Multirate

11: Multirate Systems
Multirate Systems Building blocks Resampling Cascades Noble Identities Noble Identities Proof Upsampled z-transform Downsampled z-transform Downsampled Spectrum Perfect Reconstruction Commutators Summary MATLAB routines
DSP and Digital Filters (2013-2852)
Multirate: 11 1 / 13
Multirate Systems
Multirate systems include more than one sample rate
Multirate Systems
Multirate systems include more than one sample rate Why bother?:
May need to change the sample rate

e.g. Audio sample rates include 32, 44.1, 48, 96 kHz
Multirate Systems

Can relax analog or digital lter requirements

e.g. Audio DAC increases sample rate so that the reconstruction lter can have a more gradual cutoff
Multirate Systems

Can relax analog or digital lter requirements

e.g. Audio DAC increases sample rate so that the reconstruction lter can have a more gradual cutoff
Reduce computational complexity f FIR lter length s f where f is width of transition band Lower fs shorter lter + fewer samples computation
1 2 fs
Building blocks
Downsample
y [m] = x[Km]
Building blocks
Downsample Upsample
y [m] = x[Km] v [n ] = u 0
n K
K|n
else
Building blocks
Downsample Upsample
y [m] = x[Km] v [n ] = u 0
n K
K|n
else
Example: Downsample by 3 then upsample by 4

w[n]
Building blocks
Downsample Upsample
y [m] = x[Km] v [n ] = u 0
n K
K|n
else

w[n] x[m]
Building blocks
Downsample Upsample
y [m] = x[Km] v [n ] = u 0
n K
K|n
else

w[n] x[m] y[r]
Building blocks
Downsample Upsample
y [m] = x[Km] v [n ] = u 0
n K
K|n
else

w[n] x[m] y[r]
We use different index variables (n, m, r ) for different sample rates
Building blocks
Downsample Upsample
y [m] = x[Km] v [n ] = u 0
n K
K|n
else

w[n] x[m] y[r]
We use different index variables (n, m, r ) for different sample rates Use different colours for signals at different rates (sometimes)
Building blocks
Downsample Upsample
y [m] = x[Km] v [n ] = u 0
n K
K|n
else

w[n] x[m] y[r]
We use different index variables (n, m, r ) for different sample rates Use different colours for signals at different rates (sometimes) Synchronization: all signals have a sample at n = 0.
Resampling Cascades
Successive downsamplers or upsamplers can be combined
Resampling Cascades
Successive downsamplers or upsamplers can be combined Upsampling can be exactly inverted
Resampling Cascades
Successive downsamplers or upsamplers can be combined Upsampling can be exactly inverted Downsampling destroys information permanently uninvertible
Resampling Cascades
Successive downsamplers or upsamplers can be combined Upsampling can be exactly inverted Downsampling destroys information permanently uninvertible Resampling can be interchanged iff P and Q are coprime (surprising!)
Resampling Cascades
P n if Q | n else y [n] = 0. Proof: Left side: y [n] = x Q
Resampling Cascades
P n if Q | n else y [n] = 0. Proof: Left side: y [n] = x Q P Right side: y [n] = x Q n if Q | P n.
Resampling Cascades
P n if Q | n else y [n] = 0. Proof: Left side: y [n] = x Q P Right side: y [n] = x Q n if Q | P n. But {Q | P n Q | n} iff P and Q are coprime.
Noble Identities
Resamplers commute with addition and multiplication
Noble Identities
Delays must be multiplied by the resampling ratio
Noble Identities
Delays must be multiplied by the resampling ratio Noble identities: Exchange resamplers and lters
Noble Identities
Example: H (z ) = h[0] + h[1]z 1 + h[2]z 2 +
H (z 3 ) = h[0] + h[1]z 3 + h[2]z 6 +
Noble Identities
Corrollary Example: H (z ) = h[0] + h[1]z 1 + h[2]z 2 +
H (z 3 ) = h[0] + h[1]z 3 + h[2]z 6 +
Noble Identities Proof

Dene hQ [n] to be the impulse response of H (z Q ).

Dene hQ [n] to be the impulse response of H (z Q ). Assume that h[r ] is of length M + 1 so that hQ [n] is of length QM + 1.

