Vous êtes sur la page 1sur 69

F.

Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Slide Set
Data Converters

Oversampling And Low Order Modulators

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Summary
Introduction
Noise shaping
First Order Modulator
Second Order Modulator
Circuit Design Issues
Architectural Design Issues

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Introduction
With oversampling the signal band occupies a small fraction of the Nyquist
interval.
The use an ideal digital filter possibly reduces the quantization noise power
in the signal band
2
2 2f
V

1
ref
2 =
Vn,B
B =

12
fs
12 22n OSR

(1)

EN OB = n + 0.5 log2 (OSR)

(2)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Rejection of the out-of-band noise


2fN
1
Sampled-data
Analog

2fN
2

2f'N
3

Oversampled
A/D

Digital
Filter
n-bit

Decimator
n1-bit

n1-bit

DfR
fB

fN

fB

fN

fB

fN

f'N

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Delta and Sigma-Delta Modulation


Clock

Digital
Output

n-bit
ADC

Amplitude

Analog
Input

n-bit
DAC

(a)

maximum
slopes
time
(b)

Delta modulators for increasing the effectiveness of the PCM transmission.


But, ... high pass response ...

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

From Delta to Sigma-Delta Modulation


Clock

Analog
Input
d
dt

n-bit
ADC

Clock
Digital
Output

Analog
Input
+

n-bit
ADC

Digital
Output

n-bit
DAC

(a)

n-bit
DAC

(b)

Integration (sigma) of the difference (delta) gives the name Sigma-Delta.


The Sigma-Delta become popular for the shaping of the quantization noise.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Noise shaping
The key advantage of the architecture is the noise shaping.
Differently than a normal oversampled scheme
2fN
Sampled-data
Analog

2fN

Noise
Shaping A/D

Digital
Filter
n-bit

2f'N

Decimator
n1-bit

n1-bit

DfR
fB

fN

fB

fN

fB

fN

f'N

Is the reduction of noise at given frequencies.


If the signal band is at low frequency it is desirable to reduce the noise at
low frequency high pass shaping of the noise.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

How to obtain noise shaping?


eQ
Y

X +
-

A(z)

ADC

B(z)

DAC

YD
Encoder

(a)

X
+

Y
-

A(z)

B(z)

(b)

Place the quantizer in a feedback loop.


The system has two inputs (signal and noise) and one output.
Goal is to have different suitable transfer functions (STF = signal TF, and
NTF = Signal TF).

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

STF and NTF estimation


[X Y B(z)] A(z) + Q = Y,

(3)

Study in the z-domain (it can be also done in the time-domain)

Q
X A(z)
+
.
Y =
1 + A(z)B(z)
1 + A(z)B(z)

(4)

Y = X S(z) + Q N (z)

(5)

B(z) can simply be equal to 1; A(z) should be integration-type.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

First Order Modulator


Analog
Input
+
S
X
- Y

eQ

Digital
Output
z-1

n-bit
ADC

1-z-1

YD

X
-

z-1
1-z-1

Digital
Output

Y
S

Encoder

YD

Quantized
Output
n-bit
DAC

(a)

(b)

The linear model of the modulator replaces the non-linear quantization by


the linear injection of the quantization noise.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

10

STF and NTF of the First Order Modulator


z 1
H(z) =
1 z 1

(6)

z 1
+ Q(z)
Y (z) = {X(z) Y (z)}
1
1z

(7)

Y (z) = X(z) z 1 + Q(z)(1 z 1)

(8)

Y (z) = X ST F (z) + Q(z) N T F (z).

(9)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

11

The STF is a simple delay (more than what we desire).


The NTF is (1 z 1), that using the z s transformation becomes

ejT /2 ejT /2
jT
jT
/2
N T F () = 1 e
= 2je
2j

N T F () = 2jejT /2sin(T /2)

that is a low pass response.

(10)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

12

The power of the shaped noise in a small band B is

2
Vn2 = vn,Q

Z f
B
0

2
Vn2 = Vn,Q

4 2 3 2
2
2
fB T
4 sin (f T )df ' vn,Q
3

2
3

"

fB
fs/2

#3

2
= Vn,Q

OSR3.

