You are on page 1of 7

Stochastic Time-Scale Characterization of Nonstationary Underwater

Communication Channel
Uche A.K. Okonkwo1, Razali Ngah2, Zabih Ghassemlooy3 , Tharek Abd. Rahman4
1,2,4
Wireless Communication Center Faculty of Electrical and Electronic Engineering
Universiti Teknologi Malaysia
83100 Skudai, Johor, Malaysia
E-mail: uchechude@ieee.org, razalin@fke.utm.my, tharek@fke.utm.my
3
Optical Communication Research Group,
School of Computing, Engineering & Information Sciences,
University of Northumbria, Newcastle, UK.
E-mail: fary.ghassemlooy@unn.ac.uk

Abstract sensors placed underwater, to the surface of the ocean.


From there it is possible to relay the data via a satellite to a
The underwater acoustic communication channel is one of data collection centre.
the complex and challenging channels for communication.
Due to the poor propagation capability of the
In most cases there is the need to provide communications
electromagnetic waves in sea water which is attributed
between mobile and stationary terminals. And because of
mainly to the skin effect, acoustic signals provide the most
the spherical degree of freedom for the mobile terminal,
obvious medium to enable underwater communications.
such channel is characterized as highly nonstationary. In
This limits the available bandwidth for the underwater
order to account for nonstationarity, channel
acoustic (UWA) communication to the kilo Hertz range [2].
characterization that employs the non wide-sense
More also challenging issues like the refractive properties
stationary uncorrelated scattering (non-WSSUS) approach
of the UWA channel, severe fading, multipath, rapid time-
is necessary. More also the inadequacy of Doppler shift in
variation and large Doppler spread, impedes on the
accounting for frequency shift of the channel operator
performance of the system [1], [2], [3]. Thus a good
implies that the time-frequency characterization approach
understanding of the UWA channel is important in the
is not appropriate. In this work we present the
design and simulation of the deployable component
geometrical-based stochastic time-scale characterization of
systems.
the underwater channel which emphasizes on the
nonstationary property of the channel. The effects of the The complex UWA environment remains one of the most
channel nonstationarity on the channel capacity and challenging types of channels for information transmission.
diversity gain are also addressed. From the simulated Brady and Preisig [4] described the UWA as “quite possibly
example, it is inferred that channel diversity and the nature’s more unforgiving wireless medium”. In general,
assumption of ergodic capacity depends on the number of the physical characteristics of the UWA channel are highly
independent fades which invariably depends on the dependent on the relative distance and motion of the
intervals of stationarity. terminals and the channel; the proximity and roughness of
the scattering surfaces; and the presence of ambient
Keywords: interference [5]. However, the basic channel
Underwater channel, nonstationarity, geometrical model, characterization challenges can be factored into large values
ergodic capacity, diversity, scattering function. of delay and Doppler spread. The discrepancy in terms of
the delay and Doppler spread (both are inversely
proportional to the velocity of propagation) between the
1. Introduction mobile UWA channel and the propagation in the mobile
radio channel, can be capture by τ, υ =1 / 10 3
Over the years there is the growing need for deep sea
(underwater channel), τ, υ =1 / 10 8 (mobile radio
communication among the submerged vessels and with the
surface or on-shore transceiver stations. More also the surge channel).
of ocean exploration activities has been steadily increasing. More also the non-uniform Doppler shift across the
The need for underwater wireless communication exists in composite tones (in the case of wideband signals) makes
applications such as remote control in off-shore oil industry, the evaluation of the frequency variation using Doppler
pollution monitoring in environmental systems, collection shifting as inappropriate as discussed in [6]. This issue is
of scientific data recorded at ocean-bottom stations, speech even more pronounced for the underwater OFDM
transmission between divers, and mapping of the ocean communication [5], [7]. The time-scaling defined under the
floor for the detection of objects, as well as the discovery of time-scale domain representation is a more suitable