Dene hQ [n] to be the impulse response of H (z Q ). Assume that h[r ] is of length M + 1 so that hQ [n] is of length QM + 1. We know that hQ [n] = 0 except when Q | n and that h[r ] = hQ [Qr ].

w[r ] = v [Qr ]

w[r ] = v [Qr ] =
QM s=0
hQ [s]x[Qr s]

w[r ] = v [Qr ] = QM s=0 hQ [s]x[Qr s] M = m=0 hQ [Qm]x[Qr Qm]

w[r ] = v [Qr ] = QM s=0 hQ [s]x[Qr s] M = m=0 hQ [Qm]x[Qr Qm] =
M m=0
h[m]x[Q(r m)]

w[r ] = v [Qr ] = QM s=0 hQ [s]x[Qr s] M = m=0 hQ [Qm]x[Qr Qm] = = M m=0 h[m]u[r m]
M m=0
h[m]x[Q(r m)]

w[r ] = v [Qr ] = QM s=0 hQ [s]x[Qr s] M = m=0 hQ [Qm]x[Qr Qm] = = M m=0 h[m]u[r m] = y [r ]
M m=0
h[m]x[Q(r m)]


Upsampled Noble Identity:
M m=0
h[m]x[Q(r m)]


M m=0
h[m]x[Q(r m)]
We know that v [n] = 0 except when Q | n and that v [Qr ] = x[r ].


M m=0
h[m]x[Q(r m)]
w [n ] =
QM s=0
h Q [s ]v [n s ]


M m=0
h[m]x[Q(r m)]
w [n ] =
QM s=0
h Q [s ]v [n s ] =
M m=0 hQ [Qm]v [n
Qm]


M m=0
h[m]x[Q(r m)]
w [n ] = =
QM s=0 hQ [s]v [n s] = M m=0 h[m]v [n Qm]
M m=0 hQ [Qm]v [n
Qm]


M m=0
h[m]x[Q(r m)]
w [n ] = =
M m=0 hQ [Qm]v [n
Qm]
If Q n, then v [n Qm] = 0 m so w[n] = 0 = y [n]


M m=0
h[m]x[Q(r m)]
w [n ] = =
M m=0 hQ [Qm]v [n
Qm]
If Q n, then v [n Qm] = 0 m so w[n] = 0 = y [n] If Q | n = Qr , then w[Qr ] =

M m=0
h[m]v [Qr Qm]



M m=0
h[m]x[Q(r m)]
w [n ] = =
M m=0 hQ [Qm]v [n
Qm]
=
h[m]v [Qr Qm] M m=0 h[m]x[r m] = u[r ]

M m=0


M m=0
h[m]x[Q(r m)]
w [n ] = =
M m=0 hQ [Qm]v [n
Qm]
=
h[m]v [Qr Qm] M m=0 h[m]x[r m] = u[r ] = y [Qr ]

