Вы находитесь на странице: 1из 11

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO.

5, MAY 2007

1769

61

Comb-Based Decimation Filters for


A/D
Converters: Novel Schemes and Comparisons
Massimiliano Laddomada, Member, IEEE

61

AbstractThis paper proposes several novel decimation filtering


analog-to-digital (A/D) conschemes tailored to the design of
verters. On the one hand, it extends the class of modified-sinc filters introduced by Lo Presti, in order to obtain decimation filters
capable of decoding high-order
modulators, say up to the fifth
order; on the other hand, it introduces the class of sharpenedmodified comb filters (SMCF) aimed at increasing the rejection of the
quantization noise around the folding bands and reducing the
passband drops of the designed decimation filters with respect to
the classic comb structures. Design criteria leading to optimized
structures are described. The paper compares classic comb filters
with modified and SMCF in terms of performance metrics widely
modadopted for comb filters. Finally, it derives the maximum
ulator order decodable by any proposed filter cell in the class of the
SMCFs.

61

61

61

61

Index TermsAnalog-to-digital (A/D) converter, CIC-filters,


comb, decimation, decimation filter,
, sinc filters.

Fig. 1. General architecture of a decimation and filtering chain along with a


pictorial representation of the frequency response of a classic comb decimation
filter used as first stage in the chain. The figure highlights the main performance
criteria widely adopted to compare decimation filter performances.

I. INTRODUCTION

VERSAMPLING
analog-to-digital (A/D) converters
represent a valid alternative to conventional data converters operating close to the Nyquist frequency limit [1].
Their relative simplicity and robustness against circuit imperfections and component mismatch make them especially
tailored to high-speed applications. Among their various advantages compared to conventional data converters, perhaps
the most important one is the possibility of implementation on
a single-chip high-speed Very large scale integrated (VLSI)
data converters can be found in
circuit. Details about
[1][3] and references therein.
Consider an analog input signal with maximum frequency
, sampled by a
A/D converter. The
A/D converter
much greater than
.
samples the input signal with rate
. Usually,
The oversampling ratio is defined as
the decimation and filtering of an oversampled signal is accomplished through two (or more) stages such as the ones shown
in Fig. 1, whereby the first stage is a comb filter of order ,
while the second stage is a finite-impulse-response (FIR) filter
guaranteeing the required frequency selectivity on the sampled
signal spectrum at baseband [4].
Manuscript received August 2, 2005; revised June 17, 2006. The associate
editor coordinating the review of this manuscript and approving it for publication was Dr. Yuan-Pei Lin. This work has been fully supported through funds
provided by the CAPANINA project (FP6-IST-2003-506745), as part of the EU
VI Framework Programme.
The author is with the Dipartimento di Elettronica, Politecnico di Torino,
Torino 10129, Italy (e-mail: laddomada@polito.it).
Digital Object Identifier 10.1109/TSP.2006.890822

With this background, let us provide a quick survey of the


recent literature related to the problem addressed in this paper.
This survey is by no means exhaustive and is meant to simply
provide a sampling of the literature in this fertile area.
In [5], the authors present a power efficient multirate, multistage nonrecursive comb decimation filter for 1-bit and multibit
A/D converters. Polyphase decomposition in all stages with
high decimation factor in the first stage is used to reduce the
operating rate of the nonrecursive comb filter. In [6], a lowpower fifth-order comb decimation filter with programmable
decimation factors and sampling rates suited to global system
for mobile communications (GSM) and digital enhanced cordless telecommunications (DECT) applications is presented. The
proposed architecture is based on a nonrecursive comb filter. In
[7], the authors apply filter sharpening techniques to cascaded
integrator-comb decimation filters.
The most relevant works to what is presented here are perhaps [8], [9], where the authors propose the concept of modified-sinc decimation filter for increasing the rejection of the
quantization noise (QN) around the folding bands. The design
focuses on an equivalent third-order modified-sinc decimation
filter which can lead to potential noise suppression of the order
of 8 dB compared with third-order decimation comb filters used
converters.
in second-order
In this paper, the emphasis is on the first decimation stage,
but the generalization to more stages is straightforward. We first
generalize the class of modified-sinc filters [8] by proposing

