Wireless Relay Communications using

ann Unmanned Aerial Vehicle

Pengcheng Zhan, Kai Yu, A. Lee Swindlehurst

Department of Electrical and Computer Engineering,

Brigham Young University, Provo, UT, USA, 84602
Telephone: (801) 422-3930, Fax: (801) 422-0201
e-mail: {pz8,kaiyu,swindIe}jdee.byu.edu

Abstract- Herein, we investigate the optimal deployment of an

unmanned aerial vehicle (UAV) in a wireless relay communication
system. The optimal UAV position is found by maximizing the
average data rate, while at the same time keeping the symbol
error rate (SER) below a certain threshold. We derive a closed-
form expression for the average data rate in a fixed wireless
link using adaptive modulation. By using the alternate definite ,....X
integral form for the Gaussian Q-function, the symbol error rate
(SER) of the system in the link level is evaluated. An upper bound
on the SER is also derived using the improved exponential bounds
for the Q-function. It is shown that the derived SER expression
matches the simulation results very well and the derived upper
bound is tight for a wide range of SNRs. Simulation results also
show that the system data rate matches the derived closed-form
expression.
Fig. 1. Battlefield relay communication scenario
I. INTRODUCTION
Recently, unmanned aerial vehicles (UAVs) have attracted UAV. An SER analysis is also given in this section. Numerical
considerable attention in many military as well as civilian simulation results are presented in Section IV, followed by
applications [1], [2]. Besides other advantages, one major some conclusions in Section V.
advantage of using UAVs is that they can be quickly deployed
into the battlefield, or various communication environments as II. SYSTEML MODEL
relays [3], and therefore improve the performance of wireless A. Two-hop half duplex protocal
communications systems [4], [5]. For example, in many com- In this paper, we assume that a two-hop half-duplex protocol
munication scenarios, there exist obstacles (such as mountains, is used in the system. During the first time slot, the transmitter
buildings, etc) that severely deteriorate or even block the signal sends the desired message to the UAV. The UAV decodes
between the transmitter and, the receiver. In such cases, one the message and then sends it to the receiver in the second
can deploy a UAV to help setup the communication link and time slot. Note that we assurme the UAV can not transmit and
improve the communication perfonnance, i.e. using the UAV receive simultaneously.
as a relay between the transmitter and the receiver. Using the above transmission protocol, the signal model for
In this paper, we assume there is no direct communication the first and second time slots can be written as
link between the transmitter and the receiver. A UAV is
assumed to be positioned so that it can relay messages from Yi = X His, + nL, (L
the transmitter to the receiver, as depicted in Fig. 1. Assum-
ing adaptive modulation is employed, in the communication and
system, we analyze the average data rate of the system, £2
(2)
N H2s2 n2,
Y2 = / +
and investigate the optimal position of the UAV so that the
transmission rate is maximized under the constraint that the where Yi is the received signal at the UAV HI is the
symbol error rate (SER) is below a certain threshold. channel matrix between the transmitter and the UAV, s, is the
The paper is organized as follows. Section II presents the transmitted signal, and nii is additive noise. Similar definitions
system model used in this paper In Section III we derive the are used in (2) except that in this time slot the UAV becomes
closed-form expression of the average data rate, and formulate the transmitter. We will let IM denote the number of antennas
the optimization problem to find the optimal position of the at ground stations, N the number of antennas at UAV, and we
let E£., Es2 represent the transmission power during the two D. Orthogonal space-time block coding
time slots. We assume that in both time slots, only the receiver knows
B. Channel model the channel matrix. Hence, orthogonal space-time block codes
We assume the channels H1 and H2 are Rayleigh fading (OSTBC) [11] are used to transmit the data. For example,
channels with large scale path loss, i.e. in the 2 x 2 case, the well-known Alamouti code [12] is
employed. Since adaptive modulation is used, the receiver
Hi= Hnor (3) needs to determine/predict a suitable modulation scheme and
i,l feed this information back to the transmitter. In this paper, we
where Hnorm is a normalized complex Gaussian channel assume that this feedback is perfect, i.e. the transmitter knows
which when stacked in an NMXI x 1 vector has the distribution which modulation scheme to use. Note that feeding back the
CA (O, RH), and di,u is the distance between the ith access modulation scheme costs much less than feeding back the full
point (AP) {i = 1, 2} and the UAV. In our application, channel state information.
AP 1 is the transmitter and AP2 the receiver. For free space
transmission, the path-loss exponent a equals one [6]. Values Ill. SYSTEM ANALYSIS
of a > 1 occur in obstructed environments, while a < 1 is OSTBC exploits the diversity of the MIMO channels, and
common in wave-guided environments. Note that log-normal the instantaneous
shadow fading can easily be included in the channel model and
uplink SNR at the UAV can be expressed as
the analysis below. Assume the coordinates of the ith AP and
UAV are given as [xi yi hi]TaTnd [xlt yu hu]T respectively, FH1 ()2
so that di, can be written as 2HllFP (7)
1