(11)

(12)

Assume to use a DAC that generates the quantized intervals

VDAC (i) = i

Vref
k

i = 0 k.

(13)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

13

The power of the quantization noise and the one of a sine wave are
2
Vref

2 =
Vn,Q
;
2
k 12

2 =
Vsine

2
Vref

(14)

therefore,
12 2 3
SN R,1 =
k 2 OSR3.
8

(15)

assuming n0 = log2k
SN R,1|dB = 6.02 n0 + 1.78 5.17 + 9.03 log2(OSR)

(16)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Table 6.1 - SNR improvement with Multi-level Quantizers


ADC
Thresholds
1
2
3
4
5
6
7
8
15

DAC
Levels
2
3
4
5
6
7
8
9
16

nQ
1
2

3
4

n0
extra bits
0
1
1.58
2
2.32
2.58
2.81
3
3.91

SN R
[dB]
0
6.02
9.54
12.04
13.97
15.56
16.84
18.03
23.52

14

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Example 6.1
Behavioral simulation of a first order modulator
uT

kat

i
i

ap

15

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Obtained output spectrum (vertical axis is in dB)

16

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

17

Intuitive Views
The output of an integrator is bounded if its input is, on average, zero
the output equals in average the input.
The factor 2 in the NTF indicates a worsening of the noise performances
at high frequency.
Oversampling can be viewed as a staircase with small steps between big
steps between (resolution increases but linearity remains the same).

D + ed,i+1

D=V

FS

(a)

n
/2

D + e d,i

(b)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Warning!

Therefore, remember the following ...

The feedback of a modulator does not relax the DAC


linearity.
Remind that the
method greatly reduces the
number of DAC levels but not
their accuracy requirement!

18

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Use of 1-bit Quantization


C2

C1

Vin

F2

F1
F2

F1

Digital
Output

-++

DAC
+Vref
-Vref

(a)
1.5

0.5

-0.5

-1

-1.5

50

100

150

(b)

200

250

300

19

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

20

Second Order Modulator


The use of one ontegrator around the loop is beneficial; it may be that using
two is better ...
Two integrators around a loop can be unstable; a dumping is necessary!

Analog
Input
+

Clock

One or the other?

+
P

S
-

n-bit
ADC

Digital
Output

eQ

YD
+

X
Quantized
Output
Y = R+eQ
(a)

z-1

1-z-1

1-z

n-bit
DAC
(b)

-1

Quantized
Output
S

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

21

The dumping embracing the quantizer is effective

P R
R=
s

P Y
= Y Q
s

P
Y = R + Q =
+ Q
1 + s

Y =

s Q
P
+
1 + s
1 + s

(17)
(18)

The study of the second order modulator in the z domain yields:

1
z 1
Y (z)
+ Q(z) = Y (z)
[X(z) Y (z)]
1
1
1z
1z


(19)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Y (z) = X(z) z 1 + Q(z)(1 z 1)2

22

(20)

just a delay for the signal and a more effective (second order) shaping for
the quantization noise.

N T F () = (1 ejT )2 = 4ejT {sin(T /2)}2

(21)

that, in a small band B gives

2
Vn2 = vn,Q

Z f
B
0

16 4 5 4
4
2
fB T
16 sin (f T )df ' vn,Q
5

(22)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

2
Vn2 = Vn,Q

4
5

"

fB
fs/2

#5

2
= Vn,Q

OSR5

23

(23)

Again, with k quantization intervals and a full range sine wave at the input.

12 2 5
SN R2 =
k 4 OSR5
8

(24)

that gives (with n0 = log2k)


SN R2 |dB = 6.02n0 + 1.78 12.9 + 15.05 log2(OSR)

(25)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Circuit Design Issues


The most important limits of the basic blocks used are:
Offset of the op-amp (or OTA).
Finite op-amp gain.
Finite op-amp bandwidth.
Finite op-amp slew-rate.
Non-ideal operation of the ADC.
Non-ideal operation of the DAC.