1053-587X/$25.00 2007 IEEE

1770

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 5, MAY 2007

an optimization framework for designing high-order modifiedconverters up to the fifth


comb filters (MCFs) tailored to
order. Second, we introduce the class of sharpenedmodified
comb filters (SMCFs) with the aim of increasing the rejection of
QN around the folding bands while reducing the passthe
band drops of the designed decimation filters with respect to
classic comb structures. For conciseness, we do not focus on
the practical realization of the proposed decimation filters, and
refer the interested reader to [1], [10], and [11]. Finally, we compare classic comb filters with modified and SMCFs in terms of
performance metrics widely adopted for comb filters.
The rest of the paper is organized as follows. Section II addresses the mathematical formulation adopted throughout the
paper, and briefly recalls the frequency response characteristic
of an th-order comb filter along with the main performance
criteria adopted in the design of decimation filters suitable for
converters. Section III is devoted to the design of highorder MCFs. Section IV presents the class of sharpenedmodified decimation filters. Comparisons between sharpenedmodified comb, modified-comb, and classic comb filters are given
in Section V. The conclusions are presented in Section VI. Finally, in the Appendix we present a theoretical framework for
converter order decodable by the
identifying the maximum
proposed decimation filters.
II. CLASSICAL COMB FILTERS
This section defines the mathematical terminology adopted
throughout the paper, and identifies the main performance paA/D converters.
rameters of decimation filters for
Consider, as reference, a 2-stage architecture as shown in
modulator
Fig. 1, whereby the oversampling ratio of the
is split between two decimation stages such that
(
and are the decimation factors of the first and the second filter,
as the normalrespectively). We define
ized maximum frequency contained in the input signal, so that
the digital signal at the input of the first decimation filter has
( denotes
frequency components in the range
the digital frequency).
Fig. 1 highlights the main performance parameters generally
A/D conconsidered in the design of a decimation filter for
verters, namely the passband drop and the selectivity [1],
[12]. The first parameter indicates the maximum attenuation of
the designed filter at the edge of the useful signal bandwidth
compared to the ideal low-pass filter; which is the amount of
amplitude distortion on the sampled signal that should be compensated for after decimation. The second parameter is the selectivity, , defined by Candy [12] as the ratio between the
values that the filter frequency response assumes at the edge of
the passband (i.e., at the digital frequency ) and at the lower
edge of the first folding band (see Fig. 1), i.e., at the frequency
. Performance comparison between the proposed
decimation filters will be performed in terms of both selectivity
and passband drop.
converters, particular attention
When dealing with
should be paid to the folding bands, which are the frequency
for
intervals
if
is even, and
for
odd (for

conciseness, the set of values assumed by will be denoted


throughout the paper). The reason is that the
QN
as
falling inside these frequency bands will fold down to baseband
in
because of the sampling rate reduction by a factor of
the first decimation stage, affecting the signal resolution after
multistage decimation.
The first stage of decimation in the multistage architecture
shown in Fig. 1 is usually a comb decimation filter, essentially
due to its reduced complexity compared to other options. We
recall that the transfer function of an th-order comb filter can
be expressed as [13]:
(1)
where
is the desired decimation factor. The frequency response of the th-order comb filter in (1) can be expressed as
(2)
where
is the digital frequency,
is the sampling interval, and is the analog frequency.
The frequency response in (2) is equal to zero at integer mul.
tiples of the frequencies
The passband drop introduced by an th-order comb filter on
a digital signal with maximum frequency can be evaluated as
follows:

(3)
is the residual decimation factor after the
whereby
first stage of decimation by .
The selectivity of an th-order comb filter can be evaluated
as
(4)
The aim of this paper is to propose decimation filters for
A/D converters, able to increase the
QN rejection around the
folding bands
as compared
to the classic comb filters. This topic will be addressed in detail
in the next two sections.
III. MODIFIED-COMB DECIMATION FILTERS
An efficient technique capable of increasing the attenuation
QN around the folding bands has been proposed by
of the
Lo Presti [8]. The main idea is to distribute the zeros of an
th-order comb filter in such a way as to guarantee the best
noise suppression effect around the folding bands. To this end,
a third-order MCF was proposed in [8]. Let us briefly summarize the basic rationale behind this structure.
Let us consider the transform of a first-order comb filter
. It is straightforward to
like the one shown in (1) with

LADDOMADA: COMB-BASED DECIMATION FILTERS FOR

A/D CONVERTERS

demonstrate that a clockwise rotation by an angle applied to


any zeros of the previous equation leads to the following transfer
function:
(5)
A counterclockwise rotation by the same angle applied to any
zeros of the classic first-order comb filter leads to the following
transfer function:
(6)

1771

. The value of the parameter determines the


, and therefore the filter perposition of the zeroes of
formance in terms of noise rejection within the folding bands
. The optimization goal is to
,
determine the optimal value of given the constraint
fall inside the folding bands.
which makes the zeroes of
It will be shown that the optimal choice of is generally independent of both and the decimation factor .
It is possible to optimize the parameter by maximizing the
modulator around the
attenuation of the QN in a -th order
QN power inside the folding bands is
folding bands. The
equal to [1], [3]

Multiplying (5) and (6) together, it is possible to obtain the following second-order filter:
(7)
The frequency response of the filter

(11)
where

is given by

is the power spectral density of the


can be expressed as

QN.

