Sakkarapani Balagopal

(P.29.1)A 3-bit, resistor-string DAC similar to the one shown in the fig. 29.2a was designed with
a desired resistor of 500. After fabrication, mismatch caused the actual value of the resistor to
be R
1
=500, R
2
=480, R
3
=470, R
4
=520, R
5
=510, R
6
=490, R
7
=530, R
8
=500.
Determine the maximum INL and DNL for the DAC assuming V
ref
=5V.
The equivalent value of R is given by equation-29.1-1.

8
1 i
i Equivalent
R R (29.1-1)
4000
Equivalent
R


The current flowing through the resistor string is given by equation-29.1-2.
Equivalent
ref
String R
R
V
I
_
(29.1-2)
mA
V
I
String R
25 . 1
4000
5
_

Table 29.1-1 clearly shows the INL and DNL calculation for the given problem.
Table 29.1-1 INL and DNL calculation of 3-bit DAC
Digital
input
Node
Voltages
V
Act
(V) V
Ideal
(V)
Step height(V)
DNL(V)=
S
Act
-S
Ideal

INL(V)=
V
Act
-V
Ideal

S
Act
(V) S
Ideal
(V)
0000_000 V
0
0 0 0 0 0 0
0000_001 V
1
0.625 0.625 0.625 0.625 0 0
0000_010 V
2
1.225 1.25 0.6 0.625 -0.025 -0.025
0000_100 V
3
1.8125 1.875 0.5875 0.625 -0.0375 -0.0625
0001_000 V
4
2.4625 2.5 0.65 0.625 0.025 -0.0375
0010_000 V
5
3.1 3.125 0.6375 0.625 0.0125 -0.025
0100_000 V
6
3.7125 3.75 0.6125 0.625 -0.0125 -0.0375
1000_000 V
7
4.375 4.375 0.6625 0.625 0.0375 0

From Table 29.1-1,
LSB V DNL
MAX
06 . 0 0375 . 0
LSB V INL
MAX
1 . 0 0625 . 0
Chris Gagliano

29.2 An 8-bit resistor string DAC similar to the one shown in Fig. 29.2b was fabricated with a
nominal resistance value of 1 k. If the process was able to provide matching of resistors
to within 1%, find the effective resolution of the converter. What is the maximum INL and
DNL of the converter? Assume that V
REF
= 5V.


1 LSB =
N
REF
V
2
=
8
2
5V
19.53 mV

From Eq (29.10),
INL
max
=
2
1
LSB *
N
2 *
max
R
R
k

INL
max
=
2
1
LSB *
8
2 * 0.01
INL
max
= 1.28 LSB 25 mV

From Eq (29.13),
DNL
max
=
N
REF
V
2
*
max
R
R
k

= 1 LSB * 0.01
DNL
max
= 0.01 LSB 0.1953 mV

The INL is the limiting factor for the resolution of a resistor-string DAC. To determine the
effective resolution of the DAC in this problem, we again use Eq (29.10) and the fact that the
INL must be < 0.5 LSB.
INL
max
=
2
1
LSB *
N
2 *
max
R
R
k

2
1
LSB =
2
1
LSB *
N
2 *
max
R
R
k

1 =
N
2 * 0.01

N
2 = 100
N =
) 2 log(
) 100 log(
6.64 bits

So, a resistor-string DAC with 1% matching will have a resolution of 6 bits. Another way to
determine this for the 8-bit DAC in this problem is to say the effective resolution is equal to the
number of bits minus the ceiling of INL
max
.
Effective resolution = 8 bits 2 bits = 6 bits
QAWI HARVARD
29.3 Compare the digital input codes necessary to generate all eight output values for
a 3-bit resistor string DAC similar to those shown in Fig. 29.2a and b. Design a
digital circuit that will allow a 3-bit binary digital input code to be used for the
DAC in Fig. 29.2a. Discuss the advantages and disadvantages of both
architectures.

A 3-bit resistor string DAC requires 2
N
resistors, Figure 29.2 shows a simple resistor-
string DAC and a binary switch resistor-string DAC.


















The DAC in Figure 29.2 (a) is termed a simple resistor string DAC because it is a
relatively simple structure consisting of 2
N
resistors and 2
N
switches. If we want V
0
as the
output we simply make S
0
high and ground all of the other voltages. If we want V
1
as the
output we make S
1
high and ground all of the other voltages. The corresponding voltage
on V
1
can be determined by realizing that it is simply a resistor divider:


( )
1
1 1
1 1 8
REF REF
R
V V V
N R R

= =
+

And similarly:

( )
4
4 4
4 4 8
REF REF
R
V V V
N R R

= =
+


Using this information we can create table 1 with the digital input codes necessary to
generate all eight output values for the resistor string DACs seen in Figure 29.2.





QAWI HARVARD
Table 1: Digital input code and corresponding output voltage for the DACs in Figure 29.2.

S
7
S
6
S
5
S
4
S
3
S
2
S
1
S
0
D
2
D
1
D
0
V
OUT
0 0 0 0 0 0 0 1 0 0 0 0.000
0 0 0 0 0 0 1 0 0 0 1 0.625
0 0 0 0 0 1 0 0 0 1 0 1.250
0 0 0 0 1 0 0 0 0 1 1 1.875
0 0 0 1 0 0 0 0 1 0 0 2.500
0 0 1 0 0 0 0 0 1 0 1 3.125
0 1 0 0 0 0 0 0 1 1 0 3.750
1 0 0 0 0 0 0 0

1 1 1 4.375

In order to create a circuit that generates the digital codes of table 1 using the digital input
codes of table 2 we can use three input AND gates. Figure 1 shows the circuit used to
create the input code for the DAC in Figure 29.2 (a).















The advantages and disadvantages of the two resistor string DAC structures in figure
29.2, relative to each other, are based on the output capacitance and the switch resistance.
While the simple resistor string DAC provides significantly large output capacitance due
to the 2
N
depletion capacitances it only has one switch resistance connected to the output.
The binary switch array has much less capacitance attached to the output but also has the
disadvantage of having more switch resistance connected to the output. In case of a 3-bit
DAC, with both structures using the same switches we can estimate the load as:






D
0
D
1
D
2
S
0
S
1
S
2
S
3
S
4
S
5
S
6
S
7
1R
s
8C
out
V
out
3R
s
2C
out
V
out
Simple

Binary Array

29.4 Plot the transfer curve of a 3-bit R-2R DAC if all Rs = 1.1 k and 2Rs = 2 k.
What is the maximum INL and DNL for the converter? Assume all of the switches
to be ideal and V
REF
= 5V.




DIN Ideal Vout (V) Vout (V) DNL (LSB) INL (LSB)
000 0 0 0
001 -0.625 -0.555 -0.112 0.112
010 -1.25 -1.165 -0.024 0.134
011 -1.875 -1.72 -0.112 0.245
100 -2.5 -2.416 0.114 0.131
101 -3.125 -2.971 -0.112 0.242
110 -3.75 -3.581 -0.024 0.266
111 -4.375 -4.136 -0.112 0.376

DNL
max
= DNL
100
= 0.114 LSB
INL
max
= INL
111
= 0.376 LSB
Transfer Curve
INL

1.1K 1.1K 1.1K
0.555V 1.166V 2.418V


Justin Wood

29.5 Suppose that a 3-bit R-2R DAC contained resistors that were perfectly
matched and that R=1k and V
REF
=5V. Determine the maximum switch resistance
that can be tolerated for which the converter will still have 3-bit resolution. What
are the values of INL and DNL?

For the converter to have 3-bit resolution, the INL and DNL must be within +/- 0.5 LSB.
With a 3-bit DAC and V
REF
=5, INL and DNL must be +/- 0.3125 V.

First, we need to derive the equations for the R-2R DAC. The general R-2R schematic
is:




From the schematic above:





V
OUT
= I
TOT
R
F
I
TOT
= I
D0
+ I
D1
+ I
D2
I
D0
=
V
0
2R + R
s
, I
D1
=
V
1
2R + R
s
, I
D2
=
V
2
2R + R
s
I
1
=
V
REF
V
2
R
, I
2
=
V
2
V
1
R
, I
3
=
V
1
V
0
R
, I
4
=
V
0
2R
I
TOT

I
D2
I
D1
I
D0

V
0
V
1
V
2

I
1
I
2
I
3
I
4

-


Using KCL at each node (V
2
, V
1
, V
0
) and substituting:



DNL and INL in terms of our output current are as follows:

=
, ()
+
, (1)

, ()
+
, (1)

=
, ()
+
, ()

Using the equations that we defined above, we can increase R
S
and recalculate DNL or
INL is greater than 0.3125V. R
s
was found to be 307 using excel.

Below is a sample calculation of INL1 and DNL1 @ R
s
=307

1 =
, 001
+
, 001

1 =
0,
+
1,
+
2,
+
0,
+
1,
+
2,

()

1 =

,
2+

+0 + 0 +

,
2
+ 0 +0 2 = 0.01298 ()

1 = 0.01298
2
3

5
= 0.02077()

1 =
, 1
+
, 1

, 0
+
, 0

1 =

,
2+

+ 0 +0 +

,
2
+0 +00 +0 2 = 0.01298 ()

1 = 0.01298
2
3

5
= 0.02077()
V
0
=
V
REF
2 +
1
2 +
R
s
R
|
\

|
.
|
|
|
2 +
1
2 +
R
s
R

(
(
(
1.5 +
1
2 +
R
s
R

(
(
(
1
|
\

|
.
|
|
|
1.5 +
1
2 +
R
s
R
V
1
= 1.5 +
1
2 +
R
s
R
|
\

|
.
|
|
|
V
0
V
2
= 2 +
1
2 +
R
s
R
|
\

|
.
|
|
|
V
1
V
0



The rest of the calculations are shown below:

VREF 5
R 1000
Rs 307
V0,actual 0.705961
V1,actual 1.364949
V2,actual 2.615593
V0,ideal 0.625
V1,ideal 1.25
V2,ideal 2.5

D2 D1 D0 I(TOT,ideal) I(TOT,actual)
0 0 0 0 0
0 0 1 0.0003125 0.000306008
0 1 0 0.000625 0.000591655
0 1 1 0.0009375 0.000897663
1 0 0 0.00125 0.001133764
1 0 1 0.0015625 0.001439772
1 1 0 0.001875 0.001725419
1 1 1 0.0021875 0.002031427

(V) (LSBs)
INL0 0 0
INL1 0.012984 0.020774
INL2 0.066689 0.106703
INL3 0.079673 0.127477
INL4 0.232473 0.371956
INL5 0.245457 0.392731
INL6 0.299162 0.478659
INL7 0.312146 0.499433
DNL1 0.012984 0.020774
DNL2 0.053705 0.085929
DNL3 0.012984 0.020774
DNL4 0.1528 0.244479
DNL5 0.012984 0.020774
DNL6 0.053705 0.085929
DNL7 0.012984 0.020774


Lincoln Bollschweiler
29.6 The circuit illustrated in Fig. 29.5 is known as a current-mode R-2R DAC, since the
output voltage is defined by the current through R
F
. Shown in Fig. 29.52 is an N-bit
voltage-mode R-2R DAC. Design a 3-bit voltage mode DAC and determine the output
voltage for each of the eight input codes. Label each node voltage for each input. Assume
that R = 1k and that R
2
= R
1
= 10k and V
REF
= 5V.

