Lj. D. Milić, T. Saramäki, and R. Bregović, “Multirate filters: An overview,” Proc. IEEE Asia Pacific Conf. on Circuits, Syst.

, Singapore,
Dec. 2006, pp. 914–917.

Multirate Filters: An Overiew

Ljiljana D. Milić, Senior Member, IEEE, Tapio Saramäki, Fellow, IEEE, and
Robert Bregović, Member, IEEE

fixed sampling rate is of a very high order. For multirate

Abstract— The multirate filtering techniques are widely used filters, FIR (finite impulse response) or IIR (infinite impulse
in sampling rate conversion systems, and for constructing filters response) transfer functions can be used. An FIR filter easily
with equal input and output rates. Various multirate design achieves a strictly linear phase response, but requires a larger
techniques provide that the overall filtering characteristic is
shared between several simplified sub-filters that operate at the
number of operations per output sample when compared with
lowest possible sampling rates. Design constraints for sub-filters an equal magnitude response IIR filter. Multirate techniques
are relaxed if compared to a single rate overall filter. Hence, by significantly improve the efficiency of FIR filters that makes
using the multistage approach, the total number of coefficients is them very desirable in practice.
significantly reduced. As a consequence of the reduced design The advantage of the multirate approach is the possibility to
constraints, the effects of quantization (finite word-length effects) evaluate the computations in a discrete-time system at the
in sub-filters are decreased. Multirate filters provide a practical
solution for digital filters with narrow spectral constraints that lowest possible sampling rate, thus reducing significantly the
are very difficult to solve otherwise. processing workload. The second advantage is that with a
lower sampling rate, the discrete time systems become easier
Index Terms—multirate filters, sampling rate conversion, to implement.
decimation, interpolation, multistage systems, complementary
multirate filters, halfband filters II. FILTERS IN SAMPLING RATE CONVERSION

I. INTRODUCTION Filters are used in decimation to suppress aliasing, and in

interpolation to remove imaging. The performance of the
M ULTIRATE filter can be defined as a digital filter in which
the input data rate is changed in one or more
intermediate points. Multirate filters are of essential
system for sampling rate conversion is mainly determined by
filter characteristics. Since an ideal frequency response cannot
be achieved, the choice of an appropriate specification is the
importance for communications, image processing, digital first step in filter design.
audio, and multimedia. Reducing the sampling rate by a factor of M is achieved by
Multirate filters have been developed during the past three omitting every M-1 samples, or equivalently, by keeping every
decades for implementation of digital filters with stringent Mth sample. This operation is called down–sampling. In order
spectral constraints [1], [2], [3], [4], [5], [6], [9], [10], [11], to avoid aliasing, a low-pass antialiasing filter before down-
[24], [26], [29], [38], [39], [40], sampling is needed. Therefore, a decimator is a cascade of an
Multirate filtering is used whenever two digital systems with antaliasing filter and a down-sampler, Fig. 1(a). To increase
different sampling rates have to be connected. Filtering is used the sampling rate (interpolation by factor L), L-1 zeros are
to suppress aliasing in decimation, and to remove imaging in inserted between every two samples (up-sampling). An
interpolation. With the appropriate filter a digital signal of a interpolation filter has to be used to prevent imaging in the
specified sampling rate can be converted into another signal frequency band above the low-pass cutoff frequency. An
with a target sampling rate without destroying the signal interpolator is a cascade of an up-sampler and an anti-imaging
components of interest. filter, Fig. 1(b). The role of filteing in sampling rate
Multirate techniques are used in filters for sampling rate conversion is demonstrated in Fig. 2.
conversion where the input and output rates are different, and
also in constructing filters with equal input and output rates.
Multirate filtering is one of the best approaches for solving x[n] v[n] y[m] x[n] x u[m] y[m]
H(z) M L H(z)
complex filtering problems when a single filter operating at a
(a) (b)

