Академический Документы
Профессиональный Документы
Культура Документы
..
.
..
.
bN
..
.
Two-Step+ Subranging
ADC for Ultra Wideband
+
L. Roicado
lroicado@chipidea.com
IST - Technical
University of Lisbon
V
Lisboa,
Portugal
S/H
v
ChipIdea Microelectronica
Porto Salvo, Portugal
REFN
clk
Abstract
vI
111
110
101
100
011
010
001
000
t
TS
The sampling operation performs a time discretization of the input voltage at a frequency fs = 1/Ts ,
but the amplitude of the resulting signal can still
take any value. Thus, the input range of the ADC
is divided into small regions called quantization steps
which have the width VLSB , and the sampled input
signal is approximated by one of these quantization
levels. The last step of the digitalization process - the
encoding - consists of assigning a binary number, the
output code, to each quantization level.
Keywords: Ultra Wideband, High Definition Television, Blu-Ray, Analog-to-Digital Converter, two-step
flash ADCs, Two-Step Subranging Flash ADC
VLSB =
VF S
2N
(1)
1. Introduction
1.1. Basic Concepts
The ADC is an integrated electronic circuit that
converts continuous-time / continuous-amplitude signals (usually voltages) into discrete-time / discreteamplitude signals. This is accomplished in 3 phases,
specifically Sampling, Quantization and Encoding.
Fig. 1 illustrates the digitalization of an analog signal.
|eq (t)|
VLSB
,
2
(2)
and therefore, each bit increased in the ADC resolution leads to a decrease of both VLSB and |eq (t)| to
half of their previous values.
Fase I
(durao constante)
Vin3
Fase II
(declive constante)
Vin2
The Signal-to-Noise Ratio (SNR) is the relation between the signal power and the noise power. The maximum achievable SNR of an ideal ADC is given [4] by
SN R[dB] |M AX = 6.02N + 1.76
(3)
Latched
comparators
..
.
VREFP
..
.
n=2
T HD[DB]
10
+ 10
SN R[DB]
10
VREFN
vI
clk
(6)
(7)
DN L[k] =
VT [k + 1] VT [k] VLSB
=
VLSB
b1
vIS
Clock buffers
In an N -bit flash ADC there are 2N 1 preamplifiers and latched comparators and one resistive
ladder that divides the input range in 2N quantization levels. A sample-and-hold performs the sampling
operation and the pre-amplifiers subtract the sampled
signal from the reference voltages. If the differen-111
tial input voltage of the k th pre-amplifier is positive,110
(vIS VREFk ) > 0, its differential output voltage will101
also be positive and the latched comparator after it will100
regenerate the logic level high (the opposite occurs if011
(vIS VREFk ) < 0). Thus, considering a growing ramp010
input, the outputs of the latched comparators will be
active (high logic level) one after another, from bottom001
to top, resembling the mercury scale of a thermometer.000
Finally, a ROM is usually employed to encode this output to the final binary word. Due to large number of
pre-amplifiers and latched comparators required, the
power dissipated by these ADCs makes them apropriated for resolutions no greater than 8 bit. To overcome
this limitation, two-step flash ADCs are employed and
the necessary hardware and power can be weighty
reduced. The blocks diagram of an 8-bit Two-Step
Flash ADC with DAC (Digital-to-Analog Converter)
and Subtractor is depicted in Fig. 3. A coarse flash
..
.
(5)
The Effective Number Of Bits (ENOB) indicates the
number of bits that an ideal converter would have to
achieve the measured SINAD,
SIN AD[DB] 1.76
EN OB =
.
6.02
S/H
bN
(8)
vIN
VT [k + 1] VT [k]
=
1, k = 1 . . . 2N 2,
VLSB
where VT [k] is the input voltage where the transition from code k 1 to code k occurs. From (8) we can
conclude that when the transition level of two consecutive codes is the same, VT [k + 1] = VT [k], DNL=-1.
This is an extreme case of non-linearity of the ADC.
4bit
MSB
ADC
4bit
DAC
vDAC
_+
eq
16
4bit
LSB
ADC
Amplifier
4 least
significant bits
4 most
significant bits
vI
Besides reducing the linearity requirements of its components, this architecture also reduces the necessary
hardware and power dissipation with respect to the
flash ADC.