The following LTSpice circuit schematic shows the 3-bit DAC. The output voltages see a
gain of 2 and therefore show a range which exceeds the rails of a 5V process amplifier.
As an ideal amplifier is used, the output range, shown on the following page, ranges 0V
to 8.8V. Output voltages are labeled on the simulation output for V
OUT
.






Digital Codes
000 001 010 011 100 101 110 111
8.75
7.5
6.25
5.0
3.75
2.5
1.25
0.0
Sakkarapani Balagopal
(P29.7) Design a 3-bit, current-steering DAC using the generic current-steering DAC shown in
figure 29.9. Assume that each current source, I, is 5A, and find the total output current for each
input code.

i
out
D
0
D
1
D
2
D
3
D
4 D
5
D
6
I I
I
I I I I

Figure 29.9-1 - A generic current-steering DAC.

The simple 3-bit current steering DAC is shown in the figure 29.9-1. For a 3-bit DAC, seven
equal value current sources of value I=5A will be needed. Depend on the binary control signals
from D
0
to D
6
, the current sources are either connected to i
out
or ground. The resulting currents
are shown in the table 29.7-1.

Table 29.7-1 - Output current generated from 3-bit DAC
Digital
Input
Code
D
0
D
1
D
2
D
3
D
4
D
5
D
6
i
out

000 0 0 0 0 0 0 0 0
001 1 0 0 0 0 0 0 5A
010 1 1 0 0 0 0 0 10A
011 1 1 1 0 0 0 0 15A
100 1 1 1 1 0 0 0 20A
101 1 1 1 1 1 0 0 25A
110 1 1 1 1 1 1 0 30A
111 1 1 1 1 1 1 1 35A

Chris Gagliano

29.8 A certain process is able to fabricate matched current sources to within 0.05%.
Determine the maximum resolution that a current-steering (non-binary-weighted) DAC
can attain using this process.


From Eq (29.21),

INL
I
max,
=
N
I
2


From Eq (29.26),

DNL
I
max,
=
2
I


The terms
INL
I
max,
and
DNL
I
max,
represent the maximum current source mismatch error that
will keep the INL and DNL, respectively, less than 0.5 LSB. Clearly the INL is the limiting
factor in a non-binary-weighted current-steering DAC.

If the current sources are matched within 0.05%, then
max
I = 0.05% * I.

INL
I
max,
=
N
I
2

0.05% * I =
N
I
2


N
2 =
0005 . 0
1

N =
) 2 log(
)
0005 . 0
1
log(
10.97

The maximum resolution is then 10 bits.

QAWI HARVARD
29.9 Design an 8-bit current-steering DAC using binary-weighted current sources.
Assume that the smallest current source will have a value of 1 A. What is the range
of values that the current source corresponding to the MSB can have while
maintaining an INL of LSB? Repeat for a DNL less than or equal to LSB.

Figure 29.10 gives a generic view of a binary-weighted current-steering DAC.











An 8-bit current-steering DAC using binary-weighted current sources will have N current
sources. Figure 1 shows the binary-weighted current sources used in an 8-bit DAC.


Figure 1 Block diagram of an 8-bit binary-weighted current-steering DAC.

Analyzing the integral nonlinearity of the binary weighted current steering DAC is
accomplished by realizing that the sum of current error sums to zero and that one half of the
current sources contain the maximum INL error. In this case, the MSB current will contain
the maximum error.

( )
1
,
2
N
MSB actual
I I I

= +

Subtracting the ideal current from the actual current gives the value of the maximum INL:

( ) ( )
1 1 1
, ,
2 2 2
N N N
MAX MSB actual MSB ideal
INL I I I I I I

= = + =

Realize that the LSB of the current steering DAC is simply I, and we can equate the
maximum INL to LSB:


1
2 MAX
INL LSB =
I 2I 4I 8I 16I 32I 64I
IOUT
D0 D1 D2 D3 D4 D5 D6
128I
D7
QAWI HARVARD

1
1
2
2
N
I I

=

8
1
4
2 2
N
I A
I nA

= =

In order to design a binary weighted current source with an INL less than LSB, the range
of the MSB must be:

, , , MSB ideal MSB actual MSB ideal
I I I I I

,
127.996 128.004
MSB actual
A I A

For the determination of the DNL we have to realize that the worst case occurs at the
transition from 01111111 to 10000000. We also assume that the errors all sum to zero. As an
example, consider a 3-bit binary weighted DAC with the lower two current sources
containing negative error and the MSB containing positive error. To determine the maximum
DNL we simply subtract the currents to determine the step height:

( ) ( ) ( )
2 1 0
2 2 2
MAX
DNL I I I I I I I = +

4 4 3 3 7
MAX
DNL I I I I I I = + + =

Restating equation 29.28, shows that this approach is correct:

( ) ( )
1
1 1
1
2 2
N
N k
MAX
k
DNL I I I I I

=

= +

Which results in
( )
2 1
N
MAX
DNL I = , if we wish to keep the DNL equal to less than
LSB we simply equate the two and solve for the maximum error, remembering that an LSB is
equal to I:

( )
1
2
2 1
N
I I =

( )
1
2
2 255 2 2 1
N
I A
I nA

= =



For the DNL to be less than or equal to LSB the MSB current source must have the range
of:

,
127.998 128.002
MSB actual
A I A

Comparing the two results we can see that the DNL requirement is more stringent.
Adam Johnson
29.10 Prove that the 3-bit charge-scaling DAC used in Ex. 29.6 has the same output
voltage increments as the R-2R DAC in Ex. 3 for V
REF
= 5 V and C = 0.5 pF.








From Eq. (29.33), V
OUT
= (D
0
*2
-3
+ D
1
*2
-2
+ D
2
*2
-1
)*V
REF


For DIN = 000, V
OUT
= (0*2
-3
+ 0*2
-2
+ 0*2
-1
)*5V = 0V

For DIN = 001, V
OUT
= (1*2
-3
+ 0*2
-2
+ 0*2
-1
)*5V = 0.625V

For DIN = 010, V
OUT
= (0*2
-3
+ 1*2
-2
+ 0*2
-1
)*5V = 1.25V

For DIN = 011, V
OUT
= (1*2
-3
+ 1*2
-2
+ 0*2
-1
)*5V = 1.875V

For DIN = 100, V
OUT
= (0*2
-3
+ 0*2
-2
+ 1*2
-1
)*5V = 2.5V

For DIN = 101, V
OUT
= (1*2
-3
+ 0*2
-2
+ 1*2
-1
)*5V = 3.125V

For DIN = 110, V
OUT
= (0*2
-3
+ 1*2
-2
+ 1*2
-1
)*5V = 3.75V

For DIN = 111, V
OUT
= (1*2
-3
+ 1*2
-2
+ 1*2
-1
)*5V = 4.375V


DIN Ex. (29.3) Ex. (29.6)
000 0 0
001 -0.625 0.625
010 -1.25 1.25
011 -1.875 1.875
100 -2.5 2.5
101 -3.125 3.125
110 -3.75 3.75
111 -4.375 4.375

The voltage increments (V
LSB
) of both DACs are the same in magnitude at 0.625V, but
opposite in sign.

Justin Wood
29.11! Determine the output of the 6-bit, charge-scaling DAC used in Ex. 29.7 for
each of the following inputs: D = 000010, 000100, 001000, and 010000.
With D=000010, the equivalent circuit would be:
We rst calculate VA as follows:
V
A
=
2C
6C +
8
7
C series 7C

+ 2C
V
REF
=
2
6 +
8
7
7
8
7
+ 7

+ 2
V
REF
=
2
8 +
56
57
V
REF
VOUT is then calculated as follows:
V
OUT
=
8
7
8
7
+ 7

V
A
V
OUT
=
8
7
8
7
+ 7

2
8 +
56
57

V
REF
=
1
32
V
REF
With D=000100, the equivalent circuit would be:
We rst calculate VA just like the previous code:
V
A
=
4C
4C +
8
7
C series 7C

+ 4C
V
REF
=
4
8
7
7
8
7
+ 7

+ 8
V
REF
=
4
8 +
56
57
V
REF
VOUT is then calculated as follows:
V
OUT
=
8
7
8
7
+ 7

V
A
V
OUT
=
8
7
8
7
+ 7

4
8 +
56
57

V
REF
=
1
16
V
REF
With D=001000, the equivalent circuit would be:
VOUT is calculated as follows:
V
OUT
=
1C
6C +
8
7
C series 8C

+1C

V
REF
V
OUT
=
1
8
7
8
8
7
+ 8

+ 7
V
REF
=
1
8
V
REF
With D=010000, the equivalent circuit would be:
VOUT is calculated as follows:
V
OUT
=
2C
5C +
8
7
C series 8C

+ 2C

V
REF
V
OUT
=
2
8
7
8
8
7
+ 8

+ 7
V
REF
=
1
4
V
REF
Lincoln Bollschweiler
29.12 Design a 4-bit, charge-scaling DAC using a split array. Assume that V
REF
= 5V
and that C = 0.5pF. Draw the equivalent circuit for each of the following input words
and determine the value of the output voltage: D = 0001, 0010, 0100, 1000. Assume the
capacitor associated with the MSB had a mismatch of 4 percent, calculate the INL and
DNL.

