AN9313: Circuit Considerations In Imaging Applications

Circuit Considerations In Imaging Applications
(HA-2546, HA-5020, HA-5033, HI-5700)
Application Note
November 1996
AN9313.1
Introduction
Present day image-processing systems perform many
functions such as low pass filtering and pattern recognition
in the digital domain. However, as shown in the block diagram, the analog input signal must first pass through some
signal conditioning and then be digitized by an A/D before it
can be manipulated by the Digital Signal Processing (DSP).
It is these front end components that will set the overall
dynamic range and resolution of the system and hence the
detail that can be resolved in the image. This note will
describe the considerations involved in designing and testing
the performance of this part of the system.
Video Format
RS-170 is a standard video format for monochrome
television. It was later updated to RS-170A by NTSC to
cover the requirements for color television broadcasting in
the U.S. The FCC has control over broadcast video
standards; but, since imaging processing systems are self
contained, they do not have to follow a particular standard.
For example, color cameras might provide three RGB outputs (component video) or a composite NTSC color signal.
System synchronization schemes could also vary greatly.
controls the field timing and occurs at 1/60 of a second rate.
The brightness information for the video image is transmitted
during the active line time and will vary from the reference
black level (7.5 IRE) to the reference white level (100 IRE).
RS-170 normally has an aspect ratio of 4:3. However,
because of frame buffer memory and DSP requirements
many image-processing applications will use a 1:1 aspect
ratio. Figure 2 depicts the resulting picture and timing requirements for a RS-170 video with a 1:1 aspect ratio. The active
line time is 39.44µs centered with 6.575µs of “inactive” time
on either side. Notice that for 512 active pixels per line and
485 lines one frame of digitized video information will fit into
248,320 of memory.
(1 IRE = 7.41mV)
IRE UNITS
ACTIVE LINE TIME
52.59µs
100
80
60
0.714V
40
A typical RS-170 image, or frame, is made up of two
interlaced fields. The first field scanned represents the odd
numbered lines; the second is the even numbered lines. A
total of 525 lines per frame will be scanned in 1/30 of a
second with 485 lines being visible. The number of active
elements per line, or pixels, varies from system to system
depending on the desired resolution.
The RS-170 monochrome composite video signal is shown in
Figure 1. System timing is controlled by vertical and horizontal
sync pulses. Horizontal sync controls the line by line timing
and occurs during the 10.9µs blanking period. Vertical sync
20
0
0.286V -20
-40
10.9µs
63.49µs
FIGURE 1. STANDARD RS-170 COMPOSITE VIDEO SIGNAL
Imaging System Block Diagram
ANALOG
IN
SIGNAL
CONDITIONING
1
A/D
DSP
D/A
1-888-INTERSIL or 321-724-7143 | Copyright
ANALOG
OUT
© Intersil Corporation 1999
Application Note 9313
52.59µs
39.44µs
525
LINES
485
LINES
ACTIVE VIDEO
FOR 1:1
ASPECT RATIO
RS-170
4:3 ASPECT
RATIO
512
PIXELS
FIGURE 2. MODIFIED RS-170 VIDEO
Circuit Design Considerations
The analog waveforms seen by the signal conditioning front
end to an image-processing system can be classified as
small signal or large signal. The appropriate analysis should
be used in each case. A good rule of thumb is to say any
signal of less then a 1VP-P, like RS-170 video, should be
considered as small signal.
THIS LINE
SCANNED
to frequency. That is, the second harmonic should be delayed
twice as much as the fundamental, the third three times as
much, and so on. When this occurs all the frequency components will end up having the same amount of time delay
resulting in a image that is only delayed slightly in time and
can easily be adjusted for.
For a single pole system, the attenuation factor and phase
shift at a particular frequency relative to the f-3dB can be
calculated from:
1
A ( f ) = ---------------------------------------- f 2
(EQ. 1)
1 +  -----------------
 f – 3dB
and
 f 
θ ( f ) = atan  -----------------
 f – 3dB
(EQ. 2)
Taking these error terms into account, the complete equation
for a sinewave including the effects of the system would now
be:

 f 
A
V ( t ) = ----------------------------------------- × sin  ωt + atan  ----------------- 

 f – 3dB 
 f 2
1 +  -----------------
 f – 3dB
(EQ. 3)
In Equation 1, when f equals 4MHzand the attenuation A(f)
is one 8-bit LSB (0.4%) the required small signal bandwidth
f3dB would be 40MHz. The corresponding phase shift at
4MHz from equation 2 would be 5.7 degrees.