(12)
where
(8)
The zeroes of

are
, and
veniently selected later. It
be chosen in such a way

placed at frequencies
is a parameter that can be conis worth noting that
should
as to cover the folding bands
.
Notice that the filter in (8) is a linear-phase filter and contains
real coefficients, a constraint necessary for the physical realizability of the filter.
The transfer function of a third-order MCF, denoted by
MCF , can be obtained by multiplying a first-order comb filter
with the second-order filter cell in (7), as follows:

(9)

is the spectral density of the sampled noise under the hypothesis


of representing the QN as white noise [1], is the quantization
modulator [1], [3],
level of the quantizer contained in the
is the
sampling rate. Note that (12) is valid only if
and
the noise transfer function (NTF) of the modulator is maximally
flat, i.e., it does not contain stabilizing poles. In higher order
modulators, this requires multibit feedback structures.
The results of this optimization are shown in Fig. 2 for a
converter (i.e.,
in (12)), as a function
second-order
of the gain , The extra attenuation of the quantization noise
because of the MCF is given by
(13)

The frequency response of the filter in (9) is

(10)
The frequency response in (10) is equal to zero at the frequencies
and
. Notice that, for
, filter
becomes a classic third-order comb
.
filter
It is clear that the choice of the parameter is fundamental
for achieving a good degree of noise suppression around the
folding bands with respect to an equivalent comb filter. To this
end, in what follows, we present a procedure to optimally select
. In
the parameter appearing in the filter response
order to spread the rotated zeros all across the folding bands,
so that the rotated-zeros comb
it is possible to set
filter
in (8) has zeroes located at frequencies

From Fig. 2, the following observations can be made.


The best noise suppression effect around the folding bands
is achieved when
. This value can be deduced
from Fig. 2 by observing the gain achieved for various
(one for each subplot)
normalized signal frequencies
and various values of the decimation factor as given in
the legends.
The optimal value of does not depend on either the decimation factor and the normalized maximum frequency
of the input signal. This behavior highly simplifies the
decimation filter design.
An example can clarify the way by which MCFs distribute
the zeros of a classic comb filter across the folding bands.
for
The behavior of the frequency response
, and
, is shown in Fig. 3. In the
same figure, for comparison we show the frequency response
. Note that the
of a classic third-order comb filter

1772

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 5, MAY 2007

TABLE I
OPTIMAL POSITION OF THE ZEROS OF MCF DECIMATION FILTERS

Fig. 2. Gain G (in terms of extra attenuation of P ) achieved by rotating the


(f ) by an amount =
couple of zeroes in the frequency response H
q 1 2f as compared with the case = q 1 2f = 0, with 0  q  1, for
f = 0:001; 0:00625; 0:01; 0:0125, and decimation factors D = 4; 8; 12; 16
(in all subplots). Notice that, where not evident, the curves are superimposed.

First of all, consider that a candidate decimation filter by


for
modulators must guarantee a significant noise suppression at the digital frequencies
. This, in
turn, suggests that a fourth-order MCF can be obtained by em, placing its zeploying a second-order comb filter,
, and casroes at the decimation frequencies
cading this filter with a rotated zeroes cell
obtaining
. Following similar considerations, higher
the filter
order MCFs can be defined as follows:
;

is defined in (8). The key point for


whereby the filter cell
the design of filters
, and
is the choice of the optimal positions of the respective rotated
zeros, i.e., the optimal selection of the parameters and .
We have used an optimization procedure analogous to that
based on (11), for finding the optimal
used for
and . In particular, the parameter for the filter
values
MCF has been determined by considering a third-order
modulator, while parameters and for the filters MCF and
MCF have been determined assuming, respectively, fourthmodulators.
and fifth-order
The optimal position of the zeros are summarized in Table I,
where the extra gains are shown for each filter cell MCF .
B. Passband Drops of MCFs

Fig. 3. Magnitude of the frequency response of the filter H


(f ) compared with a classic third-order comb filter, H (f ), for f = 0:01; D = 10,
and q = 0:78.

folding bands are centered, as expected, around the digital


and range from
to
,
frequencies
. The zeroes of the filter
are
with
, spanning all the folding bands and
placed at
guaranteeing an extra noise suppression of 8 dB compared
with the classic third-order comb filter.
A. High-Order MCFs
This section presents novel MCF decimation filters suitable
modulator of order
for the decimation of data generated by a
up to the fifth.