We find the value of the attenuation capacitor, using (29.35) to be (4/3)C = 1.333pF.

The complete circuit is shown as follows:


In all simplifications shown, the capacitor values will be normalized to C=1
a. The equivalent circuit for input code 0001 is as follows:


C3 in series with C4 reduces to 12/13. That in parallel with C2 is 3+12/13. That circuit is
as follows:

We can easily solve for V1 now by capacitive divider. V1 = 0.203Vref. We can use V1 in
a capacitive divider with C3 and C4 to get Vin+. Vin+ = 0.0625Vref = Vref/16. Vout =
Vin+ due to voltage follower.
b. The equivalent circuit for input code 0010 is as follows:

By similar analysis technique we can find that V1 = 0.406Vref. Vout = 0.125Vref =
Vref/8.

c. The equivalent circuit for input code 0100 is as follows:


which further reduces to:


It can be readily seen by capacitive voltage divider that Vout = Vin+ = Vref/4.

d. The equivalent circuit for input code 1000 is as follows:


It can be readily seen that Vout = Vin+ = Vref/2.

To find INL we can start by realizing that 4% matching on the MSB capacitor is the same
as having the MSB cap being off by +2% while all the others are off by -2%. Overall we
have 2% mismatch. We can use the R2-R DAC equations from chapter 30 to find INL
and DNL.

Using (30.13) and % mismatch = 2 we find:
4
1
5 0.02 0.213 0.68 .
2
DNL V LSB

Using (30.16) we find:.
5 2
0.049 0.157 .
2 100 2
INL V LSB

Sakkarapani Balagopal
(P.29.13)For the cyclic converter shown in the figure 29.17, determine the gain error for a 3-bit
conversion if the feedback amplifier had a gain of 0.45V/V. Assume that V
ref
=5V.
The output voltage at the end of nth cycle of the conversion with feedback amplifier gain of
0.5V/V is given in equation 29.13-1.
2
1
) 1 (
2
1
. ) (
1
|
.
|

\
|
+ =

n V V D n V
A ref n Out
(29.13-1)
Table 29.13-1 summarizes the resultant output voltage for different inputs. Results are plotted in
the figure 29.13-1.
Table-29.13-1 - Output voltage of 3-bit DAC with different feedback amplifier gain
Input D
0
D
1
D
2

V
out,actual
(V) for
a gain of
0.5V/V
V
out,actual
(V) for
a gain of
0.45V/V
0 0 0 0 0 0
1 0 0 1 0.625 0.4556525
2 0 1 0 1.25 1.0125
3 0 1 1 1.875 1.468125
4 1 0 0 2.5 2.25
5 1 0 1 3.125 2.705625
6 1 1 0 3.75 3.2625
7 1 1 1 4.375 3.718125


Figure 29.13-1 Transfer curve for 3-bit DAC for different feedback amplifier gain
0
1
2
3
4
5
0 1 2 3 4 5 6 7
O
u
t
p
u
t

V
o
l
t
a
g
e
(
V
)
Input Code
Transfer Curve
Gain of 0.5V/V
Gain of 0.45V/V
From page no.945 of the textbook, the gain error can be written as

d Blue
Slope Actual Slope Ideal Error Gain
Re
_ _ _ =

d Blue
x x
y y
x x
y y
Error Gain
Re
1 2
1 2
1 2
1 2
_
|
|
.
|

\
|

|
|
.
|

\
|

=
d
Blue
REF
V
Error Gain
Re
000 111
0 718125 . 3
000 111
0
8
7
_
|
.
|

\
|

|
|
|
|
.
|

\
|

=

The x and y axis in figure 29.15-1 are in different units. So, convert the denominator on above
equation into same unit or volts.

d
REF
Blue
REF
REF
V V
V
Error Gain
Re
0
8
7
0 718125 . 3
0
8
7
0
8
7
_
|
|
|
|
.
|

\
|

|
|
|
|
.
|

\
|

=

85 . 0 1
5 .
8
7
718125 . 3
1 _
Re
=
|
|
|
|
.
|

\
|
=
d
Error Gain

V
V
Error Gain 15 . 0 _ =
Chris Gagliano

29.14 Repeat Problem 29.13 assuming that the output of the summer was always 0.2V greater
than the ideal and that the amplifier in the feedback path had a perfect gain of 0.5 V/V.
















The output voltage at the end of the n
th
cycle can be expressed as:
V
out
(n) = ( ) A V n V A V D
summer offset A REF n
+ +
, 1
) 1 (

And ) (n V
A
can be expressed as:

summer offset A REF n A
V n V A V D n V
, 1
) 1 ( ) ( + + =



We can then write V
out
(n) as:
V
out
(n) = ( ) ( ) A V V n V A V D A V D
summer offset summer offset A REF n REF n
+ + + +
, , 2 1
) 2 (
V
out
(n) =
summer offset summer offset A REF n REF n
V A V A n V A V D A V D A
, ,
2 3
2
2
1
) 2 ( + + + +


V
out
(n) =


=

+
1
0
,
) ( ,
1
0
n
k
k n
summer offset
n ideal Vout
n
k
k
k n
REF
A V D A V
4 4 4 3 4 4 4 2 1


For an N-bit DAC, we sample the output voltage on the N
th
cycle. We are using a 3-bit DAC in
this problem, so the output voltage is:
V
out
= V
out
(n=3) = ( ) ( ) A A A V D A D A D A V
summer offset REF
+ + + + +
2 3
, 2 1
2
0
3

V
out
= ( ) ( ) 5 . 0 5 . 0 5 . 0 2 . 0 5 . 0 5 . 0 5 . 0 5
2 3
2 1
2
0
3
+ + + + + V D D D V
V
out
= ( ) V D D D V
ideal Vout
175 . 0 5 . 0 5 . 0 5 . 0 5
,
2 1
2
0
3
+ + +
4 4 4 4 4 4 3 4 4 4 4 4 4 2 1


The output values and transfer curve are shown below.


V
offset,summer
= 0.2 V
A = 0.5




















Gain error = slope of V
out,ideal
slope of V
out,actual


It is clear from the plot that the slopes of both lines are equal. There is only an offset error equal
of 175 mV. After removing the offset error, we would report:

Offset error = 175 mV
Gain error = 0

max
INL = 0

max
DNL = 0


QAWI HARVARD
29.15 Repeat Problem 29.13 assuming that the output of the summer was always 0.2 V
greater than the ideal and that the amplifier in the feedback path had a gain of 0.45
V/V.

29.13 For the cyclic converter shown in Fig. 29.17, determine the gain error for a 3-bit
conversion if the feedback amplifier had a gain of 0.45 V/V. Assume that V
REF
= 5 V.
















The gain error of a DAC is difference in the slope of the best fit line to the ideal transfer
curve of the DAC. Consider figure 28.17:



















A plot of the transfer curve of our 3-bit DAC versus the ideal 3-bit DAC will give us our gain
error. Realizing that the summation block has an offset of 0.2 V and the amplifier has a gain
of 0.45 V/V we can build an output table for our 3-bit DAC. An example of the output table
is shown below for a digital input of 110.


QAWI HARVARD
Table 1 Output from the 3-bit cyclic DAC used in Problem 29.15
Cycle Number, n D
n-1
v
A(n-1)
[n-1] v
out
[n]
1 0 0
( ) 0.45 0 0 0.2 0.090V + + =
2 0 0.090
( ) 0.45 0 0.09 0.2 0.131V + + =
3 0 0.131
( ) 0.45 0 0.131 0.2 0.149V + + =
1 1 0
( ) 0.45 5 0 0.2 2.34V + + =
2 1 0.090
( ) 0.45 5 2.34 0.2 3.393V + + =
3 1 2.381
( ) 0.45 5 3.393 0.2 3.867V + + =

Using the approach seen in Table 1 we are able to plot a transfer curve for each of the eight
digital input codes and compare them to the ideal.

3-Bit DAC Transfer Curve
0.000
0.625
1.250
1.875
2.500
3.125
3.750
4.375
5.000
000 001 010 011 100 101 110 111
Input Digital Code, D
O
u
t
p
u
t

V
o
l
t
a
g
e
,

V
IDEAL ACTUAL

Figure 1 Transfer curve of the 3-bit cyclic DAC of problem 29.15.

The gain error can be determined by finding the slopes of both lines. Here we simply subtract
the initial and final values.

Gain Error
3.718
0.625
1 0.15
7
LSB
LSB
= =

Adam Johnson
29.16 Design a 3-bit pipeline DAC using V
REF
= 5 V. (a) Determine the maximum and
minimum gain values for the first-stage amplifier for the DAC to have less than
LSBs of DNL assuming that the rest of the circuit is ideal. (b) Repeat for the
second-stage amplifier. (c) Repeat for the last stage amplifier.







From Eq. (29.40) we can see that

V
OUT
= (D
0
*A
0
*A
1
*A
2
+ D
1
*A
1
*A
2
+ D
2
*A
2
)*V
REF

Where A
n
is the gain of the amplifier of stage n, and A
0
= A
1
= A
2
= for the ideal case.

