+ All Categories
Home > Documents > New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf ·...

New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf ·...

Date post: 25-Jul-2020
Category:
Upload: others
View: 0 times
Download: 0 times
Share this document with a friend
48
Per Connection Performance Analysis of a Frame-based TDMA/CDMA MAC Protocol Containing both Reservation and Contention Slots Ying-Ju Chen and Jin-Fu Chang February 26, 2004 Abstract A mixture of reservation plus contention data slots is now widely implemented in TDMA/ CDMA systems to make a system flexible enough to suit various kinds of packets but preserve the good nature of frame-based protocols. We conducted perfor- mance analysis for an arbitrary connection in such a system under the assumption of MMPP (Markov-modulated Poisson process) arrivals. Accessible slot locations of this connection in a frame is made general. Success probability in accessing a contention slot is also made general. We have obtained the system size distribution which can be used to evaluate the performance of various frame-based MAC protocols. The MMPP arrival pattern can be generalized to the BMAP( batch Markovian arrival process) family to further accommodate a broader set of traffic sources. Keywords: Queueing Analysis; Contention; TDMA; CDMA Part of the work was performed when the authors were with Graduate Institute of Communication Engineering, National Taiwan University, Taipei, Taiwan 106. Corresponding author; Stern School of Business, New York University, 44 W 4th Street, KMC 8-151, New York, NY 10012, USA; [email protected]; Tel: 212-998-0489, Fax: 212-995-4003. Department of Electrical Engineering, National Chi Nan University, Puli, Nantou, Taiwan 545; [email protected]; Tel: +886-49-2910272, Fax: +886-49-2912569. 1
Transcript
Page 1: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

Per Connection Performance Analysis of a

Frame-based TDMA/CDMA MAC Protocol

Containing both Reservation and Contention Slots∗

Ying-Ju Chen† and Jin-Fu Chang‡

February 26, 2004

Abstract

A mixture of reservation plus contention data slots is now widely implemented in

TDMA/ CDMA systems to make a system flexible enough to suit various kinds of

packets but preserve the good nature of frame-based protocols. We conducted perfor-

mance analysis for an arbitrary connection in such a system under the assumption of

MMPP (Markov-modulated Poisson process) arrivals. Accessible slot locations of this

connection in a frame is made general. Success probability in accessing a contention

slot is also made general. We have obtained the system size distribution which can be

used to evaluate the performance of various frame-based MAC protocols. The MMPP

arrival pattern can be generalized to the BMAP( batch Markovian arrival process)

family to further accommodate a broader set of traffic sources.

Keywords: Queueing Analysis; Contention; TDMA; CDMA

∗Part of the work was performed when the authors were with Graduate Institute of Communication

Engineering, National Taiwan University, Taipei, Taiwan 106.†Corresponding author; Stern School of Business, New York University, 44 W 4th Street, KMC 8-151,

New York, NY 10012, USA; [email protected]; Tel: 212-998-0489, Fax: 212-995-4003.‡Department of Electrical Engineering, National Chi Nan University, Puli, Nantou, Taiwan 545;

[email protected]; Tel: +886-49-2910272, Fax: +886-49-2912569.

1

Page 2: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

1 Introduction

To coordinate the allocation of resources to competing users is the spirit of an MAC (Multiple

access control) protocol. Among numerous techniques, frame-based TDMA/CDMA (time-

division /code-division multiple access) protocols have been widely adopted, especially in

wireless communication systems. As a number of mobile terminals share the same channel,

if data packets are sent without coordination, transmission failure may result due to collision.

In the real world, MAC layer designers normally adopt a combination of reservation plus

random access or contention scheme. A slot is either in reservation or contention mode.

In the reservation mode, the slots accessible to a specific connection are guaranteed, see,

e.g. Goodman and Wei[7] and Goodman et al. [8]. In the contention mode, terminals are

competing to access a slot in a blind manner, e.g. Babich[1] and numerous papers on IEEE

802.11. Through mathematical and experimental verifications, many papers (e.g. Brand

and Aghvano[2], Nanda et al.[12], and Wu and Ma[17]) suggest that the hybrid reserva-

tion/contention scheme provides a better performance than either single mode because it

incorporates both flexibility and regularity; and consequently this hybrid scheme has been

integrated into modern TDMA/CDMA protocols.

In the past, researchers have devoted countless effort in analyzing the performance of

a pure reservation TDMA system, see Bruneel[3], Lam[10], Chang [4], Ko and Davis[9],

Rubin[14], Rubin and Zhang[15], [16]. Regarding hybrid TDMA/CDMA systems, Chen

and Chang[5] provide delay analysis for a specific connection where pure reservation is

again adopted. To our knowledge, the performance of a hybrid reservation/contention

TDMA/CDMA system has not yet been fully investigated.

This paper is an extension of Chen and Chang[5] to the hybrid reservation/contention

arrangement. Packet arrivals are assumed to follow an MMPP (Markov modulated Poisson

process), an appropriate mathematical model for the VBR (variable bit rate) data streams,

and can be generalized to the BMAP(batch Markov arrival process). The service that can

2

Page 3: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

be rendered to a connection in time and code is made general, provided that there is at least

one code reserved for this connection per frame. Recall that in Chen and Chang[5] state

transition probabilities and system size distribution counting from the prior departure point

are derived. If no reservation code is given to a connection, there then exist infinitely many

possible prior departure points and we shall find no closed-form results.

The codes open for contention come from two possible sources. Some are open for con-

tention as the connections are established; and the others are released from codes assigned

but not used by other connections which occur when their data queues have become empty.

For each contention code, the probability it is successfully captured by a connection is as-

sumed fixed; but different contention codes may have different probabilities of success. In

other words, after having observed an infinitely-long past history of the system, we consider

a steady-state scenario whereas the frame-based periodic behavior preserves.

This paper is organized as follows. In Sec. 2 we describe the traffic source and the state

parameters. Sec. 3 summarizes the main results obtained in Chen and Chang[5]. Unlike

Chen and Chang[5], the mathematical derivations can be decomposed into three categories,

whose queueing analyses are subsequently presented in Secs. 4 - 6. In Sec. 7, we show

how to integrate these derivations to fit into a general model, and discuss the complexity in

programming implementation of our mathematical results. Sec. 8 gives numerical examples

and discussions. Finally, we draw conclusions in Sec. 9.

2 Model Description

2.1 The Traffic Source

In order to incorporate real-time bursty traffic such as voice or video, we assume that arrivals

follow an MMPP process. Fig. 1 provides a graphical representation of a two-state MMPP,

and we encourage readers to consult Fischer and Meier-Hellstern[6] and Lucatoni[11] for

3

Page 4: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

detailed descriptions of an MMPP.

state 1 state 2

�1�2

�1

�2

Figure 1: The state transition rate diagram of a two-state MMPP.

A general MMPP can be characterized by the following parameters:

Q =

−σ1 σ12 . . σ1m

σ21 −σ2 . . σ2m

. . . . .

σm1 σm2 . . −σm

,

where σi =∑m

j=1,j �=i σij and Λ = diag(λ1, λ2, ...., λm). Its stationary state probability vector

π = [π1 π2 ... πm] can be obtained from equalities πQ = 0, πe = 1, where e = [1 1 ... 1]tr

and tr is abbreviated from “transpose”.

The counting process Nt of MMPP tells the number of arrivals in a time interval of length

t, and It denotes its phase at time t. Pij(n, t) = Pr(Nt = n, It = j|N0 = 0, I0 = i) represents

the probability that the number of arrivals during [0,t] is n, and the phase switches from i

to j. The PGF (probability generating function) P (z, t) of Pij(n, t) is [6]

P ∗(z, t) = e(Q−(1−z)Λ)t. (1)

2.2 The Hybrid TDMA/CDMA Protocol

In Fig. 2, we present the structure of a hybrid TDMA/CDMA protocol where the horizontal

axis represents time and the vertical axis is for code. In other words, the TDMA pattern

4

Page 5: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

is demonstrated along the horizontal direction, and the code dimension is exhibited in the

vertical direction. Each slot in the time axis has a fixed length equal to the transmission

time of a data cell or packet, and these slots are grouped into TDMA frames of fixed length

Tf , for example, 10 slots in Fig. 2.

In each slot, a number of codes are accessible to admitted connections, and the code

ceiling in Fig. 2 may vary from one slot to another to reflect different considerations of bit

error rate (BER) requirement. The two-dimensional frame pattern repeats every Tf slots.

In Fig. 2, the time slots and codes assigned to the connection we considered are marked by

X.

Code

TDMA Frame

X

X

...

...

...

...

X

X

X

X

X

Figure 2: Frame pattern of a hybrid TDMA/CDMA protocol.

In this paper, a slot may have sole reserved or contention codes or both. Since whether

a slot contains reserved codes makes great difference in our analysis, we call slots containing

reserved codes “hybrid”, while “contention” slots refer to those containing only contention

codes. That is, a hybrid slot may also contain contention codes. We use aq to denote the

position of the q-th hybrid slot within a frame, and br the position of the r-th contention

slot. In Fig. 3 of a TDMA frame with 10 slots, the connection we considered is permitted

to use M = 2 hybrid slots: the 3-rd and 6-th slot, and N = 2 contention slots: the 8-th and

10-th slot. This is to say a1 = 3, a2 = 6 and b1 = 8, b2 = 10. Without loss of generality,

we make the last contention slot also the last slot of the frame, e.g., b2 = 10 in the example

of Fig. 3. Before we go into Sec. 7 we assume that within a frame hybrid slots are always

5

Page 6: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

placed prior to contention ones, then we shall in Sec. 7 discuss how to modify our analysis

to fit into a system whose hybrid and contention slots are placed arbitrarily.

f f a a f LLLL

h­[ P h­IO[¡O

v­ v­IO O[Q

 P[T

¡O[V

¡P[ON

>>>>>>>>>>>>rbk_>d± ¬¤

Figure 3: An example to define a1, ..., aM ; b1, ..., bN ;Xn, Jn, and In.

In the q-th hybrid slot, rq denotes the number of reserved codes, and cq denotes the

number of contention codes. In a hybrid slot, we assume that reserved codes are consumed

first before a connection turns to contention codes for additional resources. In attempting a

contention code in the q-th hybrid slot we assume the attempt succeeds with probability pq.

We use c′q to denote the number of contention codes in the q-th contention slot, and p

′q the

success probability in a contention attempt. It requires only slight modification to extend

our results to a model where we allow the probabilities in a slot to vary.

We lay here a more general definition of state parameters in order to simplify the formulas

in this paper:

a0 = 0, b0 = aM ,

aq−(M+N) = aq − bN , aq+(M+N) = aq + bN , 1 ≤ q ≤M,

rq = rq mod (M+N), cq = cq mod (M+N).

