SciELO - Scientific Electronic Library Online

 
vol.19 issue4Hybrid Heuristic for Dynamic Location-Allocation on Micro-Credit Territory Design author indexsubject indexsearch form
Home Pagealphabetic serial listing  

Services on Demand

Journal

Article

Indicators

Related links

  • Have no similar articlesSimilars in SciELO

Share


Computación y Sistemas

On-line version ISSN 2007-9737Print version ISSN 1405-5546

Comp. y Sist. vol.19 n.4 Ciudad de México Oct./Dec. 2015

https://doi.org/10.13053/CyS-19-4-2075 

Articles

Sharpening Minimum-Phase Interpolated Finite Impulse Response Filters

Gordana Jovanovic Dolecek1  * 

Alfonso Fernandez-Vazquez2 

1INAOE, Department of Electronics, Puebla, Mexico. gordana.@ieee.org

2 Instituto Politécnico Nacional, School of Computer Engineering, ESCOM, Mexico. afernan@ieee.org


Abstract.

This paper presents a novel simple method for the direct design of low-pass minimum-phase (MP) filters. The method is based on a linear-phase (LP) finite impulse response (FIR) prototype, sharpening and IFIR (interpolated finite impulse response) techniques, usually used for the design of LP filters. As a result, a more complex minimum phase filter can be designed by using less complex filters. The paper presents the rules and the methodology of the design and illustrates them with an example. The advantage of the method is demonstrated by comparisons with some existing designs.

Keywords: Minimum-phase filters; frequency response; finite impulse response filters; low-pass filters

1 Introduction

Minimum-phase (MP) filters find applications in communications, speech processing, and predictive coding, among others 1, 4. The special feature of an MP filter is that it does not contain zeros outside the unit circle. As a consequence, an MP filter has several important and interesting properties revised in Section 2.

There exist two main approaches for MP filter design:

  • Methods based on the low-pass finite impulse response (FIR) prototype filter design 5, 10.

  • Methods based on complex cepstrum 4, 11, 14.

The design of an MP FIR filter starting from a low-pass FIR prototype has many advantages 5. The methods consist of the FIR prototype design and finding all zeros of the designed filter. In the next step, all zeros which are outside the unit circle are reflected to their reciprocal positions inside the unit circle. This process is called mipizing 5.

We consider here the MP FIR filter design based on LP FIR prototype design and mipizing. In our previous work 8 we proposed the IFIR (interpolated finite impulse response) filter for the design of MP filters. The sharpening technique for the design of MP filters is considered in 9, 10.

Specifically, we propose a novel procedure for the design of high order MP filters based on low complexity FIR filters using the sharpening and the IFIR techniques. We also introduce the rules for sharpening IFIR MP filters.

The paper is organized in the following way. The next section gives a brief review of mipizing and properties of MP filters. A review of IFIR filters and the sharpening technique is given in Section 3. The rules for the MP IFIR sharpening are described in Section 4. Section 5 presents the steps for the MP filter design based on the rules established in Section 4. The algorithm is illustrated with one example. Finally, the comparison with existing methods is provided in Section 6 to demonstrate the benefit of the proposed design. Finally, the conclusions are given in Section 7.

2 Mipizing and Properties of MP Filters

Mipizing consists in reflecting the zeros of LP FIR filter to their reciprocal positions inside the unit circle. This process does not change the magnitude response of the filter 4, as illustrated in the following example.

Example 1. We consider the MP filter design based on mipizing of LP FIR prototype filter, which satisfies the following specification: the normalized pass-band frequency ωρ = 0.1, the normalized stop-band frequency ωs = 0.25, the pass-band ripple is of 0.1 dB, and the minimum stop-band attenuation is of 60 dB.

The filter is designed in MATLAB using Remez algorithm. The obtained LP filter has an order of Ν = 36. The plot of its zeros is shown in Fig. 1 (a). There are three zeros outside the unit circle.

Fig. 1 (a) Zeros of LP and (b) Zeros of MP filters 

Mipizing consists in reflecting those zeros to their reciprocal positions, as shown in Fig. 1 (b). Next, the zeros of the MP filter are converted to the polynomial using the MATLAB file poly.m. The polynomial coefficients correspond to the impulse response coefficients.