V
LSB,Ideal
= A
0
*A
1
*A
2
*V
REF
= ***5V = 0.625V
V
LSB
= 0.3125V

(a) For the first-stage amplifier DNL error, we can look at the 000 to 001 transition

A
0,MIN
***5V = 0.625V - 0.3125V
A
0,MIN
= 0.25

A
0,MAX
***5V = 0.625V + 0.3125V
A
0,MAX
= 0.75

(b) In similar fashion, the second-stage amplifier DNL error can be seen with a 011 to
100 transition

V
OUT,100
V
OUT,011
= *5V (*A
1,MIN
* + *A
1,MIN
)*5V = 0.625V + 0.3125V
A
1,MIN
= 0.42

V
OUT,100
V
OUT,011
= *5V (*A
1,MAX
* + *A
1,MAX
)*5V = 0.625V - 0.3125V
A
1,MAX
= 0.58

(c) The last-stage amplifier DNL can be seen with the same 011 to 100 transition

V
OUT,100
V
OUT,011
= A
2,MIN
*5V (** + *)*5V = 0.625V - 0.3125V
A
2,MIN
= 0.44

V
OUT,100
V
OUT,011
= A
2,MAX
*5V (** + *)*5V = 0.625V + 0.3125V
A
2,MAX
= 0.56
Justin Wood
29.17! Using the same DAC designed in Problem 29.16:
(a) determine the overall error (offset, DNL, and INL) for the DAC if the S/H
amplier in the rst stage produces an offset at its output of 0.25V. Assume
that all the remaining components are ideal.
The 3-bit pipeline DAC schematic is below.
With the offset applied to the rst S/H, the output voltage from each stage would be
calculated as follows:
V
OUT1
= (D
0
V
REF
+V
OFFSET
) A
1
V
OUT 2
= (D
1
V
REF
+V
OUT1
)A
2
= (D
1
V
REF
+ (D
0
V
REF
+V
OFFSET
) A
1
)A
2
V
OUT 3
= (D
2
V
REF
+V
OUT 2
)A
3
= (D
2
V
REF
+ (D
1
V
REF
+ (D
0
V
REF
+V
OFFSET
) A
1
)A
2
)A
3
A
1
= A
2
= A
3
V
OUT 3
= V
REF
(D
2
A + D
1
A
2
+ D
0
A
3
) +V
OFFSET
A
3

For VOFFSET of 0.25V and A = 0.5, the offset would be: 0.25 (0.5)
3
= 0.03125V
In terms of LSBs, the offset is: 0.03125V
2
3
LSBs
5V
= 0.05LSBs
For the pipeline DAC, an offset doesnt contribute to DNL because it will offset all
outputs by the same amount so the step change will remain unchanged. Depending on
how INL is specied, the offset could negatively impact INL. Typically, offset is removed
when determining INL (ie. using the best-t method). Therefore, INL and DNL = 0
assuming all other elements are ideal.
(b) Repeat for the second-stage S/H.
With the offset applied to the second S/H, the output voltage from each stage would be
calculated as follows:
D2
D2
D1
D1
A2 +
S/H
+ A1 A3
S/H S/H
Do
Do
VREF VREF VREF
VOUT
V
OUT1
= (D
0
V
REF
) A
1
V
OUT 2
= (D
1
V
REF
+V
OUT1
+V
OFFSET
)A
2
= (D
1
V
REF
+ (D
0
V
REF
) A
1
+V
OFFSET
)A
2
V
OUT 3
= (D
2
V
REF
+V
OUT 2
)A
3
= (D
2
V
REF
+ (D
1
V
REF
+ (D
0
V
REF
) A
1
+V
OFFSET
)A
2
)A
3
A
1
= A
2
= A
3
V
OUT 3
= V
REF
(D
2
A + D
1
A
2
+ D
0
A
3
) +V
OFFSET
A
2

For VOFFSET of 0.25V and A = 0.5, the offset would be: 0.25 (0.5)
2
= 0.0625V
In terms of LSBs, the offset is: 0.0625V
2
3
LSBs
5V
= 0.1LSBs
Again DNL/INL are 0.
(c) Repeat for the last-stage S/H.
With the offset applied to the rst S/H, the output voltage from each stage would be
calculated as follows:
V
OUT1
= (D
0
V
REF
) A
1
V
OUT 2
= (D
1
V
REF
+V
OUT1
)A
2
= (D
1
V
REF
+ (D
0
V
REF
) A
1
)A
2
V
OUT 3
= (D
2
V
REF
+V
OUT 2
+V
OFFSET
)A
3
= (D
2
V
REF
+ (D
1
V
REF
+ (D
0
V
REF
) A
1
)A
2
+V
OFFSET
)A
3
A
1
= A
2
= A
3
V
OUT 3
= V
REF
(D
2
A + D
1
A
2
+ D
0
A
3
) +V
OFFSET
A

For VOFFSET of 0.25V and A = 0.5, the offset would be: 0.25 (0.5) = 0.125V
In terms of LSBs, the offset is: 0.125V
2
3
LSBs
5V
= 0.2LSBs
Again DNL/INL are 0.
Lincoln Bollschweiler
29.18 Design a 3-bit Flash ADC with its quantization error centered about zero LSBs.
Determine the worst-case DNL and INL if resistor matching is known to be 5%. Assume
that V
REF
= 5V.

1.5 R
0.5 R

From (29.48), and assuming V
OS
= 0,
5
0.05 0.125
2
INL V or 0.2LSBs.
From (29.52), also assuming V
OS
= 0,
3
5
0.05 0.031
2
DNL V or 0.05LSBs.
Sakkarapani Balagopal
(P.29.19)Using the ADC designed in problem 29.18, determine the maximum offset that can be
tolerated if all of the comparator have the same magnitude of offset, but with different polarities,
to attain a DNL of less than or equal to0.5LSB.
The Maximum DNL will occur, assuming R
i
is at its maximum, V
osi
is at its maximum positive
value is given in equation 29.19-1.
Max
os
Max
i
N
ref
Max
V
R
R
V
DNL 2 .
2

(29.19-1)
For the given maximum DNL of 0.5LSB and 5% resistor mismatch from previous problem, the
calculated offset voltage is

Max
os N N
V 2 05 . 0 .
2
5
2
5
1

2
05 . 0 .
2
5
2
5
1 N N
Max
os
V

mV V
Max
os
140
Chris Gagliano

29.20 A 4-bit Flash ADC converter has a resistor string with mismatch as shown in Table 29.1.
Determine the DNL and INL of the converter. How many bits of resolution does this
converter possess? V
REF
= 5 V.


For a 4-bit ADC with V
REF
= 5 V,
1 LSB =
4
2
5V
= 0.3125 V

The INL and DNL values are calculated in the table below. Note that the column labeled V
in,nulled-
offset-gain
, is used for the INL and DNL calculations since it contains the values with the static
offset and gain errors nullified.

The offset error is calculated as follows:
V
offset
= V
in,actual
(D=1) V
in,ideal
(D=1) = 0.3172 V 0.3125 V = 0.0047 V

The gain error is calculated as follows (using the V
in,no-offset
values):
Gain error = 1 -
LSB
LSB
V
V
15
3125 . 0
6813 . 4
0.0013
LSB
LSB


The V
in,nulled-offset-gain
values in the table are then calculated by the following equation:
V
in,nulled-offset-gain
(n) = V
in,no-offset
(n) + n * 1 LSB * Gain error
V
in,nulled-offset-gain
(n) = V
in,no-offset
(n) + n * 0.3125 V * 0.0013


The resolution of the ADC is calculated using the maximum
INL value from the table and knowing that it cannot be more
than half an LSB.

The resolution of the ADC is 7 bits.

N
REF
V
mV INL
2 2
1
03 . 15
max
=

33 . 166
01503 . 0 2
5
2

=
V
V
N

38 . 7
) 2 log(
) 33 . 166 log(
= N
QAWI HARVARD
29.21 Determine the open-loop gain required for the residue amplifier of a two-step ADC necessary to keep
the converter to within LSB of accuracy with resolutions of (a) 4 bits, (b) 8 bits, and (c) 10 bits.

Consider the block diagram of a two-step Flash ADC as seen in Figure 29.26.
















An advantage of the two-step Flash ADC is that each Flash ADC is only 2
N/2
bits accurate. If the residue
amplifier wants to be accurate to within LSB we can state:


1 1
2 2 2
1
1
2
REF
N
V
N
REF REF
LSB
Accuracy
V V
+

= = =

Figure 29.26 shows that the residue amplifier has gain equal to 2
N/2
, we can write the open loop gain in terms of
the closed loop gain:


1
OL
CL
OL
A
A
A
=

As A
OL
increases the closed loop gain approaches 1/. If we want the amplifier to have an accuracy of
LSB we can write the closed loop gain as:

1
1
OL
CL
OL
A
A A
A
= =

Where A is the accuracy. Solving for the open loop gain gives us:

1
1 / 2
2
2 2
N
N N
OL
A

+
+
=

Using the equation above, we can state that the requirements of the open loop gain for each number of bits:

4 bits
7
2 128 =

8 bits
13
2 8192 =

10 bits
16
2 65536 =

Adam Johnson
29.22 Assume that a 4-bit, two-step Flash ADC uses two separate Flash converters for
the MSB and LSB ADCs. Assuming that all other components are ideal, show
that the first Flash converter needs to be more accurate than the second
converter. Assume that V
REF
= 5 V.


In the ideal case, the V
2
must be between 0 and V
REF
. If we include INL, the V
2
value
can be between INL
MIN,ADC1
and V
REF
+ INL
MAX,ADC1
.

Since V
3
= V
2
*2
N/2
, and this is a 4-bit ADC, we can see that the range on V
3
will be
between (INL
MIN,ADC1
)*4 and V
REF
+ (INL
MAX,ADC1
)*4. So essentially, any error from
the first ADC conversion is multiplied along with the residual voltage by the residue
amplifier.

For this 4-bit ADC, if the INL/DNL needs to be < LSB overall, then for the first ADC,
the INL/DNL needs to be <
1
/
8
LSB, and the second ADC can be < LSB.



4x
00
01
10
11
V
REF

00
01
10
11
V
REF

ADC1
Multiplied
by 2
N/2

INL
MAX
*2
N/2

INL
MAX

ADC2
INL
MIN

INL
MIN
*2
N/2

ADC1 ADC2
29.23! Repeat Ex. 29.12 for VIN = 3, 5, 7.5, 14.75 V.
Example 29.12
Assume that the two-step ADC shown in Fig. 29.26 has four bits of resolution. Make a
table listing the MSBs, V1, V2, V3, and the LSBs for VIN = 3, 5, 7.5, 14.75 V assuming
that VREF = 16V.
With 4-bit resolution, Figure 29.26 would be as follows:
The MSB and LSB ADCs in the picture are both 2-bit ADCs and their LSBs are equal to
4V
16V
2
2
= 4V

. First, lets calculate the output for VIN=3V. Starting with the MSB ADC,
D3D2 would be 00 until VIN > 4V (LSB). V1 would then be 0, V2 would be 3V. N=4 and
V3 = 3*(2
2
) or 12V. The output of the LSB ADC would then be 11. Final output code is
0011. Output code for VIN are similarly calculated as shown in the table.
VIN D3D2(MSBs) V1 V2 V3 D1D0(LSBs)
3 00 0 3 12 11
5 01 4 1 4 01
7.5 01 4 3.5 14 11
14.75 11 12 2.75 11 10
S/H
VIN
Latches
D2 D3 D1 D0
MSBs
MSB
ADC
DAC
V1
+
-
2
N/2
LSBs
LSB
ADC
V2 V3
Lincoln Bollschweiler
29.24 Repeat Ex. 29.13 for V
IN
= 1, 3, 6, 7 V and V
REF
= 8V.

