Chapter-8: Maximally Decimated PR-FBs Marc Moonendspuser/DSP-CIS/2013...2 DSP-CIS / Chapter-8: Maximally Decimated PR-FBs / Version 2013-2014 p. 3 General `subband processing’ set-up

1

DSP-CIS

Chapter-8: Maximally Decimated PR-FBs

Marc Moonen Dept. E.E./ESAT-STADIUS, KU Leuven

[email protected] www.esat.kuleuven.be/stadius/

DSP-CIS / Chapter-8: Maximally Decimated PR-FBs / Version 2013-2014 p. 2

: Preliminaries •  Filter bank (FB) set-up and applications •  Perfect reconstruction (PR) problem •  Multi-rate systems review (=10 slides) •  Example: Ideal filter bank (=10 figures)

: Maximally Decimated PR-FBs •  Example: DFT/IDFT filter bank •  Perfect reconstruction theory •  Paraunitary PR-FBs

: DFT-Modulated FBs •  Maximally decimated DFT-modulated FBs •  Oversampled DFT-modulated FBs

: Special Topics •  Cosine-modulated FBs •  Non-uniform FBs & Wavelets •  Frequency domain filtering

Chapter-7

Chapter-8

Chapter-9

Chapter-10

Part-III : Filter Banks

2


General `subband processing’ set-up (Chapter-7) : PS: subband processing ignored in filter bank design

Refresh (1)

subband processing 3 H0(z) subband processing 3 H1(z) subband processing 3 H2(z)

3

3

3

3 subband processing 3 H3(z)

IN

F0(z)

F1(z)

F2(z)

F3(z)

+ OUT

downsampling/decimation

analysis bank synthesis bank

upsampling/expansion


PS: analysis filters Hn(z) are also decimation/anti-aliasing filters, to avoid aliasing in subband signals after decimation (downsampling) PS: synthesis filters Gn(z) are also interpolation filters, to remove images after expanders (upsampling)

downsampling/decimation upsampling/expansion

Refresh (2)


3

3

3


IN

F0(z)

F1(z)

F2(z)

F3(z)

+ OUT

3


Refresh (3) Assume subband processing does not modify subband signals (e.g. lossless coding/decoding) The overall aim would then be to have y[k]=u[k-d], i.e. that the output

signal is equal to the input signal up to a certain delay But: downsampling introduces ALIASING…


3

3

3


IN

F0(z)

F1(z)

F2(z)

F3(z)

+ OUT

y[k]=u[k-d]?


Refresh (4) Question : Can y[k]=u[k-d] be achieved in the presence of aliasing ? Answer : Yes !! PERFECT RECONSTRUCTION banks with synthesis bank designed to remove aliasing effects !


3

3

3


IN

F0(z)

F1(z)

F2(z)

F3(z)

+ OUT

y[k]=u[k-d]?

4


Refresh (5)

Two design issues : ✪ Filter specifications, e.g. stopband attenuation, passband ripple, transition band, etc. (for each (analysis) filter!) ✪ Perfect reconstruction (PR) property. Challenge will be in addressing these two design issues at

once (‘PR only’ (without filter specs) is easy, see next slides) This chapter : Maximally decimated FB’s : D = N


Example: DFT/IDFT Filter Bank

•  Basic question is..: Downsampling introduces ALIASING, then how can PERFECT RECONSTRUCTION (PR) (i.e. y[k]=u[k-d]) be achieved ? •  Next slides provide simple PR-FB example, to demonstrate that PR can indeed (easily) be obtained •  Discover the magic of aliasing-compensation….

G1(z) G2(z) G3(z) G4(z)

+

output = input 4 H1(z) output = input 4 H2(z) output = input 4 H3(z)

4 4 4 4 output = input 4 H4(z)

u[k]

y[k]=u[k-d]?