(2)

2.3 State Description

As in Chen and Chang[5], we consider within a frame the slot position in which a departure

occurs, and the system size at the beginning of the next available slot. We let Xn denote

6

Page 7: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

the number of queued packets or cells in the connection we considered at the beginning of

the n-th slot, where n = 1, 2, .... The observed point must be one of those slots accessible

by our chosen connection, but not every accessible slot needs to be observed. In Fig. 3,

Xn is observed because a departure occurs at the end of the previous slot. Jn denotes the

position of this hybrid slot within a frame, i.e., Jn = 6 in Fig. 3. We further let In denote

the phase of the MMPP at the instant Xn is observed. In discussing transitions, we may

use {Jn = aM+i} ≡ {Jn = bi}, 0 ≤ i ≤ N and {Jn = bN+i} ≡ {Jn = ai}, 0 ≤ i ≤ M for

notational convenience.

3 A Review of the Pure TDMA/CDMA System

Let us in this section review the key results in Chen and Chang[5] where a reservation-based

TDMA/CDMA scheme is treated. The analysis is divided into two cases Xn ≥ 1 and Xn = 0

to derive the elements of the state transition matrix of (Xn, Jn, In): Pr(Xn+1 = s, Jn+1 =

aq, In+1 = j|Xn = r, Jn = ap, In = i). Rearranging these probabilities Pr(Xn+1 = s, Jn+1 =

aq, In+1 = j|Xn = r, Jn = ap, In = i) into blocks Pr,s, Q(∞) can be expressed as follows :

Q(∞) =

P0,0 P0,1 P0,2 P0,3 P0,4 P0,5 . . .

P1,0 P1,1 P1,2 P1,3 P1,4 P1,5 . . .

. . . . . . . . .

Pc−1,0 Pc−1,1 Pc−1,2 Pc−1,3 Pc−1,4 Pc−1,5 . . .

A0 A1 A2 A3 A4 A5 . . .

0 A0 A1 A2 A3 A4 . . .

. 0 A0 A1 A2 A3 . . .

. . 0 A0 A1 A2 . . .

. . . . . . . . .

, (3)

where c = max1≤i≤M{ri}.

7

Page 8: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

If we further let

Bv ≡

P0,cv P0,cv+1 . . P0,cv+c−1

P1,cv−1 P1,cv . . P1,cv+c−2

. . . . .

Pc−1,cv−c−1 Pc−1,cv−c−2 . . Pc−1,cv

,

A0 ≡

A0 A1 A2 . Ac−1

0 A0 A1 . Ac−2

. 0 A0 . Ac−3

. . . . .

0 0 . . A0

,

Av ≡

Acv Acv+1 . . Acv+c−1

Acv−1 Acv . . Acv+c−2

. . . . .

Acv−c−1 Acv−c−2 . . Acv

,

then (3) can be written as follows:

Q(∞) =

B0 B1 B2 B3 B4 B5 . . .

A0 A1 A2 A3 A4 A5 . . .

0 A0 A1 A2 A3 A4 . . .

0 0 A0 A1 A2 A3 . . .

. . 0 A0 A1 A2 . . .

. . . 0 A0 A1 . . .

. . . . 0 A0 . . .

. . . . . . . . .

. (4)

By M/G/1 algorithm (Lucatoni[11] and Neuts[13]), we are able to obtain the steady

state probability vector x = [x0 x1 x2 ...], where xi = [πci πci+1 ... πc(i+1)−1] is a row

8

Page 9: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

vector of length c × m × M in which each πk is an m × M row vector. [π1,1k π2,1

k ...

πm,1k π1,2

k ... πm,2k ... π1,M

k ... πm,Mk ] with πr,s

k denoting the probability the inspected connection

has k packets at the beginning of the as+1-th slot and the MMPP’s phase is r, given that a

departure occurs at the as-th slot.

We can thus use the steady state probability vector to derive the system size distribution

at an arbitrary time. If t ∈ [aq − 1, aq),

Q∗t (z) =πq−1(z)P

∗(z, t− aq + 1)

+M−1∑i=0

πq−i−10 [I− e(Q−Λ)aM ]−1e(Q−Λ)(aq−1−aq−i−1)P ∗(z, t− aq−1 + 1),

Qt(z) = [Q∗t (1)e]

−1Q∗t (z)e ≡ Q∗

t (z)e.

(5)

The system size distribution Qt(z) where t ∈ [aq−1, aq − 1) can then be obtained as follows.

Q∗aq−1(z) = Q

∗t (z)P

∗(z, aq − 1− t). (6)

Based on the results in Eqs. (5) and (6), we obtain the first and second moments as

follows. For t ∈ [aq − 1, aq),

Q∗′t (1) = π

′q−1(1)e

−Q(t−aq+1) + πq−1(1)M1(t− aq + 1)

+

M−1∑i=0

πq−i−10 [I− e(Q−Λ)aM ]−1e(Q−Λ)(aq−1−aq−i−1)M1(t− aq−1 + 1),

Q∗′′t (1) = π

′′q−1(1)e

−Q(t−aq+1) + 2π′q−1(1)M1(t− aq + 1) + πq(1)M2(t− aq + 1)

+

M−1∑i=0

πq−i−10 [I− e(Q−Λ)aM ]−1e(Q−Λ)(aq−1−aq−i−1)M2(t− aq−1 + 1).

(7)

And for t ∈ [aq−1, aq − 1),

Q∗′aq−1(1) = Q

∗′t (1)e

−Q(aq−1−t) +Q∗t (1)M1(aq − 1− t),

Q∗′′aq−1(1) = Q

∗′′t (1)e

−Q(aq−1−t) + 2Q∗′t (1)M1(aq − 1− t) +Q∗

t (1)M2(aq − 1− t).(8)

Readers can consult [6] for explicit forms of µ1(t),M1(t), µ2(t) and M2(t). This paper

also provides the sojourn time distribution.

9

Page 10: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

4 A Model with Only Hybrid Slots

Analysis of the queueing behavior of a general TDMA/CDMA system can be divided into

three parts, each treating a special case. We shall investigate these special cases in this and

Secs. 5 - 6, and then in Sec. 7.1 integrate them into a general framework.

In this section, we consider a TDMA/CDMA protocol in which slots accessible to the

inspected connection are all of hybrid type we defined earlier, i.e., they all contain reserved

codes and N = 0. Consequently, if there are still packets in the buffer after one batch of

packet transmissions, departure will surely occur in the next accessible slot.

4.1 State Transition Matrix

Analogous to Chen and Chang[5], we establish the elements of the state transition matrix

of (Xn, Jn, In) by dividing the treatment into the following two cases.

4.1.1 Xn ≥ 1

We examine the transitions to Xn+1 conditioning on Xn ≥ 1 here. Let aq+1 denote the slot

location associated with Xn, where 0 ≤ q ≤ M − 1. Since Xn ≥ 1, the next observationpoint shall be the beginning of the aq+2-th slot. Thus

Pr(Xn+1 = k, Jn+1 = ar, In+1 = j|Xn ≥ 1, Jn = aq, In = i) = 0, r �= q + 1. (9)

From Eq. (10) to Eq. (12), base on the number of queued packets Xn = l at the beginning

of the (q+1)-th accessible slot, we divide the state transitions into : l ≥ rq+1 + cq+1 + 1,

rq+1 ≤ l ≤ rq+1+cq+1, and 1 ≤ l ≤ rq+1. If l ≥ rq+1+cq+1+1, the number of queued packets

exceeds the total number of codes in the next accessible slot, the inspected connection will

consume all reserved codes, and also compete to use contention codes. Consequently,

10

Page 11: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

Pr(Xn+1 = k, Jn+1 = aq+1, In+1 = j|Xn = l, Jn = aq, In = i)

=

cq+1∑u=max(l−rq+1−k,0)

Ccq+1u pu

q+1(1− pq+1)cq+1−uPij(k + rq+1 + u− l, aq+2 − aq+1),

l ≥ rq+1 + cq+1 + 1, k ≥ l − rq+1 − cq+1, 1 ≤ q ≤ N.

(10)

If rq+1 ≤ l ≤ rq+1+cq+1, after the connection has consumed all reserved codes, it competes

for l − rq+1 rather than cq+1 contention codes. Thus

Pr(Xn+1 = k, Jn+1 = aq+1, In+1 = j|Xn = l, Jn = aq, In = i)

=

l−rq+1∑u=max(l−rq+1−k,0)

C l−rq+1u pu

q+1(1− pq+1)l−rq+1−uPij(k + rq + u− l, aq+2 − aq+1),

rq+1 ≤ l ≤ rq+1 + cq+1, 1 ≤ q ≤ N.

(11)

Finally, if 1 ≤ l ≤ rq+1, the number of buffered packets is smaller than the number of

reserved codes, contention codes are then made available to other connections.

Pr(Xn+1 = k, Jn+1 = aq+1, In+1 = j|Xn = l, Jn = aq, In = i)

= Pij(k, aq+2 − aq+1), 1 ≤ l ≤ rq+1, 1 ≤ q ≤ N.(12)

4.1.2 Xn = 0

Now we consider the transitions when Xn = 0. In this case, buffer becomes empty at the

beginning of the next accessible slot after the n-th departure, which occurs at the end of

the aq-th slot. Therefore, the next observation point is not necessarily the beginning of

the aq+2-th slot but the next accessible slot following the departure of a new arrival which

arrives after the position associated with Xn. We let Gq(k) represent the probability that

the inspected connection has no packets to transmit during the aq−1-th slot, but does have

11

Page 12: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

packets to send in the aq-th slot; and has accumulated k packets at the beginning of the

aq+1-th slot.

Gq(k) =

rq∑u=1

P (u, aq − aq−1)P (k, aq+1 − aq)

+

k+rq+cq∑u=rq+1

min(u−rq,cq)∑v=max(u−rq−k,0)

Cmin(u−rq ,cq)v pv

q(1− pq)min(u−rq,cq)−v

P (u, aq − aq−1)P (k + rq + v − u, aq+1 − aq),

(13)

where P (u, aq−aq−1) is them×m probability matrix that u packets have arrived in (aq−aq−1)

slots taking all possible phase changes under consideration.

The first term of Gq(k) denotes the case that the number of packets queued immediately

before the aq-th slot is less than the number of reserved codes so that these packets get

transmitted in this slot, and the contention codes are not used at all. The number of packets

seen at the observation point, the beginning of the (q+1)-th slot, depends only on the arrivals

during (aq − 1, aq+1 − 1].The second term of Gq(k) is more complicated, but still can be explicitly explained.

When the number of packets seen at the beginning of the aq-th slot exceeds rq, the inspected

connection has to seek to use contention codes. The number of contention codes which

need to be acquired by this connection depends on u − rq, the number of packets since rqreserved codes are consumed first. If u − rq < cq, only u − rq packets enter competitionin the aq-th slot; otherwise, all cq contention codes have to be attempted. Let v denote

the number of packets successfully transmitted via the contention codes in the aq-th slot.

The probability P (k + rq + v − u, aq+1 − aq) specifies the number of packets required to

arrive during [aq − 1, aq+1 − 1) so that the inspected connection accumulates k packets atthe beginning of the aq+1-th slot, the next observation point.

