International Journal of Engineering Trends and Technology (IJETT) - Volume4Issue4- April 2013
Clock Harmonization during Time-Variant
Underwater Acoustic Channels
Avvaru Ravi Kiran 1, K. Phani Srinivas 2, Ch. Sree Vardhan3
1
Electronics and Communications, KL University, Guntur, Andhra Pradesh, India
2
Assistant Professor, Electronics and Communications, KL University, Guntur, Andhra Pradesh, India
3
Assistant Professor, Electronics and Communications, KL University, Guntur, Andhra Pradesh, India
Abstract— This paper focuses on the underwater acoustic
channel, accounting for the multipath arrivals, and
generating a deposit of signal processing algorithms for
estimating the necessary delay times to enable clock
harmonization protocols for adjacent nodes in the
underwater acoustic system. The projected technique
engages correlating the reactions of the bidirectional
channels to exploit the underlying reciprocity. Performed in
two stages, a progression of probe signals is first transmitted
to generate an assembly, which contains information
regarding the time-variability of the acoustic communication
channel through multipath. From this assembly, we decide
its leading time-invariant attribute and use it as a
suggestion datum for the time-delay measurements. The
second stage engages performing time-delay evaluation for
probe signals swapped between the nodes.
Index
Terms—
Acoustic
harmonization, Underwater
modem, Multipath.
communications trouble, the transmission medium
introduces environmental issues such as high latency,
multipath, refraction, transmission loss, noise and nonreciprocity, scattering. These factors have considerable
effects on the channel impulse response. The
individual multipath ar rival amplitudes and arrival times
as well vary significantly depending on the surface
circumstances and the directivity of the transducers. For a
number of environmental conditions, the first or further
multipath arrivals may be seriously attenuated or absent
altogether.
communications,
Clock
communications, Acoustic
I. Introduction
Clock harmonization between network nodes
becomes compulsory once data from multiple nodes are to
be co- processed. In the framework of surveillance
networks, harmonization of the nodes clocks is directly
associated with the correctness of the localization of a
potential objective. Clock harmonization is attained by
calculating and then correcting the skew and offset of the
inner clock of a single node. The skew of a clock is the
oscillator
frequency
differentiation
among two
synchronizing clocks, whereas the offset is the time
variation at a definite point in time between two clocks.
Clock harmonization is particularly significant for
high-latency communications such as those in underwater
acoustic networks in which data from individual nodes
takes variable times to arrive at the main processing node.
The challenge in underwater acoustic clock harmonization
lies in properly estimating the time delay among the
transmission of a signal from a node and the reception of
that signal by a different node. For the acoustic
ISSN: 2231-5381
Fig1.
The intend of this paper is to observe how timevariability of the impulse response impacts clockharmonization and how it can be kept away from or
mitigated in a spatially stationary underwater network. The
Time Synchronization for High Latency (TSHL) protocol
developed by Syed and Heidemann is used as the basis. The
protocol is tailored to account for time-variability of the
channel impulse response.
http://www.ijettjournal.org
Page 1158
International Journal of Engineering Trends and Technology (IJETT) - Volume4Issue4- April 2013
In the nonexistence of a fit-for-all underwater
statistical model, channel soundings are an excellent
alternative to the lack of prediction ability. They are straight
measurements of the time-varying impulse response of
propagation medium and substitute general assumptions
concerning the existing channel between two nodes. A
series of pulses is sent through the channel at a set
interval. The receiver examines the obtained series and
compiles its personal local channel characteristics from it.
bA the offset of clock CA
compared to its reference or master clock CB . All through
this paper B indicates the reference node and A the
synchronizing node. In an ideal situation with no errors or
drifts, we would have C(t) = t . The skew of a clock fA is the
oscillator frequency difference between CA and CB . The
offset b A is the time difference at a exact point in time
between the two clocks. Figure 2 is a graphical
representation of a perfect reference clock combined with
both a slower and faster slave with respective offset clocks.
Our approach calms down the assumption in
TSHL of a stable propagation delay among the
synchronizing nodes. It also relaxes the reciprocity
assumption for the channel. Alternatively, we repeat the
measure the response of the channel over a period of time
to statistically describe the communication channel to
estimate the propagation delay and it’s skew in an
enhanced way. This move could also be defined as adding
channel sounding to the TSHL protocol. This model
keeps the two- stage approach developed by TSHL,
correcting for the skew first followed by the offset.