L. D. Milić is with Faculty of Electrical Engineering, University of Fig. 1: Decimator and interpolator
Belgrade and with Mihajlo Pupin Institute, Belgrade, P. O. Box 15, 11001,
Serbia&Montenegro (e-mail: [email protected])
T. Saramäki is with the Institute of Signal Processing, Tampere University The efficiency of FIR filters for sampling rate conversion is
of Technology, P. O. Box 553, FIN-33101, Finland (e-mail: [email protected]) significantly improved using the polyphase realization.
R. Bregović is with the Institute of Signal Processing, Tampere University Filtering is embedded in the decimation/interpolation process
of Technology, P. O. Box 553, FIN-33101, Finland (e-mail:
[email protected]) and a polyphase structure is used to simultaneously achieve the
Lj. D. Milić, T. Saramäki, and R. Bregović, “Multirate filters: An overview,” Proc. IEEE Asia Pacific Conf. on Circuits, Syst., Singapore,
Dec. 2006, pp. 914–917.

interpolation/decimation by a given factor but running at a low [15], [19], [23], [24], [31].
data rate.
Spectrum of the Original Signal Spectrum of the Original Signal
1.5 1 III. MULTISTAGE FILTERING
1
0.5
0.5

0 0
For decimation and interpolation filters, and for multirate
1
Spectrum of the Down−Sampled Signal, M=2
1
Spectrum of the Up−Sampled Signal, L=4 narrowband filters, additional efficiency may be achieved by
0.5 0.5 cascading several stages, each of them consisting of a sub-
0 0 filter and down-sampler for decimation and an up-sampler and
Spectrum of the Decimated Signal, M=2 Spectrum of the Interpolated Signal, L=4
1 4
sub-filter for interpolation [9], [24], [26]. Design constraints
0.5 2
for sub-filters are relaxed if compared to an overall filter.
0 0
0 0.2 0.4 0.6 0.8
w/p 1 0 0.2 0.4 0.6 0.8
w/p
1
Hence, by using the multistage approach, the total number of
coefficients is significantly reduced when compared with the
(a) (b) single stage-design. The effects of finite word-length in sub-
filters are low in comparison with the single-stage overall
Fig. 2: Frequency domain representation: (a) Downsampling and decimation, filter. When a decimation/interpolation factor is expressible as
(b) Upsampling and interpolation
a power-of-two, the application of half-band filters improves
the efficiency of the system. An example of multistage
The polyphase structure is obtained when an Nth order filter
decimator is shown in Fig. 4.
transfer function is decomposed into M (L) polyphase
components, M,L<N. For FIR filters, polyphase
H1(z) M1 H2(z) M2 HK(z) MK
decomposition is obtained simply by inspection of the transfer
function [5], [9], [26], [29], [38]. Figure 3 shows the
Fig. 4: Multistage decimators
polyphase structures of M:1 decimator and 1:L interpolator.
IV. FILTERS WITH EQUAL INPUT AND OUTPUT RATES
Digital filters with sharp transition bands are difficult,
sometimes impossible, to be implemented using conventional
structures. A serious problem with a sharp FIR filter is its
complexity. The FIR filter length is inversely proportional to
transition–width and complexity becomes prohibitively high
(a) (b) for sharp filters [5], [9], [26], [29], [33], [38]. In a very long
FIR filter, the finite word-length effects produce a significant
Fig. 3: Polyphase structure derogation of the filtering characteristics in fixed-point
implementation [26]. IIR filters with sharp transition bands
Due to the polyphase multirate implementation, the number suffer from extremely high sensitivities of transfer function
of arithmetic operations in a linear phase FIR filter is poles that make them inconvenient for fixed-point
decreased by a factor M (or L). An effective method, which implementation [22]. In many practical cases, the multirate
leads to high efficiency for a high-order FIR filter is proposed approach is the only solution that could be applied for the
in [27]. implementation of a sharp FIR or IIR filter. Thus to design a
For multirate IIR filters, several approaches to polyphase multirate narrowband FIR or IIR filter, a classical time-
decomposition have been developed [2], [5], [8], [19], [31], invariant filter is replaced with three stages consisting of: (1) a
[32]. For a rational conversion factor L/M an efficient low-pass antialiasing filter and down-sampler, (2) a low-pass
decomposition of the Nth order IIR filter, based on the method kernel filter, and (3) an up-sampler and low-pass anti-imaging
given in [2] is proposed in [32]. filter [5], [9], [24], [26]. The total number of coefficients in a
Polyphase IIR filters require lower computation rates among multirate solution is considerably lower than the number of
the known decimators and interpolators [31]. If a strictly linear coefficients of a single rate time invariant filter.
phase characteristic is not requested, an IIR filter is an
adequate choice. Moreover, an IIR transfer function can be A. Multirate complementary filters
designed to approximate a linear phase in the pass-band [13], This method can be used in designing filters with any pass-
[20], [35]. An IIR decimator or interpolator is particularly band bandwidth. The multirate techniques are included to
useful in applications that cannot tolerate a considerably large reduce the computational complexity. Using the
delay of an adequate FIR decimator or interpolator. For a complementary property, the multirate, narrow pass-band filter
restricted class of filter specifications, an attractive solution designs can be used to develop high-pass and low-pass filters
based on all-pass sub-filters can be used leading to very with wide pass-bands [9], [26], [30]. When the output of a
efficient implementation [19], [31]. The most attractive low-pass multirate filter is subtracted from the delayed replica
solution is achieved for M,L=2 with the IIR half-band filter of the input signal, the result is a wideband high-pass filter.
implemented as a parallel connection two all-pass sub-filters The delay has to be selected to exactly equal the group delay
Lj. D. Milić, T. Saramäki, and R. Bregović, “Multirate filters: An overview,” Proc. IEEE Asia Pacific Conf. on Circuits, Syst., Singapore,
Dec. 2006, pp. 914–917.

