You are on page 1of 8

Adaptive Formation Control and Bio-inspired Optimization of a Cluster-based

Satellite Wireless Sensor Network

Erfu Yang
1
, Ahmet T. Erdogan
1
, Tughrul Arslan
1
, and Nick H. Barton
2
1. School of Engineering and Electronics 2. School of Biological Sciences
The University of Edinburgh
Kings Buildings, Edinburgh EH9 3JL, United Kingdom
{E.Yang, Ahmet.Erdogan, T.Arslan, N.Barton}@ed.ac.uk
Abstract
In this paper, adaptive formation control and bio-
inspired optimization are jointly addressed for a cluster-
based satellite wireless sensor network in which there are
multiple satellites ying in formation (MSFF) in the pres-
ence of unknown disturbances. The full nonlinear dynam-
ics model describing the relative positioning of the MSFF
system is used to develop an adaptive formation controller.
First, the original nonlinear system is transformed into a
linear controllable system with a perturbation term by in-
voking the input-output feedback linearization technique.
Second, by using the integral feedback design scheme, the
adaptive formation controller is presented for improving the
steady-state performance of the MSFF system in the pres-
ence of unknown disturbances. Third, as a currently popu-
lar bio-inspired algorithm, PSO (particle swarm optimizer)
is employed to minimize the total energy consumption un-
der the required quality of service by jointly optimizing the
transmission power and rate for each satellite. Simulation
results are provided to demonstrate the effectiveness of the
adaptive formation controller and the PSO-based optimiza-
tion for saving the total communication energy.
1 Introduction
The multiple satellite formation ying (MSFF) has re-
ceived considerable attention during the last decade [15].
It provides a novel concept of distributing the functionali-
ties of a large satellite among several smaller, less expen-
sive satellites to enhance the performance of future space
missions [6]. Since the sensing and communication among
the satellites in an MSFF form the basis for the coopera-
tive mission of the MSFF, the MSFF is generally viewed