VREFP
Power
MSB0
Bus
sw12=MSB0
subrange 1
875 mV
Wide Band
CDMA (5 MHz)
sw11=MSB0
comp2LSB
750 mV
500 mV
subrange 0
sw02=MSB0
MSB0
Encoder
Full range
sw10=MSB0
comp1LSB
b2
T1
b1
comp0LSB
According to [2], [7] and [3], besides the large bandwidth UWB typically requires ADCs with moderate
resolutions from 4 to 6 bit. Therefore, the two-step
subranging ADC presented before seems to be the
natural choice to equip UWB devices with. Although
it is intrinsically slower than the flash ADC, it uses
less comparators to obtain the same resolution, which
hopefully results in smaller power dissipation and
area. These are key factors when we think of portable
devices wich generally work on batteries, thus, having
limited energy resources.
VREFN
vI
Frequency
sw01=MSB0
sw00=MSB0
41.3
dBm/MHz
b3
S/H
2. General Architecture
The ADC has been implemented in a 65nm CMOS
technology and shall be able to operate between -40
and 125 . Its top-level architecture is shown in Fig. 6.
There is a 2.5 bit Coarse ADC (CADC) that makes a
rough estimation of the input signal position and quantizes the MSBs. Then, the set of reference voltages that
are closer to the input signal are selected and applied
to the Fine ADC (FADC), which quantizes the LSBs.
This architecture uses less comparators but it is intrinsically slower than the Flash ADC: after sampling the
input signal, the coarse ADC must quantize the MSBs
and the fine ADC has to wait for the settling of the reference voltages in order to correcty obtain the LSBs.
3
vI
VCM
ph1+ph4
clk
VRP
Reference Ladder
Fine ADC A
(4 bit)
Coarse ADC
(2.5 bit)
Fine ADC B
(4 bit)
clk
ph1+ph4
vIP
clk
ph2
CS
vIN
b5 ... b0
Qz
clk
clk
ph2+ph3
clk
VRN
clk
VCM
ph2+ph3
ph4
ph2
ph3
ph4
ph1
Coarse
ADC
Samples
vI[1]
Makes decision
about
vI[1]
Samples
vI[2]
Makes decision
about
vI[2]
Samples
vI[3]
Fine
ADC A
Samples
vI[1]
Regenerates
offset voltage
Connects input
capacitors to
reference
voltages
Makes decision
about
vI[1]
Samples
vI[3]
Fine
ADC B
Connects input
capacitors to
reference
voltages
Makes decision
about
vI[0]
Samples
vI[2]
Regenerates
offset voltage
Connects input
capacitors to
reference
voltages
clk
0.5
1st sample
quantized
1.5
2.5 t[ns]
2nd sample
quantized
3. Coarse ADC
The coarse ADC has a resolution of 2.5 bit, defining 7 different quantization levels by the usage of 6
comparators. The 0.5 bit extra bit is to be used as redundancy, which will be discussed in section 3.1. Fig.
8 looks inside one of those 6 comparators. The Coarse
ADC works with 2 banks of dynamic comparators in
time-interleave. This means that each capacitive network is connected to the 2 comparators, but only one of
them is sampling the input at the same time, and only
4
ph1
vIN
VREFsN
ph3
Selection according
to the MSBs
VR2
CompMSB2
vI
CompLSB15
CompLSB14
CompLSB13
CompLSB12
CompLSB11
CompLSB10
CompLSB9
CompLSB8
CompLSB7
CompLSB6
CompLSB5
CompLSB4
CompLSB3
CompLSB2
CompLSB1
VR1
CompMSB1
CCAL
are shorted and it amplifies its own offset voltage. During ph2 the FADC A is waiting for the CADC decision
and, thus, could be idle. Instead, a calibration scheme
was implemented in this phase and will be presented
in section 4.1. In ph3, the reference voltages indicated
by the coarse ADC are applied (through one of the
VREF s P switch plus one VREF s N switch) to the sampling capacitor CS , so that (vCP vCN ) is now proportional to the difference between the input signal and
the corresponding reference voltage.