At the first stage of skew correction, examination
of a sequence of chirp responses to recognize the
smallest amount time-varying multipath present in the
channel among the two nodes is performed. Based on
the known chirp transmission interval, linear regression
is performed on the calculated interval of arrivals of that
recognized multipath detected within each chirp response.
The slope of the ensuing best-of-fit straight line signifies
the skew of the clock’s receiver node compared to the
initiating or reference node.
Fig2.
2. Changeability of the underwater communication channel
The second stage consists of an exchange o f timestamp
messages between the two nodes, where we presume a
time varying one-way propagation delay between the two
nodes. The defy in this stage consists in shaping the
variation in propagation delays. This is solved by
performing, at one of the nodes, a time-delay estimate
via the chirp responses of the channel sounding of both
one-way propagation paths.
II. BACK GROUND
A. Clock Model
Each node in a sensor network has its individual
clock. The time measured by clock A generally presents
random errors, but on average it can be represented by a
linear function of the form
CA(t ) = fAt + b A
[1]
where fA represents the skew and
ISSN: 2231-5381
Fig3.
The underwater acoustic channel is a difficult
communication medium. It can be considered of as a
lengthy slim pipe, with the deprived physical link quality of
a mobile terrestrial radio channel, and the high latency of a
satellite channel. Technologies developed fo r wireless
radio communications are used as the foundation for
underwater
acoustic
communications.
Wireless
communication difficulties l i k e multipath, scattering,
noise, attenuation and fading are replicated in the ocean
because of the high latency of the medium. Acoustic signals
http://www.ijettjournal.org
Page 1159
International Journal of Engineering Trends and Technology (IJETT) - Volume4Issue4- April 2013
proliferate five-orders-of-magnitude slower in water
compared to over-the-air RF signals, with speeds of
1.5x103 m/s and 3x108m/s, respectively. Also, greater
absorption at higher acoustic frequencies restricts the
existing bandwidth, leaving only lower frequencies, which
are susceptible to channel fading due to multipath.
Underwater acoustic research regularly takes the
above-mentioned intricacies of the medium into account.
However, most realistic explanations make the hypothesis
that the environment is steady over relatively short
periods of time. On the other hand, signal fluctuations
can occur due to transceiver movement or inherent changes
within the propagation
medium. In the underwater
environment there is
always some movement present in the system. This motion
can affect the impulse response’s frequency spectrum
according to the Doppler effect, potency of the arrival
paths, or in the horrible case, the complete disappearance
of it due to one of the transceivers entering a shadow
region where no communication pathway exists between
two communicating nodes.
The Doppler effect originates frequency shifting
as well as added frequency spreading. The amount of
spectral shift is proportional to the ratio of the relative
transmitter-receiver velocity to the speed of sound. As
the speed of sound in the water is comparatively slow
(1500 m/s), a transducer’s shift from waves, currents,
tides, and platform shift will be enough to impair the
communication channel. Further, unlike wireless
communications systems in the air, it has been publicized
that assumption that the underwater acoustic channels are
subject to wide-sense stationary uncorrelated scattering
(WSSUS) does not always apply.
In addition to transducer movement, the impulse
response varies due to motion within the environment. There
are different phenomena accountable for this motion, which
drop in either long or short-term categories. Transform that
happen over an extended time scale regularly develop at a
slow pace and do not influence the signal during its
propagation through the medium. The regular happening says
that the system can accustom to the changes in the
environment. Since some of them can be cyclical in nature,
the system can be designed to expect and adapt to these
fluctuations. Long-term phenomena include but are not
limited to swell, current, internal waves, tides and daily or
seasonal water temperature variations.
Short-term changeability is articulated in loads of
diverse ways, which is the reason there is no consensus on
a numerical description of the underwater acoustic
channel. Wind energy is the central reason for the shortterm instability. It’s extremely dynamic nature leaves little
time for the system to adapt and can often result in the
ISSN: 2231-5381
outage of the communication channel.
In the nonexistence of a fit-for-all underwater
statistical model, channel soundings are an excellent
alternative to the lack of prediction capability. They are
measurements of the time-varying impulse response of
propagation medium and substitute common assumptions
regarding the channel. A chain of pulses is sent through the
channel at a set interval. The receiver analyzes the received
series and estimates the local channel characteristics from
it. This practice is an added support cost to the system, but
the result may deeply improve its overall efficiency.
