"

10

20

30

-_I

, 4

Number of contending transmitters, i

Gb(r.p ) =

erfc(

e-'''

1 -

p l r < x 8
p ~ ')

(34)

0.

otherwise

Fig. 7.

versus the number of Probability of successful reception contending transmitters i for capture model 1 with n 2 = 2.

with corresponding distance pdf

' I
Puncturedbell spatial density

__

0.8 -

Punctureduniform spatial density - - - - -

.* 0 a

A. Capture Model 1

E z
c .

:
0.4
-

For the punctured uniform spatial density function, the probability q,, i > 1, is given by

q,

I [l
1/ a
fR.u(Tt)

a-1

.* ."

4
0

f R , u ( T j ) dT.71

drt

(36)

0.2

which, after evaluation, can be written as

As expected, for p = 0, (37) reduces to (13). For the punctured bell-shaped density function, the probability q, is given by
7

B. Capture Model 2.1

?-1

The effect of puncturing for capture model 2.1 is illustrated in Fig. 8. The curves for the uniform spatial density are obtained from Monte-Carlo simulations. The 99% confidence intervals for the simulation values are shown. In the case of the bell-shaped density, following the approach in Section 111-B, we can obtain
-1

30

(38) The expressions in (37) and (38) for p = 0, 0.05, and 0.1 are plotted in Fig. 7. It can be seen that for both spatial distribution models, puncturing reduces the value of 4,. The most noticeable effect is in the asymptotic value lim-x q L . For p = 0, this limit is but for any p > 0 and (Y > 1, the limit is zero. In the latter case, the rate at which the limit is approached is governed by the value of p.

qa = z [erfc(+p2)]

/ e - ~ T ~ / ~ e rG ~ J c () ~ dx.
PL

(39) The effect of a nonzero value of p is similar to that observed for capture model 1. The throughput curves of the slotted ALOHA system described in Section V for p = 0. 0.05 and 0.1 are shown in Fig. 9. For values of y o in the normal operating range, i s . ,

5,

LAU AND LEUNG: MODELS FOR MOBILE PACKET RADIO NETWORKS

923

p=Ol-

-n--O

08

p=OO5 -

p=o

-*

J
lo3
102 PO

10'

05

1.5

Distance, r, Fig. 10. Dependence of probability of capture on distance of test mobile from the base station. Finite population model with -\-= 50, p , = lop,, and I = 4 assumed.

Fig. 9. Throughput S versus p u for capture model 2.1 with punctured = 50, and I),. = lop,,. bell spatial density, c = 4,

below the value at which the throughput is maximized, there is no noticeable difference due to puncturing. For higher values of p,, puncturing can reduce the throughput by a substantial amount.

The P ( C I r t ) curves for capture models 1 and 2.1 are plotted in Fig. 10. The same finite population model described in Section V is used here. The probability P ( n ) is given by

VII. DISTANCE DEPENDENCE OF

CAPTURE PROBABILITIES

The probability of a mobile terminal capturing the base station receiver depends on the distance, rt, between them. In this section, we calculate this probability P ( C I r t ) . For economy of space, only the bell-shaped spatial density (with possible puncturing) case is described below. Let the probability that a test packet originating at distance rt from the base station will capture the receiver in the presence of ri interferers be denoted by P( C I r t : n). For capture model 1, we have
P(CIT,,n ) = Pr(al1 n interferers are at least at distance

art from thc base station)

($ &E)

= [ erfc

(9

pa)

I'

where riTm is the probability that there are ni R-state mobiles in the system. Two sets of curves corresponding to p , = 0.01 and p , = 0.015 are shown. Each set consists of six curves, three for each capture model. The three curves are for p = 0. 0.05, and 0.1. For the p , values shown, the probability of capture does not vary much with p . It can be seen that the probability of capture drops off quite rapidly as the test mobile initially moves away from the base station. Beyond a distance of unity, the curves level off at a value determined mostly by p,. It should be emphasized that the nonzero value of the limiting capture probability is a result of the number of mobiles being held constant (at 50). If the number of mobiles per unit area were fixed and they were spread over an infinite plane, then the capture probability would decrease to zero as the test mobile moved away from the base station.

VIII. CONCLUSION
A number of different capture and spatial distribution models were applied to the problem of a number of mobile users accessing a central base station over a common nonfading channel. Methods for evaluating the probability q2 of capture when there are / contending transmitters were examined. The sensitivity of ql to both the capture model and the spatial distribution model was discussed. It was found that there is little difference between the ql values for a uniform and a bell-shaped spatial density model, with the latter yielding slightly higher values. The asymptotic values of qL for large 1 are the same for both spatial density models. The

For both capture models, we can write

P(GI rt)=
ri

P ( ~ ) P (I rt. c n)

(42)

where P(71,) is the probability of 71 interferers. In (42) it is assumed that the number of interferers is independent of the location of the test mobile. This is reasonable in most systems.