This research is funded by the UK Engineering and Physical Sciences


Research Council (EPSRC) under grant EP/C546318/1.
as a mobile wireless sensor network (WSN) in space [5, 7].
The potential applications of satellite-based WSNs include
distributed sensing for high-resolution, synthetic-aperture
radar and imaging, etc. Formation of ying satellites could
operate cooperatively to act as a sparse aperture with an ef-
fective dimension larger than what can be achieved by a
single, larger satellite. The MSFF system can also provide a
higher reliability of operation in orbit. In case of the failure
of an individual large satellite, the total mission and func-
tionality of the system will become unavailable. But, such a
catastrophic event will be a different scenario in the MSFF
system. It is very likely to recover the mission and function-
ality of malfunctional small satellite by reconguring and
distributing the missions of the other formation ying satel-
lites. Therefore, the MSFF and WSNs technology could
give a big change to the future satellite-based communica-
tion and sensing systems in space.
The research challenges from the MSFF-based WSNs
have arisen from the formation control and energy-efcient
communication among the large-scale distributed satellites.
In particular, an adaptive formation controller is always
required in the presence of orbital disturbances. For the
satellites orbiting in close formation, perturbations such as
gravity harmonics from a non-spherical Earth, atmospheric
drag, solar and lunar gravity tend to perturb the orbits from
their ideal trajectories and cause the cluster to disperse. Due
to these disturbances, each participating satellite has its own
tracking error to the desired relative position and/or attitude.
Hence, adaptive formation controllers are required to coor-
dinate the distributed satellites to minimize the total errors
of relative position and/or attitude in the presence of un-
known disturbances acting on the ying satellites.
So far, the formation control and communication prob-
lems have been separately investigated either in control eld
or the communication domain. Since the formation control
always has a signicant effect on the communication per-
formance in the MSFF, a joint study on the formation con-
NASA/ESA Conference on Adaptive Hardware and Systems
978-0-7695-3166-3/08 $25.00 2008 IEEE
DOI 10.1109/AHS.2008.60
426
NASA/ESA Conference on Adaptive Hardware and Systems
978-0-7695-3166-3/08 $25.00 2008 IEEE
DOI 10.1109/AHS.2008.60
432
NASA/ESA Conference on Adaptive Hardware and Systems
978-0-7695-3166-3/08 $25.00 2008 IEEE
DOI 10.1109/AHS.2008.60
432
trol and communication optimization is particularly desired
when we consider how to effectively save the total energy
consumption in a satellite-based WSN. Toward this end,
we jointly address the adaptive formation control and bio-
inspired optimization for a cluster-based satellite WSN. To
overcome the drawbacks of the Clohessy-Wiltshire model
[8], a full nonlinear relative dynamics which describes the
relative positioning between two satellites is utilized in this
study. First, the original nonlinear dynamic model is trans-
formed into a controllable linear system with a perturba-
tion term. As such, the established linear control theory and
techniques such as linear quadratic regulators (LQR) can be
applied to controller design.
Second, we deal with the problem on minimizing the
total energy consumption in the cluster-based WSN where
a set of satellite nodes simultaneously transmit data to the
cluster head in a way of single hop. The total energy con-
sumption needs to be minimized under the required QoS
(quality of service) by jointly optimizing the transmission
power and rate for each satellite node in the network. Com-
pared with other tradeoff problem between transmission en-
ergy and transmission time in the context of general wire-
less networks, both the transmit and circuit energy con-
sumption have to be simultaneously considered in energy-
constrained WSNs. In general, this is a non-convex opti-
mization problem and often difcult to solve analytically.
Due to this nature of the optimization problems arising from
WSNs, traditional optimization methods such as gradient-
based iteration often cannot meet the optimization require-
ments. As a result, bio-inspired optimization techniques
such as Genetic Algorithm (GA), PSO, etc. have become
promising in these applications [7, 9].
2 System Model and Network Architecture
2.1 System Model
Now let us consider a formation of (N+1) satellites and
present the relative dynamics for the MSFF system involv-
ing a leader satellite (cluster head) and follower satellites
1, , N, where N is a given positive integer. A schematic
drawing of the MSFF system is illustrated in Fig. 1. The
inertial coordinate system I is attached to the center of
the Earth. The body coordinate system B is attached to
the leader satellite with the zaxis pointing along the or-
bital angular momentum of the leader satellite, the yaxis
pointing along the direction of the vector R, and the xaxis
being mutually perpendicular to the zaxis and yaxis,
and pointing to the direction such that {x, y , z} forms a
right-hand coordinate frame. The frame B provides a ba-
sic reference coordinate system so that one can develop the
nonlinear formation model which characterizes the relative
position dynamics of the follower satellite i relative to the

Y
X
O
x
y
z
i
r
i
R

Leader Satellite
Follower Satellite
Follower Satellite
Follower Satellite
Earth
o
i
o
Fig. 1. Multiple satellite formation ying: a
schematic illustration
leader satellite.
In this study, the point-mass assumption for each satel-
lite is used and it is assumed that each satellite evolves in
free space, namely, there are no collisions among the par-
ticipating satellites. The leader satellite is supposed to be
in a circular orbit around the Earth with a constant angular
velocity . Then, referring to Fig.1 and [4], the nonlinear
position dynamics of the follower satellite i (i = 1, , N)
with respect to the leader satellite can be derived as follows:
q +S() q +T()q +H(q, R) +F
d
= u (1)
where q(t) = (x
i
(t), y
i
(t), z
i
(t))
T
R
3
denotes the rela-
tive position vector, S() R
33
is the Coriolis-like ma-
trix given by
S() =
_
_
0 2 0
2 0 0
0 0 0
_
_
(2)
T() R
33
is a matrix taking the following form
T() = diag (
2
,
2
, 0) (3)
H(q, R) R
3
is a nonlinear term dened by
H(q, R) =
_
_
M Gx
i
/R+q
3
M G
_
(y
i
+R)/R +q
3
1/R
2