Example 29.13 Assume that the pipeline converter shown in Fig. 29.30 is a 3-bit
converter. Analyze the conversion process by making a table of the following
variables: D
2
, D
1
, D
0
, V
3
, and V
2
for V
IN
= 2, 3, and 4.5 V. Assume that V
REF
=
5V, V
3
is the residue voltage out of the first stage, V
2
is the residue from the
second.


With a V
REF
of 8V, all comparators will give an output of 1 for vin
p
of 4V or greater;
output of 0 otherwise.


v
IN
(V) V
3
(V) V
2
(V) Digital Out (D
2
D
1
D
0
)
1.0 2 4 001
3.0 6 4 011
6.0 4 0 110
7.0 6 4 111

As is expected, since the reference voltage is 8 and the LSB = 8/2
2
= 1V, the digital code
is simply the input (decimal) converted to 3-bit binary.
Sakkarapani Balagopal
(P.29.25)Assume that 8-bit pipeline ADC was fabricated and that all the amplifiers had a gain of
2.1V/V instead of 2V/V. If V
IN
=3V and V
REF
=5V, what would be the resulting digital output if the
remaining components were considered to be ideal? What are the DNL and INL for this
converter?
8-bit Pipelined ADC requires 8-steps to convert the given analog signal into 8-bit digital codes.
The positive input of the comparator to the next stage is given by equation 29.25-1.Table 29.25-1
shows the conversion of the analog signal into digital codes in 8 steps.
Gain
n C V
n V n V
OUT REF
IN IN
.
2
) ( .
) ( ) 1 ( |
.
|

\
|
= + (29.25.1)
Table 29.25-1 8-bit Pipelined ADC conversion for V
IN
= 3V, V
REF
= 5V
STEP(n)
Comparators
Positive Input
V
IN
(n)
Comparators
Negative
Input(V
REF
/2)
Comparator
Output(C
OUT
(n))
1 3 2.5 1
2 1.05 2.5 0
3 2.205 2.5 0
4 4.6305 2.5 1
5 4.47405 2.5 1
6 4.145505 2.5 1
7 3.4555605 2.5 1
8 2.00667705 2.5 0

8-bit Pipelined ADC output = 1 0 0 1 1 1 1 0

We can write the input analog voltage in terms of digital codes with a gain of the amplifier of 2.1
V/V for N-bit Pipelined ADC as
1 0 2 1 2 1 ,
.
2
1
.
2
1
...... .
2
1
.
2
1

+ + + + =
N
REF
N
REF REF
N REF N N IN
Gain
V
D
Gain
V
D
Gain
V
D V D V (29.25-2)
Using the above equation, DNL and INL of 8-bit pipelined ADC is plotted in figure 29.25-1 and
2 respectively. Calculation of INL and DNL of the converter is shown in Table 29.25-2 for
digital outputs ranging from 0 to 6.



Sakkarapani Balagopal
Table 29.25-2 8-bit Pipelined ADC INL and DNL Calculation
Possible
Digital
O/P (n)
D0 D1 D2 D3 D4 D5 D6 D7
Corresponding
Analog
Input(Xn)
Calculation
Using
Equation
29.25-2
Xn in
LSB (Yn)
Yn
Shifted
by
0.5LSB
(Zn)
Ideal
Analog
Input in
LSB (In)
DNL(n) =
(Zn+1-Zn) -
1LSB
INL(n)
=Zn- In
0
0 0 0 0 0 0 0 0 0 0 0.5
0.5 - 0
1 1 0 0 0 0 0 0 0 0.013880495 0.71069 1.21069 1.5 -0.28931
-
0.28930
2 0 1 0 0 0 0 0 0 0.029149039 1.49245 1.99245 2.5 -0.218241
-
0.50755
3 1 1 0 0 0 0 0 0 0.043029534 2.20314 2.70314 3.5 -0.28931
-
0.79685
4 0 0 1 0 0 0 0 0 0.061212982 3.134145 3.634145 4.5 -0.068996
-
0.86585
5 1 0 1 0 0 0 0 0 0.075093476 3.844835 4.344835 5.5 -0.28931
-
1.15516
6 0 1 1 0 0 0 0 0 0.090362021 4.626595 5.126595 6.5 -0.218241
-
1.37340

DNL of the 8-bit Pipelined ADC:

Figure 29.25-1 DNL of the 8-bit Pipelined ADC





-2
0
2
4
6
8
10
12
1
1
6
3
1
4
6
6
1
7
6
9
1
1
0
6
1
2
1
1
3
6
1
5
1
1
6
6
1
8
1
1
9
6
2
1
1
2
2
6
2
4
1
L
S
B
Analog Input
DNL
DNL
Sakkarapani Balagopal
INL of the 8-bit Pipelined ADC:

Figure 29.25-2 INL of the 8-bit Pipelined ADC
From the figure 29.25-1, 2 the maximum and minimum INL and DNL are given as

DNL INL
Maximum 11.28259 0.001638
Minimum -0.28931 -11.281

-12
-10
-8
-6
-4
-2
0
2
1
1
4
2
7
4
0
5
3
6
6
7
9
9
2
1
0
5
1
1
8
1
3
1
1
4
4
1
5
7
1
7
0
1
8
3
1
9
6
2
0
9
2
2
2
2
3
5
2
4
8
L
S
B
Analog Input
INL
INL
Chris Gagliano

29.26 Show that the first-stage accuracy is the most critical for a 3-bit, 1-bit per stage pipeline
ADC by generating a transfer curve and determining DNL and INL for the ADC for three
cases: (1) The gain of the first-stage residue amplifier set equal to 2.2 V/V, (2) the
second-stage residue amplifier set equal to 2.2 V/V. For each case, assume that the
remaining components are ideal. Assume that V
REF
= 5V.

















D
2
will transition high when
REF p
V V =
2
1
1
and since
IN p
V V =
1
we can write the input voltage at
which D
2
transitions high as:

REF IN
V V =
2
1
1 ,


D
1
will transition high when
REF p
V V =
2
1
2
and we can write
2 p
V as:

1 2 1 2
2
1
A D V V V
REF p p

(

=

We can then solve for the input voltage at which D
1
transitions high, keeping in mind that
IN p
V V =
1
:

1 2 2 ,
2
1
2
1
A D V V V
REF IN REF

(

=

2
1
2 ,
2
1
2
1
D V V
A
V
REF REF IN
+

=



3-bit Pipeline ADC (based on Fig. 29.30)
D
0
will transition high when
REF p
V V =
2
1
3
and we can write
3 p
V as:

2 1
2
1 2 1 2 1 2 3
2
1
2
1
2
1
A D V A D V V A D V V V
REF
Vp
REF p REF p p

(
(
(
(

\
|
=
(

=
4 4 4 4 3 4 4 4 4 2 1


We can then solve for the input voltage at which D
0
transitions high, again keeping in mind that
IN p
V V =
1
:

2 1 1 2 3 ,
2
1
2
1
2
1
A D V A D V V V
REF REF IN REF

(

\
|
=

2 1
1 2 1
3 ,
2
1
2
1
2
1
D V D V
A
V
A A
V
REF REF REF IN
+

+

=

So, for a given output code, we can write a corresponding expression for
IN
V and use it to
generate a table of values for the two cases presented in the problem statement. We can then
calculate the INL and DNL values for both cases to see how they compare.

The output code values and corresponding input voltages for both cases are shown below. The
transfer curve shows that the second case (A
1
= 2.0, A
2
= 2.2) is more optimal (closer to ideal
curve) than the first case (A
1
= 2.2, A
2
= 2.0). The INL and DNL values are also better for the
second case when compared to the first case. We can then conclude that the first stage accuracy
is the most critical.

+ +

+

=
2 1
1
0
2 1
2
1
2
1
2
1
D V D V
A
D V
A A
V
REF REF REF IN
Residue



















INL
-180.000
-160.000
-140.000
-120.000
-100.000
-80.000
-60.000
-40.000
-20.000
0.000
0 1 2 3 4 5 6 7
Output Code
INL,case1 (mV)
INL,case2 (mV)

DNL
-100.000
-50.000
0.000
50.000
100.000
150.000
200.000
1 2 3 4 5 6 7
Output Code
DNL,case1 (mV)
DNL,case2 (mV)

Transfer Curve
0
1
2
3
4
5
6
7
8
0.0000 1.0000 2.0000 3.0000 4.0000 5.0000
O
u
t
p
u
t

C
o
d
e
Vin,case1 (V) Vin,case2 (V) Vin,ideal (V)

QAWI HARVARD
29.27 An 8-bit single-slope ADC with a 5V reference is used to convert a slow-moving
analog signal. What is the maximum conversion time assuming that the clock
frequency is 1 MHz? What is the maximum frequency of the analog signal? What
is the maximum value of the analog signal which can be converted?

Figure 29.32 depicts a block diagram of a single-slope integrating ADC.











As discussed in section 29.2.4 the single-slope integrating ADC operates by placing a
negative (with respect to the input) voltage on the inverting side of the comparator and
the sampled input on the other input of the comparator. The counter is used to count how
many clock cycles it takes to integrate the reference current before the V
C
voltage exceeds
the sampled input voltage. When the comparator switches states the count value is
latched and the integrator is reset. Figure 29.33 shows one conversion cycle.