of the multirate filter. For a low-pass wideband filter the The accurate FIR half-band filter design methods can be found
multirate narrowband high-pass filter has to be used. in [33], [36], [39]. For the IIR half-band filter design see [24],
Efficient FIR filters with an arbitrary bandwidth can be [25], [34].
designed using multirate and complementary filtering [9], [16]
[30]. The overall design is evaluated by cascading VI. COMPLEMENTARY FILTER PAIRS
complementary multirate filtering two-ports composed of two Complementary filter pairs are used to split the input signal
series branches and one parallel branch. The cascade is in two adjacent bands, and also are of importance for
terminated with a simple kernel filter. One series branch of the constructing complex multirate systems and filter banks. A
cascade is a decimator (filter and down-sampler), while the low-pass/high-pass filter pair can be designed to exhibit all-
other is an interpolator (up-sampler and filter). The parallel pass complementary, power complementary, or magnitude
branch is a delay. The most efficient solution is obtained when complementary properties [26]. In the most applications, half-
half-band filters are used in the cascade. band filter pairs are used to divide frequency band in two
Recently, the complementary filtering approach is extended equal sub-bands. This solution benefits the possibility to
to IIR filters [17]. The overall filter makes use of an IIR filter implement a complementary filter pair at the cost of a single
as a kernel filter, the periodic all-pass filters for constructing FIR (IIR) filter [9], [26], [33]. Figure 6 shows the example
complementary pair, and linear phase FIR filters for the magnitude responses of two efficient complementary filter
sampling rate alterations. pairs: (a) low-pass/high-pass linear phase FIR filter pair: filter
Figure 5 illustrates a solution of sharp wide band filter. order N=26 implemented with 7 multiplication constants, and
Magnitude response
0
7 delays (memory save implementation); (b) low-pass/high-
−20
pass IIR filter pair: filter order N=7 implemented with only 3
Gain [dB]

multiplication constants and 3 double delays. When the

−40
sampling rate is to be changed by 2, all the arithmetic
−60

0 0.2 0.4 0.6 0.8 1

operations can be evaluated at the lower rate.
ω/π

Fig. 5: Example of a sharp wideband filter. Magnitude Response (FIR)