PIXEL CURRENT
OUTPUT
TIME
PIXEL RATE
FIGURE 3. VIDEO
Figure 3 is an example of a worst case light pattern that might
be seen by a small section of a Charge Coupled Device (CCD)
array inside a video camera. Each square represents an individual pixel. The rate at which these pixels are scanned will
determine the bandwidth requirements. If it takes 39.44ms to
scan the 512 active pixels, then the pixel clock rate would be
12,981,174 elements per second. The video signal is a squarewave with a fundamental at half the pixel rate of 12,981,174/2
or 6.4MHz. To pass this signal undistorted would require a
great deal of small signal bandwidth however, a bandwidth of
4MHz has been found to be adequate for video.
Insufficient high frequency response and phase distortion of a
video signal will result in blurring of the fine detail in a picture
and the overall image will look darker than normal. Therefore,
the first requirement for the signal conditioning circuit is that its
small signal bandwidth needs to be considerably wider than the
bandwidth of the incoming video. This will ensure a constant
gain over the frequency band of interest and avoid a loss of
dynamic range as the input to the A/D rolls off.
The second requirement is that the system should have zero
phase shift over its entire frequency range. Because this is
impractical, a realistic goal is a phase shift that is proportional
2
By the time the video has reached the input to the A/D it has
been amplified enough so that large signal parameters such as
slew rate and full power bandwidth (FPBW) have to be considered. The required slew rate can be found by taking a conservative approach and forcing the video signal to slew through its
range in 25% of the pixel clock. Therefore, if the pixel clock is
12.98MHz (pixel time is 77ns) and the reference for the A/D is
4V, then the minimum slew rate required would be:
V
4V
SR MIN = ----------------------------------- = 208 ------µs
0.25 × 77ns
(EQ. 4)
Now that the slew rate has been determined the minimum
required full power bandwidth of the signal conditioning block
and the converter can be calculated from:
SR MIN
208
FPBW = ------------------------ = ----------------------- = 8.3MHz
2×π×V 2×π×4
(EQ. 5)
Trying to relate the above equation for FPBW to the FPBW
quoted for a converter should be done with caution by the
user and requires a knowledge of the way it has been
defined by the A/D manufacturer. One method used will test
for the presence of sparkle codes. These are anomalous
codes that show up when the input slew rate exceeds a
certain value. The term sparkle code comes from the fact
they will cause bright pixels in a video display.
Figure 4 is a plot of a sinewave input to a converter that has
been digitally reconstructed and shows evidence of sparkle
codes. The FPBW is then determined from the maximum
fullscale input frequency that has sparkle free performance.
Application Note 9313
The equation for the total RMS noise over the bandwidth of
interest is:
V
TOT
=
1.57
(EQ. 6)
× f BW ×
2
( V N ) 2 ×  1 + R----2- + ( R 2 ) 2 × ( I NN ) + ( R 3 ) 2 × ( I NP ) 2 ×  1 + R----2-


1
1
R
FIGURE 4. RECONSTRUCTED SINEWAVE
Some manufacturers will quote FPBW as the point in the
frequency domain where the fundamental is 3dB down from
the low frequency value. This method will tend to average
out sparkle codes. Other A/Ds will have the FPBW specified
as the point where a reconstructed sinewave in the time
domain is 3dB down from the low frequency value. The user
must determine if the test method used by the manufacturer
to determine the quoted FPBW will ensure accurate sparkle
free performance at their operating frequencies.
Sparkle codes can also occur if the maximum conversion
rate of the A/D is exceeded. For example, the HI-5700 is an
8-bit CMOS flash converter that has datasheet limit for
conversion rate of 20MHz. However, as is the case for many
flash converters, by skewing the duty cycle of the sampling
clock the part can be made to operate at 25MHz and higher.
The theoretical best dynamic range that can be expected from
an A/D with n bits of resolution can be calculated from:
DR(dB) = 20 log (2n) = 6.02(n). An image-processing system
would be expected to have a dynamic range of at least 36dB
or about 6 bits.