M Gz
i
/R+q
3
_
_
(4)
427 433 433
where M and Gare the Earths mass and the universal grav-
ity constant, respectively. The total, time-invariant, un-
known disturbance vector F
d
is given by
F
d
= f
di
/m
i
f
d0
/m
0
(5)
and the total control input vector u is dened as
u = (u
x
, u
y
, u
z
)
T
= f
i
/m
i
f
0
/m
0
(6)
where m
0
, m
i
are the masses of the leader and follower
satellites, F
d0
, F
di
R
3
are the vectors of the unknown
disturbances, and f
0
, f
i
R
3
are the actual control force
vectors of the leader and follower satellites, respectively.
The disturbances (F
dx
, F
dy
, and F
dz
) in F
d
are the net
specic forces applied to the leader and follower satellite i.
These disturbance forces are only the overall effects of dif-
ferent disturbance sources, such as gravity harmonics from
a non-spherical Earth, solar radiation pressure, differential
drag, unmodeled dynamics, etc. It should be noted that the
open-loop spacecraft relative position dynamics are inher-
ently unstable [6]. In the absence of any control input, a
nonzero initial state or a nonzero exogenous disturbance
will cause the cluster to disperse.
2.2 Network Architecture and Energy
Model
In the following, we discuss how to apply the cluster-
based WSN in [7, 9, 10] to the aforementioned MSFF sys-
tem. If we let the leader satellite serve as a cluster head,
then the MSFF becomes a typical cluster-based satellite
WSN. Within each cluster, the follower satellites directly
send their data to the leader (cluster head). The leader satel-
lite will then transmit all the data it has received to the leader
of another MSFF system or ground station by way of one or
several hops.
Let P
ti
denote the transmission power for satellite i
(i = 1, , N) to transmit B
i
bits during a transmission
duration T
i
.
i
is used to denote the minimum bit-energy-
to-interference-ratio threshold for the received signal from
the sensor node i. T
limit
i
and P
max
represent the upper
limit on the transmission delay and the maximum transmit
power, respectively for the follower satellite i. Thus, the
QoS requirement for the satellite i can be characterized by
the triple (
i
, T
limit
i
, P
max
) [7, 9, 10].
The energy consumption E
i
of the ith satellite node dur-
ing the transmission time T
i
is dened by [10]
E
i
=
1

(P
ti
T
i
+
i
T
i
) =
1

(P
ti
+
ci
) T
i
(7)
where
ci
is the equivalent circuit power consumption, is
the efciency of power amplier. For the N active satellite
nodes in the cluster, the total energy consumption is com-
puted by summing all the energy consumption of nodes, i.e.
E =
N

i=1
E
i
=
1

i=1
(P
ti
+
ci
) T
i
(8)
3 Statement of Control and Optimization
Problems
3.1 Control Problem
Problem 1 (Formation Control) Given a desired relative
position q
d
= (x
d
i
, y
d
i
, z
d
i
)
T
R
3
for the follower satellite
i with respect to the leader satellite, design an adaptive for-
mation controller u for the system in (1) in the presence of
the unknown disturbance vector F
d
such that the follower
satellite i is forced from an initial state x(0) to the desired
relative position q
d
as t .
3.2 Optimization Problem
For the cluster-based WSN [7, 9, 10], it is crucial to nd
the optimal transmission power P

ti
and transmission dura-
tion T

i
of each satellite node i such that the total energy
consumption of the whole cluster is minimized to trans-
mit
N

i=1
B
i
bits. Let us denote P
t
= (P
t1
, , P
tN
)
T
and
T = (T
1
, , T
N
)
T
.
Problem 2 (Optimization problem) Find the optimal P

t
and T

to minimize the total energy consumption, i.e.


min f
1
(P
t
, T) =
1

i=1
(P
ti
+
ci
) T
i
(9)
Subject to:
g
i
(P
t
, T)
i
, 0 < T
i
T
limit
i
, 0 < P
ti
P
max
i = 1, , N
where g
i
(P
t
, T) is the received bit-energy-to-interference-
density ratio at the cluster head for the ith satellite node,
and given by [10]
g
i
(P
t
, T) =
_
E
b
Ic
_
i
=
W
ri
hiPti

j=1,j=i
hjPtj+NcW
=
W
Bi
hiPtiTi

j=1,j=i
hjPtj+NcW
(10)
where r
i
= B
i
/T
i
represents the transmission rate and is
the orthogonality factor denoting multiple access interfer-
ence (MAI) from the imperfect orthogonal spreading codes
428 434 434
and the asynchronous chips across simultaneous transmit-
ting satellite nodes. N
c
is the single-sided power spectrum
density of AWGN (additive white Gaussian noise), W is the
spread spectrum bandwidth, and h
i
is the channel gain of
the ith satellite node.
In this study the channel gain h
i
is simply dened as
h
i
=
G
ti
G
rc

