IEEE COMMUNICATIONS LETTERS, VOL. 13, NO.

6, JUNE 2009 393

VoIP Capacity Analysis in Cognitive Radio System

Howon Lee and Dong-Ho Cho

Abstract—In this letter, we analyze a voice over IP (VoIP) p10

capacity in a cognitive radio system. We formulate the system
as a two-dimensional discrete time Markov chain (DTMC).
The VoIP traffic and wireless channel in the cognitive radio
system are described as a Markov modulated Poisson process
p00 p11
(Unoccupied) (Occupied)
(MMPP) model and a Markov channel model, respectively. We
demonstrate various numerical and simulation results, such as
packet dropping probability and VoIP capacity.
p01
Index Terms—Cognitive radio, VoIP service, Markov modu-
lated Poisson process, discrete-time Markov chain.
Fig. 1. Markov channel model for cognitive radio system.

I. I NTRODUCTION

C OGNITIVE radio is a promising and challengeable tech- off) in the simple on-off model can be obtained by pof f =
nology for maximizing radio resource utilization in a β −1 /(α−1 +β −1 ) , and pon = 1−pof f . Here, 1/α and 1/β are
future wireless communication system because conventional the mean values of the on and off periods which are distributed
systems exploit most available frequency bands for wireless exponentially. Furthermore, all the traffic generated by the
communications and these frequency bands are not always VoIP users in the cell can be modeled as a two-state MMPP
fully utilized in general [1]. In addition, Voice over IP (VoIP) model [6]. This MMPP model is highly suitable for formulat-
will be an essential service in the future because, through ing the multi-user VoIP traffics because the MMPP captures
VoIP technology, wireless users can utilize voice services more the interframe dependency between consecutive frames. Here,
cheaply. Therefore, supporting as many voice users as possible the transition rate matrix (R) and the Poisson arrival rate
while using limited radio resources is a very important issue matrix (Λ) of the MMPP can be expressed as follows:
that could be a key to the success of the future systems [2][3].    
−r1 r1 λ1 0
In [4], the authors proposed a discrete-time Markov chain R= , Λ= . (1)
r2 −r2 0 λ2
(DTMC) framework based on a Markov modulated Poisson
process (MMPP) traffic model to analyze VoIP performance. In order to utilize the MMPP model, we should match the
However, they did not consider the cognitive radio system. MMPP parameters (r1 , r2 , λ1 and λ2 ) in Eqn. (1) with the pa-
Also, in [5], Q. Bi et al. analyzed the VoIP capacity of 1xEV- rameters of the simple on-off model (α and β). We here adopt
DO system. In this letter, we analyze VoIP capacity in a the IDC (index of dispersion for counts) matching technique
cognitive radio system through a queuing model based on the because it yields adequate results for the matching of param-
MMPP traffic model and a Markov channel model. Here, VoIP eters and has appropriate computation complexity compared
packets can be transmitted when the wireless channel is not with other matching techniques [4]. Then, r1 , r2 , λ1 , and λ2 in
utilized by a primary user. To the best of our knowledge, the 2(λ2 −λavg )(λavg −λ1 )2
Eqn. (1) can be calculated by r1 = (λ2 −λ 1 )λavg (IDC(∞)−1)
,
VoIP capacity analysis has not been studied yet in the cognitive 2(λ2 −λavg )2 (λavg −λ1 )
 Nact avg
i·πi
radio system. r2 = (λ2 −λ1 )λavg (IDC(∞)−1) , λ1 = A· i=0
 Nact avg
, and
j=0 πj
N
The remainder of this letter is organized as follows. In i=N +1 i·πi
Section II, we present the VoIP traffic model based on the λ2 = A ·  N act avg π . Here, N is the total number of
j=N act avg +1 j
MMPP and the Markov channel model for cognitive radio VoIP users in the system, and A is the emission rate in the on-
1
system. In Section III, we propose a two-dimensional DTMC state (A = Tbasic ). Tbasic is a frame duration of voice codec,
framework considering multi-user VoIP queuing and wireless and the average arrival rate is λavg = N × A × pon . Also,
channel occupancy variations. In Section IV, we demonstrate the average number of active users is Nact avg = N × pon ,
numerical and simulation results and conclude this letter. and the steady-state probability of an one-dimensional Markov
chain when considering N independent simple on-off voice
II. VO IP T RAFFIC AND C HANNEL M ODELING users can be calculated by πi =N Ci · pion · (1 − pon )N −i .
Moreover, IDC(∞) is given in [6].
1) VoIP Traffic Modeling: In general, we can formulate
the VoIP traffic of a single user as a simple on-off model. 2) Channel Modeling in Cognitive Radio System: In a
The probability that the status of the users is inactive (= cognitive radio system, a wireless channel can be modeled
as a two-state Markov process, as shown in Fig. 1 [7]. An
Manuscript received December 23, 2008. The associate editor coordinating occupied state means that the wireless channel is utilized by a
the review of this letter and approving it for publication was H.-H. Chen. primary user. Given that the channel status is ‘Occupied’, the
The authors are with KAIST, Republic of Korea (e-mail:
hwlee@comis.kaist.ac.kr). cognitive user cannot use the channel. In this letter, we assume
Digital Object Identifier 10.1109/LCOMM.2009.082189 that there are ‘M ’ wireless channels. Then, the transition
1089-7798/09$25.00 
c 2009 IEEE

Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on May 31,2010 at 14:02:45 UTC from IEEE Xplore. Restrictions apply.
 394 IEEE COMMUNICATIONS LETTERS, VOL. 13, NO. 6, JUNE 2009

⎧ ⎫
M ⎨
 min(m,M −n)


⎬
m   M − m y  M −m−y 
B(i,m),(j,n) = U · D(j −max(i−k, 0)) · Ps (k | xc = m) · px01 pm−x p10 p11 . (5)
⎩ x 00 y 
⎭
k=0 x =max(0,m−n)

probability (Pm,n ) that there are m unoccupied channels (xc ) element of D(m) means the probability that m VoIP packets
in the current frame, and there will be n unoccupied channels arrive at the BS during the MAC frame duration (Tf ) in each
in the next frame can be represented by phase of the two-state MMPP.
Through the transition matrix (P) in Eqn. (3), we can
Pm,n =
obtain the steady-state probability matrix (πp ) for our two-

min(m,M−n)

m   M − m y M−m−y dimensional DTMC model, which can be calculated by solving

px01 pm−x p10 p11 . (2) equations πp · P = π p and π p · 1 = 1. Here, πp is ‘1’ by
x 00 y
x =max(0,m−n)
‘2 × (M + 1) × (Lmax + 1)’ matrix. Therefore, the probability
Here, y  = n− m+ x , where x and y  denote the numbers of that k VoIP packets are queued in the BS can be expressed as
channels whose status are altered from ‘Unoccupied’ to ‘Oc- 2(M+1)−1
π(k) = l=0 πp (2 · (M + 1) · k + l). (6)
cupied’ and from ‘Occupied’ to ‘Unoccupied’, respectively.
Through Eqn. (6), the ‘1’ by ‘(Lmax + 1)’ steady-state proba-
III. S YSTEM M ODELING bility matrix of the DTMC model can be obtained as follows:
1) Assumptions: Perfect information for channel occupancy π = [π(0) π(1) π(2) · · · π(Lmax )]. Therefore, by using
of primary users is available in the base station (BS), and this steady-state probability, we can present various analytical
one VoIP packet can be transmitted through one unoccupied results, such as the average number of arrived packets, the
wireless channel. If all the channels are used by the primary average queue-length, the average number of serviced VoIP
users, the VoIP packets cannot be sent. Also, we assume that packets, the average VoIP throughput, and the packet dropping
there are no packet retransmissions. That is, unreceived and probability. First, the average queue-length (Lavg ) and the
destroyed packets are not sent again. average arrival rate (ρ) can be calculated by
2) System Modeling: We can formulate a multi-user queu-  max
Lavg = L i · π(k). (7)
ing model for VoIP services in a cognitive radio system as i=0 
N ×Amax
a two-dimensional DTMC. In other words, we can analyze ρ=s· m=0 m · D(m) · 1. (8)
the behavior of queuing packets in the BS with this DTMC
model. Then, the transition matrix (P) can be defined as In Eqn. (8), s is calculated by solving s · U = s, and 1 is a
⎡ ⎤ column matrix of ones. Also, Amax is the maximum number
A0,0 A0,1 ··· A0,Lmax
.. .. ..
of packets that can arrive from each VoIP user during Tf .
P=⎣ . .
..
. .
⎦ (3) Similar to the average arrival rate, the average number of
ALmax ,0 ALmax ,1 ··· ALmax ,Lmax serviced VoIP packets (κavg ) can be expressed as
where Lmax is maximum queue length and submatrix (Ai,j ) κavg =
is expressed as
⎡ ⎤ M 
M L
max
B(i,0),(j,0) B(i,0),(j,1) ··· B(i,0),(j,M ) min(j, k) · π(k) · Ps (j | xc = i) · πch (i). (9)
.. .. ..
Ai,j =⎣ . .
..
. .
⎦. i=0 j=0 k=0
B(i,M ),(j,0) B(i,M ),(j,1) ··· B(i,M ),(j,M ) Here, πch (i) is the steady-state probability that the number of
(4)
unoccupied channels are i, and can be expressed as
In Eqn. (4), Ai,j represents the variation of the number of