24

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

25

The offset of the first integrator is added to the input signal and gives rise
to an equal offset at the digital output.
The offset of the second integrator is referred to the input of the modulator
by dividing it by the transfer function of the first block (an integrator).
The high-pass transfer function cancels out the effect.
The offset of the DAC is added to the input and causes, similar to the offset
of the first integrator.
The offset of the ADC is referred to the input by dividing it by the transfer
function of one or more integrators and does not limit the dc operation of
the modulator. The feature enables the flexibility of positioning the ADC
thresholds around the more convenient voltage level.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

26

Finite Op-Amp Gain


The basic block of a sigma delta modulator is the integrator of a difference.
Use the switched capacitor technique we have

C1

V1

F2
F1

C2

-Vout/A0

F2

F1

A0
+

Vout

F1
F2

V2
nT

nT+T

With OTA (or op-amp) finite gain the virtual ground is not at ground.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

1
C2Vout(nT + T ) 1 +
A0

1
= C2Vout(nT ) 1 +
+
A0

"

+C1 V1(nT ) V2(nT + T )


1
Vout(z 1) 1 +
A0

Vout(nT )
A0

"

(26)

C1
zVout
=
V1 zV2
C2
A0

Vout
C1
A0
z 1
=
V1 z 1 V2
C2 A0 + 1 + C1/C2 1 (1+A0)C2 z 1
C +C +A C
"

27

(27)

0 2

(28)

Chapter 6

OVERSAMPLING AND LOW ORDER SD

N T F ' (1 zp1 z 1)(1 zp2 z 1)


A +1
N T F = 1 z 1 0
A0 + 2

20
40

80
100

A =98
0

A =998
0

120
140
160
4
10

A0=9998
3

10

10
Normalized Frequency

10

(29)
(30)

60

28

!2

NTF of a Second Order with OpAmp Finite Gain

20

NTF [dB]

F. Maloberti
DATA CONVERTERS
Springer
2007

10

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

29

The finite gain affects the SNR only if the flat region contributes with a
significant fraction of noise power. The corner frequency is
espT =

A0 + 1
A0 + 2

1
fs
ln 1
fc =
2
A0 + 2


'

(31)
fs
2(A0 + 2)

(32)

The finite gain does not affect the SNR if fB >> fc


(A0 + 2) >> OSR

(33)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Example 6.2

30

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Comparison of the SNR with Gain 100 and Gain 100,000

A0 =100K
A0 =100

31

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

32

Finite Op-Amp Bandwidth


If the effect of the non-dominant poles is negligible, the step response of
the integrator is an exponential that, for large finite gain A0, is
Vout(t + nT ) = Vout(nT ) + VstepU (nT )(1 et/d )

(34)

b = VstepeTs/(2d)

(35)
eQ

Vin
+

S
-

Vout(z) =

1-eb,1

-1

1-z

(1-eb,2)z
1-z

-1

-1

Vout
S

Vinz 1(1 b,1)(1 b,2) + Q(1 z 1)2


1 z 1(

b,1 + 2b,2 b,1 b,2

) + z 2

b,2

(36)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

33

Finite Op-Amp Bandwidth


The finite op-amp bandwidth changes the signal and the noise transfer
function because if the denominator with two poles.
The effect is not very relevant when considered alone. The limit is more
significant when it is considered together the finite gain and the limited
slew-rate.
The study of the limits is conveniently done with behavioral simulators and
suitable behavioral models.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

34

Finite Op-Amp Slew-Rate


tsett1

tsett2

C
DV
F2

F1

A
F1

F2

Vin
tslew

DVin(0)
(0-)

T/2

T/2

F1

F2

time

tslew =

Vout
.
SR

(37)

V = SR ;

(38)

SR = V e(T /2tslew )/

(39)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

35

Example
Determine the minimum slew-rate required for the op-amps used in a singlebit second-order . Study the combined effect of slew-rate and finite
bandwidth.
1 V fs = 50 MHz, Vin = 6 dBF S .