2
16
2
d
2
i

(11)
where G
ti
and G
rc
are the gain of the transmitter and the
receiver, respectively. = c
l
/f is the wavelength, f is the
frequency, c
l
is the speed of light, is the systemloss factor,
d
i
is the communication distance between the ith satellite
and the cluster head, is an engineering coefcient.
4 Adaptive Formation Controller
4.1 Input-output Feedback Linearization
To facilitate the control design, let us rewrite (1) into the
following form:
x = f(x) +(x) + [g
1
(x), g
2
(x), g
3
(x)] u
:= f(x) +(x) +g(x) u
y = (x
i
x
d
i
, y
i
y
d
i
, z
i
z
d
i
)
T
:= h(x)
(12)
where x= (x
i
, x
i
, y
i
, y
i
, z
i
, z
i
)
T
R
6
, y= (y
1
, y
2
, y
3
)
T

R
3
, and the control input u is the same as in (6). f(x) is
the smooth vector eld dened as
f(x) =
_

_
x
i
2 y
i
+
2
x
i
H
x
y
i
2 x
i
+
2
y
i
H
y
z
i
H
z
_

_
(13)
with H
x
, H
y
, H
z
being the components of the nonlinear
term H(q, R) in (4). H
x
, H
y
, and H
z
are dened by
H
x
= M Gx
i
/[(R +z
i
)
2
+x
2
i
+z
2
i
]
3/2
(14)
H
y
= M G{(R +z
i
)/[(R +z
i
)
2
+x
2
i
+z
2
i
]
3/2
1/R
2
}
(15)
and
H
z
= M Gz
i
/[(R +z
i
)
2
+x
2
i
+z
2
i
]
3/2
(16)
respectively. The smooth vector elds g
1
(x), g
2
(x),
g
3
(x) have the following forms
g
1
(x) = (0, 1, 0, 0, 0, 0)
T
, g
2
(x) = (0, 0, 0, 1, 0, 0)
T
g
3
(x) = (0, 0, 0, 0, 0, 1)
T
(17)
In (12), (x) represents the disturbances applied to the
nominal system (f, g, h) and is dened as
(x) = (0, F
dx
, 0, F
dy
, 0, F
dz
)
T
(18)
with F
dx
, F
dy
, F
dz
being the components of the unknown
disturbance vector F
d
.
First, let us dene the change of coordinates as follows:
z
1
= y
1
= x
i
x
id
, z
2
= y
1
= x
i
z
3
= y
2
= y
i
y
id
, z
4
= y
2
= y
i
z
5
= y
3
= z
i
z
id
, z
6
= y
3
= z
i
(19)
In this new coordinates, the nominal system (f, g, h)
in (12) is exactly feedback linearized and described by the
three chains of integrators and no extra equations are in-
volved. When the perturbation term is considered, the feed-
back linearized system can be derived as follows:
z
1
= z
2
+L

h
1
(x), z
2
= v
1
+L

L
f
h
1
(x)
z
3
= z
4
+L

h
2
(x), z
4
= v
2
+L

L
f
h
2
(x)
z
5
= z
6
+L

h
3
(x), z
6
= v
3
+L

L
f
h
3
(x)
(20)
where v
1
= y
1
, v
2
= y
2
, and v
3
= y
3
. L

h
i
(i = 1, 2, 3)
denotes the Lie derivative of a real-valued scalar function
h
i
along a vector eld .
Next, denoting by w= (L