The digital output code will count from 00000000 (0) to 11111111 (255). The minimum
code refers to the minimum voltage and the maximum count refers to the maximum input
voltage which is equal to V
REF
1 LSB. Using equation 29.79 we can say:


t
c
=
v
in
V
REF
2
N
T
CLK



t
c
=
V
REF
1
1
2
N ( )
V
REF
2
N

1
f
CK
=
2
N
1
f
CK
=
255
1MHz
= 255s

QAWI HARVARD
The equation above shows that the full sampling time of the ADC is 255 s. Using
Nyquists Criteria we can say that the sampling frequency must be equal to twice the
input frequency, therefore we can say:


1
255s
> 2 f
in



f
in
2kHz

The maximum value of the input signal is simply equal to V
REF
1 LSB or ~4.98 V,
while the minimum voltage is 0 V.
29.28 An 8-bit single-slope ADC with a 5V reference uses a clock frequency of 1MHz.
Assuming that all of the other components are ideal, what is the limitation on the
value of RC? What is the tolerance of the clock frequency which will ensure less
than 0.5 LSB of INL?



We want the V
C
node to fully charge within a single clock cycle. From Eq. (29.81), we
can find the value of RC for a full-scale analog voltage of 5V, assuming that max input
voltage is equal to the reference voltage.


RC =
VREF t
c
V
C
=
5V
5V f
c
=1us
RC 1us


To find the clock frequency of the clock jitter, we start with finding that 0.5 LSB is 0.5 *
5V / 2
8
, or 9.77mV. Using Eq. (29.81) again we can find the t that is equivalent to this
error.


t =
RC V
C
VREF
=
1us 9.77mV
5V
=1.95ns

So the clock period must be 1us +/- 1.95ns, or 1.00195MHz f
CLK
.99805MHz.
Justin Wood
29.29! An 8-bit dual slope ADC with a 5V reference is used to convert the same
analog signal in Problem 29.27. What is the maximum conversion time assuming
that the clock frequency is 1 Mhz? What is the minimum conversion time that can
be attained? If the analog signal is 2.5V, what will be the total conversion time?
The analog signal from 29.27 was dened as a slow-moving signal. Therefore, the
discussion on pages 998 -1002 is valid (ie. fCLK >> fIN). Since we are dealing with a
dual-slope topology, the total conversion time is the sum of the charging time (T1) and
the discharging time (T2).
T1 is xed and always takes 2
N
*tCLK before the counter overows. For our 8-bit ADC
with 1Mhz clock frequency, T1 is equal to 256s. The discharge time can then be
calculated from equation 29.86 shown below:
v
in
T
1
= V
REF
T
2
The maximum conversion time occurs when vin equals -VREF. The discharge time, T2, is
then equal to the charging time, T1. The conversion time is then 2T1 or 512s.
The minimum conversion time is achieved when vin is equal to 0. The conversion time
is then just T1 or 256s (T2=0).
If the analog signal is 2.5V, the conversion time is (knowing T1=256s):
v
in
T
1
= V
REF
T
2
2.5 256s = 5 T
2
T
2
= 128s
T
TOTAL
= T
1
+ T
2
= 384s
Lincoln Bollschweiler
29.30 Discuss the advantages and disadvantages of using a dual-slope versus a single-
slope ADC architecture.

One of the more obvious disadvantages to the dual slope is the fact that the required
charging period adds time to the conversion over the single-slope. For a full scale signal
this can mean twice the time (2
N
clocks up, 2
N

clocks down) . Another disadvantage is the
use of additional circuit components which mean a larger layout area required.

An advantage of the dual-slope converter is the fact that the same integrator and clock are
used to produce both slopes. This means that any non-idealities inherent to the converter
will be added to each conversion (slope) equally and, ideally, cancel out. This makes for
a much more accurate converter than the single-slope, making design specifications less
stringent (lower-power, smaller devices, etc).
Sakkarapani Balagopal
(P.29.31)Repeat Ex. 29.15 for a 4-bit successive approximation ADC using V
REF
=5V for V
IN
=1,
3 and full-scale.
Let D
B
3
, D
B
2
, D
B
1
and D
B
0
are the initial outputs of SAR before the comparator makes its
decision. B
3
, B
2
, B
1
and B
0
are the inputs of SAR. D
3
, D
2
, D
1
and D
0
are the final output of the
SAR ADC after comparator decision. V
OUT
is the output of 4-bit DAC and C
OUT
is the output of
comparator.
When V
REF
= 5V, V
IN
= 1V,
STEP B
3
B
2
B
1
B
0
D
B
3
D
B
2
D
B
1
D
B
0
V
OUT
(V) C
OUT
D
3
D
2
D
1
D
0

1 1 0 0 0 1 0 0 0 2.5 1
0 0 0 0
2 0 1 0 0 0 1 0 0 1.25 1
0 0 0 0
3 0 0 1 0 0 0 1 0 0.625 0
0 0 1 0
4 0 0 0 1 0 0 1 1 0.9375 0
0 0 1 1

4-bit ADC output = 0 0 1 1

When V
REF
= 5V, V
IN
= 3V,
STEP B
3
B
2
B
1
B
0
D
B
3
D
B
2
D
B
1
D
B
0
V
OUT
(V) C
OUT
D
3
D
2
D
1
D
0

1 1 0 0 0 1 0 0 0 2.5 0
1 0 0 0
2 0 1 0 0 1 1 0 0 3.75 1
1 0 0 0
3 0 0 1 0 1 0 1 0 3.125 1
1 0 0 0
4 0 0 0 1 1 0 0 1 2.8125 0
1 0 0 1

4-bit ADC output = 1 0 0 1

When V
REF
= 5V, V
IN
= 5V,
STEP B
3
B
2
B
1
B
0
D
B
3
D
B
2
D
B
1
D
B
0
V
OUT
(V) C
OUT
D
3
D
2
D
1
D
0

1 1 0 0 0 1 0 0 0 2.5 0
1 0 0 0
2 0 1 0 0 1 1 0 0 3.75 0
1 1 0 0
3 0 0 1 0 1 1 1 0 4.375 0
1 1 1 0
4 0 0 0 1 1 1 1 1 4.6875 0
1 1 1 1

4-bit ADC output = 1 1 1 1


Chris Gagliano

29.32 Assume that v
in
= 2.49 V for the ADC used in Problem 29.31 and that the comparator,
because of its offset, makes the wrong decision for the MSB conversion. What will be the
final digital output? Repeat for v
in
= 0.3025 V, assuming that the comparator makes the
wrong decision on the LSB.















As in Example 29.15, we designate D
3
D
2
D
1
D
0
as the initial output of the SAR before the
comparator makes its decision. The final value is designated as D
3
D
2
D
1
D
0
.

For the case where
IN
V = 2.49 V, lets first look at the ideal case where the comparator does not
make any wrong decisions and then look at the case where the comparator makes the wrong
decision on the MSB.














The final output is 1000.

We can see that the final output is 1000 when the comparator makes the wrong decision on the
MSB whereas the final output should be 0111.




Next, lets consider the case where
IN
V = 0.3025 V and look at the results when the comparator
does not make any wrong decisions compared to when the comparator makes a wrong decision
on the LSB.

















The final output is 0001 when the comparator makes the wrong decision on the LSB whereas the
final output should be 0000 because the input voltage is less than 1 LSB.


QAWI HARVARD
29.33 Design a 3-bit, charge-redistribution ADC similar to that shown in Fig. 29.39 and
determine the voltage on the top plate of the capacitor array throughout the conversion
process for v
IN
= 2, 3, and 4 V, assuming that V
REF
= 5 V. Assume that all components
are ideal. Draw the equivalent circuit for each bit decision.

Figure 29.39 is shown below:


Using Figure 29.39 it is possible to design a 3-bit charge redistribution ADC, the completed
design is seen in the figure below.














F1 3-bit charge redistribution ADC

In order to draw an equivalent circuit for each bit we simply connect each capacitor to V
REF
while
grounding all of the other capacitors. The final 1 pF capacitor remains connected to ground for
each bit.







C C 2C 4C
v
IN
V
REF
SAR
RESET
C = 1 pF
VTOP
QAWI HARVARD












You have to notice that when the conversion begins the MSB capacitor is connected to V
REF
and a
voltage of V
REF
/2 (or 2.5 V) is added to V
TOP
. For a 2 V input this leaves a positive 0.5 V on V
TOP
,
if the capacitance is not reset to v
IN
the comparator output will remain in a low state. Due to this,
the control logic will need to save the state of each bit (connected to V
REF
if the comparator output
stays high, or connect to GND if the comparator output goes low), reset the top voltage, and
continue the conversion process with the next bit. In an attempt to understand this the tables
below show the conversion process for an input voltage of 2 V, 3 V, and 4 V.

For 2 V input:

Step D
2
D
1
D
0
V
TOP
(V) Comparator Out
--- --- - 2

---
1 100 - 2 + 2.5 = 0.5 Low (reset)
2 010 -2 + 1.25 = -0.75 High (save)
3 011 -2 + 1.25 + 0.625 = -0.125 High (save)

Output code = 011

For 3 V input:

Step D
2
D
1
D
0
V
TOP
(V) Comparator Out
--- --- - 3 ---
1 100 - 3 + 2.5 = -0.5 High (save)
2 110 -3 + 2.5 + 1.25 = 0.75 Low (reset)
3 101 -3 + 2.5 + 0.625 = 0.125 Low (reset)

Output code = 100

For 4 V input:

Step D
2
D
1
D
0
V
TOP
(V) Comparator Out
--- --- - 4 ---
1 100 - 4 + 2.5 = -1.5 High (save)
2 110 -4 + 2.5 + 1.25 = -0.25 High (save)
3 111 -4 + 2.5 +1.25 + 0.625 = 0.375 Low (reset)

Output code = 110
4C
4C
V
REF
6C
2C
V
REF
7C
C
V
REF

D
2


D
1


D
0

Adam Johnson
29.34 Determine the maximum INL and maximum DNL of the ADC designed in Problem
29.33 assuming that the capacitor array matching is 1%. Assume that the
remaining components are ideal and that the unit capacitance, C, is 1pF.

