TanReport

1
Distributed MAC Protocol Design for Full–Duplex
Cognitive Radio Networks
Le Thanh Tan and Long Bao Le
Abstract—In this paper, we consider the Medium Access
Control (MAC) protocol design for full-duplex cognitive radio
networks (FDCRNs). Our design exploits the fact that full-duplex
(FD) secondary users (SUs) can perform spectrum sensing and
access simultaneously, which enable them to detect the primary
users’ (PUs) activity during transmission. The developed FD
MAC protocol employs the standard backoff mechanism as
in the 802.11 MAC protocol. However, we propose to adopt
the frame fragmentation during the data transmission phase
for timely detection of active PUs where each data packet is
divided into multiple fragments and the active SU makes sensing
detection at the end of each data fragment. Then, we develop
a mathematical model to analyze the throughput performance
of the proposed FD MAC protocol. Furthermore, we propose
an algorithm to configure the MAC protocol so that efficient
self-interference management and sensing overhead control can
be achieved. Finally, numerical results are presented to evaluate
the performance of our design and demonstrate the throughput
enhancement compared to the existing half-duplex (HD) cognitive
MAC protocol.
Index Terms—MAC protocol, spectrum sensing, optimal sens-
ing, throughput maximization, full-duplex cognitive radios.
I. INTRODUCTION
Engineering MAC protocols for efficient sharing of white
spaces is an important research problem for cognitive radio
networks (CRNs). In general, a cognitive MAC protocol must
realize both the spectrum sensing and access functions so that
timely detection of the PUs’ activity and effective spectrum
sharing among different SUs can be achieved. Most existing
research works on cognitive MAC protocols have focused on
the design and analysis of half-duplex CRNs (e.g., see [1],
[2] and references therein). Due to the half-duplex constraint,
SUs typically employ the two-stage sensing/access procedure
where they sense the spectrum in the first stage before ac-
cessing available channels for data transmission in the second
stage [3] – [6]. These HD MAC protocols may not exploit
the white spaces very efficiently since significant sensing
time can be required, which would otherwise be utilized for
data transmission. Moreover, SUs may not timely detect the
PUs’ activity during data transmission, which causes severe
interference to active PUs.
Thanks to recent advances in the full-duplex technologies,
some recent works propose more efficient full-duplex (FD)
spectrum access design for cognitive radio networks [7] where
each SU can perform sensing and transmission simultaneously
[8]. In general, the self-interference1
due to simultaneous
sensing and access may lead to degradation on the SUs’
spectrum sensing performance. In [7], the authors consider
the cognitive FD MAC design where they assume that SUs
The authors are with INRS-EMT, University of Quebec, Montréal, Québec,
Canada. Emails: {lethanh,long.le}@emt.inrs.ca.
1Self-interference is due to the power leakage from the transmitter to the
receiver of a full-duplex transceiver.
perform sensing in multiple small time slots to detect the
PU’s activity during transmission, which may not be efficient.
Moreover, they assume that the PU can change its idle/busy
status at most once during the SU’s transmission, which
may not hold true if the SU’s data packets are long. Our
FD cognitive MAC design overcomes these limitations where
we propose to employ frame fragmentation with appropriate
sensing design for timely protection of the PU and we optimize
the sensing duration to maximize the network throughput.
Specifically, our FD MAC design employs the standard
backoff mechanism as in the 802.11 MAC protocol to solve
contention among SUs for compatibility. However, the win-
ning SU of the contention process performs simultaneous
sensing and transmission during the access phase where each
data packet is divided into multiple data fragments and sensing
decisions are taken at the end of individual data fragments.
This packet fragmentation enables timely detection of PUs
since the data fragment time is chosen to be smaller than
the required channel evacuation time. We then develop a
mathematical model for throughput performance analysis of
the proposed FD MAC design considering the imperfect
sensing effect. Moreover, we propose an algorithm to configure
different design parameters including data fragment time, SU’s
transmit power, and the contention window to achieve the
maximum throughput. Finally, we present numerical results
to illustrate the impacts of different protocol parameters on
the throughput performance and the throughput enhancement
compared to the existing HD MAC protocol.
II. SYSTEM AND PU ACTIVITY MODELS
A. System Model
We consider a network setting where n0 pairs of SUs
opportunistically exploit white spaces on a frequency band
for data transmission. We assume that each SU is equipped
with one full-duplex transceiver, which can perform sensing
and transmission simultaneously. However, SUs suffer from
self-interference from its transmission during sensing (i.e.,
transmitted signals are leaked into the received PU signal).
We denote I as the average self-interference power where
it can be modeled as I = ζ (Ps)
ξ
[8] where Ps is the SU
transmit power, ζ and ξ (0 ≤ ξ ≤ 1) are some predetermined
coefficients which capture the self-interference cancellation
quality. We design a asynchronous MAC protocol where no
synchronization is assumed between SUs and PUs. We assume
that different pairs of SUs can overhear transmissions from the
others (i.e., collocated networks). In the following, we refer
to pair i of SUs as secondary link i or flow i interchangeably.
B. Primary User Activity
We assume that the PU’s idle/busy status follows two
independent and identical distribution processes. Here the