L
f
h
1
(x), L

L
f
h
2
(x),
L

L
f
h
3
(x ))
T
R
3
the new disturbance term then yields
(x) +w +(x) u = v (21)
where
(x) =
_
_
2 y
i
+
2
x
i
H
x
2 x
i
+
2
y
i
H
y
H
z
_
_
(22)
Because the decoupling matrix (x) is globally invert-
ible, the actual input u can be easily computed from (21)
if the disturbance term w can be directly measured or esti-
mated. However, this requires a highly precise estimator for
w. It is often unavailable in practice. So it is highly desired
to have an adaptive controller which is purely designed in
the new coordinate system such that the virtual control v
can adaptively remove the effects of the disturbance termw
and the actual input u can be simply obtained as follows:
u =
1
(x) [v (x)] (23)
Finally, the perturbed system (20) can be arranged into a
standard linear system form that facilitates the subsequent
design and stability analysis of the adaptive formation con-
trol law. From (20) it follows that
z = Az +Bv +w
y = C z
(24)
where C R
66
, the matrices A R
66
, B R
63
, and
R
33
have the so-called Brunowsky canonical form. It
follows that rank(B AB A
5
B) = 6 which is equiv-
alent to the number of state variables in (24). Hence, the
linearized system (A, B) is controllable.
429 435 435
4.2 Controller Design
The disturbance forces and the satellite masses in the
preceding linear model (1) vary slowly in time. As such,
m
0
, m
i
can be considered as two constant parameters and
F
d
is a constant vector. Moreover, it is assumed in this study
that the continuous-time control forces can be provided by
the satellite actuators.
Under the aforementioned assumptions, there only ex-
ists a constant, unknown disturbance vector in (24). Thus,
in what follows we can employ the integral feedback tech-
nique [11] fromlinear control systemtheory to the design of
an adaptive formation controller for improving steady-state
performance of the system in the presence of the unknown,
constant disturbance vector w.
First, consider the following nominal linear system
z = Az +Bv
y = C z
(25)
and the stabilizing control law
v = Kz, K R
36
(26)
The closed-loop system is described by
z = (A+BK)z (27)
Since z converges to zero as t , z = 0 at steady-
state. However, the state of the perturbed linear system de-
scribed by (24) cannot tend to zero as t . At steady-
state, z = 0 then it follows
z = (A+BK)z +w = 0 (28)
which yields
z
s
= (A+BK)
1
w (29)
Thus, the existing disturbance will result in an off-set in
the systems steady-state, i.e., the system state reaches to a
nonzero state z
s
at steady-state. In view of (29), it is found
that the off-set magnitude depends upon the magnitude of
the disturbance. To suppress this off-set, the so-called inte-
gral state R
3
is needed and dened by
=
_
t
0
y() d (30)
Differentiating (30) with respect to time then yields

= y = C z (31)
Adding the differential equation (31) into the original
system (24), the resulting augmented system is as follows:
_
z

_
=
_
A 0
C 0
_ _
z

_
+
_
B
0
_
v +
_

0
_
w
:=

A z +

Bv +

w
y =
_
C 0

_
z

_
:=

C z
(32)
where the new state vector z:= (z, )
T
R
9
.
The preceding integral feedback-based design scheme
establishes the following proposition:
Proposition 1 The augmented linear system (

A,

B) is
controllable from the input v. If a stabilizing controller
v =

Kz R
39
is given for the augmented system, then
the output y of (32) will be driven to a no-offset condition
by integrating action in (30), i.e.
lim
t
y(t) = 0 (33)
Proof 1 In the last section, it has been shown that the orig-
inal linear system (A, B) is controllable. To prove the
controllability of the augmented linear system(

A,

B), one
only needs to check the controllable rank condition of the
augmented linear system (

A,

B). The controllability ma-
trix of the augmented linear system (

A,

B) is
U =
_
B AB A
2
B A
8
B
0 C B C AB C A
7
B
_
(34)
After a tedious calculation, one has rank(U)= 9, which
indicates that the augmented linear system (

A,

B) is con-
trollable.
Having had a test for the controllability of the augmented
linear system (

A,

B), one can design a stabilizing control
law v =

Kz which forces the augmented system with a
constant disturbance to a steady state as t . At steady-
state

z = 0, therefore
lim
t

(t) = 0, i.e., lim


t
y(t) = 0 (35)
In the following, we use the LQR technique to design
an adaptive formation controller in the following sense of
optimality:
J =
1
2
_

0
[z
T
(t) Qz(t) +v
T
(t) Sv(t)] dt (36)
where Q R
99
and S R
33
.
A unique optimal control v
opt
that minimizes J exists
and is given by
v
opt
=