5



First attempt to design a perfect reconstruction filter bank - Starting point is this : convince yourself that y[k]=u[k-3] …

4

4

4

4

u[k]

Δ

Δ

Δ

4

4

4

4Δ

+ Δ

+ Δ

+ u[k-3]

u[0],0,0,0,u[4],0,0,0,...

u[-1],u[0],0,0,u[3],u[4],0,0,...

u[-2],u[-1],u[0],0,u[2],u[3],u[4],0,...

u[-3],u[-2],u[-1],u[0],u[1],u[2],u[3],u[4],...



- An equivalent representation is ... As y[k]=u[k-d], this can already be viewed as a (1st) perfect reconstruction filter bank (with lots of aliasing in the subbands!) All analysis/synthesis filters are seen to be pure delays, hence are not frequency selective (i.e. far from ideal case with ideal bandpass filters….) PS: Transmux version (TDM) see Chapter-7

4 4 4 4

+ 1−z2−z3−z

1

u[k-3] 4 4 4

1−z

2−z

3−z4

1

u[k]

6



-now insert DFT-matrix (discrete Fourier transform) and its inverse (I-DFT)... as this clearly does not change the input-output

relation (hence perfect reconstruction property preserved)

4 4 4 4

+ u[k-3]

1−z

2−z

3−z

1

1−z2−z3−z

1

4 4 4

4 u[k] F 1−F

IFF =− .1


Example: DFT/IDFT Filter Bank - …and reverse order of decimators/expanders and DFT-

matrices (not done in an efficient implementation!) :

=analysis filter bank =synthesis filter bank This is the `DFT/IDFT filter bank’. It is a first (or 2nd) example

of a maximally decimated perfect reconstruction filter bank !

4 4 4 4

1−z2−z3−z

1u[k] 4

4 4

4

F + u[k-3]

1−z

2−z

3−z

1

1−F

7



What do analysis filters look like? (N-channel case) This is seen (known) to represent a collection of filters Ho(z),H1(z),..., each of which is a frequency shifted version of Ho(z) : i.e. the Hn are obtained by uniformly shifting the `prototype’ Ho over the frequency axis.

Fu[k]

Δ

Δ

Nj

N

F

NNN

N

N

N

eW

z

zz

WWWW

WWWWWWWWWWWW

zH

zHzHzH

/2

1

2

1

)1()1(210

)1(2420

1210

0000

1

2

1

0

:

1

.

...::::

...

...

...

)(:

)()()(

2

π−

+−

−

−

−−−

−

−

−

=

⎥⎥⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢⎢⎢

⎣

⎡

⎥⎥⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢⎢⎢

⎣

⎡

=

⎥⎥⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢⎢⎢

⎣

⎡

Hn (ejω ) = H0 (e

j (ω+n.(2π /N )) ) 1210 ...1)( +−−− ++++= NzzzzH

: k

:



The prototype filter Ho(z) is a not-so-great lowpass filter with significant sidelobes. Ho(z) and Hi(z)’s are thus far from ideal lowpass/bandpass filters. Hence (maximal) decimation introduces significant ALIASING in the decimated subband signals Still, we know this is a PERFECT RECONSTRUCTION filter bank (see

construction previous slides), which means the synthesis filters can apparently restore the aliasing distortion. This is remarkable!

Other perfect reconstruction banks : read on..

Ho(z) H1(z)

N=4

H3(z) H2(z)

8



Synthesis filters ? synthesis filters are (roughly) equal to analysis filters… PS: Efficient DFT/IDFT implementation based on FFT algorithm (`Fast Fourier Transform’).

...

)(:)()()(

1

2

1

0

=

⎥⎥⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢⎢⎢

⎣

⎡

− zF

zFzFzF

N

+ 1−z

z−N+1

1

1−F

*(1/N) N=4

: :


Perfect Reconstruction Theory Now comes the hard part…(?) ✪  2-channel case: Examples… ✪ N-channel case: Polyphase decomposition based approach

9


Perfect Reconstruction : 2-Channel Case

It is proved that... (try it!)