M m=0
Upsampled z-transform
V (z ) =
n v [ n ] z n
V (z ) =
n v [ n ] z = n
n n u [ n:K |n K ]z
V (z ) = =
n v [ n ] z = n Km u [ m ] z m
n n u [ n:K |n K ]z
V (z ) = =
n v [ n ] z = n
n n u [ n:K |n K ]z
Km K u [ m ] z = U ( z ) m
V (z ) = =
n v [ n ] z = n
n n u [ n:K |n K ]z
Km K u [ m ] z = U ( z ) m
Frequency Spectrum:
V (ej ) = U (ejK )
V (z ) = =
n v [ n ] z = n
n n u [ n:K |n K ]z
Km K u [ m ] z = U ( z ) m
Frequency Spectrum:
V (ej ) = U (ejK )
Frequency spectrum is horizontally shrunk and replicated K times.
V (z ) = =
n v [ n ] z = n
n n u [ n:K |n K ]z
Km K u [ m ] z = U ( z ) m
Frequency Spectrum:
V (ej ) = U (ejK )
Example: Asymmetric real spectrum ( complex signal)
0.5
-2
V (z ) = =
n v [ n ] z = n
n n u [ n:K |n K ]z
Km K u [ m ] z = U ( z ) m
Frequency Spectrum:
V (ej ) = U (ejK )
Example: Asymmetric real spectrum ( complex signal) K = 3: three images of the original spectrum in all.
0.5
0.5
-2
-2
V (z ) = =
n v [ n ] z = n
n n u [ n:K |n K ]z
Km K u [ m ] z = U ( z ) m
Frequency Spectrum:
V (ej ) = U (ejK )
Example: Asymmetric real spectrum ( complex signal) K = 3: three images of the original spectrum in all. Energy unchanged: 21
1
U (ej ) d =
1
1 2
V (ej ) d
0.5
0.5
-2
-2
V (z ) = =
n v [ n ] z = n
n n u [ n:K |n K ]z
Km K u [ m ] z = U ( z ) m
Frequency Spectrum:
V (ej ) = U (ejK )
Frequency spectrum is horizontally shrunk and replicated K times. Upsampling normally followed by a LP lter to remove images. Example: Asymmetric real spectrum ( complex signal) K = 3: three images of the original spectrum in all. Energy unchanged: 21
1
U (ej ) d =
1
1 2
V (ej ) d
0.5
0.5
-2
-2
Downsampled z-transform
Dene cK [n] = K |n [n]
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n 0 Kn
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
XK (z ) =
x K [n ]z n
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
n
XK (z ) =
x K [n ]z
1 K
K 1 k=0
j 2kn K
x [n ]z n
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
n
XK (z ) = =
1 K
x K [n ]z
K 1 k=0
1 K
x [n ] e
K 1 j 2kn K x [n ]z n n k=0 e n j 2k
K
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
n
XK (z ) = =
1 K
x K [n ]z
K 1 k=0
1 K
x [n ]
K 1 j 2kn K x [n ]z n n k=0 e n j 2k j 2k K 1 1 K e z = K k=0 X (e K z )
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
n
XK (z ) = =
1 K
x K [n ]z
K 1 k=0
1 K
x [n ]
From previous slide:
XK (z ) = Y (z K )
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
n
XK (z ) = =
1 K
x K [n ]z
K 1 k=0
1 K
x [n ]
XK (z ) = Y (z K ) 1 Y (z ) = XK (z K )
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
n
XK (z ) = =
1 K
x K [n ]z
K 1 k=0
1 K
x [n ]
XK (z ) = Y (z K ) 1 Y (z ) = XK (z K ) =
1 K
K 1 k=0
X (e
j 2k K
zK)
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
n
XK (z ) = =
1 K
x K [n ]z
K 1 k=0
1 K
x [n ]
XK (z ) = Y (z K ) 1 Y (z ) = XK (z K ) =
Frequency Spectrum:
1 K
K 1 k=0
j ( 2k) K
X (e )
j 2k K
zK)
Y (ej ) =
1 K
K 1 k=0
X (e
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
n
XK (z ) = =
1 K
x K [n ]z
K 1 k=0
1 K
x [n ]
XK (z ) = Y (z K ) 1 Y (z ) = XK (z K ) =
Frequency Spectrum:
1 K
K 1 k=0
j ( 2k) K j
X (e )
2
j 2k K
zK)
Y (ej ) = =
1 K 1 K
K 1 k=0
X (e
X (e K ) + X (e K K ) + X (e K K ) +
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
n
XK (z ) = =
1 K
x K [n ]z
K 1 k=0
1 K
x [n ]
XK (z ) = Y (z K ) 1 Y (z ) = XK (z K ) =
Frequency Spectrum:
1 K
K 1 k=0
j ( 2k) K j
X (e )
2
j 2k K
zK)
Y (ej ) = =
X (e K ) + X (e K K ) + X (e K K ) + Horizontally expanded by a factor of K and aliased K times.
1 K 1 K
K 1 k=0
X (e
1 Dene cK [n] = K |n [n] = K
K 1 k=0
j 2kn K
Now dene xK [n] =
x [n ] K | n = c K [n ]x [n ] 0 Kn
n
XK (z ) = =
1 K
x K [n ]z
K 1 k=0
1 K
x [n ]
XK (z ) = Y (z K ) 1 Y (z ) = XK (z K ) =
Frequency Spectrum:
1 K
K 1 k=0
j ( 2k) K j
X (e )
2
j 2k K
zK)
Y (ej ) = =
X (e K ) + X (e K K ) + X (e K K ) + Horizontally expanded by a factor of K and aliased K times.

