You are on page 1of 13

Chapter 11: Multirate Digital Signal Processing

Discrete-Time Signals and Systems

Multirate Digital Signal Processing: Part IV

Reference:
Sections 11.10 and 11.11 of

Dr. Deepa Kundur

John G. Proakis and Dimitris G. Manolakis, Digital Signal Processing:
Principles, Algorithms, and Applications, 4th edition, 2007.

University of Toronto

Analysis Filter Bank
LTI Filter

y0 (n)
LTI Filter
Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV

Chapter 11: Multirate Digital Signal Processing

1 / 49

11.10 Digital Filter Banks

Chapter 11: Multirate Digital Signal Processing

Filter Banks

I

Two types: analysis and synthesis

I

consist of a parallel bank of filters used for:
I
I

Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV

2 / 49

11.10 Digital Filter
Banks
LTI Filter

Analysis and Synthesis Filter Banks
Analysis Filter Bank

Synthesis Filter Bank

LTI Filter

LTI Filter

y0 (n)

signal analysis, DFT computation, etc.
signal (re-)synthesis

y0 (n)

LTI Filter

LTI Filter

LTI Filter

LTI Filter

Synthesis Filter Bank
LTI Filter
Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV

3 / 49

Deepa Kundur
yDr.
0 (n)

(University of Toronto) Multirate Digital Signal Processing: Part IV

4 / 49

Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing Hk (ω) 5 / 49 11. . prototype filter: H0 (z) Hk (z) 11.10 Digital Filter Banks = Hk (e j2πω ) = Hk (ω) = =  −j2πk/N For k = 0. .10 Digital Filter Banks Uniform DFT Filter Bank Analysis Filter Bank LTI Filter PROTOTYPE FILTER y0 (n) LTI Filter F h0 (n) ←→ F x(n) ←→ LTI Filter Note: hk (n) = h0 (n)e j2πnk/N Z ←→ Hk (z) = H0 ze  −j2πk/N e +j 2πkn N e −j 2πkn N hk (n) = h0 (n) · e j 2πkn N F ←→ F ←→ F ←→ H0 (ω) X (ω)   2πkn δ ω− N   2πkn δ ω+ N   2πk Hk (ω) = H0 (ω) ∗ δ ω − N Synthesis Filter Bank Dr. N − 1} 3. 1. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 8 / 49 . Deepa Kundur (University of Toronto) Multirate DigitalLTI Signal Processing: Part IV Filter 7 / 49 Dr. .Chapter 11: Multirate Digital Signal Processing 11. N filters {Hk (z). N:  = hk (n) ←→ H0 F  F H0 ze   H0 e jω e −j2πk/N   2πk H0 ω − N   2πk H0 (ω) ∗ δ ω − N h0 (n) ←→ h0 (n)e j2πnk/N ∴ Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing 6 / 49 11. . . . k = 0. 2.10 Digital Filter Banks hk (n) 2πk ω− N  Hk (ω) F H0 (ω)   2πk H0 ω − N = h0 (n)e j2πnk/N ←→ Dr. 1. analysis filter bank 2. .10 Digital Filter Banks Chapter 11: Multirate Digital Signal Processing Analysis Filter Bank Prototype Filter Consider Uniform DFT Filter Bank 1. .

Chapter 11: Multirate Digital Signal Processing 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: LTI Filter 12 / 49 .10 Digital Filter Banks Chapter 11: Multirate Digital Signal Processing 11.10 Digital Filter Banks Therefore. Synthesis Filter Bank Dr.10 Digital Filter Banks Dr. at synthesis bank     z        }| {   2πk 2πk H0 (ω) · X (ω) ∗ δ ω + ∗ δ ω−   N N     | {z }    compensate at analysis bank x(n) · e| B −j2πkn/N {z F ←→ } = e −jωk n    2πk X (ω) ∗ δ ω + N where ωk = 2πkn/N . Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing 10 / 49 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV LTI Filter 11 / 49 Synthesis Filter Bank Part IV Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing 9 / 49 11.10 Digital Filter Banks LOWPASS PROTOTYPE FILTER Analysis Filter Bank LTI Filter y0 (n) Analysis Filter Bank LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter Note: downsampling reduces redundancy without loss. Uniform DFT Filter Bank Yk (ω) = Hk (ω) · X (ω) = A z }|  {   2πk 2πk H0 (ω) · X ω + ∗δ ω − N N {z } | PROTOTYPE FILTER PROTOTYPE FILTER = A B   compen.