•  U(-z) represents aliased signals, hence A(z) referred to as àlias transfer function’’.

•  T(z) referred to as `distortion function’ (amplitude & phase distortion).

H0(z)

H1(z) 2

2

u[k] 2

2

F0(z)

F1(z) + y[k]

)(.

)(

)}()()().(.{21)(.

)(

)}()()().(.{21)( 11001100 zU

zA

zFzHzFzHzU

zT

zFzHzFzHzY −−+−++=

D = N



•  Requirement for àlias-free’ filter bank :

If A(z)=0, then Y(z)=T(z).U(z), hence the complete filter bank behaves as a linear time invariant (LTI) system (despite up- & downsampling)!

•  Requirement for `perfect reconstruction’ filter bank (= alias-free + distortion-free):

H0(z)

H1(z) 2

2

u[k] 2

2

F0(z)

F1(z) + y[k]

0)( =zA

δ−= zzT )(0)( =zA

10



•  A first attempt is as follows….. : so that For the real coefficient case, i.e. which means the amplitude response of H1 is the mirror image of the amplitude response of Ho with respect to the quadrature frequency hence the name `quadrature mirror filter’ (QMF)

H0(z)

H1(z) 2

2

u[k] 2

2

F0(z)

F1(z) + y[k]

H1(z) = H0 (−z), F0 (z) = H0 (z), F1(z) = −H0 (−z)

)}()({21...)( 2

020 zHzHzT −−==0...)( ==zA

)(...)()

2(

0

)2(

1

ωπ

ωπ

−+==

jjeHeH

2π

L J



`quadrature mirror filter’ (QMF) :

hence if Ho (=Fo) is designed to be a good lowpass filter, then H1 (=-F1) is a good high-pass filter.

H0(z)

H1(z) 2

2

u[k] 2

2

F0(z)

F1(z) + y[k]

π2π

Ho H1

11



•  A 2nd (better) attempt is as follows: [Smith & Barnwell 1984] [Mintzer 1985] i) so that (alias cancellation) ii) `power symmetric’ Ho(z) (real coefficients case) iii) so that (distortion function) ignore the details! This is a so-called`paraunitary’ perfect reconstruction bank (see below), based on a lossless system Ho,H1 : 1)()(

2

1

2

0 =+ ωω jj eHeH

This is already pretty complicated…

)()( ),()( 0110 zHzFzHzF −−=−=

1...)( ==zT

0...)( ==zA

1)()(2

)2(

0

2)

2(

0 =+−+ ω

πω

π jjeHeH

][.)1(][ 01 kLhkh k −−= J

J


Perfect Reconstruction : N-Channel Case

It is proved that... (try it!)

•  2nd term represents aliased signals, hence all àlias transfer functions’ An(z) should ideally be zero (for all l )

•  T(z) is referred to as `distortion function’ (amplitude & phase distortion). For perfect reconstruction, T(z) should be a pure delay

Y (z) = 1N.{ Hn (z).Fn (z)

n=0

N−1

∑ }

T (z)

.U(z)+ 1N. { Hn (z.W

n ).Fn (z)}n=0

N−1

∑

An (z) n=1

N−1

∑ .U(z.Wn )

H2(z) H3(z)

4 4

4 4

F2(z) F3(z)

y[k] H0(z) H1(z)

4 4 u[k]

4 4

F0(z) F1(z)

+

D = N

Sigh !!…

12



•  A simpler analysis results from a polyphase description : n-th row of E(z) has polyphase components of Hn(z)

n-th column of R(z) has

polyphase components of Fn(z)

4 4 4 4

+ u[k-3] 1−z

2−z

3−z

1

1−z2−z3−z

1u[k] 4

4 4

4 )( 4zE )( 4zR

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

=

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

−−−−−

−

−)1(

1|10|1

1|00|0

1

0

:1

.