In order to make use of the full dynamic range available from
the converter, the overall system noise should be less than
the theoretical quantization noise of the converter q/(√12) (q
is the A/D lsb size). Figure 5 is a typical voltage feedback or
current feedback op amp circuit that will be used to illustrate
some basic noise calculations. The voltage noise (VN) and
noise current (IN) sources have been modeled, but the
Johnson noise of the resistors will be neglected because of
the low resistor values normally used in high-speed circuits.
R2
1k
INN
R1
1K
VN
-
VOUT
+
R3
500
HA-5020
INP
VN = 4.5nV / √Hz
INN = 25pA / √Hz ; INP = 2.5pA / √Hz
FIGURE 5. OP AMP NOISE MODEL
3
2
R
2
(EQ. 7)
Where:
VTOT is the total RMS noise voltage.
R1 is the feedforward resistor.
R2 is the feedback resistor.
R3 is the noninverting input resistor.
VN is input voltage noise spectral density.
INN is the inverting input current noise spectral density.
INP is the noninverting input current noise spectral
density.
fBW is the bandwidth over the region of interest.
As is normal for current feedback op amps, the HA-5020 has
unequal INN and INP.
For the values given in the figure over the 18MHz FPBW specified for the HI-5700 8-bit flash A/D the RMS noise VTOT is
found to be equal to 106mV. The peak noise value will be about
five times this value or 530mV. This is significantly less than the
4.5mV of quantization noise for an 8-bit ADC with a 4V range.
This example has shown how to model the noise of a
opamp. If there are other noise sources present, then the
total noise can be found by taking the RMS sum of all the
individual noise sources.
A wide dynamic range is usually required of a signal
conditioning block to accommodate large incoming signal
variations. Automatic Gain Control (AGC) will compensate
for these variations and allow the user to design with a lower
resolution A/D.
The AGC circuit should be considered a control loop, and its
frequency and phase characteristics plotted. A slow AGC
loop could compensate for slow offset or gain changes over
temperature while a faster AGC loop could compensate for
signal overload conditions.
There are a number of opinions on where the AGC should be
applied. An easy way to do it is to vary the reference on the A/D
depending on signal strength. This will work fine if the converter
has been thoroughly characterized over the range of reference
voltages it will see. Unfortunately this is usually not the case.
Most datasheets will not specify the performance of the converter versus reference voltage. Therefore, the user is taking a
significant chance that the part performance will stay the same
over the life of the system for various manufacturing lots of the
A/D. The second option is to let the AGC vary the gain of the
signal conditioning circuitry while leaving the reference to the
A/D at the value where the performance is guaranteed by the
datasheet. This approach will guarantee the long term success
of a circuit. The design section of this note will discuss a technique using a multiplier chip to accomplish this.
Application Note 9313
+12V
R16
18
15
14
13
12
11
10
9
8
7
6
5
4
16
DIN1
DIN2
DIN3
DIN4
FROM PC DIN5
DIN6
DIN7
DIN8
DIN9
DIN10
DIN11
DIN12
17
IN1
IN2
IN3
IN4
IN5
IN6
IN7
IN8
IN9
IN10
IN11
IN12
CS
VCC
10K
REF IN
19
VREF
+12V
3
U1
OUT1
RFB
1
2
-
C1
30pF
20
U2
7
6
+
4
HA5177
-12V
WR DGND AGND
AD7545
2
3
+12V
7
+12V
13
16
VX+ VCC
A
12
14
VXB
5
VY+
U3
6
15
C
VY10
8
VZOUT
9
VZ+
ANALOG IN
C2
0.1µF
3
2
R4
1K
R3
7
-12V
ANALOG
OUT
HA5020
-12V
R6
R5
57.2
HA2546
1
-
7 U4
6
4
VEE GND
R1
+
402
R7
402
+12V
R2
1K
15K
15K
-12V
+12V
+5V
+12V
R8
2K
3
7
2 +
D2
ICL8069
-
+C
4
10µF
C5
0.1µF
U5
6
R10
10
R9
C6
33pF
220
4
CA3193
-12V
1
7
26
23
21
18
VREF
(4.43V to 3.76V)
17
8
U7
+5VD
+5VA
+5VA
+5VA
+5VA
C8 0.1µF
5
4
6
28
HA5033
R15
1K
C3
33pF
-12V
CLOCK
C7 10µF
+
8K
R14
3.9K
ANALOG OUT
R11
10
R12
2K
R13
+12V
Q1
2N3904
6
22
9
1
16
VREF+
D7
D6
D5
D4
D3
D2
D1
D0
VIN
1/4R
1/2R
U6
3/4R
CLK
CE1
OVF 14
HI5700
8
19
20
24
25
27
+5VDC 15 CE2
2
3
4
5
10
11
12
13
FIGURE 6. DESIGN FOR VIDEO IMAGING FRONT END
4
DOUT7
DOUT6
DOUT5
DOUT4 TO PC/DSP
DOUT3
DOUT2
DOUT1
DOUT0
OVERFLOW