2
Sensing
DATA
Sensing Sensing
DATA DATA
FD FD
CTS
time
Contention
. . .
DIFS
RTS
SIFS
Active sensing SU
FD
Backoff
HD
L
SIFS
. . .
DATA
SIFS
ACK
time
Inactive sensing SU . . .
T
Sensing
T T T
Fig. 1. Timing diagram of the proposed full-duplex MAC protocol.
channel is available and busy for the secondary access if
the PU is in the idle and busy states, respectively. Let H0
and H1 denote the events that the PU is idle and active,
respectively. To protect the PU, we assume that SUs must
stop their transmission and evacuate from the channel within
the maximum delay of Teva, which is referred to as channel
evacuation time.
Let τac and τid denote the random variables which represent
the durations of channel active and idle states, respectively.
We assume that τac and τid are larger than Teva. We denote
probability density functions of τac and τid as fτac
(t) and
fτid
(t), respectively. In addition, let P (H0) = ¯τid
¯τid+¯τac
and
P (H1) = 1−P (H0) present the probabilities that the channel
is available and busy, respectively.
III. MAC PROTOCOL DESIGN
For contention resolution, we assume that SUs employ
the backoff as in the standard CSMA/CA protocol [9]. In
particular, SU transmitters perform carrier sensing and start
the backoff after the channel is sensed to be idle in an interval
referred to as DIFS (DCF Interframe Space). Speciﬁcally,
each SU chooses a random backoff time, which is uniformly
distributed in the range [0, 2i
W − 1], 0 ≤ i ≤ m and starts
counting down while carrier sensing the channel where W
denotes the minimum backoff window and m is the maximum
backoff stage. For simplicity, we assume that m=1 in the
throughput analysis in the next section and we refer to W
simply as the contention window.
Let σ denote a mini-slot interval, each of which corre-
sponds one unit of the backoff time counter. Upon hearing
a transmission from other secondary links or PUs, all SUs
will “freeze” its backoff time counter and reactivate when
the channel is sensed idle again. Otherwise, if the backoff
time counter reaches zero, the underlying secondary link wins
the contention. To complete the reservation, the four-way
handshake with Request-to-Send/Clear-to-Send (RTS/CST)
exchanges will be employed to reserve the available channel
for transmission in the next stage. Speciﬁcally, the secondary
transmitter sends RTS to the secondary receiver and waits
until it successfully receives CTS from the secondary receiver
before sending the data. All other SUs, which hear the RTS and
CTS from the secondary winner, defer to access the channel
for a duration equal to the data packet length, L. Furthermore,
the standard small interval, namely SIFS (Short Interframe
Space), is used before the transmissions of CTS, ACK and
data frame as described in [9].
We assume that the data length of the SU transmitter, L
(L = KT) is larger than the evacuation time, Teva. Hence, SUs
divide each packet into K equal-size data fragments, each of
which is transmitted in duration T. Moreover, T is chosen to
be smaller than Teva so that timely evacuation from the busy
channel can be realized. In addition, the active SU transmitter
simultaneously senses the PU activity and transmits its data
in each fragment where it makes one sensing decision on the
idle/active channel status at the end of each data fragment.
Furthermore, if the sensing outcome at one particular fragment
indicates an “available” channel then the active SU transmitter
performs concurrent sensing and transmission in the next
fragment (called full-duplex (FD) sensing); otherwise, it only
performs sensing without transmission (referred to as half-
duplex (HD) sensing). This design allows to protect the PU
with evacuation delay at most T if the sensing is perfect. We
assume that SU’s transmit power is set equal to Ps where we
will optimize this parameter to achieve good tradeoff between
self-interference mitigation and high communication rate later.
The timing diagram of this proposed FD MAC protocol is
illustrated in Fig. 1.
IV. THROUGHPUT ANALYSIS
We perform throughput analysis for the saturated system
where all SUs are assumed to always have data to transmit [9].
Let Psucc (i0) denote the probability that the SU successfully
reserves the channel (i.e., the RTS and CTS are exchanged
successfully), Tove (i0) represent the time overhead due to
backoff and RTS/CTS exchanges, and T i0
denote the average
conditional throughput in bits/Hz for the case where the back-
off counter of the winning SU is equal to i0 (i0 ∈ [0, W − 1]).
Then, the normalized throughput can be written as
NT =
W −1
i0=0
Psucc (i0) ×
T i0
Tove (i0) + KT
. (1)
In this expression, we have considered all possible values of
the backoff counter of the winning SU in [0, W − 1]. In what
follows, we derive the quantities Psucc (i0), Tove (i0), and T i0
.
A. Derivations of Psucc (i0) and Tove (i0)
The event that the SU successfully reserves the channel for
the case the winning SU has its backoff counter equal to i0
occurs if all other SUs choose their backoff counters larger
than i0. So the probability of this event (Psucc (i0)) can be
expressed as follows:
Psucc (i0) = n0
1
W
W − 1 − i0
W
n0−1
. (2)
Moreover, the corresponding overhead involved for successful
channel reservation at backoff slot i0 is
Tove (i0) = i0 × σ + 2SIFS + RTS + CTS + DIFS (3)
where σ, SIFS, DIFS, RTS and CTS represent the du-
ration of backoff slot, the durations of one SIFS, one DIFT,
RTS, and CTS control packets, respectively.
B. Derivation of T i0
The quantity T i0
can be derived by studying the trans-
mission phase which spans K data fragment intervals each
with length T. Note that the PU’s activity is not synchronized
with the SU’s transmission; therefore, the PU can change its