Dr. . Upsampler Dr. thus Interpolator Decimator downsampling of v (m) is lossless compression. Decimator LOWPASS PROTOTYPE FILTER LTI Filter Downsampler MAXIMUM FREQ = I I I γk (n) is bandlimited such that it is oversampled by a factor of N 1 H0 (ω) behaves as a anti-aliasing filter prior to decimation. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Synthesis Filter Bank 16 / 49 .10 Digital Filter Banks LOWPASS PROTOTYPE FILTER Recall for D = 3.Chapter 11: Multirate Digital Signal Processing 11. Y (ωy ) = Chapter 11: Multirate Digital Signal Processing 13 / 49 11.10 Digital Filter Banks Recall.10 Digital Filter Banks ωy  D Note: y (m) contains the same information as v (m). Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing Downsampler 14 / 49 11. Analysis Filter Bank LTI Filter LTI Filter Y (ωy ) = − Dπ ≤ ωx ≤ π D 1 X D ωy  D LTI Filter of X (ωx ) is stretched into −π ≤ ωy ≤ π for Y (ωy ). Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 1 X D LTI Filter LTI Filter Dr.10 Digital Filter Banks Chapter 11: Multirate Digital Signal Processing 11. . Recall. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 15 / 49 Dr. .

Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 20 / 49 .10 Digital Filter Banks Chapter 11: Multirate Digital Signal Processing 11. LTI Filter y0 (n) Dr.Chapter 11: Multirate Digital Signal Processing 11.10 Digital Filter Banks Chapter 11: Multirate Digital Signal Processing 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV LTI Filter 19 / 49 Dr.10 Digital Filter Banks Synthesis Filter Banks LTI Filter Synthesis Filter Bank LTI Filter y0 (n) = γk (mD) Xk (m) LOWPASS PROTOTYPE FILTER −jωk n γk (n) = [x(n) · e ] ∗ h0 (n) ∞ X = x(l) · e −jωk l h0 (n − l) l=−∞ ∞ X = l=−∞ LTI Filter Analysis Filter Bank h0 (n − l)x(l) · e −jωk l LTI Filter Synthesis Filter Bank LTI Filter ∞ X Xk (m) = γk (mD) = LTI Filter Analysis Filter Bank h0 (mD Filter − LTI n)x(n) ·e −j2πnk/N Xk (m) = P∞ n=−∞ LTI Filter y0 (n) n=−∞ LTI Filter LTI Filter h0 (mD − n)x(n)e −j2πnk/N ≡ kth DFT coeff Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV LTI Filter 18 / 49 LTI Filter LTI Filter Chapter 11: Multirate Digital Signal Processing 11.10 Digital Filter Banks Analogous structure to analysis filter bank: Analogous structure to analysis filter bank: Synthesis Filter Bank Analysis Filter Bank Synthesis Filter Bank LTI Filter LTI Filter y0 (n) LTI Filter y0 (n) LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter Application-based processing may occur between the analysis and synthesis filter banks. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 17 / 49 Dr. Synthesis Filter Bank Note: compensation for baseband processing in analysis bank.