Having obtained Gq(k), we are able to derive the transitions for ap+1 ≤ aq−1 and ap+1 >

aq−1, respectively, which are exactly identical to those in [5]. For ap+1 ≤ aq−1,

12

Page 13: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

Pr(Xn+1 = k, Jn+1 = aq, In+1 = j|Xn = 0, Jn = ap, In = i)

= [[I− e(Q−Λ)aM ]−1e(Q−Λ)(aq−1−ap+1)Gq(k)]i,j

= etri [[I− e(Q−Λ)aN ]−1e(Q−Λ)(aq−1−ap+1)Gq(k)]ej ,

(14)

where []i,j is the element on the i− th row and j − th column of the associated matrix andei = [0 ... 0 1 0 ... 0]

tr where the only 1 appears at the i − th place. Note that the term[I− e(Q−Λ)aM ]−1e(Q−Λ)(aq−1−ap+1) in Eq. (14) is the summation of a geometric series, because

the occurrence of the new arrival may cross one frame boundary. For ap+1 > aq−1, a similar

equation can be obtained as follows.

Pr(Xn+1 = k, Jn+1 = aq, In+1 = j|Xn = 0, Jn = ap, In = i)

= etri [I− e(Q−Λ)aM ]−1e(Q−Λ)(aN +aq−1−ap+1)Gq(k)ej.

4.1.3 Construction of the M/G/1-type Transition Matrix

The procedure to construct the transition matrix from probabilities Pr(Xn+1 = k, Jn+1 =

aq, In+1 = j|Xn = l, Jn = ap, In = i) is the same as Sec. 3 with one modification

c = max1≤i≤M(ri + ci). From that we are able to obtain its steady state probability

x = [x0 x1 x2 ...].

4.2 System Size Distribution at an Arbitrary Time

Let t ∈ [aq, aq+1) be an arbitrary time within a frame, where q = 1, ...,M with [aM , aM+1) =

[0, a1). We first obtain the system size distribution for t ∈ [aq−1, aq) and that of t ∈ [aq−1, aq−1) follows Eq. (6), regardless of code allocation policies. Then

13

Page 14: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

Q∗t (z) =πq−1(z)P

∗(z, t− aq + 1)

+M−1∑i=0

πq−i−10 [I− e(Q−Λ)aM ]−1e(Q−Λ)(aq−1−aq−i−1)P ∗(z, t− aq−1 + 1),

(15)

where πq−1(z) is the z-transform of the probabilities associated with the aq-th slot in x′is of

Sec. 4.1.3 and πq−i−10 = [π1,q−i−1

0 π2,q−i−10 ... πm,q−i−1

0 ] is the probability that no packet is

awaiting at the beginning of the aq−i-th slot while the MMPP′s phase varies from 1 to m.

Moments can be obtained by Eqs. (7)-(8).

5 A Model with Single-Code Slots

In this section, we consider a system in which all slots contain no more than one code

accessible to the inspected connection. Since a hybrid slot contains at least one reserved code,

in this system no contention codes are accessible. In other words, rq = 1, cq = 0, 1 ≤ q ≤M,and c

′q = 1, 1 ≤ q ≤ N.Restricting to the single accessible code scenario makes us easier to concentrate on the

treatment of multiple successive contention slots. Note that a TDMA system subject to pure

contention in each slot is a subclass of this model; while a hybrid TDMA/CDMA system

may also be thus described if codes accessible by the inspected connection happen to be

mutually non-overlapping.

5.1 State Transition Matrix

In the following we derive the state transition matrix for Xn ≥ 1 and Xn = 0.

14

Page 15: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

5.1.1 Xn ≥ 1

A. Transitions to a Hybrid Slot other than the a1-th Slot

Let us first consider departures occurring in a hybrid slot except the very first. WhenXn ≥ 1,the next departure can occur at either the next hybrid slot, or a contention slot prior to the

next hybrid slot. In other words, it cannot occur after the time has passed a hybrid slot

since packets have been waiting in the buffer. In particular, when the location associated

with Xn is bq, 1 ≤ q ≤ N , the next departure cannot pass the a1-th slot in the upcoming

frame even if the inspected connection fails to acquire any of the residual contention slots in

the current frame. Hence

Pr(Xn+1 −Xn = k, Jn+1 = ar, In+1 = j|Xn ≥ 1, Jn = aq, In = i) = 0, r �= q + 1,

P r(Xn+1 −Xn = k, Jn+1 = ar, In+1 = j|Xn ≥ 1, Jn = bq, In = i) = 0, r �= 1.(16)

Now we turn to discuss possible transitions. Suppose that the slot position associated

with Xn is aq, where 1 ≤ q ≤ M − 1. A packet gets transmitted in the aq+1-th slot via the

reserved code, and consequently the next observation point is the beginning of the aq+2-th

slot.

Pr(Xn+1 −Xn = k,Jn+1 = aq+1, In+1 = j|Xn ≥ 1, Jn = aq, In = i)

= Pij(k + 1, aq+2 − aq+1), 0 ≤ q ≤ M − 1.(17)

B. Transition to a Contention slot and the a1-th slot

The cases where the (n+1)-st departure occurs at the end of a contention slot and the a1-th

slot are very similar, and therefore we treat them together here. We identify impossible

transitions first. If the n-th departure occurs in a hybrid slot but not the very last, i.e., the

aq-th slot where q �= M , the next departure cannot occur in a contention slot since there

15

Page 16: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

exist hybrid slots in between. That is, it is impossible to observe a state transition from

aq, q �=M to a contention slot as well as the a1-th slot in the next frame. Therefore,

Pr(Xn+1 −Xn = k, Jn+1 = br, In+1 = j|Xn ≥ 1, Jn = aq, In = i) = 0,

q �=M, 1 ≤ r ≤ N + 1.(18)

The state transition bq → br, where r ≤ q, is also impossible since the closest slot that

can be specified as br is the br-th slot of the next frame. If it were the case, the packet would

have bypassed several hybrid slots. Thus

Pr(Xn+1 −Xn = k, Jn+1 = br, In+1 = j|Xn ≥ 1, Jn = bq, In = i) = 0, r ≤ q. (19)

If the n-th departure occurs at the end of the bq-th slot, the next departure occurs at

the br-th slot, where r > q, another contention slot in the current frame, if and only if

the inspected connection fails to use the bq+1-th, bq+2-th, ..., br−1-th slots, and succeeds to

transmit in the br-th slot. Referring to the graphical illustration in Fig. 4, we obtain

Pr(Xn+1 −Xn = k, Jn+1 = br, In+1 = j|Xn ≥ 1,Jn = bq, In = i)

= [p′r

r−1∏u=q+1

(1− p′u)]Pij(k, br+1 − bq+1), 0 ≤ q ≤ N − 1, q ≤ r ≤ N.(20)

Similarly, the next departure occurs in the a1-th slot of the next frame only when the

inspected connection fails to compete in the bq+1, bq+2, ..., bN -th slots. Thus

Pr(Xn+1 −Xn = k, Jn+1 = a1, In+1 = j|Xn ≥ 1, Jn = bq, In = i)

= [

N∏u=q+1

(1− p′u)]Pij(k, a2 + bN − bq+1), 0 ≤ q ≤ N.(21)

When Xn ≥ 1, the only possible transitions from a contention slot to a hybrid slot arethe cases bq → a1 because no hybrid slot lies in between.

16

Page 17: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

...... ...... ...... ...... ............

br+1

Xn+1Xn � 1

bq bq+1 bq+2 br

fail fail fail success

Jn = bqJn+1 = br

br�1

Figure 4: An example to show the state transition bq → br when Xn ≥ 1.

5.1.2 Xn = 0

Analogous to the treatment of Xn ≥ 1, we present first the derivations for aq, q �= 1.

A. Transitions to a Hybrid (but not the a1-th) Slot

The transitions associated with aq, q �= 1, can be regarded as a special case of Sec. 4.1.2,where rq = 1, cq = 0. Thus, H

Rq (k), the degenerated probability function of Gq(k) is

HRq (k) = P (k + 1, aq+1 − aq−1)− P (0, aq − aq−1)P (k + 1, aq+1 − aq), 1 ≤ q ≤ M. (22)

Therefore, according to Sec. 4.1.2, the transition probabilities are

Pr(Xn+1 = k, Jn+1 = aq, In+1 = j|Xn = 0, Jn = ap, In = i)

= etri [[I− e(Q−Λ)aM ]−1e(Q−Λ)(aq−1−ap+1)HR

q (k)]ej, 2 ≤ p+ 1 ≤ q − 1, q ≤M,and

Pr(Xn+1 = k, Jn+1 = aq, In+1 = j|Xn = 0, Jn = ap, In = i)

= etri [I− e(Q−Λ)aM ]−1e(Q−Λ)(bN +aq−1−ap+1)HR

q (k)ej, q − 2 < p ≤ M +N, 2 ≤ q ≤ M.

B. Transitions to a Contention Slot and the a1-th slot

To derive transitions to a contention slot, we introduce a similar probability function HCq (k)

for convenience.

17

Page 18: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

HCq (k) =

q−1∑i=0

[p′q

q−1∏u=i+1

(1− p′u)]P (0, bi − aM)

[P (k + 1, bq+1 − bi)− P (0, bi+1 − bi)P (k + 1, bq+1 − bi+1)], 1 ≤ q ≤ N.(23)

The i-th term ofHCq (k) represents the probability that the inspected connection has no packet

arrivals during [aM , bi), but has packets to send in the bi+1-th slot; and this connection fails

in the competition in the bi+1-th ,bi+2-th ,...,bq−1-th slots, but succeeds in the bq-th slot; and

it has accumulated k packets at the beginning of the bq+1-th slot.

Needless to say, HCq (k) is more complicated than the H

Rq (k) in Sec. 4. The additional

complexity comes from the more diversified arriving points of the first arrival following the

observation point of Xn. If the next departure occurs in the bq-th slot, unlike the case of pure

hybrid slots, the first arrival does not have to appear right after the bq−1-th slot. Rather,

it can emerge in any one of [aM , b1), [b1, b2), ..., and [bq−1, bq) subject to the restriction that

before the bq-th slot, the inspected connection fails in the competitions in all contention slots

falling in between. Fig. 5 presents a diagrammatical visualization of this event.

...... ...... ...... ...... ............

Xn+1

bqbq+1

fail fail success

Jn+1 = bq

arrivals

aMbi bq�1bi+1

no cell

Figure 5: An example to express the meaning of HCq (k).

Likewise, a probability function HR1 (k) ≡ HC

N+1(k) for transitions to the a1-th slot is

defined as follows.

18

Page 19: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

HR1 (k) =

N∑q=0

[N∏

u=q+1

(1− p′u)]P (0, bq − aM)

[P (k + 1, a1 + bN − bq)− P (0, bq+1 − bq)P (k + 1, a1 + bN − bq+1)].

(24)

Now we are ready to discuss state transitions to the br-th slot. If the n-th departure

occurs at the end of a non-last hybrid slot, the result is the same as that in Sec. 4, but we

replace HRr (k) by H

Cr (k).