The maximum output changes of first and second integrator is 0.749 V and
3.21 V; therefore:

Vout,1
Vout,2
SR1 >
= 74.9 V /s; SR2 >
= 321 V /s
T /2
T /2

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Results of simulations
0

PSD with SR of the First OpAmp 200 V/s


SNR = 71.9dB, OSR=64
Rbit= 11.65 bits,OSR=64

20
40

40

60

60

80

100

120

120
5
10

6
10
Frequency [Hz]
(a)

7
10

SNR = 70.3dB, OSR=64

Rbit= 11.39 bits,OSR=64

80

100

140

PSD with SR of the First OpAmp 73 V/ s

20

PSD [dB]

PSD [dB]

140

5
10

6
10
Frequency [Hz]
(b)

7
10

36

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

37

ADC Non-ideal Operation


The static and dynamic limitations of a real ADC degrade the modulator
performances.

ADCout = Vin,ADC + Q + ADC ,

(40)

Shaping of the modulator acts on both Q and ADC . Therefore,


ADC < Q.
DNL and INL of less than 1 LSB is easily verified: the number of thresholds
is small and the dynamic range is large.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

38

DAC Non-ideal Operation


The DAC errors are not shaped by the NTF : they are added to the input
and transferred to the output through the STF.
A switched capacitor DACs divides a total capacitance into parts. One limit
associated is the kT/C noise.
Since the kT/C noise is white oversampling limits its power in the band of
interest.

2
V
kT
ref
2
<
vn,kT
=
/C
OSR Cin
8 22n

(41)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

39

Architectural Design Issues - Integrator Dynamic Range


The integrator voltage swing depends on signal amplitude and quantization
noise.
The dynamic range of operational amplifiers and quantizer must accommodate both the signal and the noise.
Integrator outputs that exceed the dynamic range are clipped thus causing
a loss of feedback control.
C2
C1

Vin

F2
F1

es,1

Vvg

F2

F1

Vout

Sat

S
-

es,2
+

1
1-z

-1

z-1
1-z-1

es,Q

Vsat

(a)

eQ

(b)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

40

The input capacitance is charged to Qres, the virtual ground starts moving
and when the left terminal of C1 reaches zero the fraction QresC2/(C1 +
C2) is transferred into C2 leaving a fraction QresC1/(C1 +C2) in the input
capacitance.
Qres
s =
(42)
C1 + C2
s,1 is transferred to the output multiplied by z 1; s,2 is shaped by a firstorder high-pass transfer function; s,Q is shaped by the NTF.
Y = Xz 1 + s,1z 1 + s,2(1 z 1) + (Q + s,Q)(1 z 1)2 (43)
Vn2 =

2
Vn,1

OSR

2
+ Vn,2

"

2 +
+
V
n,Q
3 OSR3
12

4
5 OSR5

(44)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

41

Example
Loss of resolution caused by the hard saturation of the op-amp outputs and
the quantizer.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

20

20

40

40

60

60

80

80

PSD [dB]

PSD [dB]

PSD with Saturation of the First Integrator at 1.85

PSD with Ideal Integrator

100
120
140

120
140

160

160
SNR = 67.6dB @ OSR=64
Rbit = 10.94 bits @ OSR=64

180
200

100

10

10
10
Frequency [Hz]

10

200

20

20

40

40

60

60

80

80

PSD [dB]

PSD [dB]

PSD with Saturation of the Second Integrator at 2.5 V

100
120
140

10

10
10
Frequency [Hz]

10

PSD with Saturation in Both Integrators

100
120
140

160

160
SNR = 67.5dB @ OSR=64
Rbit = 10.91 bits @ OSR=64

180
200

SNR = 64.4dB @ OSR=64


Rbit = 10.40 bits @ OSR=64

180

10

10
10
Frequency [Hz]

10

SNR = 60.2dB @ OSR=64


Rbit = 9.71 bits @ OSR=64

180
6

200

10

10
10
Frequency [Hz]

10

42

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD


First Integrator Output

90

Second Integrator Output

80

80

70

70

60
Occurrences

60

50

50

1000

2000

3000

4000

5000

6000

Second Integrator Output

7000 8000

40

40

30

30

20

20

10

10

0
1.5 1 0.5 0 0. 5 1
Voltage [V ]