Fig. 2 shows the corresponding impulse and gain responses. Note that the gain responses of LP filter and the mipized filter are equal.

Fig. 2 Impulse and gain responses for LP and MP filters 

The MP filters have a number of interesting and important properties 4 illustrated in the following subsection using the filters from Example 1.

2.1 The Minimum Phase-Lag Property

Mipizing decreases the phase-lag function (the negative of the phase), as shown in Fig. 3.

Fig. 3 Phase-lag functions for MP and LP filters 

2.2 The Minimum Energy Delay Property

The partial energy of MP systems is most concentrated around n=0,

1

where hm(n) and h(n) are impulse responses of the MP and LP filters, respectively. For this reason MP filters are also called minimum energy delay systems.

The group delay (grd) of a system Η(ejω) is defined as a negative derivation of the corresponding phase:

2

The group delay of an MP system is always less than that of non-minimum phase systems having the equal magnitude responses, as shown in Fig. 4.

Fig. 4 Group delays for MP and LP filters 

3 Review of IFIR and Sharpening

The IFIR filter technique designs a more complex LP filter using two less complex filters 15. The structure of an IFIR filter is shown in Fig. 5.

Fig. 5 IFIR filter 

The model filter G(z) is designed using the following specification 15:

ωp,G=Lωp;ωs,G=Lωs;Rp,G=Rp2;As,G=As 3

where ωρ and wsare the pass-band and stop-band frequencies, RP is a maximum pass-band ripple, and As is a minimum attenuation of the prototype filter.

From (3), it follows that the transition band is L times higher than that of the prototype filter, resulting in approximately L times less order of the model filter. Next, the model filter is expanded by a factor L, becoming G(zL), so that the specification of the model filter coincides with that of the prototype filter. In this process, each delay z 1 of the designed filter G(z) is replaced by zL, or alternatively, L-1 zeros are inserted between the consecutive samples of the impulse response of the filter G(z). However, the changing from G(z) to G(z L ) has a consequence that L-1 images of the original spectrum are introduced between 0 and 2π. That is the reason why one needs another filter called interpolation filter /(z), which has to eliminate images introduced by G(z L ). The specification of the filter /(z) is given as 15:

ωp,I=ωp;ωs,I=2πL-ωs;Rp,I=Rp2;Rs,I=Rs 4

More details can be found in 15.

The sharpening technique is introduced in 16 for simultaneous improvements of both the pass-band and stop-band of a linear-phase FIR filter. The sharpening technique uses the amplitude change function (ACF) which is a polynomial relationship of the form Ηsh = f(H) between the amplitudes of the sharpened and the prototype filters, Hsh and H, respectively.

The improvement in the gain response near the pass-band edge H=1, or near the stop-band edge H=0, depends on the order of tangencies m and n of the ACF at H=1, or at H=0, respectively.

We will consider here the most popular sharpening polynomial (m=1, n=1):

Hshz=3H2z-2H3(z) 5

However, the sharpening technique can only be applied to LP filters. In 10 it is shown how the sharpening can be applied to MP filters. In this paper we show how we can also apply sharpening to MP IFIR filters.

4 Rules for Sharpening and Mipizing IFIR filters

Sharpening of an IFIR filter is practically equivalent to the cascade of sharpened expanded model and interpolation filters:

Sh{GzLI(z)}Sh{G(zL)}Sh{I(z)} 6

where Sh{.} means sharpening.

Example 2. The IFIR filter is designed satisfying the following specification: ωρ=0.2, ω5=0.3, Z=2, Rp=0.2dB, As=40dB.

Fig. 6 (a) shows the gain responses for the expanded model, interpolation, and IFIR filters. Similarly, Fig. 6 (b) compares the sharpened filters from left and right sides of (6).

Fig. 6 (a) IFIR filter (b) Sharpened IFIR 

The sharpening of interpolation and expanded model filters can be presented in the following forms 10:

7

8

where

9

10

From (9) and (10), it follows that the order Ν of both model and interpolator filters must be even. The corresponding impulse responses are

11a

11b

12a

12b

In what follows we explain how to obtain the corresponding MP filters from (9) and (10).