3
Fragment
00H 01H 10H 11H
00H 01H 10H 11H10H 11H00H 01H
00H 01H 10H 11H 00H 01H 10H 11H 00H 01H 10H 11H 00H 01H 10H 11H
00H 01H1
2
3
4
Fragment 1 Fragment 2 Fragment 3 Fragment 4Contention
2t
1 1l 2 3lPU activity
> @1 0 0( ), ( ) ;ove ovet T i T i T @2 0 1 0 12 3( ) , ( ) ;ove ovet T i T t T i T t
3t

3 0 1 23( ) ,ovet T i T t t f
1t
Fig. 2. PU’s activity patterns.
active/idle status any time. It can be verified that there are four
possible events related to the status changes of the PU during
any particular data fragment, which are defined as follows. Let
H00 be the event that the PU is idle for the whole fragment
interval; H10 denote the event that PU is first active and then
becomes idle by the end of the fragment; H11 be the event
that the PU is active for the whole fragment; and finally,
H01 capture the event that PU is first idle and then becomes
active by the end of the fragment. Here, there can be at most
one transition between the active and idle states during one
fragment time. This holds because we have τac and τid are
larger than the fragment time T (since Teva is larger than T;
τac and τid are larger than Teva).
The average throughput achieved by the secondary net-
work depends on the PU’s activity and sensing outcomes at
every fragment. For any particular fragment, if the sensing
outcome indicates an available channel then the winning SU
will perform concurrent transmission and sensing in the next
fragment (FD sensing); otherwise, it will perform sensing only
in the next fragment (HD sensing) and hence the achievable
throughput is zero. Moreover, the throughput and sensing
outcome also depend on how the PU changes its state at one
particular fragment.
In what follows, we present the steps to calculate T i0
.
We first generate all possible patterns capturing how the
PU changes its idle/active status over all K fragments. For
each generated pattern, we then consider all possible sensing
outcomes in all fragments. Moreover, we quantify the achieved
throughput conditioned on individual cases with corresponding
PU’s statuses and sensing outcomes in all K fragments based
on which the overall average throughput can be calculated.
Let Si denote one particular pattern i capturing the cor-
responding changes of the PU’s idle/active status and AS
denote the set of all possible patterns. There are 2K
possible
patterns since the PU either changes or maintains the status
(idle or active status) in each fragment. Note that each pattern
Si comprises a sequence of state changes represented by
possible events Hij. For convenience, we define a pattern Si
by the corresponding four sets S00
i , S10
i , S01
i , and S11
i whose
elements are fragment indexes during which the channel state
changes corresponding to H00, H10, H01 and H11 occur,
respectively. For example, if one pattern has the PU’s state
changes as H00, H01, H11, H11 in this order then we have
S00
i = {1}, S10
i = ∅, S01
i = {2}, and S11
i = {3, 4}. Fig. 2
shows all possibles patterns for K = 4. Then, the conditional
throughput T i0
can be written as
T i0
=
|AS|
i=1
T i0
{Si} (4)
where T i0
{Si} is the throughput (bits/Hz) for the pattern Si.
Consider one particular pattern Si. Let Ωi be the set
with cardinality ρi whose elements are fragment indexes
li
j in which the PU changes its state. Moreover, let Γi
be another set also with cardinality ρi whose elements
ti
j are time intervals between consecutive PU’s state
changes. In the following, we omit index i in parameters
li
j and ti
j for brevity. We show one particular pattern Si
with the parameter tj in Fig. 2 (the corresponding Si is
indicated by dash lines). For convenience, we denote
b
t=a
as t∈R
where R ≡ [a, b] is the range of t. Moreover,
we define R1 = [Tove (i0)+(l1 −1)T, Tove (i0) + l1T] to
be the range for t1. Similarly, the range for tj is Rj =
Tove(i0)+(lj −1)T −
j−1
r=1 tr, Tove (i0)+ljT −
j−1
r=1 tr ;
then the range for tρi
can be also expressed as Rρi
=
Tove(i0)+(lρi
−1)T −
ρi−1
r=1 tr, Tove(i0)+lρi
T −
ρi−1
r=1 tr ;
and finally the range for tρi+1 is written as
Rρi+1 = Tove (i0) + KT −
ρi
r=1 tr, ∞ .
Using these notations, T i0
{Si} can be written as
T i0
{Si} = P (H0)
t1∈R1
. . .
tj ∈Rj
. . .
tρi
∈Rρi
tρi+1∈Rρi+1
2K
k=1 j1∈Φ0
k
Pj1
(ti
)
j2∈Φ1
k
Pj2
(ti
)
j3∈Φ
0
k
Tj3
(ti
) (5)
ft1
(t1). . .ftj
(tj). . .ftρi
(tρi
)ftρi+1
(tρi+1)dt1. . .dtj. . .dtρi
dtρi+1 (6)
where ¯x = 1−x and ftj
(tj) in (6) is the pdf of tj, which can
be either fτid
(.) or fτac
(.) depending on whether the underlying
interval is idle or active one, respectively.
In this expression, we have averaged over the possible
distribution of the time interval vector ti
whose elements ti
j
vary according to the exact state transition instant within the
corresponding data fragment. The quantity in (5) accounts for
2K
different possible sensing outcomes in K fragments. More-
over, the term j3∈Φ
0
k
Tj3
(ti
) represents the corresponding
achieved throughput for the underlying pattern Si and sensing
outcomes. Note that the data phase must start with the idle
state, which explains the factor P (H0) in this expression.
Moreover, the first channel transition must be from idle to
active, i.e., ft1
(t1) = fτid
(t1). In general, if j is odd, then
ftj
(tj) = fτid
(tj); otherwise, ftj
(tj) = fτac
(tj).
We now interpret the term in (5) in details. For particular
sensing outcomes in all fragments, we have defined two
subsets, namely a set of fragments Φ0
k where the sensing
indicates an available channel and the complement set of
fragments Φ1
k where the sensing indicates a busy channel.
Moreover, we have defined the set Φ
0
k whose elements are
indices of fragments each of which is the next fragment of
one corresponding fragment in Φ0
k (e.g., if Φ0
k = {1, 3} then
Φ
0
k = {2, 4}). We need the set Φ
0
k since only fragments in
this set involve data transmissions and, therefore, contribute to
the overall network throughput. In (5), the first two products