new resources [1]. Coupled with this increase in ocean measure of frequency variation in a wideband signal. For
exploration is the need to transmit data, collected by the above reasons, the representation and characterization
of the UWA channel in the time-scale domain is more variations in some interval (acquisition interval) J ∈ ℜ .
appropriate than the Fourier domain counterpart. If we define a partition of B as the countable collection
While travelling through the underwater channel, the of subintervals Pq ⊂ B, q =1,2,.., Q , then we can
transmitted signal experiences sever distortions induced by state that:
multipath propagation. The distortions become more sever
when either or both the transmitter and the receiver are in i. Pi ∩Pj ≠ {0}; ∀i ≠ j; i, j ∈q
motion. The resultant time-varying multipath imposes sever
limitations on the system performance [1]. Since this ii. J i ∩J j
≠ {0}; ∀ i ≠ j; i, j ∈k
channel is nonstationary, the large time-variability cannot
iii.
be ignored, thus the wide-sense stationary (WSS)
assumption is violated. On the other hand, large scale Hence we can define the interval J vk , k = 1,2,.., K for
smoothness of the scattering surfaces (as well as variations
which some statistical properties of the associated process
in the mean angular spread due to motion) contributes to
under observation are assumed to be stationary. An
correlation among the multipath components from the same
important process often used to characterized and simplify
surface. Hence uncorrelated scattering (US) assumption is
slowly varying wireless channel is the wide-sense
also violated. Therefore channels with such rapid time-
stationary process.
variability and the inter-path correlations mentioned above
cannot be modelled as WSSUS [8]. Instead a non-WSSUS Definition 1: A process is called wide-sense stationary
approach is employed in the UWA channel characterization. (WSS) if it’s first two moments, the mean and
autocorrelation are independent of time t on . Such process
The approach in this paper follows a purely intuitive path
is defined on if there exist some partitions for which all
that hinges on the appropriation of varied statistical
intervals, other than very complex mathematical intervals J k provide time independence with respect to
derivations. More also instead of using the acronym ‘non- the mean and ACF.
WSSUS’ which ordinarily embodies all processes that Definition 2: A process is called local-sense stationary
cannot be defined within the WSSUS assumption (grossly (LSS) if there exist some partition P for which at least
nonstationarity inclusive), a ‘local sense’ statistical basis is
defined. The result of this local sense statistics is the local one interval say J i is considered stationary. Within some
sense stationary and uncorrelated scattering (LSSUS) valid locally stationary interval J vi the mean and the
assumption. The parameters derived from the LSSUS autocorrelation (and the associated spectral property) are
statistics are shown to be more appropriate and informative approximately independent of time and frequency, and vary
than the WSSUS in obtaining the nonstationary slowly in time and frequency across all other intervals
information. They also capture long-term channel
J k ≠i . Thus the autocorrelation and spectral characteristic
properties necessary for performance analysis.
are WSS at J vi but vary slowly across with respect to all
This paper is organized as follows. Section 2 presents the
stochastic time-scale characterization and presents the other intervals {J k ≠i } .
concept behind LSSUS assumption. In Section 3, the
For all other processes with gross time varying statistical
geometrical-based scattering model that typifies the
propagation in the UWA channel and considers terminal properties over all intervals for which no J v can be
mobility is presented. Finally examples simulations and ascertained for practical purposes, nonstationarity is
results are presented and discussed in Section 4. defined. From the above discussion, we can see that a little
above the strict-sense stationarity, the wide-sense stationary
channel is defined, and a little below the nonstationarity, the
2. Stochastic Time-Scale Characterization local-sense stationary channel is define.

The time scale representation of the time-varying channel Hence for stochastic representation of the time-varying
can be given by [3]: channel we can then define three different time instants; t
, t ′ and t ′′ . Within the quasi-stationary (WSSUS) area
∞ ∞
 t −τ  dτ ds for two time instants t and t ′ , the channels statistics are
y (t ) = ∫ ∫W (τ, s) a(t ) x
−∞−∞

