- ...Considerations
 - This work was
                                                         supported by the
                                                         Natural Sciences and
                                                         Engineering Research
                                                         Council of Canada
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...Mann
 - Massachusetts Institute of Technology
                          Room E15-389, 20 Ames Street, Cambridge, MA 02139
                          e-mail steve@media.mit.edu
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...Haykin
 - Communications Research Laboratory,
                           McMaster University
                           Hamilton, Ontario, Canada L8S 4K1.
                          e-mail haykin@synapse.crl.mcmaster.ca
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...transform
 - When using a
                        multidimensional parameter space, it is often impossible
                        to establish frame boundsdaubechies:CBMS
                        on the energy in the parameter space.
                        With only one parameter, we cannot always reconstruct
                        the signal.  With two effective parameters,
                        we can reconstruct the signal, and also bound the
                        energy of the representation.
                        With three or more parameters,
                        the energy in the transform space will be infinite.
                        To the extent that multi-dimensional parameter
                        spaces are still useful, we will not let
                        this infinite energy hinder our progress.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...``time-frequency-scale-chirprate
 - Traditionally, the
                                                           term chirp-rate
                                                           (with a hyphen)
                                                           is used, but in this
                                                           paper, we use
                                                           the single word
                                                           ``chirprate'', to
                                                           avoid
                                                           confusion arising
                                                           out of hyphens in
                                                           compounded parameter
                                                           lists.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...wavelet
 - The term ``wavelet'' will appear in quotes
                 when it is used in this less restrictive sense.
                 In particular, a ``wavelet'' will be permitted to
                 have a non-zero DC component.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...otherwise
 - In general, g(t) is a complex-valued function
                      of a real variable,
                      and has finite energy: 6#6.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...affine
 - Segal [44]
                and others sometimes refer to these coordinate
                transformations as symplectomorphisms.  It is
                well-knownfolland,guilleminsymplectic
                that the actual geometry of phase space is symplectic
                geometry, and that it is a coincidence that
                28#28 corresponds to area-preserving affine
                geometry.  Therefore, we must
                keep in mind, that if we desire to extend
                our thinking to the
                analysis of signals of dimension n>1,
                then we must consider the
                symplectic geometry of 29#29.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...space
 - Note that, if we were interested in exploiting the
phase of this representation, we would need to add a fourth parameter,
to account for the extent to which the operators
do not commute.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...rectangular
 - The term ``rectangular''
is used here in the context of ``rectangular window'', meaning a 1-D function
that is unity in a certain frequency interval and zero outside that interval,
not to be confused with our later use of ``rectangular'' which will be more
consistent with its everyday usage to specify a 2-D shape.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...elsewhere
 - In actual fact, there is a small
amount of frequency smearing, but zero time smearing, as the energy is
entirely contained in the time interval under consideration.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...function
 - We do not attempt to
                           address issues of discretization
                           in this paper, except to the extent to which
                           they have influenced the development of the
                           continuous chirplet transform.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...unit
 - For this experiment, we positioned the radar horn
              facing upward and held a volleyball two meters above
              the horn, and released the ball after the recording began.
              We recorded only the in-phase component, and ignored
              the quadrature component of the radar.
              The sampling rate was 8kHz.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...sinusoidal
 - Here we are simplifying the description.
The dynamics of the sea are more fully described inpierson1.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 - ...time
 - Although there are devices, known as pitch transposers,
that attempt to perform such an operation in a highly nonlinear way.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.