Application Note 9313
System Design
HA-2546
Figure 6 is a design for a signal conditioning and A/D front end
to an image-processing system. The video input to the system
will be assumed to have a positive picture phase. That is, the
blanking and sync pulses will be the most negative portion of
the video waveform. When the video is ac coupled, the black
reference level has to be reinserted prior to the A/D. If this is not
done, then, as the amplitude of the video signal is reduced, due
to a reduced contrast image, the blanking level moves more
positive. The resulting image will now appear a light shade of
gray, rather than the preferred black level. Also, the DSP
becomes more sensitive to coupled noise and may for example, during edge detect, show an edge where none exists.
HI-5700
OVF
VIN
V
V
Y
X
ADC
µP
DAC
AD7545
Figure 7 is a simple circuit to DC restore the video. This circuit clamps the most negative point of the signal to -0.7V
which can now be offset by the HA-2546 to provide a stable
black level during changing contrast. Another 8-bit 20 MSPS
converter from Intersil Corporation, the HI1176, has an internal circuit which will clamp the back porch of a video signal
to a voltage input on the reference pin.
HA2546
1µf
VIDEO
IN
VY+
+
75
100k
1N914
-12V
FIGURE 7. DC RESTORE CIRCUIT
The HA-2546 is a wideband two quadrant analog multiplier
which makes the implementation of AGC offset and gain
correction easy. It is configured in this design to give the
transfer function:
(V × V )
X
Y
V OUT = ----------------------------- – V Z
2
(EQ. 8)
The VZ pin can be used to correct for system offset as long
as it does not exceed ±5V. The initial offset adjustment is set
by pot R2. The VX pin can be used to adjust system gain.
FIGURE 8. SLOW AGC LOOP
The HA-5177 op amp acts as an I/V converter for the DAC.
Its feedback is set so that at all ones to the DAC the output
voltage will be -2V which is the maximum voltage that is
allowed on the VX pin. At the normal operating point for the
system the DAC will be at midscale and the overall system
gain will result in a full scale swing to the A/D. Since the DAC
is at midscale the system has an equal amount of gain
correction range up and down.
The HA-5020 is a high-speed current feedback op amp
which provides additional gain so that a nominal 1VP-P signal input to the system the HI-5700 flash will see its full 0V to
4V swing. If the sync has been stripped from the video
before it is digitized [8], then the gain could be adjusted so
that the video reference black to reference white level will
span the full range of the converter.
A high-speed unity gain op amp (HA-5033) buffers the input
to the HI-5700 and provides the necessary low output
impedance over frequency required by flash converters.
Although the HA-5020 can drive the HI-5700 directly, the
HA-5033 has superior current drive, lower output
impedance, and better bandwidth.
U5 is part of a reference circuit in Figure 6 that provides the
4V reference required by the A/D and the DAC. It is capable
of 8-bit performance over the industrial temperature range.
Pot R12 will set the initial overall system gain.
The pixel clock of 12.98MHz will usually determine the minimum sampling rate of the A/D. In order to relax the filter
requirements on the front end to the system the actual
sampling rate used in this note is 15MHz. This will be more
than adequate to cover all established sampling rates
specified for the various published standards.
For the reasons outlined above, it was decided to leave the
reference to the flash at its nominal datasheet value and let
the AGC adjust the gain of the signal conditioning
components prior to the converter. A AD7545 12-bit DAC is
used as part of a slow AGC loop which uses the VX pin of
the HA-2546 to control the gain of the system.
Additional timing circuitry might be added to gate the pixel
clock so that it is only on during the active line period thereby
conserving frame buffer memory size. If the system uses an
interlaced video format then the circuitry could also define
the even and odd fields of the image frame and update the
memory accordingly.