4.1 Expanded Filter Roots

In the expanded filter, each zero of the original filter zi = ri ejψ is mapped into L zeros 8.

As an example, Fig. 7 presents zeros of the model filter G(z) and the corresponding expanded filter G(zL), for Z=2.

z´i=ri1/Lej(φ+2πm)/L,m=0,...,L-1 13

Fig. 7 (a) Zeros and poles of model and (b) Expanded model filters 

The mipizing of the expanded filter can be performed in two steps:

  • Mipize original filter,

  • - Generate zeros using (13).

Fig. 8 (a) shows zeros of the mipized filter G(z), while Fig. 8 (b) shows the zeros of the mipized expanded filter. Fig. 9 presents the corresponding impulse responses obtained using MATLAB file poly.m.

Fig. 8 (a) Zeros of mipized model and (b) Expanded model filters 

Fig. 9 Impulse responses of MP filters Gm(z) and Gm( zL

4.2 Rouche's Theorem

We apply the Rouche's theorem 17 to find the number of zeros inside the unit circle for polynomials Gs(z) and ls(z). The theorem states the following 17: "Let f(z) and g(z) be analytic functions inside and on a positively oriented simple closed contour C, and if \f(z)\ > \g(z)\ at each point z on C, the functions f(z) and f(z)+ g(z) have the same number of zeros, counting multiplicities, inside C."

To this end we rewrite Gs(z) as

14

We denote

15

The corresponding absolute values are

16

Considering that the contour C is a unit circle, we have |z|=1, resulting in

17

Observe that the condition in Rouche's theorem is satisfied because

18

on the unit circle.

Since f(z) has N/2 zeros, counting multiplicities, interior to the unit circle |z|=1, the polynomial f(z)+g(z), or Gs(z) has also N/2 zeros inside the unit circle. Knowing that the filter Gs(z) is a linear-phase filter, the rest of its N/2 zeros are outside the unit circle in the reciprocal positions of zeros that are inside the unit circle.

Similarly, we can also show that half of zeros of a linear-phase filter ls(z) are inside the unit circle, and the other half of zeros are in the reciprocal positions, i.e., outside the unit circle.

These statements result in a simple rule for the mipizing the filters Gs(z) and ls(z):

-Eliminate zeros outside the unit circle, and double the zeros which are inside the unit circle.

5 Description of the Method

The sharpened MP filter Hm(z) is obtained using the results from section 4 as

19

Where

  • Gm (zL) is the mipized expanded model filter.

  • lm(z) is the mipized interpolation filter.

  • Gsm(z) is the mipized filter Gs(z).

  • Ism(z) is the mipized filter ls(z).

The procedure for the design is described in the following steps:

Step 1. Design the model filter G(z) and the interpolator l(z) according to the given specification.

Step 2. Mipize the filters G(z) and l(z) in order to obtain the MP filters Gm(z) and lm(z).

Step 3. Obtain the expanded filter Gm(zL) from the MP filter Gm(z).

Step 4. Using (11) and (12), find the LP filters Gs(z) and ls(z).

Step 5. Delete all zeros of Gs(z) and U(z) which are outside the unit circle, and double those which are inside.

Step 6. Find Gsm(z) and lsm(z).

Step 7. Using (19), find the sharpened IFIR MP filter.

Example 3. We illustrate the method by sharpening the IFIR filter which satisfies the following specification: ωρ=0.12, ωs=0.15, L=5, Rp=1dB, As=38dB.

Step 1. We design the model and interpolation filters and find their zeros, as shown in Fig. 10. The orders of the model and interpolation filters are 18 and 24, respectively, and the conditions in (9) and (10) are satisfied. Observe that the orders of filters are even.

Fig. 10 Zeros of model and interpolation filters 

Step 2. We mipize the model and interpolation filters, as shown in Fig. 11.

Fig. 11 Mipized model and interpolation filters 

Step 3. Expand the filter Gm(z) using (13). Fig. 12 shows the zeros and impulse response of the MP expanded model filter.

Fig. 12 (a) Zeros of expanded model filter, (b) Impulse response of MP expanded model filter 

Step 4. Find LP filters Gs(z) and ls(z).