s  s2 constant over ∆t = t ′ −t . However the statistics vary
(1) across the quasi-nonstationary (non-WSSUS) area over
∆t ′ = t ′ −t ′′ . Thus it can easily be shown that we can
where { y (t )} = y(t ) is the channel realization for a
given input x (t ) , and W (τ, s ) is the delay-scale define the time, scale (or frequency), scale shift (Doppler)
and delay for the LSSUS channel as:
(wideband) spreading function.
Let U be the universal set of all stochastic
processes/channels, there exist subsets of whose statistical
properties varies with certain degrees in respect to the
t ′′ − t = ∆(t + ∆t ) + ∆t ; n k Bc .n N Tc
n = = n k .n N
s ′′ − s = ∆( s + ∆s ) + ∆s; Tc Bc
~
s ′′ − ~
s = ∆(~s + ∆~s ) + ∆~
s; (6)
τ ′′ − τ = ∆(τ + ∆τ ) + ∆τ Using the expression in [9], the n -dependent ergodic
(2) capacity can be given by:
The correlation function is then given by N −1

(τ , s, ∆τ , ∆s; ∆r ) = E[W (τ , s ) W * (τ ′′, s ′′)] Cerg = ∑ ( log 2 (1 + qn ) p n ( q) )


 (7)
RW,LSSUS
n =1

[ ]
. It can easily be shown that:
2
 Pav . χn (t , f )
RW, LSSUS (τ, s, ∆τ, ∆s; ∆r ) = where qn = with probability
τ,s
N0
. Ry , X ∆(τ +∆τ ) +∆τ , ∆( s +∆s ) +∆s δ, X τ , s
∆( t +∆t ) +∆t t distribution p n (q) .
(3)
It is evident that diversity performance improves
where monotonically with increasing number of i.i.d [10]. In fact
τ, s
X∆(τ +∆τ ) +∆τ , ∆( s +∆s ) +∆s = a l ( ∆(t + ∆t ) + ∆t +t ) as the number of i.i.d approaches infinity, the performance
of coherent diversity reception converges to the
 ( ∆(t + ∆t ) + ∆t +t ) −( ∆(τ + ∆τ ) + ∆τ +τ )  performance over a non-fading AWGN channel [12], [13].
. x
 
 By decoupling the stationarity region onto the time and
 ∆( s + ∆s ) + ∆s + s 
frequency region, the number of the i.i.d n or diversity
τ ,s  t −τ  order can be approximately given as:
and X = al (t ) x  . The first inner product term
 s  Ts
in (3) is called the local-sense Scattering function (LSF): nTD = → Time diversity (8)
Tc
 τ, s
PLSSUS (τ, s ) = R y , X ∆(τ+∆τ) +∆τ, ∆( s +∆s ) +∆s ∆( t +∆t ) +∆t
Bs
(4) nTD = → Frequency diversity (9)
Bc
Implicitly:

PWSSUS

(τ, s ) =PLSSUS (τ, s ) Ts B s
∆t ,∆τ, ∆s→0 nTD = → Time-Frequency diversity (10)
(5) Tc B c
The channel is then defined by the coherence bandwidth The expressions (8)-(10) imply that as Ts and B s are
Bc , coherence time Tc , stationarity bandwidth B s and reduced by virtue of decrease in the correlation among
stationarity time Ts : channel realizations, the diversity order reduces. Hence the
stationarity intervals set threshold and point of reference for
1 employing different diversity schemes.
Bc =  ,
5 Prms ,τ ,WSSUS
0 .4 1
Tc ≈  Bs =  ,
Prms , s ,WSSUS f c 5 ∆Prms ,τ ,WSSUS
0.4
Ts ≈  .
∆Prms , s ,WSSUS f c

 
where Prms ,τ,. and Prms ,s ,. are the respective delay
and scale profiles,

∆Prms , s ,WSSUS ( 
= (Prms , s , LSSUS

) −(Prms , s ,WSSUS ) )
. The parameters B s and Ts describe the extent of
channel variation and tends to infinity in the case of
WSSUS
Hence for the flat-fading slowly varying channel, the
number of i.i.d n is approximately given by the
stationarity and coherence regions:
3. Geometrical-based UWA Channel Model within this context the transmission loss is [11]:

Tloss = 10 log 10 ( r 2 )
A geometrical-based single bounce scattering model is
(11)
provided below in order to model and simulate the UWA
channel. Unlike the elliptical and the circular models 5. The frequency dependent of the propagation paths
described for the conventional terrestrial MRC in which is not taken into account.
mobility is restricted to the azimuthal angles and the
scattering region is defined over a circular or elliptical The geometric distribution f x , y of the N scatterers can
volume, the approach in the case of UWA channel is be defined using any of the appropriate known statistical
slightly different. Often scatterers in UWA channels are distribution functions where f x , y is independent of
located at the top and bottom of the water volume, thus the frequency. To obtain the delays associated with all
water volume can be consider as being a large rectangular multipath components (MPCs), the total path lengths have
volume with the scatterers distributed on the top and bottom to be obtained by considering Figure 1. Let the reference
lids. And the mobility of the mobile unit (MU) involves
point (0,0) be the receiver position MU (0,0) . The
both the azimuthal and the polar angles of movement, thus
the spherical coordinates are more appropriate for its path length R from MU (0,0) to BS through
position descriptions. In the ensuing discussion the s n ( x n , y n ) is given by:
geometrical-based single bounce sphero-rectangular
scattering (GBSBSRS) model for the UWA channel is Rn = { f n + g n } , n =1,2,.., N
introduced as shown in Figure 1. (12)
where:

( )
1
f n = H n2 + ( X − x n ) 2 2 (13)

(
en = a 2 + ( x n ) 2 ) (14)

and X is the distance between the MU and the BS


projected on the x-axis. The angle-of-arrival (AOA) θ is
given by:

Figure 1- Illustration of Geometrical-based single bounce ( (


θ n = co s− 1 ( 2 f n D ) − 1 D 2 + f n2 − g n 2 )
sphero-rectangular scattering (GBSBSRS) model for the (15)
underwater channel.
For the MU moving with a velocity v , its position at any
given time can be described as MU (r , φ, Φ) . The
Each scatterer is defined as a vector s n in a hypothetical evolved path length R ′ through the evolved scatterers’
spatial coordinate ( x, y , z ) . For simplicity let z = 0 , position s n ( x n′ , y n′ ) with reference to its position at B is
hence the scatterers coordinates can be specified by given by:
s n ( x, y ) bounded by the depth of the water and some
horizontal length determined by physical constraint or Rn′ = { f n′ + g n′ } , n =1,2,.., N
assumed channel length. For the model above, the (16)
following assumptions are made:
where
1. The temperature of the water volume is constant 1
over the period of simulation. 

1 
2 2

f n′ =( q 2 −( r cos φ) 2 ) 2  +( y ′ +r cos Φ)2 
  n
2. The wind speed v is very small such that the   

average height (meters) of the one-third highest (17)
waves expressed by H 1 / 3 = 0.566 ×10 −2 v 2
[11] is approximately zero. (
e′n = a ′ 2 + ( x′n ) 2 ) (18)
3. The floor of the water volume is smooth, non-
absorptive and homogeneous. and:
4. The water volume is isotropic, i.e., there is no
absorption effect. Hence the sound intensity I int
falls off as the inverse of the range r , so that
Figure 2- WSSUS scattering function for the UWA channel

( )
1 at ∆t s and 5 m/s.

p = x′n2 + r 2 − 2x′n2 r c
a ′ = a − r cos( φ) ,
oΦ s2 1

Normalized Power Delay Profile


0.9

0.8

0.7

4. Numerical Results and Discussion 0.6

0.5

0.4

Consider an underwater communication between the 0.3

0.2
hydrophone which serves as the base station (BS) and the 0.1
mobile unit (MU) located 50 m apart on the average as 0
0.033 0.034 0.035 0.036 0.037 0.038 0.039 0.04
depicted in Figure 1. Assume that the operating bandwidth Delayτ ( sec)

is 10 KHz and the MU is moving from an initial position A Figure 3- Normalized power delay profile against the
(spherical coordinate of A is MU (0,0,0) ) at a constant
delay for the UWA channel at 5 m/s and ∆t s .
velocity of 5 m/s to another position B ( MU ( r , φ, Φ)
). The following parameters are also defined for this
communication channel:
Water depth = 20 m; Vertical distance of hydrophone from
the surface = 10 m; Initial vertical distance of the MU from
the surface = 10 m, and the spatial extension r ,