4
TABLE I
DETERMINATION OF Pj AND Tj FOR FRAGMENT j ∈ Φ0
k
Pj Tj Conditions
P00
f T 00 j ∈ S00 (j − 1) ∈ Φ0
k (j − 1 = 1)
P00
f,h 0 j ∈ S00 (j − 1) ∈ Φ1
k
P10
f T 10 j ∈ S10 (j − 1) ∈ Φ0
k (j − 1 = 1)
P10
f,h 0 j ∈ S10 (j − 1) ∈ Φ1
k
P11
d T 11 j ∈ S11 (j − 1) ∈ Φ0
k (j − 1 = 1)
P11
d,h 0 j ∈ S11 (j − 1) ∈ Φ1
k
P01
d T 01 j ∈ S01 (j − 1) ∈ Φ0
k (j − 1 = 1)
P01
d,h 0 j ∈ S01 (j − 1) ∈ Φ1
k
represent the probability of sensing outcomes for all fragments
in the k-th sensing outcome.
In the following, we present the derivations of Pj and
Tj. Let P11
d , P01
d , P00
f , and P10
f denote the probabilities
of detection and false alarm for the following channel state
transition events H11, H01, H00 and H10 using FD sensing,
respectively. Similarly, we denote P11
d,h and P01
d,h, P00
f,h and
P10
f,h as the probabilities of detection and false alarm for
events H11, H01, H00 and H10 using HD sensing, respectively.
Derivations of the probabilities of detection and false alarm for
all events and all sensing schemes are given in Appendix A.
Let T00
, T10
, T01
and T11
denote the number of bits per
Hz under the state transition events H00, H10, H01 and H11,
respectively. These quantities are derived in Appendix B.
For particular fragment j ∈ Φ0
k, the quantities Pj and Tj
are given in Table I where and denote AND and OR
operations, respectively. For example, in the first line, if the
sensing outcome indicates an available channel in fragment
j − 1 ((j − 1) ∈ Φ0
k) or fragment j − 1 is the first one in
the data access phase then the SU will perform concurrent
sensing and transmission in fragment j. Moreover, fragment
j belongs to the set S00
; therefore, we have Pj = P
00
f and
Tj = T00
. Similarly, we can interpret the results for Pj and Tj
in other cases. For fragment j ∈ Φ1
k, to calculate the quantities
Pj, we use the results in the first column of Table I but we
have to change all items to Pkl
x (k, l ∈ {0, 1}, x represents f,
d, fh, and dh). Note that all the quantities depends on time
instant t when the PU changes its state. For example, at the
time point tj corresponding to fragment lj, t =
j
k=1 tk −
[Tove (i0) + (lj − 1)T]. We have omitted this dependence on t
in all notations for brevity (details can be found in Appendices
A and B).
In summary, we have derived T i0
{Si}, which can be sub-
stituted into (4) to obtain T i0
. Finally, applying the result of
T i0
to (1), we can calculate the secondary network throughput
NT .
V. CONFIGURATION OF MAC PROTOCOL FOR
THROUGHPUT MAXIMIZATION
A. Problem Formulation
We are interested in determining optimal configuration of
the proposed MAC protocol to achieve the maximum through-
put while satisfactorily protecting the PU. Specifically, let
NT (T, W, Ps) denote the normalized secondary throughput,
which is the function of fragment time T, contention window
W, and SU’s transmit power Ps. Suppose that the PU requires
that the average detection probability achieved at fragment i
be at least Pd,i. Then, the throughput maximization problem
can be stated as follows:
max
T,W,Ps
NT (T, W, Ps)
s.t. ˆPd,i εi
, T ≥ Pd,i, i = 1, 2, · · · , K
0 T ≤ Teva, 0 W ≤ Wmax,
0 Ps ≤ Pmax,
(7)
where Wmax is the maximum contention window, Pmax is the
maximum power for SUs and the fragment time T is upper
bounded by Teva. In fact, the first constraint on ˆPd,i εi
, T
implies that the spectrum sensing should be sufficiently reli-
able to protect the PU where the fragment time (also sensing
time) T must be sufficiently large. Moreover, the optimal
contention window W should be set to balance between
reducing collisions among SUs and limiting protocol overhead.
Finally, the SU’s transmit power Ps must be appropriately set
to achieve good tradeoff between the network throughput and
self-interference mitigation in spectrum sensing.
B. Configuration Algorithm for MAC Protocol
Algorithm 1 MAC CONFIGURATION ALGORITHM
1: for each value of W ∈ [1, Wmax] do
2: for each searched value of T ∈ (0, Teva] do
3: Find optimal P∗
s as P∗
s = argmax
0≤Ps≤Pmax
NT (T, W, Ps).
4: end for
5: The best (T∗
, P∗
s ) for each W is (T∗
, P∗
s ) =
argmax
T,P ∗
s
NT (T, W, P∗
s ).
6: end for
7: The final solution (W∗
, T∗
, P∗
s ) is determined as
(W∗
, T∗
, P∗
s ) = argmax
W,T ∗,P ∗
s
NT (W, T∗
, P∗
s ).
We assume the shifted exponential distribution for τac and
τid where ¯τac and ¯τid are their corresponding average values
of the exponential distribution. Specifically, let fτx
(t) denote
the pdf of τx (x represents ac or id as we calculate the pdf of
τac or τid, respectively) then
fτx
(t) =
1
¯τx
exp(−
t−T x
min
¯τx
) if t ≥ Tx
min
0 if t Tx
min
(8)
For a given T, we would set the sensing detection threshold
ε and SU’s transmit power Ps so that the constraint on
the average detection probability is met with equality, i.e.,
ˆPd (ε, T) = Pd as in [3], [4]. In addition, the first constraint
in (7) now turns to the two constraints for the average
probabilities of detection under FD and HD spectrum sensing.
First, the average probability of detection under FD sensing
can be expressed as
ˆPd =
P11
d P (H11) + P
01
d P (H01)
P (H11) + P (H01)
(9)
where P11
d is the probability of detection for H11; P
01
d is the
average probability of detection for H01, which is given as
P
01
d =
T +T id
min
T id
min
P01
d (t)fτid
t Tid
min ≤ t ≤ T + Tid
min dt (10)

