IEEE SIGNAL PROCESSING LETTERS, VOL. 10, NO. 1, JANUARY 2003 15
Finite-Channel Chromatic Derivative Filter Banks
M. Cushman, Member, IEEE, M. J. Narasimha, Fellow, IEEE, and P. P. Vaidyanathan, Fellow, IEEE
Abstract—Two recent contributions discussed the theory of
perfect-reconstruction (PR) chromatic derivative filter banks
comprising an infinite number of channels. This letter extends
the theory to the case of finite channels. A novel time domain
procedure is delineated for designing the synthesis filters that
achieve PR in this case.
Index Terms—Biorthogonal filter banks, chromatic derivatives,
perfect reconstruction.
I. INTRODUCTION
I
N THIS LETTER, we present the problem of multichannel
sampling discussed in [3] and [4] from the view of a contin-
uous-time biorthogonal filter bank, and we show that this view-
point leads to new insights. A particular filter bank of this kind,
called the chromatic derivative filter bank, is explored in greater
detail. See [6] for a comprehensive survey of single and multi-
channel sampling theory.
Chromatic derivatives are linear combinations of the ordi-
nary derivatives, where the coefficients of the combination are
derived from orthogonal polynomial theory. Surprisingly, by
taking combinations of a notoriously difficult operation such as
differentiation, we obtain perfectly stable filters. This is because
of the well-known orthogonal properties of the underlying poly-
nomial set. The weighting function employed in the definition
of these polynomials basically determines the characteristics of
the filters. We believe that these derivatives are preferable to
the ordinary derivatives in situations where wide-band channel
splitting is needed. Multichannel analog-to-digital conversion
of high-bandwidth signals is one such example.
The theory of perfect reconstruction (PR) in chromatic
derivative filter banks with an infinite number of channels is
discussed in [1] and [2]. We extend that theory to the finite-
channel case here. Although one can obtain an approximate
reconstruction of the original signal by truncating the infinite
series expansions, as explained in [1], we show that it is in
fact possible to achieve PR by proper choice of the synthesis
filters. We delineate a time domain procedure for designing
such filters.
Manuscript received May 28, 2002; revised August 1, 2002. The associate
editor coordinating the review of this manuscript and approving it for publica-
tion was Dr. Markus Pueschel.
M. Cushman was with Kromos Technology, Los Altos, CA 94022 USA. He
is now with Deephaven Capital Management, Santa Clara, CA 95054 USA
(e-mail: mcushman@alumni.uchicago.edu).
M. J. Narasimha is with Stanford University, Stanford, CA 94305 USA
(e-mail: sim@nova.stanford.edu).
P. P. Vaidyanathan is with the California Institute of Technology, Pasadena,
CA 91125 USA (e-mail: ppvnath@systems.caltech.edu).
Digital Object Identifier 10.1109/LSP.2002.806706
II. BIORTHOGONALITY IN FINITE-CHANNEL ANALOG
FILTER BANKS
Fig. 1 shows an -channel filter bank, where the input
signal is bandlimited to , and the sampling rate of
the subbands is times smaller than the Nyquist rate of
one sample per second. The output signal can be expressed
as . We as-
sume that the input belongs to the space spanned by
the doubly indexed set of (linearly independent) functions
, , i.e.,
for some set of con-
stants . For such an input, following the procedure in [1]
and [2], we can show that the PR property holds if the analysis
and synthesis filters satisfy the biorthogonality condition
(1)
where is the Kronecker delta function. If denotes the
cascaded impulse response of the th analysis filter and the th
synthesis filter, then (1) can equivalently be written as
.
III. CHROMATIC DERIVATIVES BASED ON
ORTHOGONAL POLYNOMIALS
Consider a sequence of polynomials in the interval [ 1,
1], which are orthogonal with respect to a nonnegative even
weighting function that is not identically zero over the defining
interval. It is well known [5] that such polynomials can be
obtained using the recursion
(2)
with and . The values of the constants
, , , and depend on the weighting function that defines
the polynomial family. As in [1], we identify the th analysis
filter of the filter bank using the th orthogonal polynomial as
follows:
(3)
The factor in the definition of the analysis filters guarantees
that their impulse responses will be real for all values of .
The output of the th analysis filter in Fig. 1 is known
as the th chromatic derivative of evaluated at .
Using (2), it can be shown that the analysis filters are poly-
nomials in . Specifically, we can express them as
(4)
1070-9908/03$17.00 © 2003 IEEE