This section derives the passband drops of MCFs, using the


same definition used for classic comb filters, as follows:

Let us define the following functions:


(14)
(15)
and

(16)

LADDOMADA: COMB-BASED DECIMATION FILTERS FOR

A/D CONVERTERS

whereby is the decimation factor of the first stage,


is the residual decimation factor after the first stage of decimation by , and is the zeroes rotation parameter for the filter
cell
.
With this setup, after some algebra, it is possible to obtain the
following expressions:

1773

, by
the folding bands
appropriately replacing second-order classic comb cells with
second-order, rotated-zeros filter cells. The starting point is the
class of sharpened filters proposed by Kaiser and Hamming [14]
for sharpening the response of a digital filter by using multiple
realizations of a low-complexity (or simple) filter. We recall
that the family of sharpened filters
satisfies the following equation [14]:

(21)

(17)
, and
in the previous
Finally, note that setting
expressions, one easily obtains the passband drops of classic
comb filters.

is the simple filter (which will be denoted as


where
represent the
basic filter). The nonnegative integers and
number of nonzero derivatives of
at the points where
and
, respectively.
In this section, the use of the family of filters defined by (21)
is first extended to the case of decimation schemes for
A/D
converters for orders and equal to 1 and 2, and then modQN rejection around the
ified in order to further improve
folding bands. With this setup, consider the following filters:

C. Selectivity of MCFs
In this section, we derive the selectivities of MCFs, using the
same definition used for classic comb filters:
(22)

Let us introduce the following functions:

(18)
and
(19)
With this setup, after some algebra, it is possible to obtain the
following expressions:

(20)

IV. SHARPENEDMODIFIED COMB FILTER DESIGN


In a companion paper [10], we introduced the class of SMCFs
modulators, by focusing on a specific decimation filter
for
QN rejection around
cell. The basic idea is to increase the

Before proceeding any further, we have to define the basic


appearing in (22). To this end, we note the
filter
following.
is a noncausal version of the actual physi
cally realizable final digital filter
. Since we
-tap FIR filters, when the final
are interested in
will be substituted by
filter is implemented,
.
In order to reduce the implementation complexity in the
proposed design, the considered basic filter will be a comb
filter as defined in (1).
Taking causality into account, terms like
in
[see (22)] should become
. Notice that in order to have an integer delay in
, an even number of taps, , is required.
This same consideration applies to all the filters in (22).
Based on the previous two observations, a second-order
comb structure will be used as basic filter by setting
. A similar approach will
be used for all the filters in (22).
The paper is aimed at reducing as much as possible the effect
QN aliasing occurring during the decimation of the
of the
digital output stream from the
modulator. To this end, first
of all note that from the way filters in (22) are obtained from
(21), the terms in square brackets in (22) are responsible for
the passband drop reduction since they impose a th-order of
tangency, where
. This, in turn, suggests that imQN rejection around the folding bands
provements on the
and
can be achieved by modifying only the filter cells

1774

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 5, MAY 2007

that in (22), cause noise attenuation of the high frequen.


cies outside the baseband
In order to further attenuate the QN, a modified-comb struclike the one shown in (8) is employed. The key
ture
concept in modifying the filter architecture in (22) is to substi, which has zeroes
tute a second-order comb filter cell
, with a modified-comb cell,
at the frequencies
in (8), whose zeroes are better distributed in the folding
.
bands
Based on the considerations given above, the class of SMCFs
can be written as follows:

TABLE II
OPTIMAL POSITION OF ZEROS OF SMCF DECIMATION FILTERS.
G IS THE EXTRA DENOISING OBTAINED BY ROTATING
THE ZEROS OF THE FILTER CELLS H (f ); i = 1; 2