queuing packets. That is, there are i packets in the current M p10 p01
πch (i) = ( )i · ( )M−i . (10)
frame, and there will be j packets in the next frame. In Ai,j , i p01 + p10 p01 + p10
each element (B(i,m),(j,n) ) indicates the transitions between Moreover, the average throughput is represented by Savg =
the numbers of unoccupied channels when the number of κavg × lP DU . Here, lP DU is the size of VoIP PDU. Also,
queuing packets is changed from i to j. Also, B(i,m),(j,n) we can calculate the dropping probability of VoIP packets as
is a 2-by-2 matrix because our MMPP model has two phases, follows: Pdrop = 1 − κavg /ρ. Hence, by using Pdrop , we can
such as underloading and overloading. When the number of define the VoIP capacity as following
queuing packets is i, given that k packets are scheduled,
(j − max(i − k, 0)) packets should arrive so that the number CV oIP = arg max N ∈ { N | 1 − κavg /ρ ≤ Plimit }. (11)
of packets becomes j. Hence, B(i,m),(j,n) can be calculated
Here, Plimit is the upper threshold of the packet dropping
by Eqn. (5). Here, Ps (k | xc = m) is the probability that the
probability for VoIP services.
BS serves k packets when the number of unoccupied channels
is m. In this letter, since we assumed that one VoIP packet
can be transmitted through one unoccupied wireless channel, IV. N UMERICAL AND S IMULATION R ESULTS
Ps (m | xc = m) = 1. In addition, in Eqn. (5), U and We evaluated the VoIP performance of the cognitive radio
D(m) are the transition probability matrix and the diagonal system by using MATLAB. We included all the essential fac-
probability matrix of the two-state MMPP model [4]. Each tors required for performance evaluation, such as round-robin

Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on May 31,2010 at 14:02:45 UTC from IEEE Xplore. Restrictions apply.
 LEE and CHO: VOIP CAPACITY ANALYSIS IN COGNITIVE RADIO SYSTEM 395

1
sim, po=0.8
400
sim, po=0.5
350 sim, po=0.2
anal, p01=0.8, p10=0.2

Average throughput (kbps)

Packet dropping probability

300 anal, p01=0.5, p10=0.5

Packet dropping probability

0.1 anal, p01=0.2, p10=0.8
250 0.1

Plimit 200
0.02
sim, po=0.8 150
0.01 Plimit
sim, po=0.5
100 0.02
sim, po=0.2
anal, p01=0.8, p10=0.2 anal, p00=0.2, p01=0.8, p10=0.8, p11=0.2
50
anal, p01=0.5, p10=0.5 0.01
anal, p00=0.5, p01=0.5, p10=0.5, p11=0.5
anal, p01=0.2, p10=0.8 anal, p00=0.8, p01=0.2, p10=0.2, p11=0.8
0
1E-3
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
15 20 25 30 35 40 45 50
Number of users Number of users Number of users

Fig. 2. Packet dropping probability versus number Fig. 3. Average throughput (Kbps) versus number Fig. 4. Packet dropping probability versus number
of users according to variation of channel occu- of users according to variation of channel occu- of users according to variation of channel-state
pancy probability when M=10. pancy probability when M=10. transition probability when M=10.