1
1000
0

2000

3000

4000

5000

6000

7000 8000

PSD with Saturation in Both Integrators

120
100

50

80

40

100
120

30

140

20

160
3

10

10
10
Frequency [Hz]

10

0
1

80
60
40
20

10

180
200

Second Integrator Output

60

60

1.5

140

Occurrences

40

First Integrator Output

70

SNR = 77.3dB @ OSR=64


Rbit = 12.54 bits @ OSR=64

20

80

0
1.5 1 0.5 0 0.5 1
Voltage [V ]

1.5

Occurrences

PSD [dB]

Amplitude

First Integrator Output

Occurrences

Amplitude

43

0.5
0
0. 5
Voltage [V ]

0.5
0
0.5
Voltage [V]

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

44

Dynamic Ranges Optimization


A suitable dynamic range in the op-amps (or OTAs) is essential for preserving the SNR and avoiding harmonic distortion.
The critical op-amp is the one used in first integrator (no shaping).
However, even the second integrator and the quantizer are important.
Carefully estimate the voltage swings and keep them within limits: small to
avoid saturation; not so low to distinguish the electronic noise.
The above points affect the choice of the reference value.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Scaling as used in switched capacitor filters


C1
Vout,1

C1

Vin

F2
F1

F2

F1

F1

VDAC

C3
F1
F2

F2

(a)
bC1
Vout,1/b

C1

Vin

F2
F1

F2

F1

VDAC

1
1-z-1

F1

(b)
+

1-z-1

(c)
+

S
-

1/b1
1-z-1

b1

S
-

(d)

F1
F2

F2

eQ
z-1

bC3

eQ
z-1/b 2
1-z-1

Y
b2

45

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

46

Use a different architecture (two delayed integrators).


Chose A and B so that the STF and the NTF are good and the (first) op-amp swing is
reduces.
eQ
X +

A z-1
1-z

-1

B z-1
1-z

-1

(a)
X +
-

0.5 z-1
1-z

-1

+
-

2 z-1

1-z

ADC

-1

Y
(b)

DAC

YD

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

47

study with the linear model.



Az 1
Bz 1
(X Y )
Y
+ Q = Y,
1 z 1
1 z 1

(45)

X ABz 2 + Q (1 z 1 )2
Y =
.
1
2
1 (2 B)z + (1 B + AB)z

(46)

with B=2 and A = 1/2 the denominator is 1


Y = Xz 2 + Q (1 z 1 )2

(47)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Example 6.5
Histograms of the output voltages of the two integrators (A=1/2).

Occurrences

100

100
50
0
2

150
Occurrences

First Integrator Output

0
Voltage [V]

First Integrator Output

50
0
2

0
Voltage [V]

B=2

60
40
20

100

100

Second Integrator Output

80

0
4

Occurrences

Occurrences

150

0
Voltage [V]

Second Integrator Output


B=1/2

80
60
40
20
0
1

0.5

0
Voltage [V]

0.5

48

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

49

Dynamic range optimization with multi-bit quantizer


Use of feed-forward path.

S
-

1/2 z-1
1-z-1

+
P

S
-

2 z-1

Digital
Output

threshold

1-z-1

k+1
Y

Outputs

Intuitive view: remember that the input of an integrator (the second) is in average zero.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

50

(X Y )z 1
z 1 (1 + z 1 )
Q z 1 (1 z 1 )
P =
=X
+
.
2(1 z 1 )
2
2

(48)

 2

1
1
Y = X z + 2 z (1 z ) + Q (1 z 1 )2

(49)

(X Y )z 1
z 1 (1 z 1 )
Q z 1 (1 z 1 )
P =
=X
+
.
2(1 z 1 )
2
2

(50)

ST F = z 2 + 2(1 z 1 );

(51)

Notice: The STF is slightly changed (the high pass term can be possibly removed in the
digital domain).