(23)
. Notice that SMCFs are
whereby
linear-phase filters.
Comparing (22) and (23), we can make the following
observations.
have the same order as the
SMCFs
reported in
respective SCF filters,
(22).
The terms between square brackets in any SMCF in (23)
are responsible for reducing the passband drops at the edge
of the useful signal bandwidth . Delays appearing in
and
in the transfer functions, guarantee an
equal group delay between the various filter branches in
the square brackets.
introduces second-order zeroes exactly at
Filter
.
frequencies
Filters
, introduce two first-order zeros
.
placed in the digital frequencies
We have run an optimization procedure based on (11) for
finding the optimal values of and appearing in the SMCFs
in (23). The parameter belonging to the filters
and
has been determined by considering a third-order
modulator, while parameters
and
belonging to the filters
and
have been determined assuming fifthorder
modulators. The orders of the
modulators decodable by any SMCF have been presented in the Appendix.
The optimal position of the zeros, along with the extra gain
achieved through zero rotations, are summarized in Table II.
Figs. 4 and 5 show the magnitude of the frequency responses
of MCF and SMCFs, with the optimal zero positions reported
and
. In particular, the
in Tables I and II, for
lower subplots show the frequency behavior of the mentioned
to
filters around the baseband, ranging from
, in order to show the maximum passband drops introduced
by the various filters.

Fig. 4. Magnitude of the frequency responses of filters H (f ); H (f );


(f ), and H
(f ) for f = 0:01, and D = 10. MCF and SMCFs
have the optimal zero positions reported in Tables I and II. Notice that curves
related to the frequency responses of filters H (f ) and H
(f ) are
superimposed.
H

Analyzing the frequency behaviors of these filters, the following observations can be made.
The frequencies of the rotated zeros belonging to SMCFs
are fundamental for achieving good performance in
QN rejection around the folding bands
terms of
, with
. Furthermore, SMCFs
present an attenuation
which is somewhat smaller than both MCF and classic
fourth-order comb filters within the so called dont care
bands [1], i.e., outside the folding bands. However, it is
well known that the behavior of a decimation filter outside
the baseband and the folding bands is not very critical [1],
since it will be controlled by the sharp FIR filter employed
at the end of the decimation chain.
The passband drops of the SMCFs are, as expected, definitely smaller than in classic comb filters
and MCFs. MCFs have passband
drops very similar to the ones provided by classic comb
,
filters. At the edge of the baseband, i.e., for
MCF and classic comb filters introduce a distortion of
0.57 dB in Fig. 4, and 0.85 dB in Fig. 5, while SMCFs

LADDOMADA: COMB-BASED DECIMATION FILTERS FOR

A/D CONVERTERS

1775

B. Selectivity of SMCFs
This section reports on the selectivities of SMCFs, defined as:

Before proceedings any further, let us introduce the following


functions:

Fig. 5. Magnitude of the frequency responses of filters H (f ); H (f );


H
(f ), and H
(f ) for f = 0:01, and D = 10. MCF and SMCFs
have the optimal zero positions reported in Tables I and II. Notice that curves
(f ) are
related to the frequency responses of filters H (f ) and H
superimposed.

introduce a passband drop less than 0.06 dB in either cases


considered in Figs. 4 and 5.
Finally, note that SCFs can be obtained by the class of sharp, and
where
ened modified-comb filters by setting
it applies. This means that the mathematical derivations on both
passband drops and selectivities that are provided in the next
two sections can be applied to the class of SCFs as well. In this
respect, SCFs are a subclass of the SMCFs.

(25)
whereby is the decimation factor of the first stage, and
is the residual decimation factor after the first stage of
decimation by . With this setup, after some simple mathematical calculations, it is possible to obtain the following relations:

A. Passband Drops of SMCFs


This section derives the passband drops of SMCFs. Recalling
the definitions of the functions in (14)(16), after some mathematical manipulations, it is possible to obtain the following expressions for the passband drops:

(26)
Finally, note that selectivities of SCFs can be obtained by
setting
, and
in (26) where applicable.
V. COMPARISONS BETWEEN FILTER SCHEMES

(24)
As observed previously, the passband drops of SCFs can be
, and
in (24) where applicable.
obtained by setting

This section focuses on the comparisons between classic


comb, MCF, and SMCFs in terms of both passband drops and
selectivities. Where not specified, the zero-rotations adopted
in MCF and SMCFs are the optimal ones reported in Tables I
and II.
The left subplot in Fig. 6 compares the passband drops of
,
filters
when the first stage in Fig. 1 is a decimation by
and
, as a function of the residual decimation .
Note that when
, a fourth-order comb filter introduces
a signal distortion of about 3.6 dB, while a fourth-order MCF
presents a signal distortion even greater. As a rule of thumb, it
is not convenient to equalize a drop greater than 3 dB at the end
of the decimation chain [1]. This is one of the main reasons for
which a comb filter can be used for decimating oversampled signals down to no less than four times the Nyquist frequency, i.e.,
. Their performances rapidly degrade for
. SMCFs
constitute a class of decimation filters which can be used for

1776

Fig. 6. Left subplot shows the passband drops of filters H (f );


H
(f ); H (f ); H
(f ); H (f ), and H (f ) when the
first stage in Fig. 1 operates a decimation by D = 16, as a function of the
residual decimation  . Right subplot shows the passband drops of filters
H (f ) (dotted curves) and H (f ) (solid curves) for various values of the
first decimation factor D , as a function of the residual decimation  .