Kz,

K = S
1
B
T
P (37)
P is obtained by solving the following algebraic Riccati
equation
P A+A
T
P +QP BS
1
B
T
P = 0 (38)
430 436 436
5 Particle Swarm Optimization
Many computational algorithms have been inspired by
analogy with biological systems, and in particular, with evo-
lution by natural selection. Particle swarm optimization
(PSO) is one of the currently popular bio-inspired optimiza-
tion techniques. It was proposed by Eberhart and Kennedy
in 1995 [12]. The PSO mimics the social behavior of bird
ocking or sh schooling. In the PSO, the basic idea is to
follow the leader particle which is nearest to the optimum.
Thus, the individuals in the PSO always move through the
problem space by following the current optimal particles
in the population. In addition, each individual also tries to
track its own best solution so far. Individuals move at a ve-
locity which is updated so as to converge towards the local
and the global best, but at a random rate.
Let x
p,i
denote the personal best achieved so far by par-
ticle i. The global best found by the whole population is
represented by x
g
. In the basic PSO algorithm, a particle
updates its velocity and position as follows:
v
i
(t + 1) = wv
i
(t) +c
1
r
1
[x
p,i
(t) x
i
(t)]
+c
2
r
2
[x
g
(t) x
i
(t)]
(39)
x
i
(t + 1) = x
i
(t) +v
i
(t + 1) (40)
where v
i
and x
i
are the particle velocity and position at
the tth generation, respectively. c
1
and c
2
are the scaling
constants, usually taken as c
1
= c
2
= 2.0. w is the in-
ertia weight and used to control the trade-off between the
global and the local exploration ability of the swarm. Ran-
dom numbers r
1
and r
2
are uniformly distributed in [0, 1].
The main reasons for applying the PSO to solving opti-
mization problems from cluster-based satellite WSNs lie in
the following two advantages. First, the convergence rate
in the PSO is much faster than traditional evolutionary al-
gorithms, such as GA (Genetic Algorithm). Thus, the com-
putation time is usually short. Second, PSO is extremely
computation-efcient. Since the number of particles (i.e.,
population size) is very small, the required computation re-
sources can be signicantly reduced. These two advantages
make the PSOmost attractive in the applications of satellite-
based WSN systems since all kinds of resources including
computation and control are very limited and expensive.
6 Simulation Results
6.1 Adaptive Formation Controller
In the following illustrations, the formation eet consist-
ing of two satellites (i.e., a leader satellite and a follower
satellite) was simulated with the following parameters [4]:
M = 5.974 10
24
kg, m
0
= 1550 kg, m
1
= 410 kg
G = 6.673 10
11
m
3
/kg s
2
, R = 4.224 10
7
m
= 7.272 10
5
rad/s
(41)
The net specic disturbance forces were taken as
F
d
= [5.25, 6.248, 2.415]
T
10
7
N/ kg (42)
The desired relative position and the initial state in the
frame B are set to be as follows:
q
d
= [300, 200, 100]
T
m
x(0) = [250, 0.05, 170, 0.01, 140, 0.02]
T
(43)
The corresponding orbits of the leader and follower are de-
termined by Rand R+r, respectively.
The simulation example provided in this section is to il-
lustrate the control performance of the adaptive formation
controller for formation maintenance and/or formation re-
conguration in the presence of the net specic disturbance
forces given by (42). The objective is to show that the fol-
lower satellite can be forced from the initial state x(0) to
the desired position q
d
by the designed adaptive formation
controller u. In the simulation, the total simulation time
was set to be 3.0 hours.
The simulation results obtained by using the adaptive
formation controller designed via the integral feedback
scheme and LQR technique are illustrated with the follow-
ing choice of design parameters:
Q = I
99
10
5
, S = I
33
10
10
(44)
By solving the algebraic Riccati equation in (38) with
the above matrices and substituting its solution into (37),
we obtained the following optimal feedback matrix

K

K = 1.0 10
21

_

K
1
,

K
2
,

K
3

(45)
where

K1 =
_
_
1 2 161.914
2.8698 582.736 1
3.74036 12.5315 116.115
_
_

K2 =
_
_
27751.7 51.4699 15483.3
2 79.8878 14963.9
8563.36 1 2
_
_

K3 =
_
_
3 0.400969 0.314176
0.00257717 3 0.210918
0.00597677 0.408286 3
_
_
with
1
= 2.0 10
16
,
2
= 6.32456 10
18
,
3
=
3.16228 10
13
.
The time histories of state x are shown in Fig. 2. Fig-
ure 3 illustrates the control input u and output y. Figure
431 437 437
0 1 2 3
350
300
250
200

