Professional Documents
Culture Documents
554-566, 1979.
A. Elfes, Sonar-based real-world mapping and navigation, IEEE J.
Robot. Automat., vol. RA-3, pp. 249-265, 1987.
P. H. Milne, Underwater Acoustic Positioning Systems. Houston, TX:
Gulf, 1983.
E. F. Rechner and D. E. Nelson, Dept. Archaeology, Simon Fraser Univ.,
private communication.
H. L. Van Trees, Detection, Estimation, and Modulation Theory, Part
Ill. New York: Wiley, 1971.
L. E. Freitag and P. L. Tyack, Passive acoustic localization of the
Atlantic bottlenose dolphin using whistles and echolocation clicks, J.
Acoust. SOC. Amer., vol. 93, pp. 2197-2205, 1993.
V. A. Del Grosso, New equation for the speed of sound in natural
waters (with comparisons to other equations), J. Acoust. Soc. Am., vol.
56, pp. 1084-1091, 1974.
Guide to the Expression of Uncertainty in Measurement, 1st ed.
Geneva: International Organization for Standardization, 1993.
The sonar sampling system has recently been described in: P. H.
Kraeutner and J. S. Bird, A PC based coherent sonar for experimental
underwater acoustics, IEEE Trans. Instrum. Meas., to be published.
I. INTRODUCTION
Many applications involve digital processing of periodic signals.
For example, both voltage and current in electricall power systems
are periodic signals containing harmonic components [ 11. There are
generally three steps associated with, the digital processing of a signal.
Firstly, the signal is uniformly sampled and converted into a discrete
bl looking at the
Of data i s
sequence. Then, a
sequence for a period of time and neglecting everything that happens
before and after this period. This period of time is referred to as the
data window or observation interval. Finally, digital signal-processing
techniques such as the DFT are applied to the samples within the data
window to get the results. There are some requirements associated
with the first two steps. First of all, the sampling frequency must
be higher than the Nyquist frequency, which is twice the highest
frequency of interelst. A particular problem arises when the sampling
frequency is high enough to satisfy the Nyquist theorem but the
sampling process is not synchronized with the signal to be processed.
There are two synchronization aspects which must be considered.
Firstly, the data window should cover an integer number ( L )of signal
cycles where each cycle has a period TO.This mearis that the length
of the data window is LTo. Secondly, the length of the data window
should be an integer multiple of the sampling period (denoted as Ts).
In other words, the following condition should be satisfied in order
to ensure synchronized sampling
828
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 45, NO. 4, AUGUST 1996
11. THEPROPOSED
ALGORITHM
Now consider the derivation of an algorithm to modify the samples
when
is not an integer. Suppose
= W , where [ ] denotes
the operation to get the closest integer value. -1-represents the number
of samples within the data window, and we assume that it is known a
priori. Our approach is to modify the practical samples . r ( n ) toward
an ideal sampling sequence .TO ( n )whose sampling frequency satisfies
= N, in which T,odenotes the ideal sampling
the condition
a?, = 2%- and 3 = N, the ratio
(or
is
period. Given [+]
T3 0
should be close to one for a large -Y.
In this case we assume
[y]
&
T,- T,o= e
where le1
<
2)
7 so
2-1-
'1.I- 1
M-1
(9)
The actual discrete signal samples after uniform sampling are obtained as
(2)
where I I denotes the modulus. Equation (2) means that the actual
samples deviate very slightly from the ideal samples. Without loss of
generality we assume that initially z(0) = .ro(O), which means that
the first sample is identical to the desired sample. For subsequent
samples, there is a time deviation between .r( n ) and .xo(n ) as a
result of differences between actual and ideal sampling periods. The
deviation increases with 11. For example, the time deviation for the
second sample is given as (Ts - T,o)
= e . For the third sample it
will be 2 e , and for the rrth sample it will be ( rz - 1)e. Consequently
we have the relation
s o ( n ) = s.(nT,o) = .r,(nT, - n e ) .
=0
m=O
17%
(10)
We define a sampling error sequence, which is the difference between
the practical sequence and the ideal sequence. Hence the sampling
error sequence before modification is given by
(3)
Expanding (3) into a series and neglecting the higher order components, we have
2 n n nl L e
Ico(rij E
z o ( n T s )- .rh(nT,) n e .