Fig. 7. Selectivities of the fourth-order filters H (f ); H (f ); H (f ),


(f ) for various values of the first decimation factor D , as a funcand H
tion of the residual decimation  .

decimating oversampled signal down to two times the Nyquist


frequency.
The right subplot in Fig. 6 compares the passband drops of
and
for various values of the first decfilters
imation factor , as a function of the residual decimation .
Other passband drop values can be easily obtained by directly
employing the mathematical derivations deduced in the previous
sections.
Let us turn our attention on the selectivity of the proposed
decimation filters. Fig. 7 compares the selectivity of fourth, and
for
order filters
various values of the first decimation factor , as a function of

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 5, MAY 2007

Fig. 8. Selectivities of MCFs compared to equivalent comb filters, H (f ),


for various values of the first decimation factor D , as a function of the residual
decimation  . Where not evident, curves are superimposed. In each subplot,
(z ) filters, while solid curves refer to H (z )
dotted curves refer to H
filters.

the residual decimation . Note that MCFs have superior performances with respect to classic and SMCFs in terms of selectivity. The reason for this improved performance has to be found
in the analysis of the behavior of SMCFs outside the baseband
, i.e., in the frequency range
. As a reference
example, consider the SMCF
. At high frequencies (i.e., for
), the cell
behaves as the constant 3, plus
samples since
, and as a consea delay of
behaves as
.
quence,
The multiplicative term 3 increases the value of
eval(i.e., the denominator
uated at the frequency
appearing in the expression for selectivity). This is the reason
for which SMCFs present a lower attenuation in the dont care
bands with respect to MCFs, as already observed in Figs. 4
and 5. In other words, the filter cell responsible for reducing
the passband drop at the edge of baseband introduces a constant term which depends, at high frequencies, upon the specific
SMCF.
Fig. 8 compares the selectivity of MCFs with the one due to
, for various values of the first
equivalent comb filters,
decimation factor , as a function of the residual decimation
. Since selectivity measures the ability of a decimation filter
to attenuate high-frequency noise accompanying the signal, we
conclude that both MCF and SMCFs present superior performance compared to classic comb filters.
For the sake of comparing the different architectures from
the point of view of the complexity of the hardware implementation, Table III briefly summarizes the number of basic
components, namely adders, delays and multipliers, needed for
implementing the proposed decimation filters. We notice that,
practically speaking, the computational complexity depends on
the specific implementation devised for the proposed decimation filters. Nevertheless, a worst case analysis of the number of
adders, multipliers and delays can be performed by adding the
complexities of the basic cells employed in the proposed decimation filters.

LADDOMADA: COMB-BASED DECIMATION FILTERS FOR

A/D CONVERTERS

TABLE III
SUMMARY OF COMPUTATIONAL COMPLEXITIES OF THE
PROPOSED DECIMATION FILTERS

1777

paper, we invite the interested reader to review [1], [8][11] for


further details on the practical realizations of the proposed decimation filters.
VI. CONCLUSION
This paper has focused on the design of novel decimation
filters suitable for
A/D converters, which show better performances in terms of passband drop, selectivity, and quantization noise rejection compared to conventional comb decimation
filters.
In particular, the paper extended the class of modified-comb
filters introduced by Lo Presti, then proposed the use of a combination of sharpened filter cells, which have the role of diminishing the filter passband drop, and modified-comb cells with
optimally rotated zeroes, which increase the quantization noise
rejection. Comparisons and tradeoffs have been presented for
A/D converters.
various classes of filters suitable for
APPENDIX
The aim of this Appendix is at demonstrating theoretically
modulator order decodable by the proposed
the maximum
SMCFs. First of all, we derive the relation between QN and
modulator with
oversampling ratio imposed by a generic
generic order .
Following the considerations stated in [1], the continuousmodulator
time equivalent transfer function of a th-order
can be expressed as
(27)
where
is the sampling period,
represents the
Fourier transform of the signal realization at the input of the
modulator, and
represents the Fourier transform of the QN
realization (modeling the effect of the quantizer contained in the
modulator). The transfer function modeling the effect of the
modulator on the QN is therefore
(28)
and we have
. Defining
as
the variance of the QN process , and in the hypothesis of white
frequency behavior (as considered in [1]), the power spectral
density of the QN can be expressed as
for
The noise power
at the output of the
tained within the frequency interval
as

