Académique Documents
Professionnel Documents
Culture Documents
8, AUGUST 2015
1021
is [1], [2]
(2)
I. INTRODUCTION
, are given by
(3)
Manuscript received August 07, 2014; revised November 18, 2014; accepted
December 10, 2014. Date of publication December 18, 2014; date of current
version December 23, 2014. The associate editor coordinating the review of this
manuscript and approving it for publication was Prof. Mathew Magimai Doss.
H. Johansson is with the Divison of Communication Systems, Department
of Electrical Engineering, Linkping University, Linkping, Sweden (e-mail:
hakanj@isy.liu.se).
F. Harris is with the Department of Electrical and Computer Engineering, San
Diego State University, San Diego CA 92182 USA (e-mail: fharris@mail.sdsu.
edu).
Digital Object Identifier 10.1109/LSP.2014.2381603
(7)
corresponds to an
Now, since the forming of
-fold downsampling (due to
), and a multiplication of
by
(due to
), we have
1In a real-time implementation of a filter approximating (4), there will be an
additional (typically integer) delay to account for causality, but it is left out as
it does not affect the basic principles dealt with in this letter.
1070-9908 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1022
VALUES OF
AND
B. Even Values of
TABLE I
IN EXAMPLE 1 WHERE
For even values of , the terms in (8) that fall into the band
are those for
, whereas the
are those for
terms that fall into the band
. In these two bands, respectively, we will now get
complex constants, say
, and their conjugates, instead of the
above. In the band
, we get
real constants
(8)
where
. Before proceeding, it is noted that,
since
represent real DT filters,
are con, with
jugate symmetric, implying that
denoting complex conjugate. We now distinguish between odd
and even values of .
A. Odd Values of
Due to the bandlimitation of (7), it is established that the
that fall into the band
only terms in (8) for odd values of
are those for
.
Utilizing
, and inserting (6) into (8), we then get
(13)
, the summation index ranges instead
In the band
from
to
. This corresponds to complex conjugation in (13). Thus, with
, we have shown that,
for even ,
(14)
where
(15)
with
(9)
and
being, again, given by (11) and (12), and with
denoting the sign of . Hence, the only difference beis that, for even values, we
tween even and odd values of
have the additional phase offsets
as given by (15).
C. Verification of the Results
(12)
This shows that, for odd ,
are FD filters with the
FD values
given by (11), weighted with the real gain constants
given by (12). Note that
and
can take on both
positive and negative values depending on and .
2The last two equalities in (12) hold when
is not an integer
(thereby avoiding division by zero). This is ensured for all noninteger values
of which is the basic assumption as we deal with FD filters. If would be an
integer, it is obvious that one of the polyphase components would be unity and
the remaining ones zero.
1023
A. Variable FD Filters
The Farrow-structure transfer function is expressible as3
(16)
Fig. 2. Gain (
ample 1 for
(17)
Fig. 3. Upper: Magnitude response of the approximate 88th-band filter in Example 3. Lower: Approximation error modulus of the corresponding polyphase
components (FD filters) in Example 2.
, but not at
are not
1024
TABLE II
IMPULSE RESPONSE VALUES OF THE SUBFILTERS
B. Design Examples
Example 2: We consider the design of a variable FD filter
with a specification comparable to that of Example 1 of [9],
viz., a 90% bandwidth (
),
, an approxidB, and the FD covering one sammation error of some
pling interval. Here, we meet the specification using (17) with
and second-order FIR subfilters
with symmetric
(anti-symmetric) impulse responses for even (odd) values of
, as seen in Table II. The filter is designed in the minimax
sense, as in Example 1, but over all possible values of
, and essentially in the same way as outlined for Specification I in [19], but using (17) instead of (16)6.
Fig. 3 (lower) plots the approximation error modulus for the FD
values used in a 88th-band filter, to be considered in Example 3.
There are only 10 free impulse response values (parameters)
in our design, due to the impulse response symmetries and the
zero-valued taps of anti-symmetric even-order filters. In contrast, as seen in Table I in [9], the regular Farrow structure has
5When realizing the filter in the form of (17),
tend to depart from
approximating their ideal responses (obtained via Taylor series expansion), esin which case
can be of very low order
pecially for larger values of
as seen in Example 2. This is different from the regular form of (16), as noted
in Footnote 4.
6Since
the order of
even, we have here used
in (17) is
, and
.
is
IN
EXAMPLES 2 AND 3
191 free parameters. In [9], this number was reduced by combining the Farrow structure and a so called two-rate approach
which utilizes efficient half-band filters and low-order Farrow
subfilters. Together with the frequency response masking approach [21], [24], [25], the number was reduced to 71, which
still by far exceeds 10. As mentioned in Section III-A, a price to
pay using the form of (17) is an increased filter order and thus
an increase of the additional integer part of the overall delay for
the corresponding causal filter. Here, the integer-delay part is
53, which should be compared with 36 and 43, respectively, for
the two above mentioned methods.