di,u= _A.Xi2 y ,2 h ,2 (4) where p is defined as p = 2 (oN is the noise power, and
r

We use the well known Kronecker approach [7], [8] to 11 lFdenotes the Frobenius norm. Plugging (3) into (7), we
model the correlation matrix RH of the MIMO wireless obtain
channels, i.e., RH = RTX 0$ RRx, where RTX are RRx Hnorm 2 (8)
Fu
are respectively the normalized transmit and receive channel
correlation matrices. If APl and AP2 are located in multipath A similar analysis can be used to find the SNR of the downlink
scattering environments, we would see low spatial correlation channel from the UAV, if we replace d1jU with d2,u and M
at APL and AP2. At the UAV side, however, high spatial with N.
correlation is expected since there are few if any scatterers In [13], using the inverse Laplace transform, the probability
near the UAV. The normalized channel matrix is expressed as density function (PDF) of lHrnorm ll2 is derived as
Hr = (RRx )'2G[(RTx) /2]T (5) p mj ki
where the stochastic V by MI matrix G contains independent
and identically distributed (IID) C.A(0, 1) elements, (.)T de-
f (x) =
j= 1k=
1£Aj,-jk,
(k )
c u(x), (9)
notes transpose, ( ) 1/2 is defined such that R1 /2 (Rl /2)H where o-j (j = 1 2, P) are the distinct non-zero eigen-
R, and (H) is the lHermitian transpose. values of RH, and mj denotes the respective multiplicities
C. Adaptive modulation of o-j. By solving a system of linear equations, Ajk can be
We assume that the system employs adaptive modulation determined [13]. By defining
based on the current channel SNR, denoted by y. For a given
desired SER, the required SNR thresholds are predetermined g(Ti, a, X) = ! axdx
using the SER expression given in [9], [10]: I ,ax n
T01
0 i .
(1 0)
Pe N_ Q( ) (6)
the cumulative distribution function (CDF) of urH,, F can
wxhere P, is the cymbol error probability, N is the number be expressed as in (ll\)
of nearest neighbor constellation points, and dmi0r is the min-
imum separation distance between points on the underlying A Ergodic normalized transmission rate
constellation. Due to the random nature of the channel matrices, the
Assume that Yk and k+l are the predetermined SNR instantaneous transmission rate is different for different chan-
thresholds for the kth and (k + 11th modulation schemes nel realizations. Therefore, we define the ergodic normalized
respectively. If k+l > -Yk, the kth modulation scheme transmission rate (ENTR) and. use it as the criteria to quantify
will be used to transmit the message. If y < 71, no transmit the performance of the link. ENTR is defined in equations (12)
scheme will be chosen, which indicates there will be no and (13),
transmission between the transmitter and. the receiver. RiP 0 (t) = 3 (loE1 2 (t) (12)
 X L mj
F(x) = f (t)dt :
E -'
[g(k 0-i
g(k 1
0-i
0)I (11)
-oo
j=l k=