As illustrated in Figure 8, the feedback loop could be closed
by a microprocessor using the overflow bit on the HI-5700
and could compensate for light intensity shifts or
temperature drift. In order to avoid any glitches the DAC
should be updated during the vertical retrace period.
The clock period for the HI-5700 8-bit flash is made up of an
autozero time and sample time. It was found that the autozero time can be reduced down to as little as 15ns while the
sampling time must remain at 24ns or greater. This timing
allows the sparkle free operation of the circuit at pixel rates
up to 25MHz.
5
Application Note 9313
There are many considerations which have to be taken into
account when using high speed converters. These involve
board layout, choosing the right op amp to drive the input,
and designing a low drift reference. Refer to references 6
and 7 for a complete discussion of these topics and others.
OUTPUT
CODE
REFERENCE
LINE
7
111
6
110
Test Results
5
101
The IEEE has various standards which address the type of
tests that need to be done on a broadcast video system to
verify the performance of a video A/D and D/A combination
(codec). Among them are DC linearity, Signal-To-Noise Ratio
(SNR), bandwidth, and differential phase and gain. Since
this note deals only with RS-170 monochrome video signals,
the tests that deal with the color information, such as
differential phase and gain, are not applicable. Also, adding
a DAC on the output of the converter in order to use the
IEEE test methods would tend to cloud its overall performance of the system with the errors of the DAC. Therefore,
the system will be evaluated using a set of tests that are similar to those recommended by the IEEE but are done by analyzing the digital data out of the converter. These tests can
also be found on a datasheet for a typical flash A/D. Hopefully, as a result of this approach the user will now also be
able to more intelligently read and compare converter
datasheets.
There is a great deal of information in the low frequency
(30Hz) content of video. Historically, the low frequency
performance of an A/D has been evaluated by the
Differential (DNL) and Integral (INL) NonLinearity specs.
DNL is a measure of the deviation of the code widths from
the ideal value of one Least Significant Bit (LSB). INL is the
deviation of the code edges from the ideal transfer curve of
the A/D. Since the A/D in this system is initially calibrated for
offset and gain, the line used as a reference will be one
drawn through the first and last transition point.
The DNL and INL errors can not be calibrated out and is the
best accuracy that can be expected of the system.
Therefore, the INL error should ideally not exceed 1/2 LSB
so that when it is combined with the inherent 1/2 LSB
quantizing error of an A/D the total error would not exceed 1
LSB. A DNL error of more than -1 LSB means a code is
missing from the transfer curve. An INL error of 1/2 LSB will
ensure a DNL error of at most 1 LSB.
Figure 9 shows a plot for the transfer function of a converter
with DNL and INL errors. The reference curve and the ideal
transitions are pointed out. Transition point 3 is offset in the
negative direction by 1/2 LSB therefore the ILE at this point is
-1/2 LSB. The ILE of all the other transitions is zero. The DLE
of code 2 is -1/2 LSB and the DLE of code 3 is +1/2 LSB.
The actual linearity test was done using a histogram
approach. A triangle wave is input to the system and the
number of occurrences of each code is kept track of. DNL
error is then calculated in LSBs from:
(P (i))
m
DNL ( i ) = ---------------------- – 1
(P (i))
i
(EQ. 9)
6
4
100
3
011
2
010
001
000
IDEAL TRANSITION
1
1
2
3
4
5
6
7
VIN
FIGURE 9. A/D TRANSFER FUNCTION
The ideal probability, Pi is a constant and is equal to the
average of the number of counts per code divided by the
total number of samples. Pm is the measured probability and
is equal to the total number of counts for a particular code
divided by the total number of samples. Once the DNL error
has been determined the INL error is calculated from the
sum of the DLE errors.
A histogram was done on the design discussed in this note
by inputting a 1VP-P 5kHz triangle wave, encoding the
HI5700 at 15MHz, and capturing the digital data. Figure 10
and Figure 11 are plots of the DNL and INL error for the total
system indicating an accuracy of better than 7 bits with no
missing codes.