Example 3: We now use
instances of the FD filter in
Example 2, with distinct FD values [3], as the polyphase
components of an approximate th-band filter7, to obtain
high-order frequency selective filters via few-parameter optimization. Here, we follow the polyphase decomposition in [26]
for even
to obtain a linear-phase lowpass filter, and with
the components being realized in accordance to [5] but using
(17) instead of (16). Fig. 3 (upper) shows the resulting lowpass
filter frequency response in the case of
. The overall
filter order is 9504 and the passband and stopband edges are
at
and
, respectively,
whereas the passband and stopband ripples are
and
, respectively.
The filter has the same stopband edge as the two filters compared in Example 5 of [4], but those filters meet a considerably
less stringent specification (passband edge at
, and passband and stopband ripples of 0.01 and 0.001, respectively). In
addition, those filters required more filter parameters to be determined in the design, namely 24 using the filters proposed in [4]
and 32 using the filters introduced in [27]. This should be compared with the 10 filter parameters that need to be determined
in our design. Again, this shows that we can design high-order
filters with very stringent requirements and yet with very few
free filter design parameters.
IV. CONCLUSION
This letter showed that the polyphase components of FD filters, for arbitrary -fold polyphase decompositions, are FD filters as well, but with different delays and gains. For even values
of , the components also have additional and different phase
offsets. The letter also discussed how these results can be used
in the optimization of high-order filters with few unknowns. In
particular, high-order variable FD filters and the corresponding
approximate th-band filters were discussed and exemplified
) through designs with only 10 unknown parame(for
ters in the optimization.
7The filter in Example 3 is an approximate th-band filter as the pure-delay
polyphase component of an th-band filter is approximated by an odd-order
linear-phase FIR filter.
REFERENCES
[1] P. P. Vaidyanathan, Multirate Systems and Filter Banks. Upper
Saddle River, NJ, USA: Prentice Hall, 1993.
[2] f. j. harris, Multirate Signal Processing for Communication Systems.
Upper Saddle River, NJ, USA: Prentice-Hall, 2004.
[3] M. G. Bellanger, G. Bonnerot, and M. Coudreuse, Digital filtering
by polyphase network: Application to sample-rate alteration and filter
banks, IEEE Trans. Acoust., Speech, Signal Process., vol. ASSP-24,
no. 2, pp. 109114, Apr. 1976.
[4] A. Eghbali and H. Johansson, On efficient design of high-order filters with applications to filter banks and transmultiplexers with large
number of channels, IEEE Trans. Signal Process., vol. 62, no. 6, pp.
11981209, Mar. 2014.
[5] H. Johansson and O. Gustafsson, Linear-phase FIR interpolation, decth-band filters utilizing the Farrow structure, IEEE
imation, and
Trans. Circuits Syst. I, vol. 52, no. 10, pp. 21972207, Oct. 2005.
[6] C. W. Farrow, A continuously variable delay element, in Proc. IEEE
Int. Symp., Circuits, Syst, Espoo, Finland, Jun. 79, 1988, vol. 3, pp.
26412645.
[7] D. Babic and M. Renfors, Power efficient structure for conversion
between arbitrary sampling rates, IEEE Signal Process. Lett., vol. 12,
no. 1, pp. 14, Jan. 2005.
[8] J. Selva, An efficient structure for the design of variable fractional
delay filters based on the windowing method, IEEE Trans. Signal
Process., vol. 56, no. 8, pp. 37703775, Aug. 2008.
[9] H. Johansson and E. Hermanowicz, Two-rate based low-complexity
variable fractional-delay FIR filter structures, IEEE Trans. Circuits
Syst. I, vol. 60, no. 1, pp. 136149, Jan. 2013.
[10] T. I. Laakso, V. Vlimki, M. Karjalainen, and U. K. Laine, Splitting
the unit delaytools for fractional delay filter design, Signal Process.
Mag., vol. 13, no. 1, pp. 3060, Jan. 1996.
[11] S. R. Dooley and A. K. Nandi, On explicit time delay estimation using
the farrow structure, Signal Process., vol. 72, pp. 5357, Jan. 1999.
[12] H. Johansson and P. Lwenborg, Reconstruction of nonuniformly
sampled bandlimited signals by means of digital fractional delay filters, IEEE Trans. Signal Processing, vol. 50, no. 11, pp. 27572767,
Nov. 2002.
[13] M. Olsson, H. Johansson, and P. Lwenborg, Time-delay estimation
using Farrow-based fractional-delay FIR filters: Filter approximation
vs. estimation errors, in Conf. Proc. XIV Eur. Signal Processing, Florence, Italy, Sep. 48, 2006.
1025