0 1 2 3
0.2
0
0.2

0 1 2 3
160
180
200
220

0 1 2 3
0.1
0
0.1

0 1 2 3
50
100
150
0 1 2 3
0.2
0
0.2
t (hr) t (hr)
x
i
(
m
)

x
i
(
m
/
s
)
y
i
(
m
)

y
i
(
m
/
s
)
z
i
(
m
)

z
i
(
m
/
s
)
Fig. 2. State x
0 1 2 3
2
1
0
1
x 10
3

0 1 2 3
5
0
5
10
x 10
4

0 1 2 3
10
5
0
5
x 10
4

0 1 2 3
50
0
50
100

0 1 2 3
40
20
0
20
0 1 2 3
50
0
50
t (hr) t (hr)
u
x
(
N
/
k
g
)
u
y
(
N
/
k
g
)
u
z
(
N
/
k
g
)
y
1
(
m
)
y
2
(
m
)
y
3
(
m
)
Fig. 3. Input u and output y
0 0.5 1 1.5 2 2.5 3
330
340
350
360
370
380
390
400
D
i
s
t
a
n
c
e
(
m
)
t (hr)
Fig. 4. Cluster distance
4 shows the changes of the cluster distance over the regu-
lation time. These gures were obtained by integrating the
disturbed closed-loop system with the integration feedback
action. From Fig. 3 it can be observed that the output of
the perturbed system can be driven to the desired relative
position of the follower satellite without any offset by the
integral feedback action in spite of the presence of the un-
known, constant disturbances.
6.2 Bio-inspired Optimization
The basic WSN system parameters used in the following
numerical experiment are given in Tab. 1.
Tab. 1. Basic system parameters
Parameter Value Parameter Value
f, GHz 2.4 0.9

ci
1.0 2/3
1.0 Nc, W/Hz 10
15
T
limit
i
, s 1.0 W, MHz 1.0
Pmax, mW 100.0 G
ti
Grc 100.0
c
l
, m/s 3.0 10
8
, m 0.125
B
i
, K 1.0
i
4.0
1.0
In this experiment, we consider the same scenario as in
the simulation for the adaptive formation controller, but we
will study how the optimization is affected by the forma-
tion controller in its regulation process. Toward this end,
we compare the optimization results using the PSO for the
two points, i.e., t = 0 and t = 0.55 hours, respectively.
The point at t = 0 is the time when the adaptive formation
controller just started to work. The second point t = 0.55 is
investigated because it is very close to the new steady state,
as shown in Figs. 2-4. The control parameters for the PSO
used in this experiment are summarized in Tab. 2.
Tab. 2. PSO Parameters
Parameter value Parameter value Parameter value
ng 200 np 50 c
1
2.0
c
2
2.0 w 0.721348 [0.01, 0.1]
[0.01, 1.0]
The best and mean tness found by the PSO are shown in
Figs. 5 and 6, respectively. The optimization results are also
given in Tab. 3. From Figs. 5-6 and Tab. 3, we can clearly
see that the PSO-based online optimizations are particularly
needed to obtain optimal transmit parameters for minimiz-
ing the total energy consumption during the regulation pe-
riod of the adaptive formation controller. When adaptively
optimizing the satellite-based WSN, the total communica-
tion energy can be signicantly saved. Since the point 2 is
much closer to the steady state, it can be used as a reference
point when there are no adaptive optimization in the forma-
tion control. So, in comparison with the point 2, the total
432 438 438
energy savings at the point 1 can be up to 19.67% when the
PSO-based adaptive optimization is employed.
0 20 40 60 80 100 120 140 160 180 200
0.92
0.93
0.94
0.95
0.96
0.97
0.98
0.99
Generation
B
e
s
t

f
it
n
e
s
s
Best fitness over generation
t=0.0 hour
t=0.55 hour
Fig. 5. Best tness for t = 0 and t = 0.55
hours, respectively
0 20 40 60 80 100 120 140 160 180 200
0
0.1
0.2
0.3
0.4
0.5
0.6
Generation
M
e
a
n