5
where fτid
(t |A) is the pdf of τid conditioned on A = Tid
min ≤
t ≤ T + Tid
min, which is given as
fτid
(t |A) =
fτx
(t)
Pr {A}
=
1
¯τid
exp(− t
¯τid
)
1 − exp(− T
¯τid
)
. (11)
Note that P11
d and P01
d (t) are derived in Appendix A. More-
over, P (H01) and P (H11) are the probabilities of events H01
and H11, which are given as
P(H01)=P(H0)Pr Tid
min≤τid≤T +Tid
min =P(H0) 1−exp(
−T
¯τid
) (12)
P (H11)=P(H1)Pr(τac≥T +Tac
min)=P (H1) exp(−
−T
¯τac
) (13)
The average probability of detection for HD sensing, ˆPd,h can
be derived similarly.
We propose an algorithm to determine (T, W, Ps) summa-
rized in Alg. 1. Note that there is a finite number of values for
W ∈ [1, Wmax]; therefore, we can perform exhaustive search
to determine its best value. Moreover, we can use the bisection
scheme to determine the optimal value of T. Furthermore, the
optimal value of Ps can be determined by a numerical method
for given T and W in step 3. Then, we search over all possible
choices of T and W to determine the optimal configuration
of the parameters (in steps 5 and 7).
VI. HALF-DUPLEX MAC PROTOCOL WITH PERIODIC
SENSING
A. Half-Duplex MAC Protocol Design
To demonstrate the potential performance gain of the pro-
posed FD MAC protocol, we also consider an HD MAC
protocol. In this HD MAC protocol, we perform the same
backoff for channel resolution but we employ periodic HD
sensing in each data fragment where the sensing duration is
TS and data transmission duration is T − TS . If the sensing
outcome in the sensing stage indicates an available channel
then the SU transmit data in the second stage; otherwise, it
will keep silent for the remaining time of fragment and wait
until the next fragment.
B. Throughput Analysis
To analyze the throughput, we employ the similar procedure
as presented in Section IV with some modifications as follows.
We generate all patterns {Si} capturing possible changes in
the idle/active states over the K fragments. Note, however,
that the PU may change its status during the SU’s sensing or
access stage, which requires us to consider more events in each
fragment. For example, there are two sub-events corresponding
to event H01 in one particular fragment, namely sub-event
1 {h00, h01} and sub-event 2 {h01, h11} where sub-event 1
represents the case that the PU is idle in the sensing stage
(h00) and becomes active by the end of the transmission stage
(h01); sub-event 2 captures the case that the PU is first idle
then becomes active by the end of the sensing stage (h01) and
maintains this active state during the transmission period (h11).
We can see that we only need to consider these sub-events