4.1. Calibration
VOS1
S1
vI
gm1
S2
i1
vOA
VOSl
+
-
CP
+
-
CompMSB3
R0
VOS2
if CompMSB2
decides low
VMAX
VMIN
meaning
min. abs. value of VOS
max. abs. value of VOS
average value of VOS
standard deviation of VOS
S3
vCAL
gm2
i2
S5
CP
CS
Selection
Logic
+
-
S4
if CompMSB2
decides high
unit
The latched comparator makes a decision based on
mV
mV the preamplifier output voltage at the end of ph1. The
mV result is determined by the offset voltage of the commV plete comparator chain, and is used to activate the
control logic that adjusts the calibration voltage at the
input of an auxiliary differential pair. This is made by
pre-charging CP to either VMAX or VMIN, depending on the comparators decision, and then switching
4. Fine ADC
CP to CCAL. CP is not an explicit capacitance, it is
The 4 LSBs are obtained from a Fine ADC that
a parasitic one, thus, CCAL can be made much larger
has 2 sub-Fine ADCs working in time-interleave: the
(CCAL>>CP) and the calibration voltage is adjusted
FADC A and the FADC B. In this way it is possible to
in small steps. After some hundreds of clock cycles, the
have an overall sampling frequency of 1 GHz although
auxiliary differential pair injects a current that cancels
both FADC A and FADC B sample the input signal
the offset voltages of both the pre-amplifier and latched
only at each 2 ns. Fig. 10 depicts one of the comparacomparator. This is a process that is always running
tors used in the FADC A (FADC B is equal, except for
in background during ph2 and does not affect the basic
the timming).
functions of the ADC. Fig. 12 shows the response of
the ADC to a ramp input.
Preamplifier
Input Capacitive Network
At the beginning the calibration is OFF and output
codes of the ADC are far from having a monotonic evolution. After 128 clock cycles (128 ns) the calibration
Auxiliary Differential Pair
is switched ON and we can see that the results arent
instantly better: it takes a while to vCAL to rise and
bias the auxialiary differential pair. In the end we have
an output characteristic of the ADC much linear as a
result of this calibration scheme.
value
0.4
19.6
-0.9
6.3
VDD
VREFsP
ph3
vIP
vIN
R0
VCM
ph4
S1a
ph1
S2
clk
vCP
ph1
ph1
CS
M1
clk
M2
VMAX
vCN
IB
VCALP
VMIN
VCALN
VMIN
VMAX
S1b
ph4
VREFsN
ph3
Latched
Comparator
R0
CP
CCAL
CCAL
CP
VCM
Selection
Logic
Selection according
to the MSBs
gm1
S2
i1
vOA
VOSl
+
-
+
-
S1
vI
R0
VOS2
VMIN
+
-
S4
VMAX
S3
vCAL
gm2
i2
S5
CP
CS
Selection
Logic
CALN
VMIN
VCM
vIN
CompLSB15
CompLSB14
CompLSB13
CompLSB12
CompLSB11
CompLSB10
CompLSB9
CompLSB8
CompLSB7
CompLSB6
CompLSB5
CompLSB4
CompLSB3
CompLSB2
CompLSB1
IB
S1b
ph4
VR3
CALP
vCN
CCAL
CP
64
and two 500 MHz clock signals in quadrature, as depicted in Fig. 14. This is a totally digital block, made
56
T=1 ns
OutputCode
48
40
clk
32
clk_half
2T=2 ns
24
clkf_half
16
50
100
150
200
250
300
ClockCycles
calibrationOFF
350
400
450
500
mostly with latches, shift registers and strings of inverters to introduce delays in the signals path.