III. PROPOSED CLOCK SYNCHRONIZATION
ALGORITHMS
The stated algorithms below stand for time
variability in the underwater communication channel when
synchronizing the clock of slave node A to reference
master node B. As mentioned in the introduction, this
approach is based on the TSHL protocol. Our approach
calms down the
assumption in TSHL of a stable propagation delay between
the synchronizing nodes. Instead, we frequently compute
the response of the channel over a period of time and
statistically illustrate the communication channel to better
decide the propagation delay and its skew.
1) Stage 1: Channel sound and skew correction
Our approach to the skew correction views deeper
into the physical attributes of the received chirps. When
considering to a multipath environment, an acoustic
modem will “pick” an arrival or path in agreement to a predefined metric. Considering the Teledyne Benthos acoustic
modems, the most vigorous path or principal peak is
preferred as the synchronizing reference datum. This
reference arrival can temporarily change within the
impulse response of a time-varying communication
channel. Such an environment will influence the arrivals
within the impulse response differently, resulting in few
arrivals being steadier over a period of time than others.
The following model effectively changes the pre-define
synchronization metric used by the acoustic modem. The
new metric is based on the formation of the impulse
response predictable by sending a series of chirps through
the channel.
The initial point of the process is the generation by
the reference node B of a digital impulse train, finally based
on its clock frequency. The length of this impulse train
should be adequate enough to provide enough chirps to the
synchronization node A for the linear regression algorithm.
The selection of the Repetition Pulse Interval (RPI) should
base on the sampling frequency of the system and the
intended measurement precision of the skew. The length of
the complete impulse train must be such that there are
http://www.ijettjournal.org
Page 1160
International Journal of Engineering Trends and Technology (IJETT) - Volume4Issue4- April 2013
adequate samples to cover the scale of the clock skew.
The characteristics of the pulse, influence how
precisely each multipath arrival can be differed from one
another within the impulse response. Because the function
of the algorithm is essentially the same as channel sounding,
reference identifies linear frequency-modulated (LFM) chirp
trains and pseudorandom (PRN) binary sequences as the
most suitable types of probe signals.
to the carrier frequency and propagated as the ith chirp (xi(t))
of n total chirps by the reference node’s transducer. The
environment influences the transmitted signal xi(t). The
received signal is then indicated as yi(t). The signal yi(t)
received by the synchronizing node’s transducer is filtered,
united back down to baseband, and digitized based on the
sampling frequency regulated by the node’s local clock.
Because the series of chirps is shaped, it is combined
The time delay between signals traced at two
spatially alienated sensors can be recovered using a
correlator, even in the presence of noise. The recognized
reference pulse x i(t) is correlated with the signal received by
the synchronizing node yi(t) to produce Pxyi(τ). The
separation between the peaks in Pxyi(τ) signify the time
delays between the multiple copies of the transmitted pulse.
These initiate from the multipath attribute of the channel.
P xyi(τ) can also be used to approximate the channel impulse
response at time t.
The peak detector then applies a threshold to sense
the location of the peaks present in the signal P xyi(τ). These
peaks correspond to the signal propagation time difference
between the different paths there in the channel at time t.
Fig4.
Each arrival time is noted as ai, j which points to
the jth received multipath arrival time from the ith out of n
propagated pulses. The arrival times are then grouped in a
matrix A as shown in Figure 4. It is significant to note
down the number of detected arrival may vary among the
received chirp responses. Therefore, m denotes the
maximum number of arrivals detected in a single received
pulse. The ending size of matrix A is then n x m.
similar multipath arrival for each one of the received
responses.
Therefore the arrivals of each multipath must be
properly grouped prior to the skew can be found. To act so,
a sorting algorithm line up each multipath in a single
column of a matrix, keeping the received detected arrival
for a particular pulse on the same row.
To establish the key multipath arrivals, a
probability density function (PDF) is predicted using all of
the detected arrival in matrix A of Figure 4. The time
indexes of the peaks
of this PDF are pulled out and used to signify the main
multipath, identified individually by the index q in Figure 3,
of the channel when q = 1,2,...,Q. The maximum number of
multipaths recognized by the algorithm can be pre-set. In
Figure 5, the s major peaks of the PDF will be known as the
main multipath, when s = Q.
Each key multipath is represented by a column in a
new matrix B, an example of which is shown in Figure 5.
The time indexes are compared with the noticed arrival time
of each pulse. If a detected arrival time corresponds to one of
the resoluted chief multipath arrival times, it is transported to
the new matrix on the particular row of the pulse from which
it was detected and the column to which multipath it is
related. If no arrivals are detected, a zero is inserted in the
matrix as a placeholder. The size of matrix B can be varied
from the size of A, depending on how much the impulse
response of the channel varies between the transmission
times of the n pulses.
