4.2 Downlink MU-MIMO LTE-LAA for Coexistence with Asymmetric Hidden
4.2.4 Proposed MU-MIMO LTE-LAA System
4.2.4.4 Transmission Probability of an AHAP based on Stochas-
E[eα]to compute ˜γl,u in (72).
Letxi andy0 be a point for APiand the eNB in an XY-plane, respectively. The large-scale fading is denoted as l(d), where d is the distance between a transmitter and a receiver. The small-scale fading from the eNB to AP iis denoted by GLWeNB,i, and the small-scale fading from AP j to AP i is denoted by GWj,i. In our coexistence scenario, the eNB is placed at the origin o= (0,0). In addition,Φ denotes a set including all the locations of AHAPs.
Similarly to the work [78], we define the medium access indicator for AP iinAAH as ei = Y
xj∈Φ\{xi}
1tW
j ≥tWi + 1tW
j <tWi 1GW
j,i/l(kxj−xik)≤Γcs/PW
, (74)
where1A>B is an indicator function whereAandB are integer numbers. If the conditionA > B is true, then 1A>B = 1. Otherwise,1A>B = 0;tWj is the random backoff period of thej-th AP, which is uniformly distributed on[0,1];Γcs is the carrier sensing threshold of Wi-Fi; andPW is the transmit power of an AP.
The meaning of (74) is that APican start a new transmission if its own backoff period (tWi ) is shorter than the backoff periods (tWj ,∀j6=i) of all other APs or if the signal from each AP is less thanΓcs even thoughtWi > tWj .
The goal is to computeE[ei], which is the average medium access probability of each AHAP during eNB’s transmission.
Lemma 1 When the channel information between the eNB and each AHAP is given, the cor- responding medium access probability E[ei]is obtained as:
E[ei] = Z 1
0
"
Z
B(o,reNB)
fX,Y
(ai, bi)|khWLi,eNBk2, r≤reNB
×
Y
j∈AAH,j6=i
Z
B(o,reNB)
fX,Y
(aj, bj)|khWLj,eNBk2, r≤reNB
×
(
1−texp
−µΓcs
PWl(k(aj, bj)−(ai, bi)k) )
dajdbj
daidbi
dt,
where hWLi,eNB = [(h(1,1)i,eNB)T, . . . ,(h(N
W sub,NsubL )
i,eNB )T]T. Here, h(w,l)i,eNB ∈ CNT×1 is the channel vector from subcarrier w of Wi-Fi to subcarrier l of LTE-LAA.
Proof 1 By the law of total expectation,E[ei] is rewritten as
Et
Exi
"
EΦ\{xi}
E
h
ei|Φ, tWi =ti
|xi, tWi =t
|tWi =t
#
.
Definingξ(xj)asξ(xj) = 1tW
j ≥t+1tW j <t1GW
j,i/l(kxj−xik)≤Γcs
PW
, we first derive the part,E h
ei|Φ, tWi =ti as follows:
E h
ei|Φ, tWi =ti(a)
= E
"
Y
xj∈Φ\{xi}
ξ(xj)
#
=P
Y
xj∈Φ\{xi}
ξ(xj)
= 1
!
= Y
xj∈Φ\{xi}
P ξ(xj) = 1
(75)
(b)= Y
xj∈Φ\{xi}
P
1tW
j ≥t= 1
+P
1tW
j <t= 1
×
P
1GW
j,i/l(kxj−xik)≤Γcs
PW
= 1 )
(c)= Y
xj∈Φ\{xi}
(
1−texp
−µΓcs PW
l kxj−xik )
, (76)
where (a) can be derived by inserting the conditions (Φ, tWi = t) to (74). Note that ‘Φ is conditioned’ means all the locations of AHAPs are fixed. The step (b) follows from the fact that (tWj ≥tWi ) and (tWj < tWi ) are the opposite cases and that (tWj < tWi ) and (GWj,i/l(kxj −xik)≤ Γcs/PW) are independent cases. Finally, (c) follows from the fact that GWj,i is assumed to be Rayleigh fading and that tWj is uniformly distributed on [0,1].
