A Low-Complexity Linear-Phase Graphic Audio Equalizer based on IFIR Filters

ABSTRACT  Digital audio equalization is a very common procedure in the acoustic field that allows to improve the listening experience by adjusting the auditory frequency response. The design of graphic equalizers introduces several problems related to the necessity of implementing high performing filters with linear phase, usually characterized by high computational complexity.  This paper proposes an innovative linear-phase graphic equalizer based on interpolated finite impulse response (IFIR) filters. IFIR filters seem to be suitable for a graphic equalizer thanks to their properties. In fact, they can reach very narrow transition bands, however with low computational complexity and linear phase, avoiding ripple between adjacent bands. The proposed IFIR equalizer has been compared with some state-of-the-art methods in terms of frequency response, distortion and computational cost. The experimental results have proved the effectiveness of the proposed equalizer, that has shown a considerable reduction on the computational complexity, meanwhile preserving the performances in terms of audio quality.

 

The proposed filter-bank is based on IFIR filters. Considering M as the number of bands, the input signal is filtered by C=ceil(M/2) filters Fi(z) interpolated by a factor Li with i=1,…,C. The output of Fi(z) is the same for the i-th and the (M-i+1)-th band. Then the filters Gm(z) are applied to each band with m=1,…,M. Three different cases have been considered with different values of M: M=9, M=21 and M=31 as represented in the following figures, where a) is the FFT based approach of [1], b) is the subband approach of [2] and c) is the proposed approach.

 

 

The following table reports the respective length of the filters Fi(z) and Gm(z), the employed interpolation factors Li, and the delay τm introduced by filtering process in each band. Three different cases are considered with different values of M: M=9, M=21 and M=31.

Band

m

Length of Fi(z)

Λi

Length of Gm(z)

Γm

Interpolation factor

Li

Delay

τm

M=9

M=21

M=31

M=9

M=21

M=31

M=9

M=21

M=31

M=9

M=21

M=31

#1

244

148

124

65

69

67

6

10

12

761

769

771

#2

726

243

187

13

51

59

2

6

8

731

751

773

#3

725

362

248

23

35

53

2

4

6

735

739

767

#4

730

364

371

63

61

29

2

4

4

760

756

754

#5

184

726

370

53

15

41

8

2

4

758

732

758

#6

-

723

370

63

19

61

-

2

4

760

731

768

#7

-

724

739

23

23

15

-

2

2

735

734

745

#8

-

724

737

13

31

15

-

2

2

731

738

743

#9

-

726

738

65

51

19

-

2

2

761

750

746

#10

-

726

738

-

147

21

-

2

2

-

798

747

#11

-

122

739

-

61

27

-

12

2

-

756

751

#12

-

-

739

-

147

33

-

-

2

-

798

754

#13

-

-

738

-

51

45

-

-

2

-

750

759

#14

-

-

740

-

31

73

-

-

2

-

738

775

#15

-

-

745

-

23

211

-

-

2

-

734

849

#16

-

-

124

-

19

49

-

-

12

-

731

762

#17

-

-

-

-

15

211

-

-

-

-

732

849

#18

-

-

-

-

61

73

-

-

-

-

756

775

#19

-

-

-

-

35

45

-

-

-

-

739

759

#20

-

-

-

-

51

31

-

-

-

-

751

753

#21

-

-

-

-

69

27

-

-

-

-

769

751

#22

-

-

-

-

-

21

-

-

-

-

-

747

#23

-

-

-

-

-

19

-

-

-

-

-

746

#24

-

-

-

-

-

15

-

-

-

-

-

743

#25

-

-

-

-

-

15

-

-

-

-

-

745

#26

-

-

-

-

-

61

-

-

-

-

-

768

#27

-

-

-

-

-

41

-

-

-

-

-

758

#28

-

-

-

-

-

29

-

-

-

-

-

754

#29

-

-

-

-

-

53

-

-

-

-

-

767

#30

-

-

-

-

-

59

-

-

-

-

-

773

#31

-

-

-

-

-

67

-

-

-

-

-

771

 

-       The filters coefficients are available at the following link with .mat format: https://www.dropbox.com/sh/xdripmjddcjeiku/AABajltHa2LtU4isLDQU5kaJa?dl=0

 

-       The proposed equalizer has been compared with different state-of-the-art equalizers. In particular, the FFT-based equalizer of [1], the equalizer of [2] based on a subband approach, the IIR-based equalizer of [3] and the equalizer of [4] also based on IFIR filters, considering two different model filter length Nh. The following table reports the memory occupation in bytes of each method for the saving of the respective filters.

Equalizer

Bytes

M=9

M=21

M=31

FFT-based [1]

98984

231064

345464

Subband of [2]

10368

10752

10912

IIR of [3]

3600

8400

12400

IFIR of [4] Nh=13

144

-

-

IFIR of [4] Nh=161

1328

-

-

Proposed

23992

53392

80504

 

 

[1]      H. Schopp and H. Hetzel, “A Linear Phase 512 Band Graphic Equalizer using the Fast Fourier Transform”, in Proc. 96th Audio Engineering Society Convention, Amsterdam, The Netherlands, Feb. 1994.

[2]      S. Cecchi, L. Palestini, E. Moretti, and F. Piazza, “A new approach to digital audio equalization,” in 2007 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, 2007, pp. 62–65.

[3]      V.  Välimäki and J. Liski, “Accurate cascade graphic equalizer,” IEEE Signal Processing Letters, vol. 24, no. 2, pp. 176–180, 2017.

[4]      R.  Hergum, “A low complexity, linear phase graphic equalizer,” in Proceedings of the 85th Audio Engineering Society Convention, 1988.