modulator, con, can be expressed

(29)
The table is organized as follows. For any given decimation
filter, we show the number of adders, delays and multipliers operating at the higher rate (labeled with ) and the ones operating
at the lower rate (labelled with
).
Since a thorough analysis of the computational complexities
of the proposed decimation filters is out of the scope of this

whereby
is the oversampling ratio. Note that
(19) corresponds tantamount to filter the signal with an ideal
low-pass filter whose passband extends in the range
.
Since the QN produced by a -bits quantizer can be modeled
as a random variable with uniform distribution
within the
range
, where is the level spacing between the

1778

quantization steps [1], the variance


obtained as

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 5, MAY 2007

of the QN

can be

We recall now the expressions [15], [16]

(33)

(30)
is the quantizer input dynamic range.
where
It is known that classic th-order comb filters can decode
the data coming from
modulators of order
[12]. In
other words, th-order comb filters are as good as ideal FIR
modulators of order
filters decoding the data coming from
, in terms of the tradeoff between the noise power
and
the oversampling ratio in (29). Based on these considerations,
it is interesting to answer the following question: What is the
modulator decodable by SCFs and
maximum order of the
SMCFs?
To answer this question, we consider linear time and stationary decoding filters since they can be dealt with in a simple
way. Furthermore, since SMCFs introduce a constant noise suppression with respect to equivalent SCFs (as observed in Tables I
and II), for simplicity we will answer the question by only focusing on SCFs. Suffice to say that the amount of noise suppression guaranteed by SMCFs does not depend upon and .
Based on these observations, we will expect that the relation
for an SMCF will only differ for a constant, multiplicafor an SCF.
tive term from the relation
The average QN power at the output of the first decimation
filter by
can be written as follows:

and

(34)
whereby
is the classic binomial coefficient, and is a positive integer.
To solve the integrals, let us consider a third-order modulator,
. This way, the first integral in (32) becomes
i.e.,

The latter integral can be solved by imposing


, by
substituting (33) and (34) with
and
, respectively,
and by noting that

whereby

is the set of integer values of and such that


, or if one between or is zero then
.
This way, after some algebra it is possible to obtain

(35)

(31)
whereby
is the digital frequency, and
is the
frequency response of the sharpened decoding filter. Notice that,
in the digital domain, it is
, and that the inequality between integrals in (31) follows from the fact that the
integrand function is everywhere positive.
moduLet us start with the analysis of the maximum
lator order decodable by the filter
. Considering, for
simplicity,
(i.e.,
), and replacing the frequency
response of
in (31), it is possible to obtain

Following similar considerations, it is easily seen that the


second integral in (32) is equal to
(36)
Finally, the third integral solves to

(37)
Summation of the three integrals yields
(38)
is the big-O function.
whereby
Let us consider
. Following similar considerations, and
employing relations (33) and (34), after some algebra one easily
obtains

(39)
(32)

whereby

is a generic constant term.

LADDOMADA: COMB-BASED DECIMATION FILTERS FOR

A/D CONVERTERS

Looking at the expressions (38) and (39), and considering that


modulator goes as
for a thirdthe QN output from a
for a fourth-order modulator, we
order modulator, and as
can only decode
modulators of
conclude that filter
order up to 3. This is because the dependence of
on is
the same as for an ideal low-pass filter, as the one considered in
modulator of order up to 3. In other words,
(29), only for a
cannot be used for decoding a fourth-order (or higher)
order modulator since it will penalize the bit resolution of the
decoded data.
Following similar considerations as above, it is not difficult
can be used for decoding a
to verify that even filter
modulator of order up to 3. The only difference in (38) is in the
.
constant term that multiplies
Let us now turn the attention on the filter
. For
,
the QN power in (31), considering
, is

while

Looking at the previous expressions, and considering that the


QN output from a
modulator goes as
for a fifthorder modulator, and as
for a sixth-order modulator, we
can only decode
modulators up
conclude that filter
to the fifth order.
Finally, following similar considerations, it is not hard to
demonstrate that filter
can decode
modulators up
to the fifth order.
REFERENCES
[1] S. R. Norsworthy, R. Schreier, and G. C. Temes, Delta-Sigma Data
Converters, Theory, Design, and Simulation. Piscataway, NJ: IEEE
Press, 1997.
[2] R. M. Gray, Oversampled sigma-delta modulation, IEEE Trans.
Commun., vol. COM-35, no. 5, pp. 481489, May 1987.
[3] P. W. Wong and R. M. Gray, Two-stage sigma-delta modulation,
IEEE Trans. Acoust., Speech, Signal Process., vol. 38, no. 11, pp.
19371952, Nov. 1990.
[4] R. E. Crochiere and L. R. Rabiner, Multirate Digital Signal Processing. Englewood Cliffs, NJ: Prentice-Hall PTR, 1983.