calibrationON
input15
input14
input13
input12
input11
input10
input9
input8
input7
input6
input5
input4
input3
1.2
input2
input1
1.2
input0
1.2
[lsb3]
6.7u
[lsb2]
6.03u
[lsb1]
1.2
[lsb0]
1.2
volt (lin)
volt (lin)
volt (lin)
(lin) volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
volt (lin)
2n
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
1.2
0.6
0
4n
6n
8n
10n
12n
14n
6n
8n
2:8.7n
10n
(lin)
12n
3:13.7n
14n
16n
18n
16n
18n
-4n
-9.03n
2n
4n
1:4.7n
5. Phases Generator
This ADC has a top-level block that receives the
external clock signal and generates the 3 clock phases
that control other blocks: one clock signal at 1 GHz
6
400p
600p
800p
1n
1.2n
1.4n
1.6n
1.8n
2n
2.2n
vindiff
2
RT
65.1p
FT
63.4p
1.5
[outDig]
195n
200n
205n
190n
195n
200n
205n
210n
215n
210n
215n
220n
60
volt (lin)
volt (lin)
190n
0.5
0.4
0.3
0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
volt (lin)
ph:avddk=1:avddhvk=12.5
DC 49.9%
50
40
30
20
10
0.5
[outSin]
60
50
(lin)
ph:avddk=1:avddhvk=1
ph:avddk=0.9:avddhvk=1.1
ph:avddk=1.1:avddhvk=0.9
2.5
ph:avddk=0.9:avddhvk=0.9
ph:avddk=1.1:avddhvk=1.1
30
20
10
RT
60p
FT
[outSin]
[outDig]
63.1p
60
50
(lin)
volt (lin)
40
1.5
40
30
20
10
DC 50.7%
0.5
(lin)
0
400p
600p
800p
1n
1.2n
(lin)
1.4n
1.6n
1.8n
2n
220n
Sandwork Design, Inc. (c) 2000-2006
2.2n
Min
40
36
297
294
49.0
Typ
65
63
438
438
49.9
Max
101
104
631
652
52.2
Indicator
THD
SNR
ENOB
Unit
ps
ps
ps
ps
%
Initial
-37.1
29.9
4.7
Final
-38.8
33.3
5.2
Unit
dB
dB
bit
each one supplying different sections, as shown in Table 4. Thus, it was possible to draw the pie charts
showing the obtained results, as depicted in Fig. 18.
The total power dissipation of the ADC is 79 mW and
its overall current consumption is about 46 mA.
Name
vavddref
vdvdd
#
#1
#2
Volt.
1.2
1.2
vavdd
#3
1.2
vavddhv
#4
2.5
Sections
CADC/FADC resistive ladders
digital correction, CADC
and FADCs digital blocks
CADC/FADC analog blocks,
IBias, top-level phases generat.
top/local level phases generat.,
samp. networks input switches
5.25
5.2
5.15
5.1
ENOB(bit)
5.05
5
4.95
4.9
4.85
4.8
4.75
4.7
4.65
27
47
67
87
107 127 147 167
Clockcyclesaftercalibrationstarts
187
207
VDD
27%
#3
VDD
8.5%
#2
VDDHV
57%
#4
VDD
7.5%
#1
I(VDD)
5.6mA
#2
I(VDD)
17.9mA
#3
8. Acknowledgements
I(VDDHV)
18.0mA
#4
I thank Dr. Joao Vital for the excellent opportunity to accomplish this work at ChipIdea, Dr. Pedro
Figueiredo for all the support and transmitted knowledge, Engs. Ana Lopes, Goncalo Minderico and Paulo
Cardoso for their help, and my colleagues for their
friendship and technical discussions at lunch time.
I(VDD)
4.9mA
#1
References
Figure 18: Power dissipated and supply current distribution in the ADC, according to sections presented in
Table 4.
intrinsically slower than the Flash ADC, thus, 2 timeinterleaved Fine ADCs were employed to overcome this
speed limitation. The time constant of the reference
voltages during its settling period is about 60 ps. This
is short enough to ensure a right settling much before
the next sampling phase.
The offset voltage specifications of the Coarse ADC
is relaxed by the usage of redundancy. The maximum
offset voltage allowed to its comparators is 62.5 mV,
and Monte Carlo simulations on the typical PVT corner indicated (VOS )=6.3 mV, which is low enough.
The 2 Fine ADCs have implemented a calibration
mechanism that cancels offset voltages within a range
of 50 mV. Monte Carlo simulations showed that the
offset voltage of these comparators are lower than 5
mV. The calibration was also responsible for an increase of the ENOB from 4.7 bit to 5.2 bit and an
increase of the SNR greater than 3 dB.
To overcome the problem of high resistance of the
sampling networks input switches, a new switch with
a single thick-oxide transistor was studied and implemented. With this new switch, the maximum ON resistance has decreased from 87 k to 1.3 k on the
worst case PVT corner.
To drive these switches a new topology of level converter had to be studied and implemented to transform
1.2 V clock phases into 2.5 V clock phases. Common
level converters introduce great distortion on the duty
cycle at high frequencies. This level converter is ro8