+0.59 LSB
-0.74 LSB
FIGURE 10. DIFFERENTIAL LINEARITY ERROR vs CODE
Application Note 9313
The Effective Number Of Bits (ENOB) of the system can be
found by:
+0.91 LSB
SINAD dB – 1.76
ENOB = -------------------------------------------------6.02
FIGURE 11. INTEGRAL LINEARITY ERROR vs CODE
Due to various dynamic effects such as slew rate limiting and
bandwidth rolloff the static DNL and INL will degrade as the
input frequency approaches the 4MHz bandwidth requirement
of video. DNL will show up as an increase in the quantization
noise which will tend to elevate the noise floor of the A/D. INL
is a bend in the transfer curve of the converter and will generate harmonics. Both result in a loss of dynamic range of the
system.These effects are usually evaluated in the frequency
domain by finding the SIgnal-to-Noise-And-Distortion (SINAD)
in dB.
The SINAD test requires performing a fourier transform on
the data obtained by sampling a continuous time input
waveform. The Discrete Fourier Transform (DFT) can be
thought of as a frequency selective filter that calculates the
RMS voltage at a particular frequency and will work for any
number of samples.
ENOB is a global indication of the accuracy of the system
and, along with INL and DNL will degrade as the input
frequency is increased. The low DNL and INL errors indicate
the excellent low frequency performance of the design. This
was again verified by inputting a 1VP-P sinewave at 20kHz,
encoding the part at 15MHz, and performing an FFT on the
data. The SINAD was calculated to be 44.5dB for an ENOB
of 7.1 bits. An indication of the overall low noise in the
system.
The high frequency performance of the system was evaluated by changing the input frequency to 4MHz and again
performing an FFT. Figure 12 is a spectrum plot of the
system output. The SINAD for this plot was determined to be
38.2dB for an ENOB of 6.05 bits.
0
DECIBELS
-0.89 LSB
(EQ. 12)
The coefficient for a particular frequency can be found from:
N–1
Xd ( k ) =
∑
x ( n ) × e – j2πk ( n ⁄ N )
-110
(EQ. 10)
FREQUENCY
FIGURE 12. HIGH FREQUENCY SPECTRAL PLOT OF SYSTEM
(fI = 4MHz)
n=0
N is the number of samples.
n is the time sample index (n = 0, 1, 2, ... , N-1).
k is the index for the computed frequency components (k=0,
1, 2,...,N-1).
The Fast Fourier Transform (FFT) is an algorithm that will
compute all the DFT coefficients at one time; but, unlike the
DFT it will only work for sample sizes that are a power of two.
The FFT will output the coefficients for N/2 discrete
frequency bins that will have a resolution of Fsample/N.
The full power bandwidth and slew rate capability of the
system was checked by inputting a fullscale sinewave at
8MHz and sampling it at a 15MHz rate. Figure 13 shows the
resulting reconstructed waveform. Notice the lack of
distortion and sparkle codes.
Once the FFT has been performed SINAD can be calculated
from:
 RMS SIGNAL
SINAD dB = 20 × log  ------------------------------------------
 RMS NOISE 
(EQ. 11)
Where RMSSIGNAL is the measured RMS signal in the
fundamental bin and RMSNOISE is the sum of all other
spectral components below the Nyquist frequency excluding
DC. It is important that the distortion components be
included in this calculation in order to take into account all
the system errors.
7
FIGURE 13. RECONSTRUCTED SINEWAVE (fI = 8MHz)
Application Note 9313
Time Division Multiplexed Systems
This note is mainly concerned with RS-170 type video
signals. However, it is instructive to briefly discuss the factors
to consider when dealing with other time division multiplexed
signals that might be seen from some types of CCD arrays, a
multiplexed input, or an infrared sensor array.
The output of CCD arrays many times will have the signal of
interest riding on a large DC offset. Figure 14 is an example
of an inverting buffer that can be used to remove large
offsets. Notice that since resistor R3 sees a virtual ground
Voffset can take on a value much higher than the supply
voltage.
VIN
R2
R1
R3
VOFFSET
-
Once the circuits have settled then the A/D must digitize the
level it sees at its input. The accuracy with which this can be
done is a function dynamic range of the system and will be
determined by the low frequency accuracy of the converter,
the noise generated in the signal conditioning circuits, and
the noise added by the converter. The INL, DNL, and low
frequency SINAD specifications can be used to predict
performance of the system with a particular converter.