Problem 29.33 is a 3-bit charge distribution ADC like Figure 29.39



From Eq. (29.91) we get the INL for

C
C
= 1%:


| INL |
max
=
V
REF
2

C
C
= 25mV = 0.04LSB

From Eq. (29.93) we get the DNL:


DNL
max
=
(2
N
1) V
REF
2
N

| C |
C
= 43.75mV = 0.07LSB

Justin Wood
29.35! Show that the charge redistribution ADC used in Problems 29.32 and 29.33
is immune to comparator offset by assuming an initial offset voltage of 0.3 and
determining the conversion for VIN=2V.
With VOS = 0.3V and VIN = 2V, the conversion process for the charge redistribution ADC
would be as follows (using equations 29.88, 29.89, and 29.90):
Step D2D1D0 VTOP(V) Comparator Output
-- -- -2V --
1 000 -2 + Vref/2 +0.3 = 0.8 0
2 010 -2 + Vref/4 + 0.3 = -0.45V 1
3 011 -2 + Vref/4 + Vref/8 + 0.3 = 0.175V 1
The key takeaway here is that with a 0.3V comparator offset, the comparator output
changes at Vtop > 0.3V instead of GND. Therefore, even though in step 3 VTOP is
positive, the comparator output remains high because it is not > 0.3V. You can see that
the comparator output is the same as shown in 29.33 where no offset was present.
Justin Wood
29.36! Discuss the differences between Nyquist rate ADCs and oversampling
ADCs.
Nyquist rate ADCs, as the name implies, sample the analog input signal at 2X the input
bandwidth. Oversampling ADCs sample the input at a rate much faster than the Nyquist
rate. The higher sampling frequency of the oversampling ADCs results in some distinct
advantages over the Nyquist rate ADCs.
In general, the oversampling ADCs end up providing a simplied implementation.
Figure 29.41 (page 1008) shown below shows the block diagram for the two ADCs.
As can be seen in the gure, the oversampling ADC doesnt need a dedicated S/H or
quantizer. The modulator takes care of the quantizing and the implementation usually
includes switched-capacitor circuits, which eliminates the need for a S/H.
The oversampling ADC requires more sampling time and has a lower throughput than
the Nyquist rate ADC. Because the sampling rate is much greater than the input
bandwidth, the frequency spectra are spaced much farther apart and there is little or no
aliasing. This results in a much simpler anti-aliasing lter if one is needed at all.
The one drawback of the oversampling ADC is the throughput. The oversampling ADC
requires more sampling time have a lower throughput than the Nyquist rate ADCs.
Additional discussion is available in section 29.2.6.
Adam Johnson
29.37 Write a simple computer program or use a math program to perform the analysis
shown in Ex. 29.16. Run the program for k = 200 clock cycles and show that the
average value of v
q
(kT) converges to the correct answer. How many clock cycles
will it take to obtain an average value if v
q
(kT) stays within 8-bit accuracy of the
ideal value of 0.4V? 12-bit accuracy? 16-bit accuracy?

Ex. 29.16: Using a general first order modulator, assume that the input to the
modulator v
x
(kT) is a positive DC voltage of 0.4 V. Show the values of each variable
around the modulator loop and prove that the overall average output of the DAC
approaches 0.4V after 10 cycles. Assume that the DAC output is at +/- 1 V, and that the
integrator output has a unity gain with an initial output voltage of 0.1 V, and that the
comparator output is either +/-1 V.




v
a
(kT)
Perl code:
#!/usr/bin/perl

# User changable parameters
$vref = 1.0;
$loops = 200;
$vx = 0.4; # DC input value

# Variables of interest
$k = 0; # iterator
$va = 0; # integrator input
$vu = 0; # integrator output
$va_last = 0; # previous integrator input
$vu_last = 0.1; # previous integrator output
$vq = 0; # feedback/output
$vq_sum = 0; # feedback/output sum

printf("k\t va\t vu\t vq\t Qe\tvq(avg)\tQe(avg)\n");
for ($k=0; $k<=$loops; $k++) {
$vu = $vu_last + $va_last;
if ($vu <= 0) {
$vq = - $vref;
} else {
$vq = $vref;
}
$va = $vx - $vq;
$vq_sum = $vq_sum + $vq;
printf("%03d\t%6.2f\t%6.2f\t%6.2f\t%6.2f\t%7.3f\t%9.6f\n",
$k, $va, $vu, $vq, $vq - $vu, $vq_sum/($k+1), ($vq_sum/($k+1))-$vx);
$va_last = $va;
$vu_last = $vu;

}

200 Loops:
k va vu vq Qe vq(avg) Qe(avg)
000 -0.60 0.10 1.00 0.90 1.000 0.600000
001 1.40 -0.50 -1.00 -0.50 0.000 -0.400000
002 -0.60 0.90 1.00 0.10 0.333 -0.066667
003 -0.60 0.30 1.00 0.70 0.500 0.100000
004 1.40 -0.30 -1.00 -0.70 0.200 -0.200000
005 -0.60 1.10 1.00 -0.10 0.333 -0.066667
006 -0.60 0.50 1.00 0.50 0.429 0.028571
007 1.40 -0.10 -1.00 -0.90 0.250 -0.150000
008 -0.60 1.30 1.00 -0.30 0.333 -0.066667
009 -0.60 0.70 1.00 0.30 0.400 0.000000
010 -0.60 0.10 1.00 0.90 0.455 0.054545
011 1.40 -0.50 -1.00 -0.50 0.333 -0.066667
012 -0.60 0.90 1.00 0.10 0.385 -0.015385
013 -0.60 0.30 1.00 0.70 0.429 0.028571
014 1.40 -0.30 -1.00 -0.70 0.333 -0.066667
015 -0.60 1.10 1.00 -0.10 0.375 -0.025000
.
.
185 -0.60 1.10 1.00 -0.10 0.398 -0.002151
186 -0.60 0.50 1.00 0.50 0.401 0.001070
187 1.40 -0.10 -1.00 -0.90 0.394 -0.006383
188 -0.60 1.30 1.00 -0.30 0.397 -0.003175
189 -0.60 0.70 1.00 0.30 0.400 0.000000
190 -0.60 0.10 1.00 0.90 0.403 0.003141
191 1.40 -0.50 -1.00 -0.50 0.396 -0.004167
192 -0.60 0.90 1.00 0.10 0.399 -0.001036
193 -0.60 0.30 1.00 0.70 0.402 0.002062
194 1.40 -0.30 -1.00 -0.70 0.395 -0.005128
195 -0.60 1.10 1.00 -0.10 0.398 -0.002041
196 -0.60 0.50 1.00 0.50 0.401 0.001015
197 1.40 -0.10 -1.00 -0.90 0.394 -0.006061
198 -0.60 1.30 1.00 -0.30 0.397 -0.003015
199 -0.60 0.70 1.00 0.30 0.400 0.000000
200 -0.60 0.10 1.00 0.90 0.403 0.002985

8-bit accuracy would require Qe(avg) < 1/2LSB or 0.5*V
REF
/2
8
, or Qe(avg) < 1.95mV.
by modifying the code above to only print out errors greater than this threshold, we can
see that this requires >300 cycles.

(Ran 1000 times)
k va vu vq Qe vq(avg) Qe(avg)
200 -0.60 0.10 1.00 0.90 0.403 0.002985
203 -0.60 0.30 1.00 0.70 0.402 0.001961
210 -0.60 0.10 1.00 0.90 0.403 0.002844
220 -0.60 0.10 1.00 0.90 0.403 0.002715
230 -0.60 0.10 1.00 0.90 0.403 0.002597
240 -0.60 0.10 1.00 0.90 0.402 0.002490
250 -0.60 0.10 1.00 0.90 0.402 0.002390
260 -0.60 0.10 1.00 0.90 0.402 0.002299
270 -0.60 0.10 1.00 0.90 0.402 0.002214
280 -0.60 0.10 1.00 0.90 0.402 0.002135
290 -0.60 0.10 1.00 0.90 0.402 0.002062
300 -0.60 0.10 1.00 0.90 0.402 0.001993

12-bit accuracy would require Qe(avg) < 122uV, which requires >4910 cycles.

(Ran 10000 times)
k va vu vq Qe vq(avg) Qe(avg)
4770 -0.60 0.10 1.00 0.90 0.400 0.000126
4780 -0.60 0.10 1.00 0.90 0.400 0.000125
4790 -0.60 0.10 1.00 0.90 0.400 0.000125
4800 -0.60 0.10 1.00 0.90 0.400 0.000125
4810 -0.60 0.10 1.00 0.90 0.400 0.000125
4820 -0.60 0.10 1.00 0.90 0.400 0.000124
4830 -0.60 0.10 1.00 0.90 0.400 0.000124
4840 -0.60 0.10 1.00 0.90 0.400 0.000124
4850 -0.60 0.10 1.00 0.90 0.400 0.000124
4860 -0.60 0.10 1.00 0.90 0.400 0.000123
4870 -0.60 0.10 1.00 0.90 0.400 0.000123
4880 -0.60 0.10 1.00 0.90 0.400 0.000123
4890 -0.60 0.10 1.00 0.90 0.400 0.000123
4900 -0.60 0.10 1.00 0.90 0.400 0.000122
4910 -0.60 0.10 1.00 0.90 0.400 0.000122


16-bit accuracy would require Qe(avg) < 7.63uV, which requires >78630 cycles.