for events H01 and H10 (i.e., there is a status change in the
fragment). Furthermore, depending on the specific sub-event
in one particular fragment, there are three possible sub-events
in the next fragment. Therefore, there are |AS| = 3K
possible
patterns and Fig. 3 shows all possibles patterns for K = 3.
01H
10H10H 11H00H 01H
00H
01h
11h
00h
01h
00H
00h
00h
01h
11h
00h
01h
00h
00h
00h
00h
11h
11h
11h
10h
10h
00h
01H
01h
11h
00h
01h
10H
11h
10h
10h
00h
00H
00h
00h
01H
01h
11h
00h
01h
11H
11h
11h
11h
10h
10h
00h
11H
11h
11h
Sensing 1Contention
2t
1 11 1,l V
2 22 0,l VPU activity
3t1t
Data 1 Sensing 2 Data 2 Sensing 3 Data 3
Fragment
1
2
3
Fragment 1 Fragment 2 Fragment 3
Fig. 3. PU’s activity in HD MAC protocol.
Similar to the analysis for the FD MAC protocol, we also
denote Ωi as the set with cardinality ρi whose elements are
fragment indexes li
j in which the PU changes its state, Γi
as the set with cardinality ρi whose elements ti
j are time
intervals between consecutive PU’s state changes. In addition,
Σi denotes the set with cardinality ρi whose elements σi
j
is defined as follows: if the PU changes its state in the
transmission stage then σi
j = 1, otherwise σi
j = 0 (i.e., the
PU changes its state in the sensing stage). Moreover, we define
σi
j = 1 − σi
j. In the following analysis, we omit the index i
in li
j, ti
j and σi
j = 1 for brevity.
We again denote
b
t=a
as t∈R
where R ≡ [a, b]
is the range of t. Moreover, we define R1 =
[TC1 + σ1TS, TC1 + σ1TS + σ1T] as the range for t1,
where TC1 = Tove (i0) + (l1 − 1)T, which means that
if the PU changes its state in the sensing stage, then
R1 = [TC1, TC1 + TS] (because σj = 0); otherwise,
R1 = [TC1 + TS, TC1 + T] (because the PU changes
state in transmission stage and σj = 1). Similarly, the
range for t2 is R2 = [TC2 + σ2TS, TC2 + σ2TS + σ2T]
where TC2 = Tove (i0) + (l2 − 1)T − t1; the range for
tj is Rj = [TCj + σjTS, TCj + σjTS + σjT] where
TCj = Tove (i0) + (lj − 1)T −
j−1
r=1 tr; the range for
tρi
is Rρi
= [TCρi
+ σρi
TS, TCρi
+ σρi
TS + σρi
T] where
TCρi
= Tove (i0) + (lρi
− 1)T −
ρi−1
r=1 tr; and finally the
range for tρi+1 is Rρi+1 = Tove (i0) + KT −
ρi
r=1 tr, ∞ .
In addition, the throughput T i0
is also calculated as in (4)
and T i0
{Si} is given as
T i0
{Si}=P (H0)
t1∈R1 t2∈R2
. . .
tj ∈Rj
. . .
tρi
∈Rρi
tρi+1∈Rρi+1
2K
k=1 j1∈Φ0
k
Pj1
(ti
)
j2∈Φ1
k
Pj2
(ti
)
j3∈Φ0
k
Tj3
(ti
) (14)
ft1
(t1). . .ftj
(tj). . .ftρi
(tρi
)ftρi+1
(tρi+1)dt1. . .dtj. . .dtρi
dtρi+1 (15)
where ¯x = 1 − x and ftj
(tj) in (15) is the pdf of tj, which
can be either fτid
(.) or fτac
(.), which can be specified as in
Section IV.