0

B. Pipelining/Interleaving techniques −20

Gain [dB]

The pipelining/interleaving (P/I) architectures when used in −40

constructing multirate systems can significantly improve the −60

hardware efficiency. The P/I technique developed in [14] is −80

0 0.2 0.4 0.6 0.8 1

based on the repetitive use of the same filter with the clock Magnitude Response (IIR)
0
rate two or three times higher than a data rate, and with
−20
interleaved feedback of the output samples. The resulting
Gain [dB]

−40
filtering is equivalent to the cascade of identical filters.
−60

−80
V. HALF-BAND FILTERS 0 0.2 0.4 0.6 0.8
ω/π
1

Fig. 6: Efficient half-band filter pairs

Half-band filters are basic building blocks in multirate
systems. A half–band filter divides the basis band of a
As shown recently, an IIR filter pairs with the arbitrary
discrete-time system in two equal bands with symmetry
crossover frequency can be easily obtained by simple
properties. The FIR filters are most often used as half-band
transformation of the half-band filter prototypes [41], [42].
filters. For a linear phase FIR half-band filter, half of the
This new filter pair retains the complementary properties of
constants are zero valued when the filter order is an even
the start-up half-band filter pair.
number [26], [33]. A half-band IIR filter can have fewer
multipliers than the FIR filter for the same sharp cutoff
VII. MULTIPLIERLESS SOLUTIONS
specification. An IIR elliptic half-band filter when
implemented as a parallel connection of two all-pass branches The efficiency of multirate filters is greatly improved by
is an efficient solution [24]. The main disadvantage of elliptic simplifying arithmetic operations. This is achieved by
IIR filters is their very nonlinear phase response. To overcome replacing a multiplier with a small number of shifters-and-
the phase distortion one can use optimization to design an IIR adders. Generally, implementing multiplierless design
filter with an approximately linear phase response [35] or one techniques in sub-filters, at the cost of a slight derogation of
can apply the double filtering with the block processing filtering performances, increases the efficiency of the overall
technique for real-time processing [21], [28]. multirate filter. For instance, one can use the optimization
For the appropriate usage of digital filter design software in technique [18], the multiplier block approach [7], or design
half-band filter design, it is necessary to calculate the exact based on EMQF (Elliptic Minimal Q-Factors) transfer
relations between the filter design parameters in advance [24]. functions [21], [23], [25]. A well-known solution for large
Lj. D. Milić, T. Saramäki, and R. Bregović, “Multirate filters: An overview,” Proc. IEEE Asia Pacific Conf. on Circuits, Syst., Singapore,
Dec. 2006, pp. 914–917.

