home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky sci.physics:22462 alt.sci.physics.new-theories:2726
- Newsgroups: sci.physics,alt.sci.physics.new-theories
- Path: sparky!uunet!well!sarfatti
- From: sarfatti@well.sf.ca.us (Jack Sarfatti)
- Subject: Wavelets, coherent states 10, sampling-reconstruction
- Message-ID: <C0o8nK.KtF@well.sf.ca.us>
- Sender: news@well.sf.ca.us
- Organization: Whole Earth 'Lectronic Link
- Date: Mon, 11 Jan 1993 04:04:32 GMT
- Lines: 84
-
-
- 10.Reconstruction of signal by sampling.
- The h[v,s] are highly redundant (over-complete, not linearly independent as
- kets etc - see earlier notes). We want to find discrete subsets that are
- complete frames. Kaiser gives an algorithm on p.38.
-
- Let T > 0 be a fixed time interval. Sample output signal Fs(v) only at
- instants s = nT, n is integer. Discretize frequency as well.
-
- fnT(t) = hnT(t)*f(t) (74) (* is cc of h not a "convolution" command)
-
- has compact support in interval nT-& <=t<= nT, hence expand it in a Fourier
- series
-
- fnT(t) = S(m)[e^-i2pimt/& cmn] (75)
-
- cmn = (1/&)Integral(from nT-& to nT)[e^i2pimt/& h(t-nT)* f(t)
-
- = (1/&)F(m/&|nT) (76)
-
- It can be shown (p.39) that
-
- &hnT(t)fnT(t) = S(m)h[m/&,nT](t)<h[m/&,nT]|f> (77)
-
- sum over n, assume convergence limit g(t) where
-
- g(t) = @ S(n)[|h(t-nT)|^2] (78)
-
- 0< A <= g(t) <= B (79)
-
- then double discrete sum
-
- S^2(n,m)[|<h[m/&,nT]|f>|^2 = Integral dt g(t)|f(t)|^2 (80)
-
- hence the subset
-
- H(M)[T,&} = {h[m/&,nT]|m,n in Z} (81)
-
- forms a "discrete subframe" with frame constants A nd B. This subframe is
- not generally tight, but the "metric" operator G is simply multiplication
- by g(t) . Note how the metric is built from sampled windows in equation
- (76). The time signal is reconstructed as
-
-
- f(t) = g(t)^-1 S^2(n,m)[h[m/&,nT](t) F(m/&|nT)] (82)
-
- *The idea here is that we can recover the continuous signal by discrete
- sampling at time intervals T and frequency intervals 1/&. The Heisenberg
- uncertainty principle is here. Heisenberg's principle is the condition that
- the discrete subset h[m/&,nT] forms a frame. We cannot satisfy g(t) +> A >0
- unless the supports of successive windows like h(t) and h(t-T) overlap so
- that T < & which is the uncertainty principle going the wrong way! It is
- the same as the condition on "virtual particles" fluctuating out of and
- back into the vacuum in Yukawa force picture. It is really consistent with
- Heisenberg because, if we use quantum pictures, we are asking the
- "complementary" question. We are not measuring conjugate (i.e.
- incompatible) observables on real particles but estimating the minimal
- noise levels of virtual particles that allow us to measure the real
- particles.
-
- The closer T gets to &, the more difficult for the window function h to be
- smooth. Does it break into a kind of chaotic fractal at a T-& resonance?
- When T = & h(t) must be discontinuous if the frame condition is obeyed.
- This means that the Fourier transform of the window function H(v) is no
- longer concentrated near v = 0. This suggests some kind of a phase
- transition on computer analysis of natural time series of complex systems
- in which we vary T on the computer and see if there are any natural time
- scales & in the complex system!
-
- **Above two paragraphs are Sarfatti's half-baked conjectures not Kaisers!**
-
- This means that for nice windows h(t) we can get a good "perturbative"
- solution to the reconstruction of the signal by truncating the discrete
- double sampling sum in eq.(82) (like Feynman diagrams in QED? Sarfatti
- note). But the convergence of the perturbation series breaks down at a T-&
- resonance.
-
- If we are only interested in a particular time interval and a particular
- frequency interval because we want to eliminate high frequency static in a
- sound system, for example, then the appropriate truncation includes an area
- in the time-frequency plane only slightly larger than the area of interest.
- In effect we are getting an least-squares approximation f1 to the original
- signal. (p.41)
-
-