4 24

IEEE TRANSACTIONS ON COMMUNICATIONS. VOL. 40, NO. 5, MAY 1 9 Y 1

TABLE I

Using the fact that crf((y) 5

tjP2Y/fi

for all y, we can write

2 4 6

8 10

0.59033 0.52492 0.51200 0.507 12 0 50474

1.3O.j x 7.473 x 3.210 x 1.22G x 4 3s9 x

IO-' lo-"
l(l-'3'i

10-"
l(I-5-

probability of capture is more sensitive to the capture models considered, namely capture models 1 and 2(PL). However, variants of capture model 2(PL) result in ql values which are not significantly different. The use of capture model 2(PL) in estimating q, for a simplified NCFSK system was described. This model can be applied to other systems in which the P( S 1 z ) curve is available. The capture probabilities were used to calculate the throughput of a finite population slotted ALOHA system. In most practical systems, there is a minimum distance separation between the base station and the mobiles. It was found that such a constraint reduces q, and can have a significant negative effect on the throughput under heavy traffic conditions. Finally, the probability of a test mobile capturing the base station receiver was obtained as a function of their distance separation. APPENDIX A In this Appendix, we discuss an upper bound on the truncation error which results in the numerical integration of (14). The error which results from setting the upper limit of the integral in (14) to b is given by

For large 1 , the upperbound reduces to LowerBound: Using the fact that 0 2 x 2 n. for .r CY 2 1, we obtain
m
1-1

5.

20

and

y =CY2
'

(49)

APPENDIX C expression for q, for the simplified NCFSK system discussed in Section IV is derived in this appendix. Substituting (29) and (20) in (151, we can write

2,i(,i - 1)

+ (1

1)*:r,2

(50)

(45) follows since erfc(y) is a decreasing function of y. To illustrate how effective the bound in (46) is, we consider Table I shows the values of a typical example with (Y = y, and the bound on t for different values of I with b = 2. It can be seen that the truncation error is negligible even for a small value of b.

a.

Upper Bound: Using the fact that (1 can upper bound q, by

+ e-"/')
1

5 1, we

APPENDIX B In this appendix, we obtain upper and lower bounds on q, as given by (14). Upper Bound: Since f - K r 2 1 4 5 1 for .I' 2 0, we have (47)
0

These bounds can be used to evaluate q, accurately by an appropriate choice of b. This is illustrated in Table I1 where b = G and L = 500.

LAU AND LEUNG: MODELS FOR MOBILE PACKET RADIO NETWORKS

925

TABLE I1 BOUNDS ON (1, lower bound on

[ 141 E. Zainal and R. Garcia, The effects of Rayleigh fading on capture phenomenon in ALOHA channels, /E INFOCOM 1987, pp. 888-893,
upper bound on
(1, (52)

(51) 2 4 6 8 10 30

0 35989080 0 24592095 0 22178242 0 21 I34094 0 20551828 0 19140641

0 35989160 0 24592149 0 22 17829 1 0 21134141 0 20551873 0 19 140683

It can

be shown that % is lower and upper-bounded by

%-1 i

5 qi L % - 1SI,

(53)

ACKNOWLEDGMENT The authors wish to thank the reviewers for their helpful comments.

Mar. 1987. [IS] I. Cidon, H. Kodesh, and M. Sidi, Erasure, capture, and random power level selection in multiple-access systems, IEEE Trans. Commun., vol. COM-36, pp. 263-271, Mar. 1988. [ I61 I. M. I. Habbab, M. Kdvehrdd, and C. E. W. Sundberg, ALOHA with capture over slow and fast fading radio channels with coding and diversity, IEEE J . Select. Areas Commun., vol. SAC-7, pp. 79-88, Jan. 1989. [I71 Y. Onozato, J. Liu, and S. Noguchi, Stability of a slotted ALOHA system with capture effect, I Trans. Vehic. Technol., vol. VT-38, pp. 31-35, Feb. 1989. [I81 J.-P. M. G. Linnartz and R. Prasad, Near-far effect on slotted ALOHA channels with shadowing and capture, Proc. 39th IEEE Vehic. Technol. Con& San Francisco, CA, Apr.-1989, pp. 809-813. [ 191 D. F. Lyons and P. Papantoni-Kazakos, A window random access algorithm for environments with capture, IEEE Trans. Commun., vol. COM37, pp. 766-770, July 1989. [20] A. Shwartz and M. Sidi, Erasure, capture, and noise errors in controlled multiple-access networks, I Trans. Commun., vol. COM-37, pp. 1228-1231, Nov. 1989. [21] K. Bullington, Radio propagation for vehicular communications, IE Trans. Vehic. Technol., vol. VT-26, pp. 295-308, Nov. 1977. [22] L. W. Couch 11, Digital and Analog Communicarion Systems, 2nd Edition. New York: MacMillan, 1987. [23] 1. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Prodacts. New York: Academic, 1980. [24] J.S. Bird, Error performance of binary NCFSK in the presence of multiple tone interference and system noise, IEEE Trans. Commun., vol. COM-33, pp. 203-209, Mar. 1985.