The HI5800 is a low noise 12-bit 3 MSPS converter that is
perfect for the applications which require a higher dynamic
range at slower pixel rates. It is a complete sampling
converter with on board sample and hold and reference. The
low frequency (20kHz input) SINAD of typically 70dB reflects
its outstanding low noise performance. The high frequency
(1MHz input) SINAD number of 68dB illustrates how the
performance is maintained at higher input frequencies.
VOUT
+
Conclusion
FIGURE 14. INVERTING BUFFER
The circuit gain can be calculated from:
V OUT = ( – R 2 ⁄ R 1 ) × V IN – ( R 2 ⁄ R 3 ) × V OFFSET
(EQ. 13)
The circuits that process large signal pulse type waveforms
must slew and settle quickly so, as depicted in Figure 15, the
A/D can then accurately digitize the pixel information. Given
the ever increasing pixel rates this can become quite a
challenge.
This note has discussed the various considerations involved
in designing the analog front end to an image-processing
system. A system design was presented and proved to have
accurate sparkle free performance at typical video
frequencies.The methodology presented could be used to
analyze the system requirements for systems with higher
pixel rates.
References
[1] Joey Doernberg, Hae-Seung Lee, David A. Hodges, “Full
Speed Testing of A/D Converters,” IEEE Journal of Solid
State Circuits, Vol. SC-19, No. 6, DEC. 1984.
[2] Fredrickson, Thomas M.,”Intuitive Operational Amplifiers,”
McGraw-Hill Inc., New York, NY, 1988.
CCD
OUTPUT
[3] Demler, Michael J., “High-Speed Analog-To-Digital
Conversion,” Academic Press Inc., 1992.
A/D
ENCODE
FIGURE 15. TIME DIVISION MULTIPLEXED SIGNAL
The overall system settling time is made up of two parts.
Initially the signal must slew until it enters a region where
small signal analysis takes over. Similar slew rate
requirements as discussed in the design considerations
section apply in this case also. For a single pole system, the
error will then decay with a time constant determined by the
small signal bandwidth of the system. The settling time in an
actual system is very much a function of the circuit parasitics
and the overall frequency response of the circuit. As such, it
is difficult to calculate an accurate number beforehand.
Reference 2 has a more thorough discussion of settling time
and the calculations involved.
Additional large signal time domain converter specifications
such as overvoltage recovery time and transient response
time become important in these types of applications. As in
the case of full power bandwidth, there are many ways to
define these tests so be aware of the method used on the
datasheet and how it applies to a particular application.
8
[4] “IEEE Standard for Performance Measurements of A/D
and D/A Converters for PCM Television Video Circuits,”
IEEE Standard 746-1984.
[5] “High Speed Design Seminar,” Published by Analog
Devices, 1990.
[6] “High Speed Signal Processing Applications Seminar”,
Published by Intersil Corporation, 1992.
[7] “Using Intersil High Speed A/D Converters,” Application
Note AN9214, Published by Intersil Corporation, 1992.
[8] “Video Amplifier with Sync Stripper and DC Restore,”
Intersil Corporation, Application Note AN9514.
Application Note 9313
All Intersil semiconductor products are manufactured, assembled and tested under ISO9000 quality systems certification.
Intersil semiconductor products are sold by description only. Intersil Corporation reserves the right to make changes in circuit design and/or specifications at any time without notice. Accordingly, the reader is cautioned to verify that data sheets are current before placing orders. Information furnished by Intersil is believed to be accurate and
reliable. However, no responsibility is assumed by Intersil or its subsidiaries for its use; nor for any infringements of patents or other rights of third parties which may result
from its use. No license is granted by implication or otherwise under any patent or patent rights of Intersil or its subsidiaries.
For information regarding Intersil Corporation and its products, see web site www.intersil.com
Sales Office Headquarters
NORTH AMERICA
Intersil Corporation
P. O. Box 883, Mail Stop 53-204
Melbourne, FL 32902
TEL: (321) 724-7000
FAX: (321) 724-7240
EUROPE
Intersil SA
Mercure Center
100, Rue de la Fusee
1130 Brussels, Belgium
TEL: (32) 2.724.2111
FAX: (32) 2.724.22.05
9
ASIA
Intersil (Taiwan) Ltd.
7F-6, No. 101 Fu Hsing North Road
Taipei, Taiwan
Republic of China
TEL: (886) 2 2716 9310
FAX: (886) 2 2715 3029