 c 
r =  + v∆t 
2
 c f 
For ∆t ={0.08 , 0.1, 0.2, 0.5, 0.8, 1.0, 3.0}
sec, φ =120 0 and Φ = 60 0 , if we assume that the
speed of propagation of sound in water is 1500 m/s, the
resultant channel responses are shown in the Figure 2. The (a) (b)
stationarity time is obtained using ∆t s = λ / 2v , where v
is the speed of the MU.
In simulating the above synthesized channel, the test signal
used is also the Mexican hat wavelet. The resultant delay-
scale scattering functions PWSSUS (τ , s ) and
PLSSUS (τ , s ) are shown in Figure 2, and Figure 4,
respectively. From the scattering function, the power delay
profile (PDP) for the WSSUS case is derived and shown in
Figure 3. The PDP P (τ) is obtained by taking the
normalized power values at P (τ , s ) = P (τ , s min ) over
the delay bins. To obtain the equivalent scale profile
P (s ) the normalized power values is taken at (c) (d)
P (τ , s ) = P (τ min , s ) over the scale bins. The plots of
Figure 4- LSSUS scattering functions at (a) 0.5 sec (b) 0.8
PLSSUS (τ , s ) for ∆t ={ 1.0, 3.0, 5.0, 10 .0} are sec (c) 1.0 sec (d) 3.0 sec (at 5 m/s).
shown in Figure 4. The values of the corresponding coherence and stationarity
parameters are shown in Table 1. And using (8)-(10) the
available iids for the channel at different time variations are
tabulated in Table 2.
Table 1- Channel condensed parameters for the UWA
∆ t (sec) Bc(Hz) Fc (ms) Bs (kHz) Fs (s)

∆t s 159 152.97 ∞ ∞
0.08 159 152.97 18.165 1.887
0.1 159 152.97 12.392 1.037 1 .8

0.2 159 152.97 2.3418 0.145 1 .6 W SSU S


t = 0 .8 sec
0. 5 159 152.97 0.2025 0.186 1 .4 t = 0 .1 sec
t = 0 .2 sec

Channel capacity (bit/sec/Hz)


0.8 159 152.97 0.0556 0.049 1 .2 t = 0 .5 sec
t = 1 .0 sec
1
1.0 159 152.97 0.0377 0.080
0 .8
3.0 159 152.97 0.0267 0.283
0 .6

0 .4
Table 2: Number of identically independent fading channels
0 .2
∆t (se nTFD nTD nFD 10 11 12 13 14
A vera ge S N R (dB )
15 16 17

c)
Figure 5- Ergodic channel capacity versus signal-to-noise
∆t s ∞ ∞ ∞ ratio (SNR) for the UWA channel at 5 m/s for different time
scales.
0.08 1409 12 114