conversion factors in decimation is a cascaded integrated comb [22] M. D. Lutovac, D. V. Tošić, and B. L. Evans, Filter design for
signal processing using MATLAB and Mathematica, Upperside River,
(CIC) filter, which performs multiplierless filtering [11], [12], New Jersey: Prentice Hall, 2000.
[26]. [23] L. D., Milić and M. D. Lutovac, “Design of multiplierless elliptic
IIR filters with a small quantization error,” IEEE Transactions on
REFERENCES Signal Processing, vol. 47, pp. 469-479, Feb. 1999.
[24] L. D., Milić and M.D. Lutovac, “Efficient multirate filtering,” in
[1] R. Ansari and B. Liu, “Multirate signal processing,” in Sanjit. K. Gordana Jovanović-Doleček, (ed.), Multirate Systems: Design &
Mitra and James F. Kaiser (ed.), Handbook for Digital Signal Applications, Hershey, PA: Idea Group Publishing, pp. 105-142, 2002.
Processin, 981-1084, John Wiley, 1993 [25] L. D. Milić, and M. D. Lutovac, “Efficient algorithm for the design
[2] M.G. Bellanger, G. Bonnerot, and M. Coudreuse, “Digital filtering of high-speed elliptic IIR filters,” International Journal of Electronics
by polyphase network: application to sample-rate alteration and filter and Communications, vol. 57, pp. 255-262, No. 4, 2003.
banks,” IEEE Transactions on Acoustics, Speech, and Signal [26] S. K. Mitra, Digital signal processing: A computer based
Processing, vol. 24, pp. 109-114, April1976. approach, New York: The McGrow-Hill Companies, 2005.
[3] M. Bellanger, Digital processing of signals: Theory and practice, [27] S. Muramatsu and H. Kiya, “Extended overlap-add and save
New York: John Wiley & Sons, 1989. methods for multirate signal processing,” IEEE Transactions on
[4] Crochiere, R.E., and Rabiner, L.R., “Interpolation and decimation Acoustics, Speech, and Signal Processing, vol. 45, pp. 2376-2380,
of digital signals - A Tutorial Review,” Proceedings of the IEEE, 1997.
vol.78, pp. 56-93, March 1981. [28] S. Poweland M. Chau, “A technique for realizing linear phase IIR
[5] R.E., Crochiere and L.R. Rabiner, Multirate digital signal filters,” IEEE Transactions on Signal Processing, vol. 39, pp. 2425-
processing Englewood Clifs, Prentice-Hall, 1983. 2435, Nov. 1991.
[6] D.J. DeFata, J.G. Lucas, and W.S. Hodgkiss, Digital Signal [29] J. G. Proakis and D.G. Manolakis, Digital signal processing:
Processing: A System Design Approach, New York: John Wiley & Principles, algorithms, and applications, London: Prentice Hall, 1996.
Sons, 1988. [30] T.A. Ramstad and T. Saramäki, “ Multistage, multirate FIR Filter
[7] A.G. Dempster and M.D. Macleod, “General algorithms for structures for narrow transition-band filters,” in Proc. 1990 IEEE Int.
reduced-adder integer multiplier design,” Electron. Letters, vol. 31, pp. Symp. Circuits and Systems, New Orleans, Louisiana, pp. 2017 – 2021,
261-264, Oct. 1995. May 1990.
[8] W. Drews and L. Gaszi, “A new desig method for polyphase filters [31] M. Renfors and T. Saramäki, (1987, January). Recursive Nth-band
using all-pass sections,” IEEE Transactions on Circuits and Systems, digital filters - Part I: Design and properties. IEEE Transactions on
vol. 33, pp. 346-348, March 1986. Circuits and Systems, vol. 34, vol. 24-39, Jan. 1987.
[9] N. J. Fliege, Multirate digital signal processing, New York: John [32] A.I. Russel, “Efficient rational sampling rate alteration using IIR
Wiley & Sons, 1994. filters,” IEEE Signal processing Letters, vol. 7, pp. 6-7, Jan. 2000.
[10] F. J. Harris, Multirate Signal Processing, Upper Saddle River: [33] T. Saramäki, “Finite impulse response filter design,” Chapter 4 in
Prentice Hall, 2004. Handbook for Digital Signal Processing, Edited by S. K. Mitra and J.
[11] T. Hentchel, Sample rate conversion in software configurable F. Kaiser, John Wiley and Sons, New York, pp. 155 – 277, 1993.