Downsampling is normally preceded by a LP lter to prevent aliasing.
DSP and Digital Filters (2013-2852) Multirate: 11 8 / 13
1 K 1 K
K 1 k=0
X (e
Downsampled Spectrum
Y (e ) =
1 K
K 1 k=0
X (e
j ( 2k) K
Y (e ) =
Example 1:
1 K
K 1 k=0
X (e
j ( 2k) K
K=3
Not quite limited to K
0.5
-2
Y (e ) =
Example 1:
1 K
K 1 k=0
X (e
j ( 2k) K
K=3
Not quite limited to K Shaded region shows aliasing
0.5
0.5
-2
-2
Y (e ) =
Example 1:
1 K
K 1 k=0
X (e
j ( 2k) K
K=3
0.5
0.5
-2
-2
Energy decreases: 21
Y (ej ) d
1 2K
X (ej ) d
Y (e ) =
Example 1:
1 K
K 1 k=0
X (e
j ( 2k) K
K=3
0.5
0.5
-2
-2
Energy decreases: 21 Example 2:
Y (ej ) d
1
1 2K
X (ej ) d
K=3
| | 2 K Limited to K
0.5
-2
Y (e ) =
Example 1:
1 K
K 1 k=0
X (e
j ( 2k) K
K=3
0.5
0.5
-2
-2
Y (ej ) d
1
1 2K
X (ej ) d
1
K=3
| | 2 K Limited to K No aliasing:
0.5
0.5
-2
-2
Y (e ) =
Example 1:
1 K
K 1 k=0
X (e
j ( 2k) K
K=3
0.5
0.5
-2
-2
Y (ej ) d
1
1 2K
X (ej ) d
1
K=3
0.5
0.5
-2
-2
Criterion for no aliasing:

Normally quoted Nyquist criterion is | | K
Y (e ) =
Example 1:
1 K
K 1 k=0
X (e
j ( 2k) K
K=3
0.5
0.5
-2
-2
Y (ej ) d
1
1 2K
X (ej ) d
1
K=3
0.5
0.5
-2
-2
Criterion for no aliasing:

Normally quoted Nyquist criterion is | | K | | (r + 1) K Actually OK if spectral energy is restricted to r K
Perfect Reconstruction
x [n ]
x [n ] u [m ]
x [n ] u [m ] p [n ]
x [n ] u [m ] p [n ] v [m ]
x [n ] u [m ] p [n ] v [m ] q [n ]
x [n ] u [m ] p [n ] v [m ] q [n ] w [m ]
x [n ] u [m ] p [n ] v [m ] q [n ] w [m ] y [n ]
x [n ] u [m ] p [n ] v [m ] q [n ] w [m ] y [n ]
Input sequence x[n] is split into three streams:
u[m] = x[3m], v [m] = x[3m 1], w[m] = x[3m 2]
x [n ] u [m ] p [n ] v [m ] q [n ] w [m ] y [n ]
u[m] = x[3m], v [m] = x[3m 1], w[m] = x[3m 2]

Following upsampling, the streams are aligned by the delays and then added to give:
y [n] = x[n 2]
x [n ] u [m ] p [n ] v [m ] q [n ] w [m ] y [n ]
u[m] = x[3m], v [m] = x[3m 1], w[m] = x[3m 2]

Following upsampling, the streams are aligned by the delays and then added to give:
y [n] = x[n 2]
Perfect Reconstruction: output is a delayed scaled replica of the input
Commutators
x [n ] u [m ] v [m ] w [m ]
y [n ]
Commutators
x [n ] u [m ] v [m ] w [m ]
y [n ]
The combination of delays and downsamplers can be regarded as a commutator that distributes values in sequence to u, w and v .
Commutators
x [n ] u [m ] v [m ] w [m ]
y [n ]
The combination of delays and downsamplers can be regarded as a commutator that distributes values in sequence to u, w and v . 2 1 Fractional delays, z 3 and z 3 are needed to synchronize the streams.
Commutators
x [n ] u [m ] v [m ] w [m ]
y [n ]
The combination of delays and downsamplers can be regarded as a commutator that distributes values in sequence to u, w and v . 2 1 Fractional delays, z 3 and z 3 are needed to synchronize the streams. The output commutator takes values from the streams in sequence.
Commutators
x [n ] u [m ] v [m ] w [m ] v [m + 1 3] w [m + 2 3] y [n ]
The combination of delays and downsamplers can be regarded as a commutator that distributes values in sequence to u, w and v . 2 1 Fractional delays, z 3 and z 3 are needed to synchronize the streams. The output commutator takes values from the streams in sequence. For clarity, we omit the fractional delays and regard each terminal, , as holding its value until needed.
Commutators
x [n ] u [m ] v [m ] w [m ] v [m + 1 3] w [m + 2 3] y [n ]
The combination of delays and downsamplers can be regarded as a commutator that distributes values in sequence to u, w and v . 2 1 Fractional delays, z 3 and z 3 are needed to synchronize the streams. The output commutator takes values from the streams in sequence. For clarity, we omit the fractional delays and regard each terminal, , as holding its value until needed. Initial commutator position has zero delay.
Commutators
x [n ] u [m ] v [m ] w [m ] v [m + 1 3] w [m + 2 3] y [n ]
The combination of delays and downsamplers can be regarded as a commutator that distributes values in sequence to u, w and v . 2 1 Fractional delays, z 3 and z 3 are needed to synchronize the streams. The output commutator takes values from the streams in sequence. For clarity, we omit the fractional delays and regard each terminal, , as holding its value until needed. Initial commutator position has zero delay.
The commutator direction is against the direction of the z 1 delays.