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Example 6.6
Hystogram of the first op-amp output with the feedforward branch (3-bit quantization)

51

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

52

Spectrum of the output voltage of the first op-amp

Notice the small signal tone (high pass filter) and the first order shaped quantization noise.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

53

Sampled-data Circuit Implementation


2C1
C1

Vin

F2

F2

2C2

F1

F2
F1

C2

F2

F2

F1

F1

ADC

DAC

(a)
C2

C1
F1

C1

Vin

F2

F1
F1

F2

F2

C2
F2

F1
F2

_
ADC

C2
F2
F1

F1
F2
DAC

(b)

The DAC can be realized with a separate SC structure or by sharing the input capacitor.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

54

Noise Analysis
The noise generators are due to the on-resistance of the switches and the noise of the
op-amp (described by an input referred noise generator).
2CU
Vin F2

CU

F1
Ron

F1
F2
Ron

CU
F2

F2

2CU

F1

F1

vn,A1

VDIG

Ron

F2
Ron

+
2
vn,A2

VDAC

VRef
-VRef

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

55

There are two different circuit configurations during the two phases
It is assumed that at the end of the phase (before sampling) the voltages across capacitors
reach the stationarity.
PHASE 2

2CU
CU
2
vn,R

A1

2CU

+
2
vn,A1
CL

Ron

2CU
CU

2
vn,R

2
vn,R

A1

Ron Ron

(a)

CL

2
vn,R

2CU

A2
_

vx

vout

+
2
vn,A2

vout Cf
gm vn,A

1/gm

CU

2
CL Ron
vn,A1

2
2CU vn,R

gm vn,A1

PHASE 1

Cin

CL

2
vn,R

vx

gm2 v2x
CL

Ron

(b)

Ron

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

56

The input referred noise is proportional to 1/gm


2
vn,A1
= A1

4kT
;
gm,A1

2
vn,A2
= A2

4kT
gm,A2

(52)

Estimate the transfer functions between noise input and output


HA1,in2 =

vn,Cin2
1
=
vn,A1
1 + s(0 + 0 2CU /CL + R ) + s2 0 R

(53)

after sampling
Vn,A1,in2 = A1

kT
CL

(54)

another transfer function


HR,in2 =

1 + s0
1 + s(0 + 0 2CU /CL + R ) + s2 0 R

(55)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Vn,R,in2 =

vout (Ron +
vx =

Ron +

kT
2CU

1
1
) + vn,R sC
sCin
f
1
1
+ sCin
sCf

gm (vn,A vx ) = vout sCL + (vout vx )sCf

(56)

(57)
(58)

CL (vx vout )
Cin

(59)

CL
vn,A + (1 + s0 )vn,R
Cf 1 + (0 / + 0 Cin /CL + R )s + 0 R s2

(60)

vCin =

vCin =

57

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

2
vn,2

2
vn,out



kT
2
vn,1
= 2Ts 2
+ A1 kT /CL
CU


kT
= 2Ts
+ A1 kT /CL + A2 kT /CL
CU


2kT
2
vn,3
= 2Ts
+ A2 kT /CL
CADC

2
vn,1
|z 2 |2

2
vn,2
2z 1 (1

2


2
1 2 2

) + vn,3 (1 z )

58

(61)

(62)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

Use of a transistor level simulator

on

HR,in2

HA1,in2

59

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

60

Summing up. the various noise generators cause, after the sampling, a noise voltage
which spectrum is white (over the Nyquist interval).
It is necessary to distinguish between noise contribution at the end of phase 1 and at the
end of phase 2.
All the noise terms are uncorrelated and must be superposed quadratically.
Table 6.2 - Noise Power Terms of the Second Order Modulator
Phase

Source

2
2
1
1

4kT Ron
Ai 4kT /gm
4kT Ron
Ai 4kT /gm

2
Vn1
[V 2 ]
kT /CU

kT /CU
A1 kT /CL

2
Vn2
[V 2 ]
kT /(2CU )
A1 kT /CL
kT /(2CU )
A2 kT /CL

3
Vn1
[V 2 ]
kT /CADC
A2 kT /CL

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

61

Noise at the output of the modulator


After the sampling the noise is transferred to the output with a transfer function that depends on the injection point.
vn,1
+