radios, Morwood, MA: Artech House, 2002. [34] H. W. Schussler, H.W and Stefen, P. Halfband filters and Hilbert
[12] E.B. Hogenauer, “An economical class of digital filters for transformers. Circuits Systems Signal Processing, vol. 17 137-164, ,
decimation and interpolation,” IEEE Trans. Acoustics, Speech, and No. 2, 1998.
signal processing, vol. 29, pp. 155-162, 1981. [35] K. Surma-Aho and T. Saramäki, “ A systematic technique for
[13] M. Jaworski, and T. Saramäki, “Linear phase IIR filters composed designing approximately linear phase recursive digital filters,” IEEE
of two parallel all-pass sections,” in Proc of IEEE Int. Symposium on Transactions on Circuits and Systems-II, vol. 46, pp. 956-963, July
Circuits and Systems, 2, London, U. K., 1994, pp. 537-540. 1999.
[14] Jiang, Z. and A.N. Willson, Jr., “Efficient digital filtering [36] P.P. Vaidyanathan and T.O. Nguen, (1987, March). A trick for the
architectures using pipelining/interleaving,” IEEE Transactions on design of FIR half-band filters. IEEE Transactions on Circuits and
Circuits and Systems-II: Analog and Digital Signal Processing, vol. 44, Systems, vol. 34, pp. 297-300, March 1987.
pp. 110-119, Feb. 1997. [37] P.P. Vaidyanathan, “Multirate digital filters, filter banks, polyphase
[15] H. Johansson and L. Wanhammar, “High-speed recursive filter networks, and applications: A Tutorial,” Proceedings of the IEEE, vol.
structures composed of identical all-pass subfilters for interpolation, 78, pp.56-93, Jan. 1990.
decimation, and QMF banks with perfect magnitude reconstruction,” [38] P.P. Vaidyanathan, Multirate systems and filter banks. Englewood
IEEE Transactions on Circuits and Systems-II: Analog and Digital Cliffs, NJ: Prentice Hall, 1993.
Signal Processing, vol. 46, pp. 16-28, Jan. 1999. [39] A.N. Wilsson, Jr. and H. J. Orchard, “A design method for half-
[16] H. Johansson and L. Wanhammar, “Design and implementation of band FIR filters,” IEEE Transactions on Circuits and Systems-I:
multirate digital filters,” in Gordana Jovanović-Doleček, (ed.), Multirate Fundamental Theory and Applications, vol. 45, pp. 95-101, Jan. 1999.
Systems: Design & Applications, Hershey, PA: Idea Group Publishing, [40] G. Zelniker and F.T. Taylor, (1994). Advanced digital signal
257-292, 2002. processing: Theory and application, New York, NJ: Marcel Dekker,
[17] H. Johansson, “Multirate IIR filter structures for arbitrary 1994.
bandwidth,” IEEE Transactions on Circuits and Systems-I: [41] Lj. Milić and T. Saramäki, ”Power-complementary IIR filter pairs
Fundamental Theory and Applications, 50, 1515-1529, Dec. 2003. with an adjustable crossover frequency,” Facta universitatis (Niš), Ser.:
[18] J. Kaakinen and T. Saramäki, ”Design of very low-sensitivity and Elec. and Energ., vol. 16, pp. 295-304, Dec. 2003.
low-noise recursive filters using a cascade of low-order lattice wave [42] Lj. Milić and T. Saramäki, ” Three classes of IIR complementary
digital filters,” IEEE Transactions on Circuits and Systems-II: Analog filter pairs with an adjustable crossover frequency,” in Proc. of the 2003
and Digital Signal Processing, vol. 46, pp. 906-914, July 1999. Int. Symp. on Circ. and Sys., vol. 4, pp. IV-145 - IV-148, May 2003.
[19] A. Krukowski and I. Kale, DSP system design: Complexity reduced
IIR filter implementation for practical applications, Boston: Kluwer
Academic Publishers, 2003.
[20] S. S. Lawson, “Direct approach to design of PCAS filters with
combined gain and phase specification,” IEEE Proceedings of Vision,
Image and Signal Processing, vol. 141, pp. 161-167, June 1994.
[21] M. D. Lutovac, and L. D. Milić, “Approximate linear phase
multiplierless IIR half-band filter,” IEEE Signal Processing Letters, vol.
7, pp. 52-53, March 2000.