(4)
+ S ) T , } = .r,{(n + -\-j(T70+ e ) }
+ .VT,o + n e + -Y,)
.r,(nT,o + rce + Y e ) = s,{r~(T,o+ e ) + .\-e}
where C,,,
,, = -4nL[Bo,n,(7~)
- 11 and &,,(n)
= e'=.
From
(11) it is clear that C,,,,,
corresponds to the sampling error for mth
harmonic component. For simplicity, we use the relative value of the
sampling error for m th harmonic component, given by
r ( n + -V)= za{(rt
= z,(nT,o
-
= .c,(nT, + N e ) .
(5)
A.
Hence
Substituting (6) into (4) gives the following formula which can be
used to modify the input samples
Now we evaluate the sampling error after the proposed algorithm has
been applied. The new samples according to (7) are given by
.Tl(/l)
=x(u)
+ XN { r ( n )- . T ( n + N ) }
Clearly for modifying one sample this algorithm (7) only requires
one addition and one multiplication which makes it quit suitable for
on-line implementation.
Hence the sampling error sequence is given by
111. ANALYSIS
OF THE PROPOSED ALGORITHM
El(?,)
829
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 45, NO. 4, AUGUST 1996
where Dm,n =
2TnmLe
- 11 and B I m ( n ) = e J r [ l
Am[Bl,,(72)
2nml e
F(1- r ' T T ) ] .
TABLEI
COMPARISON
OF LEAKAGE
COEFFICIENTS
obtained as
Rl,7n(7L)
=
Dm,n
A, = [BI,,(n)
11.
(17)
49.8
83.33%
50.0
0.00
0.00
50.2
0.12
0.02
83.33%
50.5
0.32
0.13
59.38%
(18)
It can be shown from (18) that Bl,,,(n) decreases with
value of Bl,vt(n)is bounded with the range given by
B1,m(N)5
Bl,m(71)
71..
Hence the
5 Bl,m(O)
(19)
or
The second term in (28) reflects the noise power after compensation.
It is clear from (28:i that the proposed algorithm is very sensitive to the
noise, because the noise power increases by a factor of (1+$ $).
To prevent the noise from blowing up, the ratio (7b,'X) should be as
small as possible. ,4s the result, n should be limited within the range
10, N - 11.
It is also interesting to investigate the ideal case where e = 0. In
this case the samplle sequence is periodic, that is, ~ ( n=) .c(n X).
Clearly from (25) the proposed algorithm has IIO effect on the
actual sample. This result is very important, since in such cases the
actual sample sequence is the same as the ideal s:equence, and no
modification should be made.
Therefore we have
and
Given that
e < &,
<< -V
IV. SIMULATION
RESULTS
we have
s(n)
+ ,u(n)
(26)
+ ?[.u(n)
N
+ <v
u(n + N ) ] .
(27)
gL;-'
I X ( k ) ( - (X,,,X(
I-XInaxI
(29)
X,,,
830
I
-2
50
100
150
-2
(4
100
50
150
(b)
Fig. 1. Comparison between the processed and unprocessed data: (a) unprocessed data. (b) processed data, (c) amplitude spectrum of unprocessed data.
and (d) amplitude spectrum of processed data.
C. Digital Wattmeters
Microprocessor-based (or digital) wattmeters [4]-[SI offer improved accuracy and speed of response over electromechanical instruments for electrical power measurement. Since both voltage, C 7 ( t ) .
and current, I ( t ) ,are periodic, the average power is evaluated as
0.47
0.20
49.8
0.16
0.03
50 0
0.00
0.00
50.2
0.13
0.08
50.5
0 24
0.23
49.5
+ 7)
+ 10siri(l07i,ft+ 15)
(36)
(37)
and
I (t ) = 100 sin(2.irft
V. CONCLUSION
The authors would like to thank the anonymous reviewers for their
comments and suggestions which have served to enhance this paper.
where p ( n ) = u ( n ) i ( 7 ~and
) [ f ? / f o ] = S . If .fs/.fo = .1-,the
calculated average power, P , is equal to the actual average power in
(31). However, when . f s / . f o is not an integer, P will not equal Po;
the difference between P and POis referred to as truncation error,
defined as
(33)
Since the instantaneous power V ( t ) / ( t )is also a periodic function,
(25) can be applied to compensate p ( n )
REFERENCES
[I] J. Arrilaga, D. A. Bradley, and P. S. Bordger, PowerSystem Harmonics.