REFERENCES
N. Abramson, The ALOHA system-another alternative for computer communications, in Proc. AFIPS Full Joinr Compuf. Conj, Montvale, NJ, vol. 37, 1970, pp. 281-285. L. Kleinrock and S. Lam, Packet switching in a multiaccess broadcast channel: Performance evaluation, I Trans. Commun., vol. COM23, pp. 410-423, Apr. 1975. L. Kleinrock and F.A. Tobagi, Packet switching in radio channels: Part I-Carrier sense multiple-access modes and their throughput-delay characteristics, I Trans. Commun., vol. COM-23, pp. 1400- 1416, Dec. 1975. R. M. Metcalfe and D. R. Boggs, Ethernet: Distributed packet switching for local computer networks, Commun. ACM, vol. 19, pp. 395-404, July 1976. F.A. Tobagi and L. Kleinrock, Packet switching in radio channels: Part 11-The hidden terminal problem in carrier sense multiple-access and the busy-tone solution, ITrans. Commun., vol. COM-23, pp. 1417-1433, Dec. 1975. J. I . Capetanakis, Tree Algorithms for packet broadcast channels, I Trans. Inform. Theory, vol. IT-25, pp. 505-515, Sept. 1979. L.G. Roberts, ALOHA packet system with and without slots and capture, Comput. Commun. Rev., pp. 28-42, Apr. 1975. J. J. Metzner, On improving utilization in ALOHA networks, IEEE Trans. Commun., vol. COM-24, pp. 447-448, Apr. 1976. N. Abramson, The throughput of packet broadcasting channels, I Trans. Commun., vol. COM-25, pp. 117-128, Jan. 1977. F. Kuperus and J. Arnbak, Packet radio in a Rayleigh channel, IE Electron. Lett., vol. 18, pp. 506-507, June 1982. C. Namislo, Analysis of mobile radio slotted ALOHA networks, I Trans. Vehic. Technol., vol. VT-33, pp. 199-204, Aug. 1984. D. J. Goodman and A. M. Saleh, The nearifar effect in local ALOHA radio communications, ITrans. t4hic. Technol., vol. VT-36, pp. 19-27, Feb. 1987. J.C. Arnbak and W. Blitterswijk, Capacity of slotted ALOHA in Rayleigh-fading channels, IEEEJ. Select.Areas Commun., vol. SAC-5, pp. 261-269, Feb. 1987.

Chiew T. Lau (S84) received the B.Eng. degree from Lakehead University, Thunder Bay, Ont., Canada, in 1983, and the M.A.Sc. and Ph.D. degrees in electrical engineering from the University of British Columbia (U.B.C), Vancouver, B.C., in 1985 and 1990, respectively. From 1985 to 1986, he was a lecturer in the Department of Electronics and Communications at Singapore Polytechnic, Singapore. At the present time he is with the School of Applied Science, Nanyang Technological University, Singapore. His research interests are in packet radio systems.

Cyril Leung (S74-M76) received the BSc. (honors) degree from Imperial College, University of
London, England, in 1973, and the M.S. and Ph.D. degrees in electrical engineering from Stanford University in 1974 and 1976, respectively. From 1976 to 1979, he was an Aswtant Professor in the Department of Electrical Engineering and Computer Science, Mdssachusetts Institute of Technology. He wds on leave during 1978 dt Bell Laboratories, Holmdel, NJ, working on data networks During 1979-1980 he wds with the Department of Systems Engineering and Computing Science, Carleton University, Ottawd, Canada Since July 1980, he has been with the Department of Electrical Engineering, University of British Columbia, Vancouver, B.C , Canada, where he is a Professor. In 1987, he wa5 on sabbdticdl leave in the Department of Electrical Engineering, Ecole Polytechnique de Montreal, Montreal, P.Q., Canada Hi5 current rcsedrch interests are i n mobile data communications, coding and information theory