Pr(Xn+1 = k, Jn+1 = br, In+1 = j|Xn = 0, Jn = aq, In = i)

= etri [I− e(Q−Λ)bN ]−1e(Q−Λ)(aM−aq+1)HC

r (k)ej , 1 ≤ q ≤ M − 1, 1 ≤ r ≤ N + 1.(25)

If the n-th departure occurs at the end of a contention slot bq, where r ≤ q + 1, state

transition covers at least one frame. Thus

Pr(Xn+1 = k, Jn+1 = br, In+1 = j|Xn = 0,Jn = bq, In = i)

= etri [I− e(Q−Λ)bN ]−1e(Q−Λ)(bN +aM−bq+1)HC

r (k)]ej , 0 ≤ q ≤ N, 1 ≤ r ≤ q + 1.(26)

The case r ≥ q + 2 is slightly different. Since the observation point of Xn = 0 is prior to

the br-th slot, transition may occur in the current frame, and the probability depends on the

arriving point of the new first arrival from [bq+1, bq+2) to [br−1, br). Therefore the transition

probability is composed of two terms:

Pr(Xn+1 = k, Jn+1 = br, In+1 = j|Xn = 0, Jn = bq, In = i)

= etri {

r−1∑u=q

[p′r

r−1∏v=u+2

(1− p′v)]P (0, bu+1 − bq+1)

[P (k + 1, br+1 − bu+1)−P (0, bu+2 − bu+1)P (k + 1, br+1 − bu+2)]

+[I− e(Q−Λ)bN ]−1e(Q−Λ)(bN +aM−bq+1)HCr (k)}ej,

0 ≤ q ≤ N − 2, q + 2 ≤ r ≤ N + 1.

(27)

19

Page 20: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

5.1.3 Construction of the M/G/1-type Transition Matrix

After completing the derivations of the probabilities of all state transitions, we now are able

to construct the transition matrix as follows.

Q(∞) =

P0,0 P0,1 P0,2 P0,3 P0,4 P0,5 . . .

A0 A1 A2 A3 A4 A5 . . .

0 A0 A1 A2 A3 A4 . . .

. 0 A0 A1 A2 A3 . . .

. . 0 A0 A1 A2 . . .

. . . . . . . . .

, (28)

which is already an M/G/1-type transition matrix. Therefore, we can apply the M/G/1 algo-

rithm on Q(∞) of this format to obtain its steady state probability vector x = [x0 x1 x2 ...].

5.2 System Size Distribution at an Arbitrary Time

We provide the system size distribution when t belongs to an accessible slot here. Observing

that t ∈ [ai−1, ai), 1 ≤ i ≤M is exactly the same as Sec. 4, instead we focus on t ∈ [bi−1, bi),i.e., the time within a contention slot. For convenience we introduce Hi(k, t), a probability

function similar to HCi (k), and its PGF Hi(z, t),

Hi(z, t) =

∞∑k=0

Hi(k, t)zk,

Hi(z, t) =

i−1∑j=0

[

i−1∏u=j+1

(1− p′u)]e(Q−Λ)(bj−aM )[P (z, t− bj + 1)− e(Q−Λ)(bj+1−bj)P (z, t− bj+1 + 1)].

(29)

The j-th term of Hi(k, t) represents the probability that the inspected connection has

no packets in the aM -th slot, and has arrivals in the bj-th slot; it fails the competition in

the bj+1-th, bj+2-th ,..., bi−1-th slots, and has accumulated k packets at time t ∈ [bi − 1, bi).

20

Page 21: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

The physical meaning of HCi (k) is like that of H

Ci (k), except that we need not consider the

contention in the bi-th slot. Thus, for t ∈ [bi − 1, bi),

Q∗t (z) =

M−1∑q=0

πq0[I− e(Q−Λ)bN ]−1e(Q−Λ)(aM−aq+1−1)Hi(z, t)

+N−1∑q=0

πq+M0 [I− e(Q−Λ)bN ]−1e(Q−Λ)(aM +bN−bq+1)Hi(z, t)

+i−1∑q=0

πq+M0

i−1∑j=q+1

[i−1∏

u=j+1

(1− p′u)]e(Q−Λ)(bj−bq+1)[P (z, t− bj + 1)−

e(Q−Λ)(bj+1−bj)P (z, t− bj+1 + 1)]

+

i−1∑q=0

[πq+M (z)− πq+M0 ][

i−1∏u=q+1

(1− p′u)]P (z, t− bq+1 + 1),

(30)

where πq+M(z) is the z-transform generated from the probabilities belonging to the bq-th slot

and πq+M0 = [π1,q+M

0 π2,q+M0 ... πm,q+M

0 ] is the probability of seeing no packets right before

the bq+1-th slot when the MMPP′s phase varies from 1 to m.

The expression of Q∗t (z) can be divided into four terms. The first term represents the

probability that the latest departure occurs in a0 ∼ aM−1, and the observed connection has

no packets before the aM -th slot; moreover, from the aM -th slot to t, it transmits no packet,

and has accumulated k packets at t. Consequently, this event coincides with the definition of

Hi(k, t). The second term represents the event that the latest departure occurs in b0 ∼ bN−1,

but at least one frame has passed-by from the latest departure point to t.

The third term represents the probability that the latest departure occurs in either the

aM -th slot or a contention slot prior to t, i.e., b1, ..., bi−1, in the same frame as t. The index

of the first summation denotes the slot associated with the latest departure, while the index

of the second summation denotes the arrival point of the new first arrival. Since the latest

departure occurs at the end of the bq-th slot, if this connection has no packet during [bq+1, bj)

but has packets to transmit before the bj+1-th slot, it will fail the competitions in the bj+1-th,

21

Page 22: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

bj+2-th, ..., bi−1-th contention slots.

The last term of Q∗t (z) represents the event that there are packets remain in the buffer

at the latest observed point. Since packets queued at the beginning of a hybrid slot will be

transmitted, the only possible situation is that the latest departure occurs in a contention slot

or the aM -th slot. If the latest departure occurs in the bq-th slot, the inspected connection

will fail the competitions in the bq+1-th , bq+2-th, ..., bi−1-th slots. Fig. 6 is drawn to illustrate

those events of the third and the last terms.

(a)Xn � 1

...... ...... ...... ...... ............

...... ...... ...... ...... ...... ......

Xn = 0

(b)Xn = 0

bi

Qt = kXn � 1

bq bq+1 bq+2 bi�1

fail fail fail

bq bq+1 bj bi�1 bi

Qt = k

fail

fail fail

bj+1

no cell arrivals

Figure 6: Two examples to show possible locations of the latest departure

It is straight-forward to compute the moments by differentiating Qt(z), in which the

followings are needed:

H′i(1, t) =

i−1∑j=0

[i−1∏

u=j+1

(1− p′u)]e(Q−Λ)(bj−aM )[M1(t− bj + 1)− e(Q−Λ)(bj+1−bj)M1(t− bj+1 + 1)],

H′′i (1, t) =

i−1∑j=0

[

i−1∏u=j+1

(1− p′u)]e(Q−Λ)(bj−aM )[M2(t− bj + 1)− e(Q−Λ)(bj+1−bj)M2(t− bj+1 + 1)].

6 AModel with Contention Slots Having Multiple Codes

In this section we treat a model where each contention slot has multiple codes. Mathematical

derivations are much more complicated than previous sections.

22

Page 23: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

In order to address the complexity brought by such multicode contention slots, we restrict

hybrid slots to contain only reserved codes. Thus the system parameters of this model are:

rq �= 0, cq = 0, 1 ≤ q ≤ M,, and N > 1, c′q ≥ 1, 1 ≤ q ≤ N. In other words, the model

we discuss here is a hybrid TDMA/CDMA system considered in [5] followed by a string

of multicode contention slots. Such framework can easily find its place in many existing

communication protocols.

6.1 State Transition Matrix

Derivations are same as Sec. 4 except transitions corresponding to the a1-th slot and con-

tention slots b1, b2, ..., bN . Thus we concentrate on these two cases in the sequel.

Let us recall Eq. (20) where a pure contention TDMA system is considered. In that

equation, we used only a simple product term [p′r

∏r−1u=q+1(1 − p

′u)] to represent the event

that the inspected connection has failed competitions in the bq+1-th, bq+2-th, ..., br-th slots

because the number of codes available for contention is merely 1.

However, if the number of codes in these contention slots may vary, the derivation is no

longer simple. Elements of the transition matrix may depend on whether or not the number

of codes in the former slot is larger than the latter, and therefore the algorithmic “if-else”

or “switch-case” instructions must be widely employed.

6.1.1 Xn ≥ 1

If there are already packets queued in the buffer when the observation point of Xn is placed

at the beginning of the aq+1-th slot, transmission surely occurs in the aq+1-th slot. In other

words, transitions where there are hybrid slots in between can never occur. Hence

23

Page 24: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

Pr(Xn+1 = k, Jn+1 = br, In+1 = j|Xn = l, Jn = aq, In = i) = 0, q �=M, 1 ≤ r ≤ N + 1,

P r(Xn+1 = k, Jn+1 = br, In+1 = j|Xn = l, Jn = bq, In = i) = 0, 1 ≤ q ≤ N, r < q + 1.(31)

Now we consider the case when the n-th departure occurs at the end of a contention

slot bq, i.e., the observation point of Xn = l is at the beginning of the bq+1-th slot. We

first consider transitions between two consecutive contention slots, i.e., bq → bq+1. In the

bq+1-th slot, the number of codes available for competition to the inspected connection is

min(l, c′q+1). If the next departure occurs in the bq+1-th slot, the connection will have at least

successfully sent one packet in this slot; moreover, if it has successfully sent less than (l− k)packets in this slot, the number of packets accumulated right before the next observation

point, i.e., the bq+2-th slot of the next frame, will exceed k. Thus, the number of packets

transmitted in this slot shall be lower bounded by max(1, l − k), and we conclude that

Pr(Xn+1 = k, Jn+1 = bq+1, In+1 = j|Xn = l, Jn = bq, In = i)

=

min(l,c′q+1)∑

u=max(1,l−k)

Cmin(l,c

′q+1)

u p′uq+1(1− p

′q+1)

min(l,c′q+1)−uPij(k + u− l, bq+2 − bq+1), 0 ≤ q ≤ N.

We next consider transitions between two non-consecutive contention slots.

Pr(Xn+1 = k, Jn+1 = br,In+1 = j|Xn = l ≥ 1, Jn = bq, In = i)

= etri {

Vq+2∑vq+2=0

...Vs∑

vs=0

...Vr∑

vr=0

Ur∑u=max(1,l+

∑rq+2 vt−k)

CUru p

′ur (1− p

′r)

Ur−u

[r−1∏

w=q+1

(1− p′w)Uw ][r∏

w=q+2

P (vw, bw − bw−1)]P (k + u− (l +r∑

t=q+2

vt), br+1 − br)}ej ,

0 ≤ q ≤ r − 2, 1 ≤ r ≤ N,(32)

24

Page 25: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

where

Vs = k + c′r − (l +

s−1∑t=q+1

vt), q + 2 ≤ s ≤ r,

Uw = min(c′w, l +

w∑t=q+1

vt), q + 1 ≤ w ≤ r,

vq+1 = 0.