Step 5. Find the zeros of the LP filters Gs(z) and ls(z) and the zeros of the corresponding MP filters Gsm(z) and l sm(z), as shown in Figs. 13 and 14.

Fig. 13 Zeros of LP and MP filters Gs(z) and Gsm(z) 

Fig. 14 Zeros of LP and MP filters ls(z) and /sm(z) 

Step 6. Find the corresponding polynomials Gsm(z) and l sm(z).

Step 7. Using (19), find the sharpened IFIR MP filter.

Fig. 15 compares the gain responses of MP IFIR and MP sharpened IFIR filters. Note that the resulting MP filter has approximately 50% more stop-band attenuation and two times less pass-band ripple than the MP IFIR filter.

Fig. 15: MP IFIR and MP sharpened IFIR filters 

The corresponding MP filter obtained by mipizing an LP equiripple FIR filter has an order of 158. However, in the proposed design, the MP filter is obtained using filters of orders 18 and 24.

6 Comparison with the Method in ( 4 )

Finally, we compare our proposed design with the method described in 4, which uses the cepstrum technique. Fig. 16 shows the resulting gain responses. Observe that our proposed approach has better gain response than the approach given in 4. It is worth to point out that our approach is based on the design of low order digital filters, while method in 4 is based on the design of a high order FIR filter and the complex cepstrum technique.

Fig. 16 Method in (4) and MP sharpened IFIR filters 

7 Conclusions

A simple procedure for sharpening MP FIR filters is presented. It is demonstrated that the sharpening and IFIR design techniques, developed for LP narrow-band filters, can also be used for the MP narrow-band filter design. The only restriction is that the orders of the model and interpolation filters must be even.

Additionally, our method, like the IFIR technique, is useful not only for lowpass filters but also for highpass and certain bandpass filters, by a simple change of the specification of the corresponding interpolation filter. The filters are designed using MATLAB Remez algorithm and the files roots.m and poly.m. This work is devoted to the design of narrow-band MP filters. As a future work, we consider the design of minimum phase filters using frequency masking approach 2, which can be used to design wideband minimum-phase filters.

Acknowledgement

This work is supported by the CONACYT grant No. 179587.

References

1. Mitra, S.K. & Kaiser, J.F. (1993). Handbook for Digital Signal ProcessingNY: John Wiley and Sons. [ Links ]

2. Mitra, S.K. (2010). Digital Signal Processing: A Computer-Based ApproachNY: McGraw-Hill. [ Links ]

3. Haykin, S. (1990). Modern FiltersNY: McMillan Publishing Company. [ Links ]

4. Oppenheim, A.V. & Schafer, R.W. (2009). Digital Signal Processing3 ed., NJ: Prentice-Hall. Englewood Cliffs. [ Links ]

5. Kale, I. Cain, G.D., & Morling, R.C.S. (1995). Minimum-Phase Filter Design from Linear-Phase Startpoint via Balanced Model Truncation. Electronic LettersVol. 31, No. 20, pp. 1728-1729. DOI: 10.1049/el:19951228. [ Links ]

6. Herrmann, O. & Schuessler, W. (1974). Design of Nonrecursive Digital Filters with Minimum Phase. Electronic LettersVol. 6, No. 11, pp. 69-70. DOI: 10.1049/el:19700232. [ Links ]

7. Stathaki, T. & Fotinopoulos, I. (2010). Equiripple Minimum Phase FIR Filter Design from Linear Phase System Using Root Moments. IEEE Trans, on Circ and Syst-ll: Analog and Digital Signal ProcessingVol. 48, No. 6, pp. 580-587. DOI: 10.1109/82.943328. [ Links ]

8. Jovanovic-Dolecek, G. & Diaz-Carmona, J. (1997). Lowpass Minimum Phase Filter Design Using IFIR Filters. Electronic LettersVol. 33 No. 23. pp. 1933-1935. DOI: 10.1049/el:19971330. [ Links ]

9. Jovanovic-Dolecek, G. & Dolecek, V. (2001). Method for Narrowband Minimum Phase Filter Design. Electronic LettersVol. 37. No. 5. pp. 324-325. DOI: 10.1049/el:20010185. [ Links ]