0.1 529 6 78
From Figure 5, it can be observed that the graphs are
0.2 14 1 15 slightly convergent (on the WSSUS graph) up to
∆t = 0.1 sec. Hence ergodic assumption can be applied
0.5 2 1 2
over the associated distances. However, the graphs of
0. 8 1 1 1 ∆t > 0.1 sec are not convergent, hence the assumption
of and the use of ergodic capacity is invalid over the
1.0 1 1 1 corresponding distances. This implies that even at close
time displacement, the channel stationarity intervals are
3.0 1 1 1
small due to the high delay and Doppler variations.
As for diversity gain associated with this particular channel,
Discussion: Table 2 indicates that enough diversity gain especially in
with the time-frequency diversity scheme can only be
In this simulation, the coherent bandwidth of the channel achieved within the stationarity time over which WSSUS is
defined within the WSSUS range is approximately assumed.
160 Hz . Thus, for the operating bandwidth of 20 kHz
the system is highly frequency selective. For this value, the
channel has approximately equal gain and linear phase. 5. Conclusion
This value also limits the potential data rate of a system
deployed in this environment without coding and diversity In this work the nonstationary property of the underwater
to about 160 Hz . The coherent time is 152 .97 ms , acoustic communication channel was presented using time-
thus the channel response is essentially invariant over this scale domain characterization. The nonstationarity is
time. The frame length or slot time at 10 kHz is not defined using the concept of local-sense stationarity and
explicitly given. But by using the Nyquist’s theorem, it can modeled using a geometrical-based model adequate for
be inferred that for BPSK modulation, up to 5 kbit/s can be UWA channel. The resultant simulation indicates that as the
obtained. Thus to transmit a single bit takes 0.1 ms. Hence spatial displacement of the mobile unit increases, the
for a frame with 100 symbols the channel is essentially diversity gain decreases and the assumption of ergodic
slow fading. For a frame of 1000 symbols, the channel is capacity becomes invalid. Thus for diversity to be
between the boundary of slow and fast fading, but for achieved in most cases, either the coherent intervals are
frames with over 1000 symbols, the channel is essentially reduced at the expense of bandwidth and channel capacity
fast fading. or the mobile speed is reduced. In the latter, time-frequency
diversity will still the most viable option. In our future
Table 2 also shows that the ergodic durations as well as the work the exploitation of channel selectivity properties
number of iids decreases with increase in ∆t . The instead of the coherent properties in providing capacity and
resultant ergodic capacity (7) for flat-fading (assuming diversity estimates will be undertaken.
symbol duration Ts = 1 / Bc ) is shown in Figure 5.
Acknowledgments [13] A. J. Viterbi, CDMA: Principles of Spread
Spectrum Communications. Reading, MA: Addison-
The authors thank the Ministry of Higher Education Wesley, 1995.
(MOHE), Malaysia for providing financial support under
Grant (78368). The Grant is managed by Research
Management Center (RMC) Universiti Teknologi Malaysia
(UTM)

References

[1] Stojanovic, M. 2003. Underwater acoustic


communications, in: Entry in Encyclopedia of
Telecommunications, Proakis J. G., Ed., New York:
John Wiley & Sons.
[2] Stojanovic, M. 1996. Recent Advances in High-
Speed Underwater Acoustic Communications. IEEE
Journal of Ocean Engineering, 21(2): 125-136.
[3] Kilfoyle, D. B., and Baggeroer, A. B. 2000. The
State of the Art in Underwater Acoustic Telemetry.
IEEE Journal of Ocean Engineering, 25(1): 4-27.
[4] Brady, D., and Preisig, J. C. 1998. Underwater
Acoustic Communications. In Wireless
Communication: Signal Processing Perspectives,
Poor H. V., and Wornell G. W., Eds., Mass: Prentice-
Hall, 1998, 330-379.
[5] Hwang, S.J., and Schniter, P. 2008. Efficient
Multicarrier Communication for Highly Spread
Underwater Acoustic Channels. IEEE Journal.
Selected areas in Communication, 26(9): 1674-1683.
[6] Jiang, Y., and Papandreou-Suppappola A. 2006.
Discrete Time-Scale Characterization and Wideband
Time-Varying Systems. IEEE Transaction on Signal
Processing, 54(4): 1364-1375.
[7] Grythe, K., and Hakegard, J. E. 2009. Non-perfect
Channel Estimation in OFDM-MIMO-based
Underwater Communication. In Proc. Oceans 2009-
Europe,1-9, Trondheim, Norway.
[8] Smedsrud M., and Jenserud T. 2008.
Characterization of Long-Range Time-Varying
Underwater Acoustic Communication Channels. In
Proc. Acoustic’08, 6097-6102, Paris.
[9] Goldsmith A. 2005. Wireless Communications. New
York, USA: Cambridge University Press.
[10] Biglieri E. 2005. Coding for Wireless Channels.
USA: Springer.
[11] Etter P. C. 2003. Underwater Acoustic Modelling
and Simulation, London: Spoon Press, Taylor &
Francis.
[12] J. Ventura-Traveset, G. Caire, E. Biglieri, and G.
Taricco, “Impact of Diversity Reception on Fading
Channels with Coded Modulation—Part I:Coherent
detection,” IEEE Trans. Commun., vol. 45,
May1997, pp. 563–572.

You might also like