Note that except the one associated with u, there are (r−q+1) summations on the right-handside.

Figure 7: An example of transitions between two contention slots.

In Fig. 7, vs is the number of packets arriving in [bs−1, bs), and u is the number of

packets that are successfully transmitted in the br-th slot. Since the (n+1)-st departure

occurs in the br-th slot, packets competing in slots bq+1, ..., br−1 shall all fail. Therefore, the

number of codes in the bs-th slot available for competition to the inspected connection is the

minimum of the number of accessible codes c′s and the packets accumulated in the queue at

the beginning of that slot, that is, l +∑s

t=q+1 vt. We need not consider the case in which

the number of packets awaiting in the buffer is larger than k + c′r, where c

′r is the maximal

number of packets that can be transmitted in the br-th slot. Consequently, the upper bound

of vs is k + c′r − (l +

∑s−1t=q+1 vt).

Actual value of Eq. (32) can be obtained by running these (r − q) loops in computerprograms. We can further “simplify” the formulas when l +

∑st=q+1 vt ≥ maxs≤i≤r{c′i} –

although it seems even more sophisticated at the first glance– since in this case all codes in the

25

Page 26: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

bq+1-th,...,br-th slots are for contention. Consequently, we are able to deduce a deterministic

multiplicative term Πs≤i≤r−1(1− p′i)c′i in our formula regardless of the number of arrivals in

[bs − 1, br+1 − 1). In other words, we can eliminate loops associated with vs+1, ..., vr−1 to

significantly reduce computational burden. For simplicity we provide here only the formulas

equivalent to Eq. (32) when k ≥ maxq+1≤i≤r{c′i}.

Pr(Xn+1 = k, Jn+1 = br, In+1 = j|Xn = l ≥ 1, Jn = bq, In = i)

=etri {

Vq+2∑vq+2=Wq+2

c′r∑

u=uq+2

Cc′r

u p′ur (1− p

′r)

c′r−u

(1− p′q+1)Uq+1 [

r−1∏w=q+2

(1− p′w)c′w ]P (vq+2, bq+2 − bq+1)P (k + u− (l +

q+2∑t=q+2

vt), br+1 − bq+2)

+

r−1∑s=q+3

Wq+2−1∑vq+2=0

...

Ws−1−1∑vs−1=0

Vs∑vs=Ws

c′r∑

u=us

Cc′r

u p′ur (1− p

′r)

c′r−u[

s−1∏w=q+1

(1− p′w)Uw ]

r−1∏w=s

(1− p′w)c′w ][

s∏w=q+2

P (vw, bw − bw−1)]P (k + u− (l +s∑

t=q+2

vt), br+1 − bs)

+

Wq+2−1∑vq+2=0

...

Wr−1−1∑vr−1=0

Wr−1∑vr=0

Ur∑u=ur

CUru p

′ur (1− p

′r)

Ur−u

[r−1∏

w=q+1

(1− p′w)Uw ][r∏

w=q+2

P (vw, bw − bw−1)]P (k + u− (l +r∑

t=q+2

vt), br+1 − br)}ej ,

0 ≤ q ≤ r − 2, 1 ≤ r ≤ N, k ≥ maxr≤i≤r

{c′i},

(33)

26

Page 27: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

where

Vs = k + c′r − (l +

s−1∑t=q+1

vt), q + 2 ≤ s ≤ r,

Ws = maxs≤i≤r

{c′i} − (l +s−1∑

t=q+1

vt), q + 2 ≤ s ≤ r,

Uw = min(c′w, l +

w∑t=q+1

vt), q + 1 ≤ w ≤ r,

us = max(1, l +s∑

q+2

vt − k), vq+1 = 0.

A formula similar to Eq. (32) can be derived for transitions bq → a1, 0 ≤ q ≤ N, as

follows.

Pr(Xn+1 = k, Jn+1 = a1, In+1 = j|Xn = l ≥ 1, Jn = bq, In = i)

=etri {

Vq+2∑vq+2=0

...Vs∑

vs=0

...

VN+1∑vN+1=0

[N∏

w=q+1

(1− p′w)Uw ]

[

N+1∏w=q+2

P (vw, bw − bw−1)]P (k +min(0, r1 − (l +N+1∑

t=q+2

vt)), bN+1 − bN )}ej, 0 ≤ q ≤ N − 1,

(34)

where

Vs = k + r1 − (l +s−1∑

t=q+1

vt), q + 2 ≤ s ≤ N + 1,

Uw = min(c′w, l +

w∑t=q+1

vt), q + 1 ≤ w ≤ N,

vq+1 = 0.

(35)

Note that in this transition, k + min(0, r1 − (l +∑N+1t=q+2 vt)) packets shall arrive during

[a1 − 1, a2 − 1) so that k packets are accumulated in the buffer at the beginning of the a2-th

slot. Finally, transition bN → a1 is the same as that to other hybrid slot.

27

Page 28: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

6.1.2 Xn = 0

Now we treat the case Xn = 0. We first introduce the probability functions HCr (k), 1 ≤ r ≤

N, and HR1 (k). H

Cr (k) represents the probability that our chosen connection has no packets

at the beginning of the aM -th slot, but has sent packets in the br-th slot and accumulated k

packets at the beginning of the br+1-th slot of the next frame.

HC1 (k) =

k+c′1∑

v=1

min(v,c′1)∑

u=max(1,v−k)

Cmin(v,c′1)

u p′u1 (1− p

′1)

min(v,c′1)−uP (v, b1 − aM)P (k − v + u, a1), (36)

where v denotes the number of arrivals during [aM − 1, b1 − 1), and u denotes the packetssuccessfully transmitted in the b1-th slot. According to Eq. (36), in the b1-th slot the

number of packets competing for the contention codes is min(v, c′1), and the lower bound of

the number of packets transmitted in this slot is max(1, l− k).The probability functions HC

r (k) and HR1 (k) are similar to Eqs. (32) and (34) when we

make l = 0 and start at the aM -th slot. Henceforth, ∀1 ≤ r ≤ N ,

HCr (k) =

V1∑v1=0

...

Vs∑vs=0

...

Vr∑vr=0

Ur∑u=max(1,

∑r1 vt−k)

CUru p

′ur (1− p

′r)

Ur−uI{r∑

t=1

vt �= 0}

[

r−1∏w=q+1

(1− p′w)Uw ][

r∏w=1

P (vw, bw − bw−1)]P (k + u−r∑

t=1

vt, br+1 − br),(37)

where

Vs = k + c′r −

s−1∑t=q+1

vt, 1 ≤ s ≤ r,

Uw = min(c′w,

w∑t=1

vt), 1 ≤ w ≤ r,(38)

and I{} is the indicator function. Note that here ∑rt=1 vt �= 0, otherwise we would not have

any departure in the br-th slot. If∑r

t=1 vt = 0, then Ur = 0. In this case we can also define

28

Page 29: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

∑Ur

u=max(1,∑r

1 vt−k) F (u) ≡∑0

u=1 F (u) ≡ 0, ∀F (u), to eliminate the indicator function, whichis adopted in the sequel.

We may define the probability function HR1 (k) ≡ HC

N+1(k) to represent the corresponding

function for the a1-th slot. It can be expressed as follows.

HR1 (k) =

V1∑v1=0

...

Vs∑vs=0

...

VN+1∑vN+1=0

[

N∏w=1

(1− p′w)Uw ]

[N+1∏w=1

P (vw, bw − bw−1)]P (k +min(0, r1 −N+1∑t=1

vt), bN+1 − bN ),(39)

where

Vs = k + r1 −s−1∑t=1

vt, 1 ≤ s ≤ N + 1,

Uw = min(c′w,

w∑t=1

vt), 1 ≤ w ≤ N.(40)

With the help of HCr (k) and H

R1 (k), we are ready to obtain the transition probabilities.

Pr(Xn+1 = k, Jn+1 = br, In+1 = j|Xn = 0, Jn = aq, In = i)

= etri [I− e(Q−Λ)bN ]−1e(Q−Λ)(aM−aq+1)HC

r (k)ej, 1 ≤ q ≤M − 1, 1 ≤ r ≤ N + 1,

P r(Xn+1 = k, Jn+1 = br, In+1 = j|Xn = 0, Jn = bq, In = i)

= etri [I− e(Q−Λ)bN ]−1e(Q−Λ)aM +bN−bq+1HC

r (k)ej, 1 ≤ r ≤ N, q ≥ r − 1,

(41)

Pr(Xn+1 = k, Jn+1 = br, In+1 = j|Xn = 0, Jn = bq, In = i)

=etri {

Vq+2∑vq+2=0

...Vs∑

vs=0

...Vr∑

vr=0

Ur∑u=max(1,

∑rq+2 vt−k)

CUru p

′ur (1− p

′r)

Ur−u

[r−1∏

w=q+2

(1− p′w)Uw ][r∏

w=q+2

P (vw, bw − bw−1)]P (k + u−r∑

t=q+2

vt, br+1 − br)

+I− e(Q−Λ)bN ]−1e(Q−Λ)aM +bN−bq+1HCr (k)}ej , 1 ≤ r ≤ N, q ≤ r − 2,

(42)

29

Page 30: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

where

Vs = k + c′r −

s−1∑t=q+1

vt, q + 2 ≤ s ≤ r,

Uw = min(c′w,

w∑t=q+1

vt), q + 2 ≤ w ≤ r.(43)

Pr(Xn+1 = k, Jn+1 = a1, In+1 = j|Xn = l ≥ 1, Jn = bq, In = i)

=etri {

Vq+2∑vq+2=0

...

Vs∑vs=0

...

VN+1∑vN+1=0

[N∏

w=q+2

(1− p′w)Uw ][N+1∏

w=q+2

P (vw, bw − bw−1)]P (k −min(0,N+1∑

t=q+2

vt)− r1), bN+1 − bN )

+ [I− e(Q−Λ)bN ]−1e(Q−Λ)aM +bN−bq+1HR1 (k)}ej, 0 ≤ q ≤ N,

(44)

where

Vs = k + r1 −s−1∑

t=q+2

vt, q + 2 ≤ s ≤ N + 1,

Uw = min(c′w,

w∑t=q+2

vt), q + 2 ≤ w ≤ N.(45)

Note that in both Eqs. (42) and (44) there is an extra term when comparing with earlier

formulas. This additional term is due to that the n-th departure occurs in less than one

frame before the (n+1)-st departure. Since there are no packets seen at the beginning of the

bq+1-th slot, no codes are attempted and the number of packets transmitted in the br-th slot

depends on the arrivals during [bq+1 − 1, br − 1) in Eq. (42) and [bq+1 − 1, a1+ bN − 1) in Eq.(44).