)(

)(...)(::

)(...)(

)(:)(

NNNN

NN

NN

N

N z

Nz

zEzE

zEzE

zH

zH E

)(

)(...)(::

)(...)(.

1:

)(:)(

1|11|0

0|10|0)1(

1

0

Nz

zRzR

zRzRz

zF

zF

NNN

NN

NN

NTNT

N

R

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

=

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

−−−

−−−

−

Do not continue until you understand how formulae correspond to block scheme!



•  With the `noble identities’, this is equivalent to: Necessary & sufficient conditions for i) alias cancellation ii) perfect reconstruction are then derived, based on the product

4 4 4 4

+ u[k-3] 1−z

2−z

3−z

1

1−z2−z3−z

1u[k] 4

4 4

4 )(zE )(zR

)().( zz ER

13



Necessary & sufficient condition for alias-free FB is…: a pseudo-circulant matrix is a circulant matrix with the additional feature that elements below the main diagonal are multiplied by 1/z, i.e.

..and first row of R(z).E(z) are polyphase cmpnts of `distortion function’ T(z) read on->

4 4 4 4

+ u[k-3] 1−z

2−z

3−z

1

1−z2−z3−z

1u[k] 4

4 4

4 )(zE )(zR

circulant'-`pseudo)().( =zz ER

⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢

⎣

⎡

=

−−−

−−

−

)()(.)(.)(.)()()(.)(.)()()()(.)()()()(

)().(

031

21

11

1031

21

21031

3210

zpzpzzpzzpzzpzpzpzzpzzpzpzpzpzzpzpzpzp

zz ER



PS: This can be explained as follows: first, previous block scheme is equivalent to (cfr. Noble identities)

then (iff R.E is pseudo-circ.)…

so that finally..

4 4 4 4

+ 1−z

2−z

3−z

1

1−z2−z3−z

1

u[k] 4 4 4

4 )().( 44 zz ER

)(.))()()()((.

1

...)(.

1

).().()(

43

342

241

140

3

2

1

3

2

144 zUzpzzpzzpzzp

zzz

zU

zzz

zzzT

−−−

−

−

−

−

−

−

+++

⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢

⎣

⎡

==

⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢

⎣

⎡

ER

4 4 4 4 4

+ 1−z2−z3−z

1

T(z)*u[k-3] 4 4 4

1−z

2−z

3−z

1u[k]

)(zT

14



Necessary & sufficient condition for PR is…: (i.e. where T(z)=pure delay, hence pr(z)=pure delay, and all other pn(z)=0)

In is nxn identity matrix, r is arbitrary Example (r=0) : for conciseness, will use this from now on !

4 4 4 4

+ u[k-3] 1−z

2−z

3−z

1

1−z2−z3−z

1u[k] 4

4 4

4 )(zE )(zR

10 ,0.

0)().( 1 −≤≤⎥

⎦

⎤⎢⎣

⎡=

−−−

−

NrIz

Izzz

r

rNδ

δ

ER

NIzzz δ−=)().( ER



•  Example : DFT/IDFT Filter bank : E(z)=F , R(z)=F^-1

•  Design Procedure: 1. Design all analysis filters (see Part-II). 2. This determines E(z) (=polyphase matrix). 3. Assuming E(z) can be inverted (?), synthesis filters are (delta to make synthesis causal) •  Will consider only FIR analysis filters, leading to simple polyphase

decompositions (see Chapter-7). However, FIR E(z) generally leads to IIR R(z), where stability is a concern…

)(.)( 1 zzz −−= ER δ


15



PS: Inversion of matrix transfer functions ?…

–  The inverse of a scalar (i.e. 1-by-1 matrix) FIR transfer function is always IIR (except for contrived examples)