K(t) = K pu(C(t)- y) (+ (Ki+l-Ki)u(1(t)- i+l), (13)

i=

L-1 tCi+ (t)

Ri,u(t) = i {E log2~ KiK)
{log,2 f( ( x)dx+ lo2KL J CF (t) f (x)dx}
i= J i (t)
L-1
= E 1K924i[F (Ci+1 (t)) -F(Ci (t))] + 1 K92KL [1 -F(CL (t))] }
*3{Z
i=l
(14)

where is a scalar that takes into account the rate loss when Using this alternative form and interchanging the order of the
OSTBC is used. Note that for 2 x 2 Alamouti coding, = 1. integrations, the SER can be rewritten as in (19). Recalling
In (113), Ki is the number of the constellation points for the ith the definition in (10), it is straightforward to derive the SER
modulation scheme, and L is the total number of modulation expression given in (20).
schemes used in the system. Defining Ci (t) = ]L d 2 (t), it 2) SER upper bound: In order to relieve the computational
is straightforward to show that the expression for the ETNR burden when evaluating (20), an upper bound, for the SER is
of the API-UAV link can be written as in (14). A similar derived by resorting to the results of [15]. In Chiani's work,
expression can be obtained for the UAV-AP2 link. an improved exponential bound for Q function is given as:
Since the communication between APL and AP2 is through
the UAV relay, and, also due to the two-hop half-duplex charac- bix2
teristics of the communication system, the overall transmission Q(x) < E aiexp(
i=l
2
) (21)
rate is
R(t) = mint{R1, (t), R2,u,, (t) I (15) where
2(0i i- )
ai (22)
In order to improve the system performance, i.e. to increase 7r
the overall network transmission rate R(t), we position the and
UAV according to the following optimization problem: 1
bi (23)
arg max R(t) s.t. sin o
xtZ (t), Y".; (t), h.jL (t)
Note that this bound is much better than the popular Chernoff
P1 U) < PT
bound. After some manipulation, the upper bound of the SER
Pe 2, u) <- PeT (16) is found to be given by (24).
where P6(i, u) i = 1, 2 is the SER between the ith AP and
the UAV, and PIT is the quality of service (QoS) criterion. IV. NUMERICAL SIMULATIONS
However due to the specific method we have used to select the
thresholds for adaptive modulation, the inequality constraints In this section, numerical simulations are conducted on both
in (16) can be removed. the link level (i.e. AP - UAV) and system level (i.e. APl -
UAV - AP2). In the simulations, we assume both APs and
B. SE analysis the UAV are equipped with 2 antennas, and that Alamouti
This section is dedicated to the derivation of the SER
coding is used. The transmit power at hoth API and the
analysis at the link level Once the error analysis for each UAV is 1 W, and the noise power density at both AP2
link has heen performed, the SER of the whole system can he
and the UAV is 10 -16WHz The carrier frequency of the

calculated system is assumed to be IGHz and the system bandwidth is

1) Closed.form SER expression.- The SER can he expressed
assumed. to he 0kHz. This can be seen as consistent with
a narrowband system in a suburban area [6]. Seven different
as in (17).
In [14], an alternative definite integral form for the
MPSK modulation schemes are used in the simulations, i.e.
Gaussian Q function is given as

from BPSK to 128PSK. The QoS is chosen such that the

1 7 / 2 average SER is less than 10-2 Note that the altitude of the
Q(x) =- Iexp 1- 2sin2o
. dO x > 0. (18) UAV is fixed at 600m in the simulations.
 N-1 fC.+,(t)
PS = >
i= 1 i (t)
Ne(i) VQ( wp )f (x)dx + IN (t) Ne (N) Q(V tin()J(x)dx (17)

(sitd2n d))(l9)
Oc

F= Ne(i) exp(- xPin'2( )Jf(x)dx

dw + Ve (N) * exp(- -
4sin 2o
)f(x)dx
i= 1 ° ci (t) TJN (t)