6.1.3 Construction of the M/G/1-type Matrix

The procedure to construct the transition matrix is exactly the same except c = max{max1≤i≤M{ri+ci},max1≤j≤N{c′j}}.

30

Page 31: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

6.2 System Size Distribution at an Arbitrary Time

After obtaining the transition matrix and its steady state probabilities x = [x0 x1 x2 ...],

we may go on to obtain the system size distribution. When t ∈ [aq − 1, aq), 2 ≤ q ≤ M ,

derivations are same as those in Sec. 4, where the case t ∈ [b1 −1, b1) can be regarded as the(M + 1)-st hybrid slot since the slot type is irrelevant to the consideration of system size.

Similarly, the a1-th slot can be regarded as the (N + 1)-st contention slot. We shall in the

sequel focus on the cases t ∈ [br − 1, br), 2 ≤ r ≤ N + 1, where t ∈ [bN+1 − 1, bN+1) refers to

the case t ∈ [a1 − 1, a1).

We first introduce probability functions Hbr(k, t), 2 ≤ r ≤ N + 1. Hbr(k, t) represents

the probability that the connection has no packet right before the aM -th slot, and does not

transmit any packet in the contention slots b1, b2, ..., br−1; and has accumulated k packets

up to time t. Since the definition is similar to HCr (k), we immediately obtain the following

equation:

Hbr(k, t) =

V1(k)∑v1=0

...

Vs(k)∑vs=0

...

Vr−1(k)∑vr−1=0

[r−1∏w=1

(1− p′w)Uw ][r−1∏w=1

P (vw, bw − bw−1)]P (k −r−1∑i=1

vi, t− br−1 + 1),

Vs(k) = k −s−1∑i=1

vi, 1 ≤ s ≤ r − 1,

Uw = min(c′w,

w∑i=1

vi), 1 ≤ w ≤ r − 1, 2 ≤ r ≤ N + 1.

(46)

Here we denote k − ∑s−1i=1 vi by Vs(k) to emphasize that it is a function of k.

If k ≥ max1≤q≤r−1{c′q}, we can rewrite Hbr(k, t) in a manner slightly different from Eq.

(33):

31

Page 32: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

Hbr(k, t) =

V1(k)∑

v1=W′1+1

[

r−1∏w=1

(1− p′w)c′w ]P (v1, b1 − b0)P (k −

1∑i=1

vi), t− b1 + 1)

+r−1∑s=2

W′1∑

v1=0

...

W′s−1∑

vs−1=0

Vs(k)∑

vs=W′s−1+1

[s−1∏w=1

(1− p′w)Uw ]

r−1∏w=s

(1− p′w)c′w ][

s∏w=1

P (vw, bw − bw−1)]P (k −s∑

i=1

vi), t− bs + 1)

+

W′1∑

v1=0

...

W′r−1∑

vr−1=0

[

r−1∏w=1

(1− p′w)Uw ][

r−1∏w=1

P (vw, bw − bw−1)]P (k −r−1∑i=1

vi), t− br−1 + 1)

2 ≤ r ≤ N + 1, k ≥ max1≤q≤r−1

{c′q},

(47)

where

Vs(k) = k −s−1∑i=1

vi, 1 ≤ s ≤ r − 1,

W′s = max

1≤i≤r{c′i} −

s−1∑i=1

vi, 1 ≤ s ≤ r − 1,

Uw = min(c′w,

w∑i=1

vi), 1 ≤ w ≤ r − 1.

(48)

Note that when vs ≥ W′s + 1,

∑si=1 vi ≥ max1≤i≤r{c′i} + 1 > maxs≤i≤r{c′i}, and conse-

quently all the contention codes in slots bs, ..., br−1 will be attempted. We actually changed

the subscripts and superscripts of those summations associated with vs, 1 ≤ s ≤ r− 1, sincethis simplifies the expression of the PGF Hbr(z, t) =

∑∞k=0Hbr(k, t)z

k.

32

Page 33: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

Hbr(z, t) =

cmax∑k=0

V1(k)∑v1=0

...

Vs(k)∑vs=0

...

Vr−1(k)∑vr−1=0

[

r−1∏w=1

(1− p′w)Uw ][

r−1∏w=1

P (vw, bw − bw−1)]P (k −r−1∑i=1

vi, t− br−1 + 1)zk

+∞∑

k=cmax+1

V1(k)∑

v1=W′1+1

[r−1∏w=1

(1− p′w)c′w ]P (v1, b1 − b0)zv1P (k −

1∑i=1

vi, t− b1 + 1)zk−vi

+∞∑

k=cmax+1

r−1∑s=2

W′1∑

v1=0

...

W′s−1∑

vs−1=0

Vs(k)∑vs=W ′

s+1

[s−1∏w=1

(1− p′w)Uw ]

[

r−1∏w=s

(1− p′w)c′w ][

s∏w=1

P (vw, bw − bw−1)]z∑s

i=1 viP (k −s∑

i=1

vi, t− bs + 1)zk−∑si=1 vi

+∞∑

k=cmax+1

W′1−1∑

v1=0

...

W′r−1−1∑

vr−1=0

[r−1∏w=1

(1− p′w)Uw ]

[r−1∏w=1

P (vw, bw − bw−1)]z∑r−1

i=1 viP (k −r−1∑i=1

vi, t− br−1 + 1)zk−∑r−1

i=1 vi ,

(49)

where

Vs(k) = k −s−1∑i=0

vi, 1 ≤ s ≤ r − 1,

W′s = max

s≤i≤r{c′i} −

s−1∑i=0

vi, 1 ≤ s ≤ r − 1,

Uw = min(c′w,

w∑i=1

vi), 1 ≤ w ≤ r − 1,

cmax = max1≤q≤r−1

{c′q}, v0 = 0.

(50)

Observing that Vs(k) is a function of k, whereas Us and W′s are both independent of k, we

can further simplify this PGF by exchanging the orders of summations. This order exchange

is valid since all terms involved are non-negative.

33

Page 34: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

Hbr(z, t) =

cmax∑k=0

V1(k)∑v1=0

...

Vs(k)∑vs=0

...

Vr−1(k)∑vr−1=0

[

r−1∏w=1

(1− p′w)Uw ][

r−1∏w=1

P (vw, bw − bw−1)]P (k −r−1∑i=1

vi, t− br−1 + 1)zk

+[

r−1∏w=1

(1− p′w)c′w ][P (z, b1 − b0)−

W′1∑

v1=0

P (v1, b1 − b0)zv1 ]P (z, t− b1 + 1)

+

r−1∑s=2

W′1∑

v1=0

...

W′s−1∑

vs−1=0

[

s−1∏w=1

(1− p′w)Uw ][

r−1∏w=s

(1− p′w)c′w ]

[

s−1∏w=1

P (vw, bw − bw−1)zvi ][P (z, bs − bs−1)−

W′s∑

vs=0

P (vs, bs − bs−1)zvs ]P (z, t− bs + 1)

+

W′1∑

v1=0

...

W′r−1∑

vr−1=0

[r−1∏w=1

(1− p′w)Uw ]

[

r−1∏w=1

P (vw, bw − bw−1)]z∑r−1

i=1 vi [P (z, t− br−1 + 1)−cmax−

∑r−1i=1 vi∑

k=0

P (k, t− br−1 + 1)zk],

(51)

Now we derive Q∗t (z), the PGF of the system size for at t ∈ [br − 1, br).

Q∗t (z) =

r−1∑q=0

Ibq

br(z, t) +

M−1∑i=0

πi0[I− e−(Q−Λ)bN ]−1e(Q−Λ)(aM−ai+1)Hbr(z, t)

+N∑

i=0

πM+i0 [I− e−(Q−Λ)bN ]−1e(Q−Λ)(aM +bN−bi+1)Hbr(z, t),

(52)

34

Page 35: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

where

Ibr−1

br(z, t) =πM+r−1(z)P (z, t− br + 1),

Ibq

br(z, t) =

cmax∑k=0

Vq+1(k)∑vq+1=0

...

Vs(k)∑vs=0

...

Vr−1(k)∑vr−1=0

[

r−1∏w=1

(1− p′w)Uw ]πM+qvq+1[

r−1∏w=q+2

P (vw, bw − bw−1)]P (k −r−1∑i=1

vi, t− br−1 + 1)zk

+[r−1∏w=1

(1− p′w)c′w ][πM+q(z)−

W′q+1∑

vq+1=0

πM+qvq+1

zvq+1 ]P (z, t− bq+1 + 1)

+

r−1∑s=q+2

W′q+1∑

vq+1=0

...

W′s−1∑

vs−1=0

[

s−1∏w=q+1

(1− p′w)Uw ][

r−1∏w=s

(1− p′w)c′w ]πM+q

vq+1zvq+1

[

s−1∏w=q+2

P (vw, bw − bw−1)zvi ][P (z, bs − bs−1)−

W′s∑

vs=0

P (vs, bs − bs−1)zvs ]P (z, t− bs + 1)

+

W′q+1∑

vq+1=0

...

W′r−1∑

vr−1=0

[r−1∏w=1

(1− p′w)Uw ]πM+qvq+1

zvq+1 [r−1∏

w=q+2

P (vw, bw − bw−1)]

z∑r−1

i=q+2 vi [P (z, t− br−1 + 1)−cmax−

∑r−1i=q+1 vi∑

k=0

P (k, t− br−1 + 1)zk],

q ≤ r − 2.(53)

The first term of Q∗t (z) tells that the latest departure occurs in slots bq+1, ..., br−1 in the

current frame. In this case, we have to discuss case by case the points of arrivals between the

latest point to be observed and time t since they affect the number of contention codes that

are attempted in slots b0, ..., br−1. Consequently, Ibq

br(z, t)’s appearance is similar to Hbr(z, t)

except now we consider only these (r − q − 1) slots and the first term becomes πM+qvq+1, the

steady-state probability of having vq+1 packets at the beginning of the bq+1-th slot given that

a departure occurs at the bq-th slot.

The second term represents the event that the latest departure occurs in a hybrid slot

35

Page 36: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

except the aM -th. The third term represents the probability that the latest departure occurs

in slots b0, ..., bN in at least one frame prior to time t.

As before, we use Q∗t (z) where t ∈ [br −1, br) to obtain the system size distribution when

t ∈ (br−1, br−1). After routine differentiations, we compute new termsH ′br(1, t), H”

br(1, t), (Ibr

bq)′(1, t)

and (Ibr

bq)”(1, t) and then obtain the moments of system size.

7 Generalization and Limitations

In this section, we first describe how to integrate the results derived in the previous three

sections to deal with a generally-structured frame-based model, provided that the system

has at least one reserved code for the inspected connection. We then explain why our

methodology does not work for a frame-based pure-contention system. The complexity of

our analysis is also briefly discussed.

7.1 A Model with Alternating Hybrid and Contention Slots

In Sec. 5, we treated a system in which each contention slot contains only one code, and in

Sec. 6 we switched to a system having a succession of multicode slots. Along with a system

containing only hybrid slots, this section provides a procedure to combine characteristics for

a hybrid system with arbitrary slot and code allocations. A hybrid TDMA/CDMA system

may have successive contention slots in which multiple codes are accessible by the inspected

connection, hybrid slots may or may not contain contention codes, and most importantly,

hybrid and contention slots may arbitrarily appear within a frame.