Yk (ω) Uniform DFT Filter Bank = Hk (ω) · X (ω) = A z }|   {   2πk 2πk H0 (ω) · X ω + ∗δ ω − N N | {z } = 11.Chapter 11: Multirate Digital Signal Processing 11.10 Digital Filter Banks Chapter 11: Multirate Digital Signal Processing Therefore. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 11.10 Digital Filter Banks LTI Filter Analogous structure to analysis filter bank: 1/N Synthesis Filter Bank LTI Filter  γk (n) LTI Filter vk0 (n) Yk (n/D) n = 0. . Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 23 / 49 1 N 1 N N−1 X m=−∞ vk (n) = k=0 N−1 X ∞ X 1 N N−1 X vk0 (n)e j2πnk/N k=0 Yk (m)g0 (n − mD)e j2πnk/N k=0 m=−∞ Dr.10 Digital Filter Banks PROTOTYPE FILTER LOWPASS PROTOTYPE B FILTER   compen.10 Digital Filter Banks Chapter 11: Multirate Digital Signal Processing 22 / 49 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 24 / 49 . = Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV LTI Filter Chapter 11: Multirate Digital Signal Processing 21 / 49 Dr. Dr. at synthesis bank     z      {    }|  2πk  2πk H0 (ω) · X (ω) ∗ δ ω + ∗ δ ω−   N N Analysis Filter Bank     | {z }    LTI Filter PROTOTYPE FILTER A LTI Filter compensate at analysis bank r (n) · e| +j2πkn/N {z = e jωk n }    2πk ←→ LTI FilterR(ω) ∗ δ ω − N F B where ωk = 2πkn/N . . 0 otherwise ∞ X = g0 (n) ∗ γk (n) = γk (m)g0 (n − m) = m=−∞ = · · · + γk (−D)g0 (n + D) + γk (0)g0 (n) + γk (D)g0 (n − D) + · · · ∞ ∞ X X = γk (mD)g0 (n − mD) = Yk (m)g0 (n − mD) LTI Filter m=−∞ v (n) = Note: compensation for baseband processing in analysis bank. . ±2D. ±D.

10 Digital Filter Banks Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing 25 / 49 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing 26 / 49 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 28 / 49 . Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 27 / 49 Dr. fn | {z } LTI Filter ≡yn (m)≡IDFT coeff v (m) = P∞ m=−∞ g0 (n − mD) h P N−1 1 N k=0 i Yk (m)e j2πnk/N ≡ IDFT LTI Filter hk (n) = h0 (n)e j2πnk/N   2πnk Hk (ω) = H0 ω − N Dr.10 Digital Filter Banks Alternative Filter Bank Structures LTI Filter v (n) = = N−1 ∞ 1 X X Yk (m)g0 (n − mD)e j2πnk/N N k=0 m=−∞ " N−1 # ∞ X 1 X j2πnk/N g0 (n − mD) Yk (m)e | {z } N m=−∞ k=0 interpol.10 Digital Filter Banks LTI Filter LTI Filter LTI Filter Dr.10 Digital Filter Banks Chapter 11: Multirate Digital Signal Processing 11.Chapter 11: Multirate Digital Signal Processing 11.

LTI Filter Chapter 11: Multirate Digital Signal Processing 11. the synthesis filter bank is given by: LTI Filter LTI Filter LTI Filter LOWPASS PROTOTYPE FILTER LTI Filter Analysis Filter Bank LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 31 / 49 Dr. Synthesis Filter Bank LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter Dr.10 Digital Filter Banks LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter LTI Filter vs.10 Digital Filter Banks LTI Filter Chapter 11: Multirate Digital Signal Processing LTI Filter 11.10 Digital Filter Banks Alternative Filter Bank Structures Similarly. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 30 / 49 LTI Filter Chapter 11: Multirate Digital Signal Processing 11.10 Digital Filter Banks Chapter 11: Multirate Digital Signal Processing Alternative Filter Bank Structures 11. LTI Filter vs. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 29 / 49 Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 32 / 49 .

10 Digital Filter Banks Chapter 11: Multirate Digital Signal Processing Critically Sampled Filter Banks 11. HPF ANALYSIS SECTION SYNTHESIS SECTION H0 (ω) = H(ω) H1 (ω) = H(ω − π) Dr. respectively. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing I 33 / 49 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 36 / 49 .Chapter 11: Multirate Digital Signal Processing 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing Two-Channel Quadrature Mirror Filter Bank 34 / 49 11.11 Two-Channel Quadrature Mirror Filter Bank Two-Channel Quadrature Mirror Filter Bank I multirate digital filter structure that employs: D=N I I I maximizes efficiency by minimizing the number of samples in computations Dr.11 Two-Channel Quadrature Mirror Filter Bank Two-Channel Quadrature Mirror Filter Bank Q: What is a quadrature mirror filter? LPF A: Consider analysis filters H0 (ω) and H1 (ω) that are lowpass and highpass. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 35 / 49 Dr.11 Two-Channel Quadrature Mirror Filter Bank two decimators for “signal analysis” two interpolators “signal synthesis” basic building block for quadrature mirror filter (QMF) applications Dr.

Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing LPF HPF 11.11 Two-Channel Quadrature Mirror Filter Bank LPF ANALYSIS SECTION SYNTHESIS SECTION HPF ANALYSIS SECTION SYNTHESIS SECTION Suppose we connect the analysis and synthesis sections. for a downsampler: Z x(n) = u(nD) ←→ X (z) = D−1 1 X U(z 1/D WDi ) D i=0 HPF where WD = e −j2π/D .11 Two-Channel Quadrature Mirror Filter Bank Chapter 11: Multirate Digital Signal Processing LPF Recall. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing Xa0 (ω) = Xa1 (ω) = 37 / 49 11.11 Two-Channel Quadrature Mirror Filter Bank 38 / 49 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 40 / 49 .Chapter 11: Multirate Digital Signal Processing 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 39 / 49 Then. Xs0 (ω) = Xa0 (ω) Xs1 (ω) = Xa1 (ω) ˆ (ω) X = Xs0 (2ω)G0 (ω) + Xs1 (2ω)G1 (ω) = Xa0 (2ω)G0 (ω) + Xa1 (2ω)G1 (ω) Dr. Z x(n) = u(nD) ←→ X (z) =   D−1 1 X ω − 2πi U D D i=0 Xa0 (ω) = For D = 2: X (z) = = Note:   2−1 ω − 2πi 1X U 2 2 i=0      1 ω−2π ω +U U 2 2 2 Dr.11 Two-Channel Quadrature Mirror Filter Bank ANALYSIS SECTION SYNTHESIS SECTION      1 ω ω − 2π U0 + U0 2 2 2 U0 (ω) = H0 (ω)X (ω)          1 ω ω ω − 2π ω − 2π H0 X + H0 X 2 2 2 2 2          1 ω ω ω − 2π ω − 2π H1 X + H1 X 2 2 2 2 2 Dr. V0 (ω) = Xs0 (2ω) V1 (ω) = Xs1 (2ω) ˆ (ω) = X = V0 (ω)G0 (ω) + V1 (ω)G1 (ω) Xs0 (2ω)G0 (ω) + Xs1 (2ω)G1 (ω) Dr.

Therefore. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 44 / 49 .11 Two-Channel Quadrature Mirror Filter Bank Chapter 11: Multirate Digital Signal Processing 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing To Eliminate Aliasing 42 / 49 11.ELIMINATE! =Q(z) Suppose we connect the analysis and synthesis sections. H0 (−z)G0 (z) + H1 (−z)G1 (z) = 0 1 [H0 (ω − π)G0 (ω) + H1 (ω − π)G1 (ω)] = 0 2 Sufficient condition: Let G0 (ω) = H1 (ω − π) and G1 (ω) = −H0 (ω − π).11 Two-Channel Quadrature Mirror Filter Bank LPF desired QMF bank output HPF ANALYSIS SECTION z }| { 1 ˆ (ω) = X [H0 (ω)G0 (ω) + H1 (ω)G1 (ω)] X (ω) 2 1 + [H0 (ω − π)G0 (ω) + H1 (ω − π)G1 (ω)] X (ω − π) |2 {z } SYNTHESIS SECTION effect of aliasing . ˆ (ω) = Xa0 (2ω)G0 (ω) + Xa1 (2ω)G1 (ω) X      1 2ω − 2π 2ω − 2π = H0 (ω) X (ω) + H0 X G0 (ω) 2 2 2      2ω − 2π 2ω − 2π 1 H1 (ω) X (ω) + H1 X G1 (ω) + 2 2 2 1 = [H0 (ω)G0 (ω) + H1 (ω)G1 (ω)] X (ω) 2 1 + [H0 (ω − π)G0 (ω) + H1 (ω − π)G1 (ω)] X (ω − π) 2 Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing z }| { 1 ˆ (z) = X [H0 (z)G0 (z) + H1 (z)G1 (z)] X (z) 2 1 [H0 (−z)G0 (z) + H1 (−z)G1 (z)] X (−z) + |2 {z } =A(z) = Q(z)X (z) + A(z)X (−z) 41 / 49 11.11 Two-Channel Quadrature Mirror Filter Bank Mirror Image Symmetric Filters We require A(z) = 0 or . .Chapter 11: Multirate Digital Signal Processing 11.11 Two-Channel Quadrature Mirror Filter Bank Dr. 1 [H0 (ω − π)G0 (ω) + H1 (ω − π)G1 (ω)] 2 1 = [H0 (ω − π)H1 (ω − π) + H1 (ω − π)(−H0 (ω − π))] 2 = 0 LHS = Dr. . Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 43 / 49 H0 (ω) = H(ω) H0 (z) = Hz) H1 (ω) G0 (ω) = H(ω − π) = H(ω) H1 (z) = H(−z) G0 (z) = H(z) G1 (ω) = −H(ω − π) G1 (z) = −H(−z) Dr.