DSP and Digital Filters (2013-2852) Multirate: 11 11 / 13
Summary
Multirate Building Blocks 1:K Upsample: X (z ) X (z K ) Invertible, Inserts K 1 zeros between samples
Shrinks and replicates spectrum Follow by LP lter to remove images
Summary
Downsample: X (z )
Destroys information, keeps every K th sample Expands and aliasses spectrum Precede by LP lter to prevent aliases
K :1 1 K
j 2k 1 K 1 K K z ) k=0 X (e
Summary
Downsample: X (z )
K :1 1 K
j 2k 1 K 1 K K z ) k=0 X (e
Equivalences Noble Identities: H (z ) H (z K ) Interchange P : 1 and 1 : Q iff P and Q coprime
Summary
Downsample: X (z )
K :1 1 K
j 2k 1 K 1 K K z ) k=0 X (e
Equivalences Noble Identities: H (z ) H (z K ) Interchange P : 1 and 1 : Q iff P and Q coprime Commutators Combine delays and down/up sampling
Summary
Downsample: X (z )
K :1 1 K
j 2k 1 K 1 K K z ) k=0 X (e
Equivalences Noble Identities: H (z ) H (z K ) Interchange P : 1 and 1 : Q iff P and Q coprime Commutators Combine delays and down/up sampling
For further details see Mitra: 13.
MATLAB routines
resample
change sampling rate

Multirate

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Multirate

Uploaded by

Copyright:

Available Formats

11: Multirate Systems

11: Multirate Systems

DSP and Digital Filters (2013-2852)

Multirate systems include more than one sample rate

DSP and Digital Filters (2013-2852)

May need to change the sample rate

DSP and Digital Filters (2013-2852)

May need to change the sample rate

Can relax analog or digital lter requirements

DSP and Digital Filters (2013-2852)

May need to change the sample rate

Can relax analog or digital lter requirements

DSP and Digital Filters (2013-2852)

DSP and Digital Filters (2013-2852)

DSP and Digital Filters (2013-2852)

Example: Downsample by 3 then upsample by 4

DSP and Digital Filters (2013-2852)

Example: Downsample by 3 then upsample by 4

DSP and Digital Filters (2013-2852)

Example: Downsample by 3 then upsample by 4

DSP and Digital Filters (2013-2852)

Example: Downsample by 3 then upsample by 4

We use different index variables (n, m, r ) for different sample rates

DSP and Digital Filters (2013-2852)

Example: Downsample by 3 then upsample by 4

DSP and Digital Filters (2013-2852)

Example: Downsample by 3 then upsample by 4

DSP and Digital Filters (2013-2852)

Successive downsamplers or upsamplers can be combined

DSP and Digital Filters (2013-2852)

Successive downsamplers or upsamplers can be combined Upsampling can be exactly inverted

DSP and Digital Filters (2013-2852)

DSP and Digital Filters (2013-2852)

DSP and Digital Filters (2013-2852)

DSP and Digital Filters (2013-2852)

DSP and Digital Filters (2013-2852)

DSP and Digital Filters (2013-2852)

Resamplers commute with addition and multiplication

DSP and Digital Filters (2013-2852)

Resamplers commute with addition and multiplication

Delays must be multiplied by the resampling ratio

DSP and Digital Filters (2013-2852)

Resamplers commute with addition and multiplication

DSP and Digital Filters (2013-2852)

Resamplers commute with addition and multiplication

Example: H (z ) = h[0] + h[1]z 1 + h[2]z 2 +

H (z 3 ) = h[0] + h[1]z 3 + h[2]z 6 +

DSP and Digital Filters (2013-2852)

Resamplers commute with addition and multiplication

Corrollary Example: H (z ) = h[0] + h[1]z 1 + h[2]z 2 +

H (z 3 ) = h[0] + h[1]z 3 + h[2]z 6 +

DSP and Digital Filters (2013-2852)

Noble Identities Proof

Dene hQ [n] to be the impulse response of H (z Q ).

DSP and Digital Filters (2013-2852)

Noble Identities Proof

DSP and Digital Filters (2013-2852)

Noble Identities Proof

DSP and Digital Filters (2013-2852)

Noble Identities Proof

DSP and Digital Filters (2013-2852)

Noble Identities Proof