1779

[5] H. Aboushady, Y. Dumonteix, M. Lourat, and H. Mehrez, Efficient polyphase decomposition of comb decimation filters in
analog-to-digital converters, IEEE Trans. Circuits Systems II, Analog
Digit. Signal Process., vol. 48, no. 10, pp. 898903, Oct. 2001.
[6] Y. Gao, J. Tenhunen, and H. Tenhunen, A fifth-order comb decimation filter for multi-standard transceiver applications, in Proc. IEEE
Int. Symp. Circuits Systems (ISCAS), Geneva, Switzerland, May 2831,
2000, pp. III-89III-92.
[7] A. Y. Kwentus, Z. Jiang, and A. N. Willson, Application of filter
sharpening to cascaded integrator-comb decimation filters, iEEE
Trans. Signal Process., vol. 45, no. 2, pp. 457467, Feb. 1997.
[8] L. Lo Presti, Efficient modified-sinc filters for sigma-delta A/D converters, IEEE Trans. Circuits Syst. II, Analog Digit. Signal Process.,
vol. 47, no. 11, pp. 12041213, Nov. 2000.
[9] L. Lo Presti and A. Akhdar, Efficient antialising decimation filter for
converters, in Proc. of Int. Conf. Electronics, Circuits, Systems
(ICECS), Sep. 1998, vol. 1, pp. 367370.
[10] M. Laddomada and M. Mondin, Decimation schemes for
A/D
converters based on Kaiser and Hamming sharpened filters, Proc.
Inst. Elect. Eng.Vision, Image, Signal Process., vol. 151, no. 4, pp.
287296, Aug. 2004.
[11] A. Kwentus, O. Lee, and A. N. Willson, Jr., A 250 Msample/sec programmable cascaded integrator-comb decimation filter, in Proc. Workshop VLSI Signal Processing, IX, 30 Oct.1 Nov. 1996, pp. 231240.
[12] J. C. Candy, Decimation for sigma delta modulation, IEEE Trans.
Commun., vol. COM-34, no. 1, pp. 7276, Jan. 1986.
[13] E. B. Hogenauer, An economical class of digital filters for decimation
and interpolation, IEEE Trans. Acoust., Speech, Signal Process., vol.
ASSP-29, no. 2, pp. 155162, Apr. 1981.
[14] J. Kaiser and R. Hamming, Sharpening the response of a symmetric
nonrecursive filter by multiple use of the same filter, IEEE Trans.
Acoust., Speech, Signal Process., vol. ASSP-25, pp. 415422, Oct.
1977.
[15] F. Oberhettinger, Fourier Expansions: A Collection of Formulas.
New York: Academic, 1973.
[16] I. S. Gradshteyn, I. M. Ryzhik, A. Jeffrey, and D. Zwillinger, Table of
Integrals, Series, and Products. New York: Academic Press, 2000.

61

61

61

Massimiliano Laddomada (S00M03) was born


in 1973. He received the electronics engineering degree and the Ph.D. degree in communications engineering from Politecnico di Torino, Turin, Italy, in
1999 and 2003, respectively.
From June 2000 to March 2001, he was a visiting
researcher at CSLA (California State University),
Los Angeles, and a consultant engineer in Technoconcepts, Inc., Los Angeles, CA, a startup company
specializing in software radio. He is currently
Assistant Professor at Politecnico di Torino. His
research mainly focuses on wireless communications, especially modulation
and coding, including turbo codes and, more recently, networks coding.
Dr. Laddomada was awarded a five-year open-ended fellowship by Ente per
il Diritto allo Studio Universitario (E.D.S.U.) in recognition of his university
career as an electronics engineer. In 2003, he was awarded with the Premio
Zucca per lInnovazione nellICT from Unione Industriale of Turin. Currently,
he is serving as a member of the editorial board of the IEEE Communications
Surveys and Tutorials.

Вам также может понравиться