The peak detector will not essentially detect the
identical number of arrivals for each received pulse. The
time variability of the channel impulse response can
influence the energy level received from a particular
multipath and avoid it from being detected by the peak
detector. Thus, the detected arrival times present in a single
column of Figure 4 need not essentially symbolize the
ISSN: 2231-5381
http://www.ijettjournal.org
Page 1161
International Journal of Engineering Trends and Technology (IJETT) - Volume4Issue4- April 2013
demonstrated in Figure 6 by independently labeling the
one-way propagation delays.
Fig5.
Now lined up, every common ensemble of arrivals
(separate columns of matrix B in Figure 5) is processed by a
linear-regression algorithm. The predictable time delay
calculated for each chirp for a regular multipath arrival and
the well-known RPI can be used to carry out a linearregression analysis. A straight line is fit to the data in each of
the columns of B. This study results in two values Fj is the
slope of the best fit line and Ej denotes the goodness of fit of
this line, better defined as the linear correlation coefficient.
The slope Fj denotes the skew between the synchronizing
and reference clocks. The linear correlation coefficient Ej is
used in the subsequent step of the algorithm to choose the
most stable multipath arrival. The last step of the process
spots which set of arrivals is the most stable and consistent
within the sequence of chirp responses analyzed. This
verdict is based on the measured value of Ej and the number
of arrivals detected in a particular set. The linear correlation
coefficient is calculated based on the standard deviations of
the measured and predicted time delays and their related
mutual covariance. The value of the mutual coherence can
vary between 0 and 1, where 1 stands for the best possible fit
(i.e., all the points are on the line) and 0 the worst possible.
Time variability in a definite multipath cause the
arrivals to vary in the time domain, resulting in a superior
deviation of the arrivals from the best-of-fit line. We
arrange all of the correlation coefficients Ej , j = 1,2,...,m and
discover the largest one. This most stable multipath arrival
is selected to be the one with the highest correlation
coefficient (i.e., closest to 1), and this arrival has the
slightest deviation from its best-of-fit line.
Additional metric essential to corroborate this
most stable multipath arrival is the number of arrivals
included in the ensemble. A low number of arrivals in a
multipath arrival ensemble is a superior indication by itself
that it is not a reliable multipath arrival (if a column of B
contains several zeros). As well, it can negatively persuade
the result of the linear regression. For example, performing
linear regression with merely two arrivals produces a
perfect correlation coefficient of the most stable multipath
is known by the number of multipath detections and the
related time-variability of that multipath.
II. Stage 2: Offset correction
To correct the offset b A (as defined in equation (1))
of the skew-corrected clock compared to a reference is
unlike than THSL’s as it doesn’t presume that the
propagation between the two. Such a condition is
ISSN: 2231-5381
Fig6.
The definition of P is
pA,B+pB,A=P.
As well, the offset can be defined as
[2]
T2−T1−pA,B=b A
[3]
T4−T3−pB,A=−bA.
[4]
The foundation of our approach is the two-way message
exchange used by the majority clock-synchronization
protocols. Its distinction lies in its capability to
distinguish the propagation delays for both one-way
propagation channels instead of using an normal based on
the round-trip propagation time. A chirp is sent to the
front of each message allowing the chirp response of the
one-way communication channel to be determined by the
receiver. Both expected impulse responses are correlated
at the synchronizing node A to determine the difference
in propagation delay Δp between the two. This delay Δp
is the difference between the time corresponding to the
largest peak and zero lag. This method can be used
because the modem all the time synchronizes on the
largest amplitude arrival. The offset of the clock can be
calculated by using this propagation delay difference and
the four time-stamps produced by the two-way message
exchange.
The synchronizing node A begins the offset
correction process by sending a new pulse to the front of the
offset calculation request message to reference node B. The
time of the start of the transmission is time-stamped T1 and
recorded by node A.
http://www.ijettjournal.org
As it propagates through the channel, the pulse is
Page 1162
International Journal of Engineering Trends and Technology (IJETT) - Volume4Issue4- April 2013
transformed. The propagation delay to be resoluted is
denoted as p B,A in Figure 4. Time-stamp T2 is recorded by
the receiver upon initial detecting the arrival of the pulse sent
by node A. From the pulse response, the instantaneous oneway impulse response from node A to B is resoluted using a
matched filter and identifying the peaks of the resulting time
series. This impulse response is mandatory later in the
algorithm at the synchronizing node. To lessen the
bandwidth required to send it across the channel for
processing, the predictable impulse response is condensed by
retaining its main characteristics, which are the amplitudes
of the largest peaks and their related time indices. These
characteristics are symbolized as HA,B .
well as the compacted impulse response information HA,B
recorded from the probe signal preceding the first message.
