Data processing: measuring – calibrating – or testing – Measurement system – Performance or efficiency evaluation
Reexamination Certificate
1999-11-08
2002-08-27
Hoff, Marc S. (Department: 2857)
Data processing: measuring, calibrating, or testing
Measurement system
Performance or efficiency evaluation
C702S074000, C702S032000, C324S076220, C324S076240, C324S076380
Reexamination Certificate
active
06442506
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Technical Field
The present invention relates in general to signal processing and in particular to a method and apparatus for spectral analysis of a signal. Still more particularly, the invention relates to a method and apparatus for real-time spectral analysis that determine two time-dependent intervals over which a signal should be analyzed to decompose it into its instantaneous mean and random components and to determine its instantaneous frequency spectrum (IFS).
2. Description of the Related Art
Spectral analysis is the decomposition of a signal into its sinusoidal components. Just as sunlight can be decomposed into a rainbow by a prism, a sampled signal can be decomposed into frequencies by an applicable algorithm. And just as a rainbow can be blended back into sunlight by passing its constituent colors through an inverted prism, a signal can be reconstructed by combining its component sinusoids.
Different signals, like chemical elements when heated to incandescence, manifest their own characteristic spectrum. A signal's spectrum thus provides an identity or signature in the sense that no other signal has the same spectrum. From this spectrum emerge all parameters necessary to quantify the instantaneous statistical character of the signal. Phase information about the signal, however, is lost.
Hereafter, spectral analysis, unless stated otherwise, refers specifically to time-varying spectral analysis. Time-varying spectra are common in everyday life because dynamic phenomena are common in everyday life. For example, a standard musical score represents a time-varying spectrum because it indicates to a musician what notes (i.e., frequencies) should be played as time progresses. A viable time-varying spectral analysis algorithm must determine which frequencies exist in a random signal at a particular instant in time.
Conventional spectral analysis entails the decomposition of a time-varying signal into its instantaneous mean and deviations from the mean. These two components are also referred to as the systematic (or DC) component and random component, respectively. For sampled signals, the systematic and random components are independent of one another, meaning that two different signals can have the same random component but different means, and vice versa. In other words, one particular mean is not necessarily linked to one particular random part.
Presently available spectral analysis algorithms are all non-real-time and use a single user-defined averaging interval (or window) for analyzing a random signal. Specifically, conventional spectral analysis algorithms require that the user “guess” the requisite magnitude of the time-dependent averaging interval over which a signal is to be continually analyzed. The extent of the selected interval is a reflection of the user's perception of how the signal properties are changing, or will change, with respect to time. (See, for example, Jones, D. L., & Parks, T. W.:1990, ‘A High Resolution Data-Adaptive Time-Frequency Representation’, IEEE Trans. Acous., Speech, and Sig. Proc. 38, 2127-2135; Jones, G., & Boashash, B.: 1997, ‘Generalized Instantaneous Parameters and Window Matching in the Time-Frequency Plane’, IEEE Trans. Sig. Proc. 45, 1265-1275; and Katkovnik, V. & Stankovic, L.: 1998, ‘Instantaneous Frequency Estimation Using the Wigner Distribution with Varying and Data-Driven Window Length’, IEEE Trans. Sig. Proc. 46, 2315-2325.) Such guessing introduces biases which degrade IFS computation, because even educated guessing cannot reliably emulate changing features of random signals, in real-time or otherwise.
Often, a user employs an averaging interval of large constant extent for all portions of signal, in the belief that incorporation of a substantially large amount of information will produce a statistically valid “parameterized” IFS. The parameter in question is the magnitude of the overly large constant averaging interval, which the user believes can be “scaled” out later. Such approaches contaminate the produced result because changes in signal properties are due to physical forces whose time-dependent features cannot be completely characterized by changes in time scales only (e.g., recall that forces have dimensions of (mass)×(acceleration)).
The use of a large constant averaging interval also contaminates the produced result whenever “too much” information is included within the interval because signals have a natural property that defines how far back in time present characteristics are connected to earlier ones. Thus, the inclusion of too much information within the averaging interval risks the possibility that past signal behavior, which is completely unrelated to immediate behavior, is incorporated into the IFS computation. This has the same effect as two (or possibly three) pianists, for example, playing the same piece of music, but not in synchrony with one another the listener has difficulty focusing on the true melody.
Use of too much information also leads to an anomaly referred to as 1/f noise (e.g., see Keshner, M.: 1982, ‘1/f Noise’, Proc. IEEE 70, 212-218). Use of overly long intervals also increases the likelihood that significant short-lived events (intermittency) are masked by uncharacteristically long averaging (see, e.g., Gupta, A. K.: 1996, ‘Short-Time Averaging of Turbulent Shear Flow Signals’, in G. Treviño et al. (eds.), Current Topics in Nonstationary Analysis, World-Scientific, Singapore, pp. 159-173.)
Some currently available spectral analysis algorithms also use information from the future, and in so doing assume that future behavior impacts present behavior. A common example of this approach is to define frequency information from time information according to
F
(
p
,
ω
,
M
)
=
(
1
(
M
)
(
SR
)
)
∑
n
=
p
p
+
M
-
1
S
(
n
)
exp
[
-
ⅈ
(
ω
SR
)
(
n
-
p
)
]
[
1
]
where S(n) represents a signal in discrete form, p is an integer denoting the present, &ohgr; is frequency in radians per second, and M is a p-dependent integer assigned by the user consistent with some user-specified criterion.
For &ohgr;=O, the symbol
∑
m
=
p
p
+
M
-
1
denotes
(
1
(
M
)
(
SR
)
)
{
S
(
p
)
+
S
(
p
+
1
)
+
S
(
p
+
2
)
+
…
+
S
(
p
+
M
-
1
)
}
[
2
]
where m takes on the integer values p,p+1,p+2, . . . ,p+M−1; for non-zero values of &ohgr;, the symbol denotes correspondingly similar expressions. Another approach is given by
F
(
p
,
ω
,
M
)
=
(
1
(
2
M
+
1
)
(
SR
)
)
∑
n
=
p
-
M
p
+
M
S
(
n
)
exp
[
-
ⅈ
(
ω
SR
)
(
n
-
p
)
]
[
3
]
Note that both of these approaches require information beyond the present, and as such are non-causal. The formulations are also non-real-time since they require knowledge of the signal at p+1,p+2,p+3, . . . ,p+M−1, which denote instants in time later than p, to define a signal property at p. In a world governed by entropy and its companion irreversibility, assignations such as [1] and [3] cannot be relied on to produce even near real-time information.
Also note these approaches use the same user-defined averaging interval (M) for all components of the signal. The specific components intended here are the p-dependent DC component (mean), defined by multiplying equation [2] by
{square root over ((
SR
+L )/
M
+L )}
and the random component, defined by subtracting the p-dependent DC component from the signal. Since a signal is a time history of some random-varying physical phenomenon, there is no assurance the above user-selected M is compatible with (i.e., “matches”) the window peculiar to instantaneous properties
Bracewell & Patterson L.L.P.
Hoff Marc S.
Kim Paul
Russell Brian F.
LandOfFree
Spectrum analysis method and apparatus does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Spectrum analysis method and apparatus, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Spectrum analysis method and apparatus will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2881471