[g(k I-)I 1-(Pd&(i I

P =l j= 1i
k=l
1 ( Pdi (i&)
4sin2o ji Ci+ (t ) I
4sin2o
I), Ci(t))]d9
o7j

N-1- p ml- 1, _(Pdmi2n(N) 1

),CN(t))dO} (20)
Ps~~2,
2 ( IC......X Ne(N)Ajk
"I,, k
4sin2o 0-i

I
j=l k=l 1 2
-(b,pd
1 ~4min (N) + a-i -), Ci+1(t)) -g(k -1-( b,pd 4in(N) a-i)
0j
ci M)l
prnj Q
N,(N). anAjgk- 1, b-(b0pd2
( '(N) I-),N(t)0 (24)
2 »=I = n=l
Q(k a-i

We assume rich multipath scenarios at both APs, so that the r

correlation matrices at the AP-side are given by

980
I 0
RAPi
the= U , t = 1, 2.
70
At the UAV side, high spatial correlation is assumed- .4
9
1 0.8 r.
RUAV .2 960
0.8 1 0

A. Data rate simulation results 950

In this section, we investigate overall system performance.
We assume both APs are moving on circles centered at (0, Ul 0) 40
and (1000, 1000, 0) respectively. The speeds for API and AP2
are set at 20 m/s and 30 m/s, with angular velocities 0.1L rad/s
Q(:), o
and 0.2 radls respectively. The path-loss exponent is assumed 896 898 900 902 904 906 908 910
to be 1.5 and 1.7 for the links between APl - UAV and UAV UAV position X-axis (m)
- AP2, respectively. The optimal position of the UAV is then Fig. 2. Trajectory of UAV movement
obtained by solving the optimization problem (16).
The trajectory of the optimal positions of the UAV is plotted
in Fig. 2. It can be seen that the UAV always flies closer to the APs are fixed at (0, 0, 0) and (1000, 1000, 0) respectively.
AP2, due to the larger path-loss exponent for the AP2 link. Fig. 4 shows the average system data rate for different a. We
The overall ystem and, link data rates are plotted, in Fig. 3,
find that for =(1, 1.1, 1.2, 1.3), the system data rates are
a

along with the theoretical system data rate. From Fig. 3, we the same. This is because the UAV is positioned right above
conclude that the theoretical results match the simulated results API for all these a, and the system data rate is limited by the
well. Furtheirmore, the data rates for both links are close to link between the API and the UAV.
each other, which indicates the proposed optimization method
reaches the desired balance between them B. SER simulation results
We also studied the influence of the path loss exponent av In this section, we simulate a scenario in which the AP
on the system performance. In this simulation, the path loss and UAV are 1.6 km away, and each of them is equipped
exponent is fixed at 1.5 between API and UAV. From UAV with 2 antennas. The path-loss exponent is 1.5, and we run
to AP2, a is varied between L and 2. During the simulation, 100 channel realizations to simulate the SER. The transmit
 ~62 Il
Simulated Network Rate
N
-6 - - -Theoretical Network Rate
Simulated
1. 0.1 Theoretical
.t~ Chernoff
- - N= 10
.5 --
0.0O - - N= 19
52 N = 28
0 50 100 150 200 250 300 350 400 450 50(
Time (s)
0.06
33

z 3.
N
-
--- Simulated Link
- -Simulated Link 2 Rate
Rate|
'n 0.04
'n
-t. 32 -- -: --- -:- -- ---
-0
u
3.
P4 0.02