–  The inverse of an N-by-N (N>1) FIR transfer function can be FIR 1))(det(

212

2)()(

2221

)( 21

1

1

1

12

=

⎥⎥⎦

⎤

⎢⎢⎣

⎡

+−

−==⇒

⎥⎥

⎦

⎤

⎢⎢

⎣

⎡ += −−

−

−

−

−−

z

zzz

zzz

zzz

E

ERE

)2(1)()()2()( 1

11−

−−

−==⇒−=

zzzzz ERE



PS: Inversion of matrix transfer functions ?… Compare this to inversion of integers and integer matrices:

–  The inverse of an integer is always non-integer (except for È=1’)

–  The inverse of an N-by-N (N>1) integer matrix can be integer

1)det(5465

5465 1

=

⎥⎦

⎤⎢⎣

⎡

−

−==⇒⎥

⎦

⎤⎢⎣

⎡= −

E

ERE

212 1 ==⇒= −ERE

16



Question: Can we build FIR E(z)’s that have an FIR inverse? Answer: YES, ùnimodular’ E(z)’s = matrices with {determinant=constant*z^-d} e.g.

where the El’s are constant (≠ a function of z) invertible matrices

Example: E(z) = FIR LPC lattice , see Chapter-5 (not a good filter bank though… explain!)

Design Procedure: optimize El’s to obtain filter specs (ripple, etc.), etc..

E(z) = EL.IN−1 0

0 z−1

"

#$$

%

&''.EL−1.

IN−1 0

0 z−1

"

#$$

%

&''...E1.

IN−1 0

0 z−1

"

#$$

%

&''.E0

R(z) = E0−1. z−1.IN−1 0

0 1

"

#$$

%

&''.E1

−1... z−1.IN−1 00 1

"

#$$

%

&''.EL−1

−1 . z−1.IN−1 00 1

"

#$$

%

&''.EL

−1

⇒ R(z).E(z) = z−L.IN

all E(z)’s

FIR E(z)’s

FIR unimodular E(z)’s



Question: Can we build unimodular E(z)’s with additional `special properties’, where inverses are avoided so that R(z) is trivially obtained and its specs are better controlled? (compare with (real) orthogonal or (complex) unitary matrices, where inverse is equal to (Hermitian) transpose) Answer: YES, `paraunitary‘ E(z)’s See next slides…. (start by comparing p.37 to p.31)

all E(z)’s

FIR E(z)’s

FIR unimodular E(z)’s

FIR paraunitary E(z)’s

17


Paraunitary PR Filter Banks

Review : `PARACONJUGATION’ •  For a scalar transfer function H(z), paraconjugate is i.e it is obtained from H(z) by - replacing z by 1/z - replacing each coefficient by its complex conjugate Example : On the unit circle, paraconjugate corresponds to complex conjugate paraconjugate = ànalytic extension’ of unit-circle complex conjugate

)()(~ 1*

−= zHzH

zazHzazH .1)(~.1)( *1 +=⇒+= −

*1* })({)()(~ ωωω jjj ezezez

zHzHzH==

−

===



Review : `PARACONJUGATION’ •  For a matrix transfer function H(z), paraconjugate is i.e it is obtained from H(z) by - transposition - replacing z by 1/z - replacing each coefficient by is complex conjugate Example : On the unit circle, paraconjugate corresponds to conjugate transpose(*) paraconjugate = ànalytic extension’ of unit-circle conjugate transpose(*)

)()(~ 1*

−= zz THH

[ ]zbzazzbza

z .1.1)(~.1.1

)( **1

1

++=⇒⎥⎦

⎤⎢⎣

⎡

+

+=

−

−

HH

Hezez

T

ezjjj

zzz })({)()(~ 1* ωωω ==

−

=== HHH

(*) =

Her

miti

an tr

ansp

ose

18



Review : `PARAUNITARY matrix transfer functions’ •  Matrix transfer function H(z), is paraunitary if (possibly up to a scalar)