6
In this expression, we have averaged over the possible
distributions of the time interval vector ti
whose elements
ti
j vary according to the state transition instants within the
corresponding data fragment. The quantity in (14) accounts
for 2K
different possible sensing outcomes in K fragments.
The first product represents the probability of the event that
sensing outcomes indicate that channel is available at frag-
ments j1 in Φ0
k; while the second product represents the
probability of the event that sensing outcomes indicate that
channel is not available at fragments j2 in Φ1
k. Moreover, the
term j3∈Φ0
k
Tj3
(ti
) represents the corresponding achieved
throughput for the fragments in Φ0
k because the SU only
transmits data at fragments in Φ0
k. Also the data phase must
start with the idle state, which explains the factor P (H0) in
this expression.
In the following, we present the derivations of Pj and
Tj. Derivations of the probabilities of detection and a false
alarm for all events and all sensing schemes are given in
Appendix A with the following modifications: i) sensing
time T becomes TS, and ii) time instant t at fragment
lj where the PU changes its state is calculated as t =
j
k=1 tk −[Tove (i0) + (lj − 1)T + σjTS]. The throughput Tj
under different possible state transitions, which is denoted
as Tkl
h , is derived in Appendix B with the modifications as
follows: i) self-interference-plus-noise power I+N0 becomes
noise power N0, ii) data length T becomes the remaining
time in fragment, i.e., T − TS, and iii) time instant t at
fragment lj where the PU changes its state is calculated as
above.
For particular fragment j ∈ Φ0
k, the quantities Pj and Tj are
given in Table II. For example, in the third row under H01,
we consider the case where the PU changes its state in the
sensing stage, i.e., σk = 0, j = lk (sub-events {h01, h11}),
we have Pj = P
01
d,h and Tj = T11
h ; otherwise, we have
Pj = P
00
f and Tj = T01
h (i.e., the PU changes its state in
the transmission stage, σk = 1 and sub-events {h00, h01}).
Note that the specific expressions of Pj and Tj depend on the
sub-events 1 and 2 not event H01 in this example.
Moreover, we now explain the case presented in the first
line as another example. Here there is no PU’s transition in the
underlying fragment, so we do not need σk. Similarly, we can
interpret the results for Pj and Tj in other cases. For fragment
j ∈ Φ1
k, if the sensing outcome indicates a busy channel in
the sensing stage then the SU will not transmit data in the
transmission stage, i.e., Tj = 0. To calculate the quantities
Pj, we use the results in the fourth column of Table II.
Note that this throughput analysis considering all possible
patterns as in Fig. 3 is general, which can be applied to any
pdfs of τac and τid. Hence, this also holds true under the
pdfs of τac and τid as given in (8). Note, however, that the
achieved throughput for some patterns can be zero. For ex-
ample, the pattern Si = {{h00, h01} ; {h10, h00} ; {h00, h00}}
in Fig. 3 (we indicate this pattern by the red dashed lines),
we can show that T i0
{Si} = 0 as follows. For this pat-
tern, the PU arrives in the transmission stage in fragment
1 and leaves in the sensing stage in fragment 2. It im-
plies that t1 ∈ R1 = [Tove (i0) + TS, Tove (i0) + T] and
t2 ∈ R2 = [Tove (i0) + T − t1, Tove (i0) + T + TS − t1]. It
can be verified that t2 ≤ T ≤ Tac
min; therefore, we have
ft2
(t2) = 0. Substitute ft2
(t2) to (14) and (15), we finally
obtain T i0
{Si} = 0.
TABLE II
DETERMINATION OF Pj AND Tj FOR FRAGMENT j ∈ Φ0
k
Hkl Sub-events σk Pj Tj
H00 {h00, h00} P
00
f,h T00
h
H11 {h11, h11} P
11
d,h T11
h
H01 {h01, h11} 0 P
01
d,h T11
h
{h00, h01} 1 P
00
f,h T01
h
H10 {h10, h00} 0 P
10
f,h T00
h
{h11, h10} 1 P
11
d,h T10
h
100 1000
1
2
3
4
Contention window (W)
Throughput(NT,bits/s/Hz)
Throughput vs contention window
ξ = 0.04, 0.03, 0.02, 0.01
Fig. 4. Normalized throughput versus contention window W for T = 18ms,
¯τid = 1000ms, ¯τac = 100ms, K = 4 and varying ξ.
0
10
20
0
0.02
0.04
0.05
0.1
0.15
0.2
0.25
SNRs(dB)
Throughput vs SNRs and T
T(s)
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0.22
NT ∗
(20 ms, 11 dB) = 0.2347
Fig. 5. Normalized throughput versus SU transmit power Ps and length of
fragment T for W = 1024, ¯τid = 200ms, n0 = 40, ¯τac = 100ms, K = 4,
ξ = 0.95 and ζ = 0.6.
VII. NUMERICAL RESULTS
To obtain numerical results, we take key parameters for the
MAC protocol from Table II in [9]. All other parameters are
chosen as follows unless stated otherwise: mini-slot is σ =
20µs; sampling frequency is fs = 6MHz; bandwidth of PU’s
QPSK signal is 6MHz; Pd = 0.8; Teva = 40ms; Tac
min = 40ms;
Tid
min = 45ms; the SNR of PU signals at SUs γP =
Pp
N0
=
−20dB; the self-interference parameters ζ = 0.4 and varying
ξ. Without loss of generality, the noise power is normalized to
one; hence, the SU transmit power, Ps becomes Ps = SNRs;
and Pmax = 25dB.
We first consider the effect of self-interference on the
throughput performance where ζ = 0.4 and ξ is varied in
ξ = {0.01, 0.02, 0.03, 0.04}. Fig. 4 illustrates the variations
of the throughput versus contention window W. It can be
observed that when ξ decreases (i.e., the self-interference is
smaller), the achieved throughput increases. This is because