Given that A(z) = 0. If H(ω) also has linear phase.11 Two-Channel Quadrature Mirror Filter Bank Condition for Perfect Reconstruction LPF Suppose F ANALYSIS SECTION h(n) ←→ H(ω) HPF SYNTHESIS SECTION Therefore.11 Two-Channel Quadrature Mirror Filter Bank Polyphase Form of the QMF Bank Thus.Chapter 11: Multirate Digital Signal Processing 11. H0 (ω) = H(ω) H1 (ω) = H(ω − π) G0 (ω) = H(ω) G1 (ω) = −H(ω − π) =⇒ =⇒ =⇒ =⇒ h0 (n) = h(n) h1 (n) = e jπn h(n) = (−1)n h(n) g0 (n)h(n) h1 (n) = −e jπn h(n) = (−1)n+1 h(n) Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 47 / 49 See Figure 11. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV  48 / 49 . this is a sufficient condition for perfect reconstruction.11 Two-Channel Quadrature Mirror Filter Bank Chapter 11: Multirate Digital Signal Processing Mirror Image Symmetric Filters 11.3 of text . perfect Recall.11 Two-Channel Quadrature Mirror Filter Bank Chapter 11: Multirate Digital Signal Processing 46 / 49 11. X reconstruction is possible for xˆ(n) = x(n−k) or: 1 [H0 (z)G0 (z) + H1 (z)G1 (z)] = z −k 2 1 = [H(z)H(z) + H(−z)(−H(−z))] = z −k 2 ∴ H 2 (z) + H 2 (−z) = 2z −k Q(z) 45 / 49 = Dr. Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV Chapter 11: Multirate Digital Signal Processing ˆ (z) = Q(z)X (z) + A(z)X (−z). Consider polyphase filters for M = 2: H 2 (z) + H 2 (−z) = 2z −k H 2 (ω) − H 2 (ω − π) = 2e −jωk |H 2 (ω) − H 2 (ω − π)| = |2e −jωk | = 2 H0 (z) = H(z) = P0 (z 2 ) + z −1 P1 (z 2 ) H1 (z) = H(−z) = P0 ((−z)2 ) + (−z)−1 P1 ((−z)2 ) = P0 (z 2 ) − z −1 P1 (z 2 ) G0 (z) = H(z)P0 (z 2 ) + z −1 P1 (z 2 )   G1 (z) = −H(−z) = − P0 ((−z)2 ) + (−z)−1 P1 ((−z)2 ) Therefore.11. |H 2 (ω) − H 2 (ω − π)| = C > 0 is a necessary condition for perfect reconstruction. Dr. Dr.

Deepa Kundur (University of Toronto) Multirate Digital Signal Processing: Part IV 49 / 49 .11 Two-Channel Quadrature Mirror Filter Bank Return Dr.Chapter 11: Multirate Digital Signal Processing 11.