An example of a generally-structured frame is depicted in Fig. 8, where four hybrid

slots and five contention slots exist in a frame for our inspected connection. Since now slot

allocation is made general, the b5-th slot is no longer the last of a frame.

36

Page 37: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

.... .... .... .... .... .... .... .... .... ....HH H HC C C C C

a 1 a 2 a 3 a 4b 1 b2 b3 b4 b5

FrameBoundary

FrameBoundary

Figure 8: An example of arbitrarily arranged hybrid and contention slots.

7.1.1 State Transition Matrix

We regard a generally-structured system as a cascade of the models we discussed above.

While deriving the transition matrix or the system size distribution, we first classify a slot’s

type according to the number of contention slots n sitting between the slot and the previous

hybrid slot.

Let us illustrate the classification through the aid of Fig. 8. Suppose that c′1 = c

′2 = 1

but c′4 �= c′5 in Fig. 8. First we classify the types of these four hybrid slots. Referring to

Fig. 8, if the accessible slot prior to our observed hybrid slot is hybrid, i.e., n = 0, then

its derivations should follow Sec. 4. The a2-th slot in Fig. 8 belongs to this category. If

n = 1, e.g., the a3-th slot in Fig. 8, we should follow the treatment of the a1-th slot in Sec.

6, but the derivations are much simpler since the most difficult parts in , HCr (k), Hbr(z, t),

and Ibrbq(z, t) disappear when N = 1. Interested readers are encouraged to derive their own

closed-form formulas of this case. If n > 1 and contention slots prior to this slot all contain

single accessible code, e.g., the a1-th slot in Fig. 8, then derivations in Sec. 5 shall be applied

with only slight modification. Other cases such as the a4-th slot in Fig. 8 are referred to

the general procedure in Sec. 6. Note that n ≤ M +N , where M +N is the total numberof accessible slots in a frame. The inequality holds since we assume that the system has

at least one reserved code, and hence at least one hybrid slot. The cases associated with a

contention slot can also be dealt analogously.

37

Page 38: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

The procedure to construct the state transition matrix after having obtained the transi-

tion probabilities is the same as Sec. 6 and the M/G/1 algorithm can be applied to compute

its stationary probability vector, where c = max(max1≤q≤M(rq + cq),max1≤s≤N c′s).

7.1.2 System Size Distribution at an Arbitrary Time

We again combine the derivations in the previous three sections to obtain the system size

distribution for a generally-structured system. First we consider the case that time t belongs

to an accessible slot. The classification is based on the number of consecutive contention

slots prior to the slot to which t belongs. Recall that mathematical derivation is independent

of slot type, we treat hybrid and contention slots in the same manner if they both have the

same number of prior consecutive contention slots.

Referring to Fig. 8, n = 0 for slots b1, b3, b4, and a2. Therefore, we apply to these cases

the derivations in Sec. 4. Since both the a3-th and b5-th slots in Fig. 8 have only one prior

contention slot, the associated system sizes can be derived in the same manner as the b2-th

slot in Sec. 6. Similarly, we obtain system size distribution for t ∈ [a4 − 1, a4) in Fig. 8 as if

we were dealing with the b3-th slot in Sec. 6. Finally, we apply the derivations in Sec. 5 for

the cases t ∈ [a1 − 1, a1) and t ∈ [b2 − 1, b2) in Fig. 8.For t belonging to inaccessible slots, the procedure in obtaining the system size distribu-

tion from the established distribution for accessible slots still works here.

7.2 Limitations

We first illustrate the difficulties we face while applying our methodology to a model with

pure contention scheme. The remaining section is given to discuss the complexity of our

analysis.

38

Page 39: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

7.2.1 A Model with Pure Contention Scheme

Now, what if the number of consecutive contention slots prior to a chosen slot approaches

infinity? In other words, what if the system does not have any hybrid slot, i.e., M = 0? In

the sequel we discuss this pure contention model assuming that the number of contention

slots N > 1. The case M = 0, N = 1 can be regarded as a slotted system without the frame

structure, whose performance has been thoroughly investigated in the literature.

Transitions bq → br, where q ≥ r−1, can occur across several frame boundaries regardlessof Xn’s value. They could be regarded as a geometric series and be rewritten as a closed-

form formula such as Eq. (14) if there were hybrid slots. In a model with pure contention

scheme, however, the infinitely many transition probabilities do not follow the geometric

series pattern, and therefore it is invalid to first study the probability function Gq(k) and

then represent all transition probabilities as Gq(k) multiplied by a constant and e(Q−Λ)ubN ,

where u is an integer. The proposed method is fairly inefficient in this case due to summations

involving infinitely many terms.

However, tail terms of this series will vanish as the difference between two consecutive

departures approaches infinity, because the events happen only when no arrival occurs during

a long period of time. In practice, one can compute finite terms to get good approximation

of performance measures for this case.

7.2.2 Complexity Issues

The major portion of computation time is spent in deriving the moments of the system

size. Reexamine these equations we use to obtain the moments in Secs. 4, 5, and 6: Eq.

(15), Eqs. (30), and Eqs. (51)-(52), we find that the complexity of these three models

are respectively O(M), O(N × (M + N)), and O(N × (M + N) × cNmax), where cmax =

max(max1≤q≤M(rq + cq),max1≤s≤N c′s). Note that in the latter two models, although there

are also derivations for the case when t ∈ [aq − 1, aq), the complexity of these equations is

39

Page 40: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

O(M +N) and is dominated by the cases when t belongs to a contention slot.

It immediately follows that the models in Sec. 4 and Sec. 5 can be solved in polynomial

time, whereas complexity of the third grows exponentially in N, the number of contention

slots. In words, our analysis may be inefficient when the number of consecutive contention

slots becomes large. In contrast, the number of hybrid slots does not play a dominant role

regarding complexity. Note also that if we regard the problem in Sec. 5 as a special case of

Sec. 6, the problem is polynomially solvable because cmax = 1 and therefore O(N × (M +N)× cNmax) = O(N × (M +N)), which coincides with the result in the above.While reading Sec. 7.1, readers may wonder why we do not regard accessible codes as

contention codes when the success probability that the inspected connection gets to use a

reserved code is 1 and use the methodology developed in Sec. 6, provided that we maintain at

least one hybrid slot. The argument is valid, however to follow that methodology will result

in significant computational burden due to the NP-complete nature of Sec. 6. Henceforth,

it is more efficient to adopt the cascade structure and make classifications.

8 Numerical Results and Discussions

Although we have no intention to less emphasize the importance of the implications conveyed

in the numerical examples provided in this section, the prime purpose is to demonstrate

through these numerical examples that mathematical derivations throughout this paper are

implementable in computer programs. We consider as our traffic source a two-state MMPP

with the following parameters:

Q =

−0.05 0.05

0.05 −0.05

, and Λ =

0.2ρ 0

0 0.1ρ

, (54)

where ρ serves as a measure of the arrival rate since the average arrival rate of this process

is λ = π[λ1 λ2]tr = [0.5 0.5][0.2ρ 0.1ρ]tr = 0.15ρ.

40

Page 41: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

In the discussions that follow, we denote slot allocation, reserved code allocation, con-

tention code allocation, and contention probability allocation via row vectors a, r, c, and p,

respectively. If there is no contention code in a hybrid slot aq, i.e., this hybrid slot contains

only reserved codes, we let pq = 1.

In Figs. 9- 10 we consider a system with frame length 48 slots, 6 of which are assigned to

the inspected connection. These six slots all contain reserved codes and are placed uniformly

in time, i.e., a = [8 16 24 32 40 48]. Since all accessible slots are hybrid, we apply results

derived in Sec. 4 to obtain performance measures. In Figs. 9 and 10 we plot the mean and

standard deviation of system size respectively for two different contention probabilities. In

both systems, two reserved codes and one contention code are granted in each accessible slot:

denoted by r = [2 2 2 2 2 2] and c = [1 1 1 1 1 1]. In one system, the contention probabilities

are all 0.5, i.e., p = [0.5 0.5 0.5 0.5 0.5 0.5]. In the other, we use p = [0.2 0.8 0.2 0.8 0.2 0.8].

It follows immediately that the effective number of codes invested per frame in both systems

is 6× (2 + 1× 0.5) = 3× [(2 + 1× 0.2) + (2 + 1× 0.8)] = 15. Since the mean system sizesand their standard deviations almost coincide in these figures, we conclude that queueing

performance is not sensitive to the allocation of probabilities, provided that the effective

number of codes invested is fixed.

In Figs. 11- 12 we switch our attention to the effect of code allocation policies on TDMA

systems. In these figures the frame length is fixed at 40 slots, 3 of which are reserved for the

inspected connection and 7 are for contention slots. These slots are uniformly placed, i.e.,

a = [4 8 12 16 20 24 28 32 36 40]. The contention probability for one system is 0.3 and 0.5

for the other. The system performances in these figures can be obtained from equations in

Sec. 5. According to these figures, the effect of contention probability on the mean system

size is insignificant, regardless of the observation point in a frame.

In Figs. 13- 14, we compare two different code allocation policies. In both systems,

there are 12 reserved codes and 6 contention codes in a frame of 48 slots, and the contention

41

Page 42: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

0 . 1 0 . 2 0 . 3 0 . 4 0 . 5 0 . 6 0 . 7 0 . 80

0 . 5

1

1 . 5

2

2 . 5

3

3 . 5

4

ρ

Me

an

Sy

ste

m S

ize 6 h y b r i d s l o t s i n a f r a m e o f 4 8 s l o t s

t h e d i s t a n c e b e t w e e n t w o a c c e s s i b l e s l o t s : 8

r = [ 2 2 2 2 2 2 ]

c = [ 1 1 1 1 1 1 ]

Λ= [ 0 . 2ρ 0 ; 0 0 . 1ρ]

p = [ 0 . 5 0 . 5 0 . 5 0 . 5 0 . 5 0 . 5 ]p = [ 0 . 2 0 . 8 0 . 2 0 . 8 0 . 2 0 . 8 ]

Figure 9: Mean system size vs traffic load for two different probability allocation policies.

0 . 1 0 . 2 0 . 3 0 . 4 0 . 5 0 . 6 0 . 7 0 . 80

2

4

6

8

1 0

1 2

ρ

Sta

nd

ar

d D

ev

iati

on

of

the

Sy

ste

m S

ize

6 h y b r i d s l o t s i n a f r a m e o f 4 8 s l o t s

t h e d i s t a n c e b e t w e e n t w o a c c e s s i b l e s l o t s : 8

r = [ 2 2 2 2 2 2 ]

c = [ 1 1 1 1 1 1 ]

Λ= [ 0 . 2ρ 0 ; 0 0 . 1ρ]

p = [ 0 . 5 0 . 5 0 . 5 0 . 5 0 . 5 0 . 5 ]p = [ 0 . 2 0 . 8 0 . 2 0 . 8 0 . 2 0 . 8 ]

Figure 10: Standard deviation vs traffic load for two different probability allocation policies.