For a square matrix function A paraunitary matrix is unitary on the unit circle paraunitary = ànalytic extension’ of unit-circle unitary. PS: if Hx(z) and Hy(z) are paraunitary, then Hx(z).Hy(z) is paraunitary

Izz =)().(~ HH

Izz jj ezH

ez=

==})(.{})({ ωω HH

1)}({)(~ −= zz HH



- If E(z) is paraunitary then perfect reconstruction is obtained with (delta to make synthesis causal) If E(z) is FIR, then R(z) is also FIR !! (cfr. definition paraconjugation)

4 4 4 4

+ u[k-3] 1−z

2−z

3−z

1

1−z2−z3−z

1u[k] 4

4 4

4 )(zE )(zR

)(~.)( zzz ER δ−=

Izz =)().(~ EE


19



•  Question: Can we build paraunitary FIR E(z) (with FIR inverse R(z))? •  Answer: Yes!

where the El’s are constant unitary matrices

•  Example: 1-input/2-output FIR lossless lattice, see Chapter-5

Example: 1-input/M-output FIR lossless lattice, see Chapter-5

•  Design Procedure: optimize unitary El’s (e.g. rotation angles in lossless lattices) to obtain analysis filter specs, etc..

E(z) = EL.IN−1 0

0 z−1

"

#$$

%

&''.EL−1.

IN−1 0

0 z−1

"

#$$

%

&''...E1.

IN−1 0

0 z−1

"

#$$

%

&''.E0

⇒ E(z).E(z) = IN

R(z) = E0H . z−1.IN−1 0

0 1

"

#$$

%

&''.E1

H ... z−1.IN−1 00 1

"

#$$

%

&''.EL−1

H . z−1.IN−1 00 1

"

#$$

%

&''.EL

H

⇒ R(z).E(z) = z−LIN


Paraunitary PR Filter Banks Properties of paraunitary PR filter banks: (proofs omitted)

•  If polyphase matrix E(z) (and hence E(z^N)) is paranunitary, and

then vector transfer function H(z) (=all analysis filters ) is paraunitary •  If vector transfer function H(z) is paraunitary, then its components are

power complementary (lossless 1-input/N-output system)

(see Chapter 5)

Hn (ejω )

2

n=0

N−1

∑ = 1 (or other constant)

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

=

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

=−−

−−−

−

−)1(

1|10|1

1|00|0

1

0

:1

.

)(

)(...)(::

)(...)(

)(:)(

)(NN

NNN

N

NN

N

N z

Nz

zEzE

zEzE

zH

zHz

E

H

20


Paraunitary PR Filter Banks Properties of paraunitary PR filter banks (continued):

•  Synthesis filters are obtained from analysis filters by conjugating the analysis filter coefficients + reversing the order (cfr page 34):

•  Hence magnitude response of synthesis filter Fn is the same as magnitude response of corresponding analysis filter Hn:

•  Hence, as analysis filters are power complementary (cfr. supra), synthesis filters are also power complementary

•  Examples: DFT/IDFT bank, 2-channel case (p. 21) •  Great properties/designs....

fn[k]= hn*[L − k], 0 ≤ n ≤ N −1

Fn (ejω ) = Hn (e

jω ) , 0 ≤ n ≤ N −1


Conclusions

•  Have derived general conditions for perfect reconstruction, based on polyphase matrices for analysis/synthesis bank

•  Seen example of general PR filter bank design : PR/FIR/Paraunitary FBs, e.g. based on FIR lossless lattice filters •  Sequel = other (better) PR structures Chapter 9: Modulated filter banks Chapter 10: Oversampled filter banks, etc..

•  Reference: `Multirate Systems & Filter Banks’ , P.P. Vaidyanathan Prentice Hall 1993.

Documents

Chapter-8: Maximally Decimated PR-FBs Marc Moonendspuser/DSP-CIS/2013...2 DSP-CIS / Chapter-8: Maximally Decimated PR-FBs / Version 2013-2014 p. 3 General `subband processing’ set-up