7
1 100
1.5
2
2.5
3
3.5
4
4.5
Number of users (n0)
Throughput vs number of users
ξ = 0.04, 0.03, 0.02, 0.01
Fig. 6. Normalized throughput versus number of SUs n0 for T = 18ms,
W = 1024, ¯τid = 1000ms, ¯τac = 100ms, K = 4 and varying ξ.
5 10 15 20 25
0.05
0.1
0.15
0.2
0.25
Throughput vs SNRs
SNRs(dB)
ζ = 0.75, 0.45
T*
= 25, 20 ms
T
*
= 25, 20 ms
HD MAC
FD MAC
Fig. 7. Normalized throughput versus Ps for (¯τid, ¯τac) = (200, 100) ms,
n0 = 40, K = 4, ξ = 0.9 and ζ = {0.45, 0.75}.
the SU can transmit with higher power while still maintaining
the sensing constraint, which leads to throughput improve-
ment. The optimal Ps corresponding to these values of ξ
are Ps = SNRs = {25.00, 18.19, 13.56, 10.78}dB and the
optimal contention window is indicated by a star.
Fig. 5 illustrates the throughput performance versus SU
transmit power Ps and length of fragment T where ξ = 0.95,
ζ = 0.45, Teva = Tac
min = Tid
min = 40ms, Pmax = 30dB. It can
be observed that there exists an optimal configuration of SU
transmit power P∗
s = 11dB and fragment time T∗
= 20ms
to achieve the maximum throughput NT (T∗
, P∗
s ) = 0.2347,
which is indicated by a star symbol. This demonstrates the sig-
nificance of power allocation to mitigate the self-interference
and the optimization of fragment time T to effectively exploit
the spectrum opportunity. Fig. 6 illustrates the throughput
performance versus number of SUs, n0 where ζ = 0.4 and
ξ is varied in the set ξ = {0.01, 0.02, 0.03, 0.04}. Again,
when ξ decreases (i.e., the self-interference is smaller), the
achieved throughput increases. In this figure, the SNRs cor-
responding to the considered values of ξ are Ps = SNRs =
{25.00, 18.19, 13.56, 10.78} dB.
Finally, we compare the throughput of our proposed FD
MAC protocol and the HD MAC protocol with periodic
sensing in Fig. 7. For fair comparison, we first obtain the
optimal configuration of FD MAC protocol, i.e., (T∗
, W∗
, P∗
s )
((20ms, 1024, 12dB) for ζ = 0.45 and (25ms, 1024, 10dB)
for ζ = 0.75), then we use (T∗
, W∗
) for the HD MAC
protocol. Moreover, we optimize sensing time, TS to maximize
the achieved throughput for the HD MAC protocol. We can
see that when the self-interference is higher (i.e., ζ increases),
the FD MAC protocol requires higher fragment length, T but
lower SU transmit power, Ps. For both studied cases of ζ,
our proposed FD MAC protocol with power allocation out-
performs the HD MAC protocol at the corresponding optimal
power levels required by the FD MAC protocol. Moreover, we
can observe that our proposed FD MAC protocol can achieve
the maximum throughput at the transmit power level less than
Pmax while the HD MAC protocol achieves the maximum
throughput at Pmax since it does not suffer from the self-
interference.
VIII. CONCLUSION
In this paper, we have proposed the FD MAC protocol
for FDCRNs that explicitly takes into account the self-
interference. Specifically, we have derived the normalized
throughput of the proposed MAC protocols and determined
their optimal configuration for throughput maximization. Fi-
nally, we have presented numerical results to demonstrate the
desirable performance of the proposed design.
APPENDIX A
PROBABILITIES OF A FALSE ALARM AND DETECTION
Assume that the transmitted signals from the PU and
SU transmitter are circularly symmetric complex Gaussian
(CSCG) signals while the noise at the secondary links is
independent and identically distributed CSCG CN (0, N0) [3].
Under FD sensing, the probability of a false alarm for event
H00 can be derived using the similar method as the one in
[3], which is given as
P00
f = Q
ǫ
N0 + I
− 1 fsT (16)
where Q (x) =
+∞
x
exp −t2
/2 dt; fs, N0, ǫ, I are the
sampling frequency, the noise power, the detection threshold
and the self-interference, respectively; T is the duration of FD
sensing. The probability of detection for event H11 is
P11
d = Q
ǫ
N0 + I
− γP S − 1
√
fsT
γP S + 1
(17)
where γP S =
Pp
N0+I is the signal-to-interference-plus-noise
ratio (SINR) of the PU’s signal at the SU.
Similarly, we can express P10
f and P01
d as follows:
P10
f = Q


ǫ
N0+I − t
T γP S − 1
√
fsT
t
T (γP S + 1)
2
+ 1 − t
T

 (18)
P01
d = Q


ǫ
N0+I − T −t
T γP S − 1
√
fsT
T −t
T (γP S + 1)
2
+ t
T

 (19)
where t is the time instant when the PU changes its state. For
HD sensing, the expressions for the probabilities of detection
and a false alarm for the corresponding four events are similar
to the ones for FD sensing except that the self-interference-
plus-noise power N0+I becomes noise power N0 only; hence,
γP S becomes γh
P S =
Pp
N0
.

TanReport

More Related Content

What's hot (19)

Similar to TanReport (20)

More from Polytechnique Montreal (15)

TanReport