Next, we derive EΦ\{xi}
E
h
ei|Φ, tWi =ti
|xi, tWi =t
as
EΦ\{xi}
Y
xj∈Φ\{xi}
(
1−texp
−µΓcs
PWl kxj −xik )
(a)= Y
xj∈Φ\{xi}
Exj
"
1−texp
−µΓcs
PWl kxj−xik #
(b)= Y
xj∈Φ\{xi}
Z
B(o,reNB)
fX,Y
aj, bj|khWLj,eNBk2, r≤reNB
× (
1−texp
−µΓcs
PWl k(aj, bj)−xik )
dajdbj, (77)
whereB(x, r) is a closed ball with centerx and radiusr, reNB is the coverage of the eNB, andµ is the parameter of Rayleigh fading channel. Finally, fX,Y
(aj, bj)|khWLj,eNBk2, r≤reNB
is the probability density function (pdf ) of probability that AP j is placed at xj = (aj, bj) for given khWLj,eNBk2 and the condition r ≤reNB. Note that r ≤reNB means AP j should be in the eNB’s
coverage. The step (a) can be derived by using the law of total expectation, while (b) follows from the definition of expectation.
By using (77) and the definition of expectation, we have
Exi
"
EΦ\{xi}
E
h
ei|Φ, tWi =ti
|xi, tWi =t
|tWi =t
#
= Z
B(o,reNB)
fX,Y
ai, bi|khWLi,eNBk2, r≤reNB
×
Y
xj∈Φ\{xi}
(Z
B(o,reNB)
fX,Y
aj, bj|khWLj,eNBk2, r≤reNB
×
(
1−texp
−µΓcs
PWl kxj−xik )
dajdbj
daidbi. (78)
Recall that tWi is uniformly distributed over [0,1]. Thus,E[ei]can be obtained as the expectation of (78)with respect to t, which proves the lemma.
Lemma 2 The truncated distribution for the probability that APiis at(ai, bi),fX,Y
ai, bi|khWLi,eNBk2, r≤reNB
, is given by:
fX,Y
ai, bi|khWLi,eNBk2, r≤reNB
= (
q
a2i +b2i)−1µNT
khWLi,eNBk2·l(r)NT−1
·eµkhWLi,eNBk2·l(r) Γ (NT)R
B(o,reNB)fX,Y
ai, bi| khWLi,eNBk2 dxdy
, (79)
where r= q
a2i +b2i.
Proof 2 The eNB acquires hWLi,eNB by listening to APi’s beacon signals, and hence can compute khWLi,eNBk2. Note that both the long-term and short-term fading effect, denoted as l(R) and G, respectively, is taken into account within khWLi,eNBk2; that is khWLi,eNBk2 = G/l(R), where R is a random variable representing the distance between AP iand the eNB.
At this point, we assume that each element of hWLi,eNB is an independent and identically dis- tributed (i.i.d.) complex Gaussian random variable, or equivalently assume that the absolute value of each element of hWLi,eNB is a Rayleigh random variable with the parameter µ. Then, the long-term fading effect l(R)is a function of R for givenR, and the short-term fading effectGis given as the sum ofNT i.i.d. exponential random variables with the parameterµ. Consequently, G is distributed according to the Gamma distribution with the parametersα=NT andβ =µ.
Note again that khWLi,eNBk2 is known by the eNB. Then, for given khWLi,eNBk2 = G/l(R), we
have
FG
khWLi,eNBk2·l(r) | khWLi,eNBk2
(80)
=P
G≤ khWLi,eNBk2·l(r) | khWLi,eNBk2
=P
khWLi,eNBk2·l(R)≤ khWLi,eNBk2·l(r) | khWLi,eNBk2
=P
l(R)≤l(r) | khWLi,eNBk2
=FR
r | khWLi,eNBk2
, (81)
whereFG(·) is the cumulative density function (cdf ) of the short-term fadingG. SinceGfollows the Gamma distribution, from (80), the pdf of R is obtained as follows.
fR
r|khWLi,eNBk2
= µNT
khLWi,eNBk2·l(r)NT−1
·eµkhLWi,eNBk2·l(r)
Γ (NT) .
Recall that the eNB is placed at the origin(0,0), and the AP is located at(x, y). By transforming the polar coordinate form to a Cartesian coordinate form, we finally have
fX,Y
x, y| khWLi,eNBk2
=
fRp
x2+y2| khWLi,eNBk2
px2+y2 . (82) Since the APs are distributed within the coverage of the eNB with radius of reNB, we need to restrict the domain of fX,Y. For the truncated distribution, we can derive
fX,Y
x, y| khWLi,eNBk2, r≤reNB
=
fX,Y
x, y| khWLi,eNBk2 R
B(o,reNB)fX,Y
x, y| khWLi,eNBk2 dxdy
= (p
x2+y2)−1µNT
khWLi,eNBk2·l(r)NT−1
·eµkhWLi,eNBk2·l(r) Γ (NT)R
B(o,reNB)fX,Y
x, y| khWLi,eNBk2 dxdy
,
which proves the lemma.
In Section 4.2.5, the derived transmission probability shall be numerically evaluated, which turns out to be 97-percent accurate.