TABLE I
V O IP CAPACITY ACCORDING TO VARIATION OF DL/UL FRAME RATIO models for all the cases are the same. However, if the p01
and p01 are larger, the state transitions occur more frequently.
hhhh DL/UL frame ratio Thus, the VoIP capacity could be smaller. On the other hand,
hhh
Capacity hhhh DL:UL = 1:1 DL:UL = 2:1
we can apply our formulation to VoIP admission control in the
h
Downlink capacity 13 24 cognitive radio system. If the channel occupancy variations are
Uplink capacity 33 21 dynamic, we can adjust the maximum supportable number of
users based on the results demonstrated in this letter.
Furthermore, given that the total number of channels is 30,
scheduler, packet-size and packet-generation period variations the overhead caused by control signaling is 30 % [8], and
of VoIP codecs, and so on. po =0.5, we can obtain both downlink and uplink capacity
To obtain results, we assumed that M = 10, Tf = 5 through our analysis method. As shown in Table I, when
ms, Lmax = 50, and Amax = 200. We utilized the G.729B DL/UL frame ratio is 1:1, we can show that the VoIP capacity
codec where Tbasic is 10 ms. This codec has two data rates is restricted by the downlink owing to the overhead caused by
(8 kbps and 0 kbps), and voice activity factor is 0.4, in this control signaling. However, when the DL/UL ratio is 2:1, the
letter. We assumed that the size of the RTP/UDP/IPv4 header capacity is limited by the uplink due to the fact that the size
compressed by payload header suppression (PHS) is 16 bytes. of uplink frame is much smaller than that of downlink frame.
Also, the size of the MAC header is 6 bytes. In the simulation, Therefore, we can conclude that the VoIP capacity is deter-
we used an on-off source model for each VoIP user. mined by bottleneck-link, which can be different according to
As shown in Fig. 2, we can show Pdrop versus the number system parameters such as the DL/UL ratio.
of users according to variation of channel occupancy probabil-
ity (po ). In this letter, we assume that Plimit = 0.02. Here, po R EFERENCES
means the probability that the primary users utilize wireless [1] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “NeXt
channels, and also denotes the steady-state probability when generation/dynamic spectrum access/cognitive radio wireless networks:
channel status is ‘Occupied’. For a low loading condition, the a survey,” Computer Networks: The International J. of Computer and
Telecommun. Networking, vol. 50, no. 13, pp. 2127–2159, Sept. 2006.
MMPP based numerical results have lower packet dropping [2] S. McBeath, J. Smith, L. Chen, and A. C. K. Soong, “VoIP support using
probabilities than the on-off based simulation results mainly group resource allocation based on the UMB system,” IEEE Commun.
due to the characteristic of the IDC matching technique [4]. Mag., vol 46, no. 1, pp. 114–120, Jan. 2008.
[3] F. Wang, A. Ghosh, C. Sankaran, P. Fleming, F. Hsieh, and S. J. Benes,
In addition, the average throughput grows linearly according “Mobile WiMAX systems: performance and evolution,” IEEE Commun.
to the increment of the number of VoIP users up to the Mag., vol. 46, no. 10, pp. 41–49, Oct. 2008.
saturation point, as shown in Fig. 3. Since resource saturation [4] J.-W. So, “Performance analysis of VoIP services in the IEEE 802.16e
OFDMA system with inband signaling,” IEEE Trans. Veh. Technol., vol.
occurs at this point, the BS cannot assign channels to surplus 57, no. 3, pp. 1876–1886, May 2008.
users beyond the saturation point. Thus, even though the [5] Q. Bi, P.-C. Chen, Y. Yang, and Q. Zhang, “An analysis of VoIP service
number of users increases, the average throughput cannot be using 1xEV-DO Revision A system,” IEEE J. Select. Areas Commun.,
vol 24, no. 1, pp. 36–44, Jan. 2006.
enlarged without limit. Given that po is small, there exist more [6] H. Heffes and D. M. Lucantoni, “A Markov modulated characterization
unoccupied channels that can be utilized by cognitive users. of packetized voice and data traffic and related statistical multiplexer
Thus, the throughput when po is small could be higher than performance,” IEEE J. Select. Areas Commun., vol 4, no. 6, pp. 856–
868, Sept. 1986.
the throughput when po is large. Also, from Fig. 2 and Fig. 3, [7] Q. Zhao, L. Tong, A. Swami, and Y. Chen, “Decentralized cognitive
it can be seen that the performance of analysis and simulation MAC for opportunistic spectrum access in ad hoc networks: a POMDP
are almost the same. framework,” IEEE J. Select. Areas Commun., vol. 25, no. 3, pp. 589–600,
Apr. 2007.
The VoIP capacity can be varied according to the variation [8] F. Wang, A. Ghosh, C. Sankaran, P. Fleming, F. Hsieh, and S. J. Benes,
of channel-state transition probability, as shown in Fig.4. Here, “Mobile WiMAX systems: performance and evolution,” IEEE Commun.
the steady-state transition probabilities of the Markov channel Mag., vol. 46, no. 10, pp. 41–49, Oct. 2008.

Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on May 31,2010 at 14:02:45 UTC from IEEE Xplore. Restrictions apply.