42

Page 43: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

0 . 1 0 . 1 2 0 . 1 4 0 . 1 6 0 . 1 8 0 . 2 0 . 2 2 0 . 2 4 0 . 2 60

0 . 5

1

1 . 5

2

2 . 5

3

t r a f f i c l o a d r

me

an

qu

eu

e l

en

gth

(r

es

er

ve

d s

lot)

M e a n q u e u e l e n g t h ( i n a r e s e r v e d s l o t ) v s T r a f f i c l o a d

p = 0 . 5 p = 0 . 3

T D M A s y s t e m w i t h r e s e r v e d s l o t s a n d c o n t e n t i o n o n e s

3 r e s e r v e d s l o t s a n d 7 c o n t e n t i o n o n e s w i t h f r a m e l e n g t h = 4 0

Figure 11: Mean system size vs traffic load within a hybrid slot.

0 . 1 0 . 1 2 0 . 1 4 0 . 1 6 0 . 1 8 0 . 2 0 . 2 2 0 . 2 4 0 . 2 60

0 . 5

1

1 . 5

2

2 . 5

3

3 . 5

4

4 . 5

me

an

qu

eu

e l

en

gth

(c

on

ten

tio

n s

lot)

M e a n q u e u e l e n g t h ( i n a c o n t e n t i o n s l o t ) v s T r a f f i c l o a d

p = 0 . 5 p = 0 . 3

T D M A s y s t e m w i t h r e s e r v e d s l o t s a n d c o n t e n t i o n o n e s

3 r e s e r v e d s l o t s a n d 7 c o n t e n t i o n o n e s w i t h f r a m e l e n g t h = 4 0

t r a f f i c l o a d r

Figure 12: Mean system size vs traffic load within a contention slot.

43

Page 44: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

probability is 0.5. In one system, two reserved codes and one contention code are assigned to

a slot, and therefore six slots are accessible to the inspected connection. We further assume

that slot allocation vector a = [8 16 24 32 40 48]. In the other, two reserved codes are

assigned to each hybrid slot, and three contention codes are allocated to each contention

slot. In other words, there are 6 hybrid slots and 2 contention slots in a frame in this case.

Let slot allocation for this latter case be a = [6 12 18 24 30 36 42 48]. Note that the second

system can be analyzed by the mathematics in Sec. 6. We observe that the second system

provides better performance in all parameters because the distance between two consecutive

accessible slots is 6 (slots) in the second system, but 8 in the first. Code allocation policy

does affect system performance.

0 . 1 0 . 2 0 . 3 0 . 4 0 . 5 0 . 6 0 . 7 0 . 80

0 . 5

1

1 . 5

2

2 . 5

3

3 . 5

4

ρ

Me

an

Sy

ste

m S

ize

1 2 h y b r i d s l o t s a n d 6 c o n t e n t i o n s l o t s

F r a m e l e n g t h : 4 8 s l o t s

C o n t e n t i o n p r o b a bi l i t y = 0 . 5 f o r a l l co d e s

Λ= [ 0 . 2ρ 0 ; 0 0 . 1ρ]

r = [ 2 2 2 2 2 2 ] , c = [ 1 1 1 1 1 1 ]r = [ 2 2 2 2 2 2 0 0 ] , c = [ 0 0 0 0 0 0 3 3 ]

Figure 13: Mean system size vs traffic load for two different contention code policies.

Finally, we present the comparisons in mean system size, standard deviation, and corre-

lation of coefficient between two generally-structured systems subject to different contention

probabilities. There are four hybrid slots and five contention slots in a frame, and these slots

are alternatingly placed as Fig. 8 in Sec. 7.1. The distance between each two consecutive ac-

cessible slots is 6 (slots), i.e., a = [6 12 18 24 30 36 42 48 54]. The probability vectors of these

44

Page 45: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

0 . 1 0 . 2 0 . 3 0 . 4 0 . 5 0 . 6 0 . 7 0 . 80 . 4

0 . 5

0 . 6

0 . 7

0 . 8

0 . 9

1

1 . 1

1 . 2

1 . 3

ρ

Co

rr

ela

tio

n o

f C

oe

ffic

ien

t o

f th

e S

ys

tem

Siz

e

1 2 h y b r i d s l o t s a n d 6 c o n t e n t i o n s l o t s

F r a m e l e n g t h : 4 8 s l o t s

C o n t e n t i o n p r o b a bi l i t y = 0 . 5 f o r a l l co d e s

Λ= [ 0 . 2ρ 0 ; 0 0 . 1ρ]

r = [ 2 2 2 2 2 2 ] , c = [ 1 1 1 1 1 1 ]r = [ 2 2 2 2 2 2 0 0 ] , c = [ 0 0 0 0 0 0 3 3 ]

Figure 14: Correlation of Coefficient vs traffic load for two different contention code policies.

two systems are respectively p = [0.5 0.5 1 1 0.5 1 0.5 0.5 1] and p = [0.2 0.8 1 1 0.5 1 0.2 0.7 1].

Note that the effective numbers of codes in these two systems are equal. Comparing with

Figs. 9- 10, the difference becomes bigger.

9 Conclusion

Through the help of M/G/1 type algorithm, we have obtained the system size distribution of

a selected connection under a TDMA/CDMA protocol with hybrid reservation/contention

scheme. The protocol is made general to accommodate arbitrary slot and code allocations,

assuming that the inspected connection is assigned at least one reserved code. Although

throughout the paper we deal with an m-state MMPP input, the derivations can be easily

extended to the BMAP family in the same manner as Chen and Chang[5]. Our analysis

provides exact rather than approximate result. The results we have developed here are

useful in analyzing many existing protocols as long as they have a frame structure.

45

Page 46: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

0 . 1 0 . 1 5 0 . 2 0 . 2 5 0 . 3 0 . 3 5 0 . 4 0 . 4 5 0 . 50

1

2

3

4

5

6

ρ

Me

an

Sy

ste

m S

ize 4 h y b r i d s l o t s a n d 5 c o n t e n t i o n s l o t s

F r a m e l e n g t h : 5 4 s l o t s

c = [ 1 1 0 0 3 0 2 3 0 ]

r = [ 0 0 1 2 0 2 0 0 1 ]

Λ= [ 0 . 2ρ 0 ; 0 0 . 1ρ]

p = [ 0 . 5 0 . 5 1 1 0 . 5 1 0 . 5 0 . 5 1 ]p = [ 0 . 2 0 . 8 1 1 0 . 5 1 0 . 2 0 . 7 1 ]

Figure 15: Mean system size vs traffic load for two probability policies in general models.

0 . 1 0 . 1 5 0 . 2 0 . 2 5 0 . 3 0 . 3 5 0 . 4 0 . 4 5 0 . 50 . 4

0 . 4 5

0 . 5

0 . 5 5

0 . 6

0 . 6 5

0 . 7

0 . 7 5

0 . 8

0 . 8 5

0 . 9

ρ

Co

rr

ela

tio

n o

f C

oe

ffic

ien

t o

f th

e S

ys

tem

Siz

e

4 h y b r i d s l o t s a n d 5 c o n t e n t i o n s l o t s

F r a m e l e n g t h : 5 4 s l o t s

c = [ 1 1 0 0 3 0 2 3 0 ]

r = [ 0 0 1 2 0 2 0 0 1 ]

Λ= [ 0 . 2ρ 0 ; 0 0 . 1ρ]

p = [ 0 . 5 0 . 5 1 1 0 . 5 1 0 . 5 0 . 5 1 ]p = [ 0 . 2 0 . 8 1 1 0 . 5 1 0 . 2 0 . 7 1 ]

Figure 16: σµvs traffic load for two probability policies in general models.

46

Page 47: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

References

[1] F. Babich, Analysis of frame-based reservation random access protocols for microcellular

radio networks, IEEE Trans. Vehic. Technol. 46 (1990) 408-421.

[2] A. E. Brand and A. H. Aghvano, Performance of the joint CDMA/PRMA protocol for

voice transmission in a cellular environment, Proc. IEEE GLOBECOM (1995) 1060-

1064.

[3] H. Bruneel, Message delay in TDMA channels with contiguous output, IEEE Trans.

Commun. 34 (1986) 681-684.

[4] J. -F. Chang, A multibeam packet satellite using random access technique, IEEE Trans.

Commun. 31 (1983) 1143-1154.

[5] Y. -J. Chen and J. -F. Chang, Per connection delay analysis of a frame-based

TDMA/CDMA MAC protocol, Performance Evaluation, forthcoming.

[6] W. Fischer and K. Meier-Hellstern, The Markov-modulated Poisson process (MMPP)

cookbook, Performance Evaluation 18 (1992) 149-171.

[7] D. J. Goodman and S. X. Wei, Efficiency of packet reservation multiple access, IEEE

Trans. Vehic. Technol. 40 (1991) 170-176.

[8] D. J. Goodman, R. A. Valenzuela, K. T. Gayliard, and B. Ramamurthi, Packet reser-

vation multiple access for local wireless communications, IEEE Trans. Vehic. Technol.

37 (1989) 885-890.

[9] K. -T. Ko and B. R. Davis, Delay analysis for a TDMA channel with contiguous output

and Poisson message arrival, IEEE Trans. Commun. 32 (1984) 707-709.

[10] S. S. Lam, Delay analysis of a time division multiple access(TDMA) channel, IEEE

Trans. Commun. 25 (1977) 1489-1494.

47

Page 48: New York Universitypages.stern.nyu.edu/~ychen0/paper/cont040224_QS.pdf · PerConnectionPerformanceAnalysisofa Frame-basedTDMA/CDMAMACProtocol ContainingbothReservationandContentionSlots∗

[11] D. M. Lucatoni, New results on the single server queue with a batch Markovian arrival

process, Commun. Statis.-Stochastic Models 7 (1991) 1-46.

[12] S. Nanda, D. J. Goodman, and U. Timor, Performance of PRMA: a packet voice protocol

for cellular systems, IEEE Trans. Veh. Tech. 40 (1991) 584-598.

[13] M. F. Neuts, Structured Stochastic Matrices of M/G/1 Type And Their Applications,

New York; Marcel Dekker, 1989.

[14] I. Rubin, Message delays in FDMA and TDMA communication channels, IEEE Trans.

Commun. 27 (1979) 769-777.

[15] I. Rubin and Z. Zhang, Message delay and queue-size analysis for circuit-switched

TDMA systems, IEEE Trans. Commun. 39 (1991) 905-914.

[16] I. Rubin and Z. Zhang, Message delay analysis for TDMA schemes using contiguous-slot

assignments, IEEE Trans. Commun. 40 (1992) 730-737.

[17] C. S. Wu, and G. K. Ma, Performance of packet reservation MAC protocols for wireless

networks, Proc. IEEE VTC (1998) 2537-2541.

48


Recommended