The second message promulgates through the channel
from node B to A. The message is transformed by the channel.
The yet to resoluted propagation delay is denoted as pA,B in
Fig.4. Time-stamp T4 is produced by the receiver upon detecting
the arrival of the chirp sent by node B. From this received chirp,
the instantaneous one-way impulse response from node B to A
(HB,A) is predicted and its information compressed. From their
individual recorded characteristics, the main arrivals of both
estimated impulse responses are rebuild in the time-domain
from HA,B and HB,A . Their cross-correlation allows for the timedelay estimation between the two signals. Based on the
supposition that the modem will synchronize with the strongest
arrival, the difference in propagation delay Δp (eq. [5]) will be
the time index difference between the highest peak of the crosscorrelation and its zero lag:
The reference node then transmits the same new
reference pulse before transmitting the second message of the
exchange. Time-stamp T3 is recorded concurrently to the
commencement of the transmission of the second message. It
carries the two time-stamps generated by node B, T2 and T3, as
pA,B − pB,A = Δp =τ *
[5]
V. ACKNOWLEDGMENT
Adding equations (2) and (5) together, we get
P + Δp = 2 pA,B.
[6]
We thank Mr. K. Phani Srinivas, Assistant Professor,
The one-way propagation delay pA,B can be determined fromDepartment of ECE, KL University, for his valuable
contribution and guidance in helping me to prepare this paper
equation (6):
P + Δp = pA,B.
[7] and helping me out many a times when we need assistance and
Mr. Ch. Sree Vardhan Assistant Professor, Department of ECE,
2
KL University, for his encouragement and support. I indebted
The second propagation delay can be evaluated from
to the Department of Electronics and Communication
P − p A,B = pB,A.
[8] Engineering of KL University for providing the support
required.
Either one of the one-way propagation delay can be used in
equation [7] as Δp can be positive or negative. Whichever one-way
VI. REFERENCES
propagation is calculated in equation [7], the remaining one can be
[1] A.A Syed and J. Heidemann, “Time Synchronization for
determined using equation [8]. The last step of the algorithm
High Latency Acoustic Networks,” in Proc. IEEE Conference on Computer
consists of evaluating the offset bA between the two nodes’ clock
Communications (Infocom), Barcelona, Spain, Apr. 2006.
using equations [3] or [4]. The result of both equations will be the
[2] M. Stojanovic and J. Preisig, “Underwater Acoustic Communication
Channels: Propagation Models and Statistical Characterization”, IEEE
same.
Communications Magazine, Jan. 2009.
[3] P.V. Walree, Channel sounding for acoustic communications: techniques
and shallow-water examples, report from Norwegian Defence Research
IV. CONCLUSION
Establishment (FFI), Apr. 2011.
[4] X. Lurton, “Environment variability and signal fluctuations”, in An
In this paper we speak out the problem of clock Introduction to Underwater Acoustics,11 Principles and Applications,
harmonization in an underwater acoustic network. This is a Chichester, UK: Praxis, pp. 128- 129, 2002.
C.H. Knapp and C. Carter, “The Generalized Correlation Method for
very actively researched area for every form of communication [5]
Estimation of Time Delay,” IEEE Transaction on Acoustics, Speech, and
channel. This is mainly challenging in the underwater acoustic Signal Processing, vol ASSP-24, no. 4, Aug. 1976.
environment due to its high latency characteristics, and it [6] J.R. Taylor, “Least-Squares Fitting,” in An introduction to Error
requires the advancement of a exact clock synchronization Analysis, 2nd ed. Sausalito, CA: University Science Books, pp. 181-190, 1997.
The BELLHOP Manual and User’s Guide, Preliminary draft, Heat, Light,
protocol. Most of the approaches found in the literature imagine [7]
and Sound Research Inc., La Jolla, CA, 2011.
constant propagation delays between two nodes and ignore time
variability of the channel. An general idea of how and why time
variance exists in the underwater acoustic environment is
specified.
ISSN: 2231-5381
http://www.ijettjournal.org
Page 1163