(Ran 100000 times)
k va vu vq Qe vq(avg) Qe(avg)
78530 -0.60 0.10 1.00 0.90 0.400 0.000007640
78540 -0.60 0.10 1.00 0.90 0.400 0.000007639
78550 -0.60 0.10 1.00 0.90 0.400 0.000007638
78560 -0.60 0.10 1.00 0.90 0.400 0.000007637
78570 -0.60 0.10 1.00 0.90 0.400 0.000007636
78580 -0.60 0.10 1.00 0.90 0.400 0.000007635
78590 -0.60 0.10 1.00 0.90 0.400 0.000007634
78600 -0.60 0.10 1.00 0.90 0.400 0.000007633
78610 -0.60 0.10 1.00 0.90 0.400 0.000007633
78620 -0.60 0.10 1.00 0.90 0.400 0.000007632
78630 -0.60 0.10 1.00 0.90 0.400 0.000007631



QAWI HARVARD
29.38 Prove that the output of the second-order

modulator shown in Fig. 29.49 is,


y kT
( )
= x kT T
( )
+ Q
e
kT
( )
2Q
e
kT T
( )
+ Q
e
kT 2T
( )



In order to prove that the above equation is that of the second-order sigma delta modulator we
first start by determining the intermediate voltages shown above:


u
1
kT
( )
= x kT
( )
y kT
( )
+ u
1
kT T
( )



u
2
kT
( )
= u
1
kT T
( )
y kT T
( )
+ u
2
kT T
( )


The equations above have taken

q kT
( )
= y kT
( )
for an ideal 1-bit DAC. We can plug

u
1
kT
( )

into

u
2
kT
( )
and get:


u
2
kT
( )
= x kT T
( )
y kT T
( )
+ u
1
kT 2T
( )
y kT T
( )
+ u
2
kT T
( )


The quantization noise of the ADC can be written as:


Q
e
kT
( )
= y kT
( )
u
2
kT
( )


We can use the quantization noise equation above in two places. First we can rewrite

u
2
kT
( )
with
the quantization noise, and then we can plug

u
2
kT
( )
into the quantization noise equation to solve
for

y kT
( )
.


u
2
kT
( )
= x kT T
( )
y kT T
( )
+ u
1
kT 2T
( )
Q
e
kT T
( )



y kT
( )
= Q
e
kT
( )
+ u
2
kT
( )



y kT
( )
= Q
e
kT
( )
+ x kT T
( )
y kT T
( )
+ u
1
kT 2T
( )
Q
e
kT T
( )


The

u
1
kT 2T
( )
term is the only term that will not be in our final equation. In order to
change this term we can use the

u
2
kT
( )
equation:


u
1
kT ( )


Q
e
kT T ( )


u
2
kT ( )

QAWI HARVARD

u
2
kT
( )
= u
1
kT T
( )
y kT T
( )
+ u
2
kT T
( )


Solving for

u
1
kT T
( )
:


u
1
kT T
( )
= u
2
kT
( )
+ y kT T
( )
u
2
kT T
( )


Advancing time by one clock cycle gives:


u
1
kT 2T
( )
= u
2
kT T
( )
+ y kT 2T
( )
u
2
kT 2T
( )


Plugging this into our

y kT
( )
equation gives:


y kT
( )
= Q
e
kT
( )
+ x kT T
( )
y kT T
( )
+ u
2
kT T
( )
+ y kT 2T
( )
u
2
kT 2T
( )
Q
e
kT T
( )


Rearranging:


y kT
( )
= x kT T
( )
+ Q
e
kT
( )
y kT T
( )
+ u
2
kT T
( )
Q
e
kT T
( )
+ y kT 2T
( )
u
2
kT 2T
( )

Using the quantization equation gives:


y kT ( ) = x kT T ( ) + Q
e
kT ( ) 2Q
e
kT T ( ) + Q
e
kT + 2T ( )


Q
e
kT 2T ( )


Q
e
kT T ( )

Chris Gagliano

29.39 Assume that a first order ADC used on a satellite in a low earth orbit experiences
radiation in which an energetic particle causes a noise spike resulting in the comparator
making the wrong decision on the 10
th
clock period. Using the program written in
Problem 29.37, determine the number of clock cycles required before the average value
of v
q
(kT) is within 12-bit accuracy of the ideal value of 0.4 V. How many extra clock
cycles were required for this case versus the ideal conversion used in Prob. 37?

For this problem, I wrote a Perl script to perform the analysis shown in Ex. 29.16. The script was
used to solve problem 29.37 (referenced by this problem) as well so the comparison can be made
between the ideal case (problem 29.37) and the case in this problem where the comparator makes
the wrong decision on the 10
th
clock period. To determine if the average value of V
q
(kT) stays
within the chosen accuracy, I check that all of the values within a certain window size be within
the chosen accuracy. I somewhat arbitrarily selected a window size of 5 values.

The results are summarized below and the Perl code follows. In short, the wrong decision on
cycle 10 in this problem does not affect the point at which the average value of V
q
(kT) stays
within the chosen accuracy for the chosen window size of 5. However, we can see that the clock
cycle at which the average value of V
q
(kT) initially becomes exactly equal to the ideal value of
0.4 V changes from clock cycle 10 in the ideal case to clock cycle 20.

Running until Avg Vq(kT) is 12-bit accurate (between 0.3997558594 and 0.4002441406) for 5 cycles
k Vq(kT) Vu(kT) Vq(kT) Qe(kT) Avg Vq(kT)
0 -0.600 0.100 1.000 0.900 1.0000000000
1 1.400 -0.500 -1.000 -0.500 0.0000000000
2 -0.600 0.900 1.000 0.100 0.3333333333
3 -0.600 0.300 1.000 0.700 0.5000000000
4 1.400 -0.300 -1.000 -0.700 0.2000000000
5 -0.600 1.100 1.000 -0.100 0.3333333333
6 -0.600 0.500 1.000 0.500 0.4285714286
7 1.400 -0.100 -1.000 -0.900 0.2500000000
8 -0.600 1.300 1.000 -0.300 0.3333333333
9 1.400 0.700 -1.000 -1.700 0.2000000000
10 -0.600 2.100 1.000 -1.100 0.2727272727
11 -0.600 1.500 1.000 -0.500 0.3333333333
12 -0.600 0.900 1.000 0.100 0.3846153846
13 -0.600 0.300 1.000 0.700 0.4285714286
14 1.400 -0.300 -1.000 -0.700 0.3333333333
15 -0.600 1.100 1.000 -0.100 0.3750000000
16 -0.600 0.500 1.000 0.500 0.4117647059
17 1.400 -0.100 -1.000 -0.900 0.3333333333
18 -0.600 1.300 1.000 -0.300 0.3684210526
19 -0.600 0.700 1.000 0.300 0.4000000000
.
.
.

Running until Avg Vq(kT) is 12-bit accurate (between 0.3997558594 and 0.4002441406) for 5 cycles
k Vq(kT) Vu(kT) Vq(kT) Qe(kT) Avg Vq(kT)
0 -0.600 0.100 1.000 0.900 1.0000000000
1 1.400 -0.500 -1.000 -0.500 0.0000000000
2 -0.600 0.900 1.000 0.100 0.3333333333
3 -0.600 0.300 1.000 0.700 0.5000000000
4 1.400 -0.300 -1.000 -0.700 0.2000000000
5 -0.600 1.100 1.000 -0.100 0.3333333333
6 -0.600 0.500 1.000 0.500 0.4285714286
7 1.400 -0.100 -1.000 -0.900 0.2500000000
8 -0.600 1.300 1.000 -0.300 0.3333333333
9 -0.600 0.700 1.000 0.300 0.4000000000
.
.
.

P29.39: Comparator makes
wrong decision on cycle 10
Ideal Case (P29.37)
Initially converges to 0.4 V at cycle 20.
Initially converges to 0.4 V at cycle 10
Overall, the comparator making a wrong decision on cycle 10 does not affect when the average
value of V
q
(kT) finally reaches and stays within the 8-, 12-, or 16-bit accuracy as shown in the
table below.














Perl Code:
# P29.39 - N-bit accurate w/ error on 10th cycle

# Get input argument
if(exists($ARGV[0])) {
$N = $ARGV[0]; # Number of bits of accuracy
}
else {
die "ERROR: Must supply input argument (N). For example: $0 8\n";
}

$output_file = sprintf("P29.39_%d_bit_sim_output.txt", $N);

$Vref = 1.0; # Vref
$Vx = 0.4; # Input voltage is DC 0.4V

$max_delta = $Vref / (2 ** $N); # Vref/(2^N) = 1 LSB
$window = 5; # Number of cycles for which the Average Vq stays within Vx +/-
max_delta

@Va = ();
@Vu = ();
@Vq = ();
@Qe = ();
@Avg_Vq = ();

$Vu[0] = 0.1; # Initial condition. Integrator output = 0.1V.

$sum = 0;
$k = 0;
$converged = 0;
while(!$converged)
{
if($k > 0) {
$Vu[$k] = $Vu[$k-1] + $Va[$k-1]; # Eq (29.102)
}
if($Vu[$k] > 0) {
$Vq[$k] = $Vref;
}
else {
$Vq[$k] = -$Vref;
}

# Problem 29.39: The comparator makes a wrong decision on the 10th cycle
if($k == 9) {
$Vq[$k] = -$Vq[$k]; # Invert the decision made on the 10th clock cycle
}

$Va[$k] = $Vx - $Vq[$k]; # Eq (29.103)

$Qe[$k] = $Vq[$k] - $Vu[$k]; # Eq (29.104)

# Calculate Average Vq
$sum += $Vq[$k];
$Avg_Vq[$k] = $sum / ($k+1);

# Check to see if we have converged to Vx +/- max_delta for the given window size
if(($k+1) >= $window) {
$converged = 1; # Start off optimistic
for($i=0; $i < $window; $i++) {
if(($Avg_Vq[$k-$i] < ($Vx - $max_delta)) || ($Avg_Vq[$k-$i] > ($Vx + $max_delta))) {
$converged = 0; # Avg Vq is outside of Vx +/- max_delta, so keep going
}
}
}

$k++;
}

$num_clocks = $k;

# Print results (to screen and output file)
open(OUTPUT, ">$output_file");
printf( "Running until Avg Vq(kT) is %d-bit accurate (between %.10f and %.10f) for %d
cycles\n", $N, ($Vx - $max_delta), ($Vx + $max_delta), $window);
printf(OUTPUT "Running until Avg Vq(kT) is %d-bit accurate (between %.10f and %.10f) for %d
cycles\n", $N, ($Vx - $max_delta), ($Vx + $max_delta), $window);
printf( "%4s %10s %10s %10s %10s %15s\n", "k", "Vq(kT)", "Vu(kT)", "Vq(kT)", "Qe(kT)", "Avg
Vq(kT)");
printf(OUTPUT "%4s %10s %10s %10s %10s %15s\n", "k", "Vq(kT)", "Vu(kT)", "Vq(kT)", "Qe(kT)", "Avg
Vq(kT)");
for($k=0; $k < $num_clocks; $k++)
{
printf( "%4d %10.3f %10.3f %10.3f %10.3f %15.10f\n", $k, $Va[$k], $Vu[$k], $Vq[$k],
$Qe[$k], $Avg_Vq[$k]);
printf(OUTPUT "%4d %10.3f %10.3f %10.3f %10.3f %15.10f\n", $k, $Va[$k], $Vu[$k], $Vq[$k],
$Qe[$k], $Avg_Vq[$k]);
}
close(OUTPUT);

exit;