f
it
n
e
s
s
Mean fitness over generation
t=0.0 hour
t=0.55 hour
Fig. 6. Mean tness for t = 0 and t = 0.55
hours, respectively
7 Conclusions
This paper has jointly addressed the adaptive formation
control and bio-inspired optimization for a cluster-based
satellite wireless sensor network in which there are mul-
tiple satellites ying in formation in the presence of un-
known disturbances. The full nonlinear dynamics model
describing the relative positioning of the MSFF system was
used to develop the adaptive formation controller. By using
the integral feedback design scheme, an adaptive formation
controller has been presented for improving the steady-state
performance of the MSFF system in the presence of un-
known disturbances. As a currently popular bio-inspired
algorithm, PSO was also employed to minimize the total
energy consumption under the required quality of service
by jointly optimizing the transmission power and rate for
each satellite node in the network. Simulation results were
provided to demonstrate the effectiveness of the adaptive
formation controller and the bio-inspired optimization ap-
proach for saving the total communication energy in the
cluster-based satellite WSN.
Tab. 3. Optimization results with the PSO
Time,
hour
Distance,
m
Optimal
P,W
Optimal
T, s
Best
tness
Total Energy
Consumption,J
0.0 333.1666 0.028862 0.010092 0.9885297 0.0115366
0.55 371.5796 0.023239 0.012143 0.9862885 0.0138064
Acknowledgment
The authors thank all of the team members of the ES-
PACENET project, which involves the Universities of Edin-
burgh, Surrey, Essex, and Kent, Surrey Satellite Technology
(SSTL), NASA Jet Propulsion Laboratory (JPL), EPSON,
and Spiral Gateway.
References
[1] H.-H. Yeh, E. Nelson, and A. G. Sparks. Nonlinear tracking control
for satellite formations. Journal of Guidance, Control, and Dynam-
ics, 25(2):376386, 2002.
[2] J. R. Carpenter. Decentralized control of satellite formations. Inter-
national Journal of Robust and Nonlinear Control, 12(2-3):141161,
2002.
[3] M. B. Milam, N. Petit, and R.d M. Murray. Constrained trajectory
generation for micro-satellite formation ying. August 2001. AIAA
Paper 2001-4030.
[4] M. S. de Queiroz, V. Kapila, and Q. Yan. Adaptive nonlinear con-
trol of multiple spacecraft formation ying. Journal of Guidance,
Control, and Dynamics, 23(3):385390, 2000.
[5] A. Krishnamurthy and R. Preis. Satellite formation: a mobile sensor
network in space. In Proc. 19th IEEE Int. Parallel & Distributed
Processing Symp., pp. 243249, Denver, Colorado, April 3-8 2005.
[6] V. Kapila, A. G. Sparks, J. M. Bufngton, and Q. Yan. Spacecraft
formation ying: Dynamics and control. Journal of Guidance, Con-
trol, and Dynamics, 23(3):561564, 2000.
[7] E. Yang, N. Haridas, A. El-Rayis, A. T. Erdogan, T. Arslan, and
N. H. Barton. Multiobjective optimal design of MEMS-based recon-
gurable and evolvable sensor networks for space applications. In
Proc. 2nd NASA/ESA Conf. Adapt. Hardware & Systems, pp. 2734,
Edinburgh, Aug. 5-8 2007.
[8] W. H. Clohessy and R. S. Wiltshire. Terminal guidance system for
satellite rendezvous. Journal of Aerospace Science, 27(9):653658,
1960.
[9] E. Yang, A. T. Erdogan, T. Arslan, and N. Barton. Multi-objective
evolutionary optimizations of a space-based recongurable sensor
network under hard constraints. In Proc. ECSIS Symp. Bio-inspired,
Learning, and Intell. Systs. for Security, pp. 7275, Edinburgh, Aug.
9-10 2007.
[10] T. Shu and M. Krunz. Joint power/rate optimization for CDMA-
based wireless sensor networks. In Proc. 3rd Int. Workshop on Mea-
surement, Modelling, and Performance Analysis of Wireless Sensor
Networks, pp. 106115, San Diego, July 21 2005.
[11] M. Gopal. Modern Control System Theory. Wiley Eastern Limited,
New Delhi, India, second edition, 1993.
[12] R. C. Eberhart and J. Kennedy. Anewoptimizer using particle swarm
theory. In Proc. 6th Int. Symp. on Micro- machine and Human Sci-
ence, pp. 3943, Nagoya, Oct. 4-6 1995.
433 439 439

You might also like