I~~~~~~
-t. 1.
m -
LI

05;
0 50 100 150 200 250 300 350 400 450 500
Time (s) 1 2 3 5 6

Fig. 3. System and link data rates Fig. 5. Link level SER

REFERENCES
[1] K. Xu, X. Hong, M. Geerla, H. Ly, and D. L. Gu, "LLandmark routing
in large wireless batlefield networks using UAVs," in IEEE MILCOM
N
2001, September 2005, vol. 1, pp. 561-573.
z
'n
[2] P. Zhan, D. Casbeer, and A. Lee Swindlehurst, "A centralized control
In
.t. algorithm for target tracking with UAVs," in 39th IEEE Asilomar
-0
ID
Confeience, October 2005.
P4 [3] Z. Han, A. Lee Swindlehurst, and K. J. Ray Liu, 'Smart deploy-
-t.
m
ment/movement of unmanned air vehicle to improve connectivity in
MANET," in IEEE Wireless Communications and Networking Con-
0
ference, 2006, to appear.
4 [4] G. Kramer, M. Gastpar, and P. Gupta, "Cooperative strategies and ca-
pacity theorems for relay networks," IEEE Transactions on Information
Theory, vol. 51, no. 9, pp. 3037-3063, 2005.
[5] H. Bolcskei, R. U. Nabar, 0. Oyman, and A. J. Paulraj, "Capacity
scaling laws in MIMO relay networks," IEEE Transactions on Wireless
Communications, 2006, to appear.
ll 12 1 L4 l. 16 L7 L8 19 [6] T. S. Rappaport, Wireless Communications, Principles and Practice,
Pathloss Exponent of Link 2 Prentice Hall PTR7 1996.
[7] D-S. Shiu, G. J. Foschini, M. J. Gans, and J. M. Kahn, "Fading
Fig. 4. Averaged system data rate with different av correlation and its effect on the capacity of multielement antenna
systems," IEEE Transactions on Communications, vol 48, no. 3, pp.
502-513, March 2000.
[8] K. Yu, M. Bengtsson, B. Ottersten, D. McNamara, P. Karlsson, and
power for the AP is 0.1LW, and the noise power density M. Beach, "Modeling of wideband MIMO radio channels based
on NLOS indoor measurements,"' IEEE Transactions on Vehicular
and system bandwidth are assumed to be the same as in Technology, vol. 53, no. 3, pp. 655-665, May 2004.
the previous simulations. In Fig. 5, it is clearly shown that [9] John G. Proakis, Digital Communications, McGraw Hill, 2001.
the derived theoretical SER match the simulation results very [10] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless
well. Furthermore, it shows the derived upper bound using the Communications, Cambridge, 2003.
[l11] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, "Space-time block
improved exponential bounds is quite tight in our simulation, codes from orthogonal designs," IEEE Transactions on Information
and better than the conventional Chernoff bound. Theory, vol. 45, no. 5, pp. 1456-1467, July 1999.
[12] S. M. Alamouti, "A simple transmit diversity technique for wireless
V. CONCLUSIONS communications," IEEE Journal on Selected Areas in Communications,
vol. 16, no. 8R pp. 1451-1458d 1998.
In this paper we have proposed a method to optimally de- [13] R. U. Nabar, H. Bolcskei, and A. J. Paulraj, "Outage properties of space-
time block codes in correlated Rayleigh or Ricean fading environments,"
ploy a UAV to improve the quality of communications between in IEEE Inteernational Conference on Acoustics, Speech, and Signal
two obstructed APs. The optimal positions of the UAV have Piocessing 2002 vol 3 pp 2381-2384
been pursued by resorting to a min-max optimization method. [14] M.K. Simon and M-S Alouinmi "A unified approach to the performance
analysis of digital communication over generalized fading channels,"
Closed form expressions for the average data rate and SER Proceedings of The IEEE vol 86 no 9 pp 1860-1877 1998
have also been derived in the paper. An upper bound for the [15] M. Chiani, D. Dardari, and M. K. Simon, "New exponential bounds
SER is also derived for the sake of reducing the required and approximations for the computation of error probability in fading
channels," IEEE Triansactions on Wireless Communications, vol. 2, no.
computational load. The simulation and theoretical results are 4, pp. 840 - 845, July 2003.
shown to match very well.