S
-

vn,2
1/2 z-1
1-z

DAC1

-1

vn,3
2z-1
1-z

-1

Y
ADC

DAC2

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

62

Quantization Error and Dithering


modulators are good for busy signal (approximating the quantization error with a
noise is acceptable).
Constant or very slow inputs can give rise to repetitive patterns (idle channel tones or
pattern noise.).
Quantization Error
1

37

0.8
0.6

Amplitude

0.4
0.2
0
0.2
0.4
0.6
0.8
1

50

time

100

150

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

63

A key goal is to avoid repetitive patterns that determine tones that can fall in the band of
the signal. The amplitude of tones limits the SFDR.
The use of high-order modulators is beneficial.

Call up
Other solutions follow the ... ...................

Use multi-bit quantizers or dithering to destabilize the tonal behavior


of modulators especially when
the input may contain a dominant dc
component.

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

64

Dither is a suitable signal capable to destroy the limit cycles.


Dither must be effective against the tones and should not alter the signal.
Inject a sine wave or a square wave whose frequency is out of the signal band. (The
amplitude of the dither must be as low as possible.)
Noise-like signal whose contribution does not degrade the SNR.
eQ

eQ
X +

Modulator

X +

Modulator

dith

dith
(a)

(b)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

65

Single-bit or Multi-bit?
Disadvantages of using single-bit quantizers.
High SNR with a 1-bit entails the use of high order modulators (design of a stable
architecture or high OSR).
Bandwidth of the op-amps (or OTAs) higher than the clock frequency.
The usable reference voltages of 1-bit modulators is a small fraction of the supply voltage.
Assume that the linear region of the op-amp is VDD and that a 6 dBF S sine wave gives
rise to a swing Vref maximum swing at the first integrator
|Vref | <

VDD
2swing

(63)

The slew-rate of the op-amp must ensure an accurate settling. The output changes of the
first integrator (whose gain is G) can be 2.
SR =

Iout =

2G(VRef Vov )
.
Ts /2

(64)

2VRef (Cin + CL )
.
Ts /2

(65)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

66

Disadvantages of using multi-bit quantizers.


In multi-bit the analog section is more complicated than the single bit counterpart.
The multi-bit DAC is normally a capacitive MDAC with, normally, the subtraction and the
DAC functions obtained by the same capacitive array.
C2

1
C1/4
1 2e

C1/4
2a

2f

C1/4
2b

1 2g

C1/4
2c

1 2h

++

2d

Vin
-Vref
Vref
(a)

t4

2d

t4

2h

t3

2c

t3

2g

t2

2b

t2

2f

t1

2a

t1

2e

(b)

--

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

67

The input capacitance C1 is divided into equal elements, pre-charged to the input signal
during phase 1 and, during phase 2 , under the control of the thermometric codes
t1 , , t4 , connected to +Vref or Vref .
Sharing of the same array for the input injection and the DAC function (also used for a
single-bit architecture) reduces the feedback factor of the OTA.
The charge delivered by the reference voltage generator is a non-linear function of the
input signal



QRef (n) = k(n) VRef Vin (n)

(66)

Output resistance of the reference generator very small for avoiding distortion.
and, also, ....
MATCHING ACCURACY OF THE CAPACITANCES OF THE DAC MUST BE VERY
HIGH (TO ENSURE THE EXPECTED RESOLUTION)

F. Maloberti
DATA CONVERTERS
Springer
2007

Chapter 6

OVERSAMPLING AND LOW ORDER SD

68

Wrap-up
The limited benefit of the oversampling technique can be enhanced by shaping the quantization noise. The spectrum is reduced in the signal band and, possibly augmented
out-band.
A high-pass filtering of the quantization noise is achieved by closing the quantizer in a
feedback loop.
The transfer function of the signal is such that the low-frequency components are unchanged. The noise transfer function significantly attenuates the in-band region.
The performances of real modulators greatly depend on the limitations due to the
real circuit.
First-order and second-order schemes with single-bit or multi-bit quantizers have been
studied so far.

Vous aimerez peut-être aussi