10. Jovanovic-Dolecek G. & Fernandez-Vazquez, A. (2013). Sharpening Minimum Phase Filters. Journal of PhysicsVol. 410, No. 1, pp.1-4. [ Links ]

11. Pei, S.C. & Lin, H.S. (2006). Minimum-Phase FIR Filter Design Using Real Cepstrum. IEEE Transactions on Circuits and Systems-II: Express BriefsVol. 53, No. 10, pp. 1113-1117. DOI: 10.1109/TCSH.2006.882193 [ Links ]

12. Dam, H.H., Nordebo, S., & Svensson, L. (2003). Design of Minimum-Phase Digital filters as the Sum of two Allpass Functions Using the Cepstrum Technique. IEEE Transactions on Signal ProcessingVol. 51, No. 3, pp. 726-731. DOI: 10.1109/TSP.2002806983 [ Links ]

13. Nagarajan, T., Prasad K., & Murthy, H.A. (2003). Minimum Phase Signal Derived from Root Cepstrum. Electronic LettersVol. 39, No. 12, pp. 941-942. DOI: 10.1049/el:20030616. [ Links ]

14. Venkata, N.D., Evans, B.L., & McCaslin, S.R. (2000). Design of Optimal Minimum-phase Digital FIR Filters Using Discrete Hilbert Transforms. IEEE Transactions on Signal ProcessingVol. 48, No. 5. pp. 1491-1495. [ Links ]

15. Nuevo, Y., Dong, C.Y., & Mitra, S.K. (1984). Interpolated Finite Impulse Response Filters. IEEE Transactions on Acoustic, Speech, and Signal ProcessingVol. 32, No. 3, pp. 563-570. DOI: 10.1109/TASSP.1984.1164348. [ Links ]

16. Kaiser, J.F. & Hamming, R.W. (1977). Sharpening the Response of Symmetric Non-recursive Filter. IEEE Transactions on Acoustic, Speech, and Signal ProcessingVol. 25, No. 5, pp. 415-422. DOI: 10.1109/TASSP.1977.1162980 [ Links ]

17. Churchiill, R.V. & Brown, J.W. (1984). Complex Variables and Applications NY: McGraw-Hill , Inc. [ Links ]

Received: October 30, 2014; Accepted: November 11, 2015

Corresponding author is Gordana Jovanovic Dolecek

Gordana Jovanovic Dolecek received a B.S. degree from the Department of Electrical Engineering, University of Sarajevo, an M.Sc. degree from University of Belgrade, and a Ph.D. degree from the Faculty of Electrical Engineering, University of Sarajevo. She was professor at the Faculty of Electrical Engineering, University of Sarajevo, until 1993, and in 1993-1995 she was with the Institute Mihailo Pupin, Belgrade. In 1995 she joined the Institute of Astrophysics, Optics and Electronics (INAOE), Department for Electronics. Puebla, Mexico, where she works as a full professor. During 2001-2002 and 2006 she was with the Department of Electrical & Computer Engineering, University of California, Santa Barbara, as visiting researcher. She was with San Diego State University as visiting researcher on a sabbatical leave in 2008-2009. Currently she is on sabbatical leave at the University of California, UCI A, I A, USA. She is the author of three books, editor of one book, and author/coauthor of more than 300 papers. Her research interests include digital signal processing and digital communications. She is a Senior Member of IEEE, a member of Mexican Academy of Science, and a member of National Researcher System (SNI) Mexico.

Alfonso Fernandez-Vazquez received his B.S. degree in Electronics Engineering from Instituto Tecnologico de Puebla in 1998, and M.Sc. and Ph.D. degrees from the Department of Electronics, the Institute of Astrophysics, Optics and Electronics (INAOE), in 2000 and 2006. respectively. He was a postdoc researcher at the Universidad de las Americas, Puebla, from July to December 2006. In 2007, he was an Assistant Professor at the School of Electronic Engineering, Xidian University, Xian, P. R. China. He is an IEEE member and a member of SNI (National Researcher System in Mexico) Level 1. His current fields of interest include digital signal processing and digital communications.

Creative Commons License This is an open-access article distributed under the terms of the Creative Commons Attribution License