Sandau R. Digital Airborne Camera: Introduction and Technology

Подождите немного. Документ загружается.

4.4 Opto-Electronic C onverters 193

Fig. 4.4-11 Full-frame

transfer CCD principle (IAI

FZK, 2004)

prevent image distortion, they have square dimensions, up to 20 megapixels, and a

pixel size of 7–24 μm.

Since the pixel ﬁeld is used for both the exposure and readout, it is necessary

to have a mechanical shutter or some other form of synchronisation to prevent

smearing when photographing. These smearing effects are always produced when

photodiodes are being continuously exposed, i.e., they are exposed or overexposed

during readout as well.

The maximum readout speed is limited by the bandwidth of the output ampliﬁer

and by the speed of the peripheral processing electronic array (see Section 4.5).

But it can be markedly increased if the image is divided into smaller sub-images

of equal size, which are then read out simultaneously. In a subsequent operation, a

video processor reassembles the image digitally. Both the front – and the rear-side

exposures have been implemented with full-frame CCDs.

In the case of rear-side exposure, the incoming photons are converted into elec-

trical charges more effectively than in the case of front-side ones, because the light

coming from the rear does not have to cross the gate covered by the photosensitive

photodiodes.

The disadvantage of full-frame cameras is that they can only provide individual

time-delayed images and not a continuous video stream, because the shutter has to

be shut repeatedly during readout. This predestines this type of CCD for scientiﬁc

and medical applications, where little light is available and higher resolutions are

required, such as astronomy.

TDI/CCD Lines

Time delayed integration (TDI) is a technology in which electrical charges of par-

allel CCD lines are pushed at right angles to the line in synchronisation with the

motion of the object to be scanned (comparable to the vertical shift registers of an

194 4 Structure of a Digital Airborne Camera

Fig. 4.4-12 TDI-CCD principle illustration based on CCD525 from Fairchild (Fairchild, 2004)

area array sensor). In this way, the same section of the object (despite its motion)

is photographed over a number of lines (TDI stages) on the same pixel, so that the

integration time is multiplied by the number of stages. The principle is illustrated in

Fig. 4.4-12.

Or with other words: unlike linear sensors, TDI sensors have several photosensi-

tive lines arranged side by side (a markedly asymmetrical matrix, at it were), the

image data being pushed from one line to the next in synchronisation with the

motion of the object to be scanned and read out according to a deﬁned number

of TDI stages, which in many cases is even selectable. The readout registers are

subdivided (multi-tapping, see Section 4.4.2.1) to ensure very fast readout.

The advantages of TDI sensors over conventional line sensors include not only

a noise reduction of N (N = number of TDI stages), but, even more importantly,

increased responsivity (directly proportional to N). The latter is particularly beneﬁ-

cial, for instance, to applications involving fast-moving objects, when the integration

times are too short for normal linear sensors.

Interlaced or Progressive Scan as Matrix Readout Mode

For the sake of completeness, readout methods commonly used in video technology

of interlaced or frame-transfer CCD are brieﬂy reviewed to show how ﬂexible these

4.4 Opto-Electronic C onverters 195

sensors are and that the image composition is adaptable to a wide variety of (video)

requirements.

On the one hand, there are sensors that operate exclusively with interlaced tech-

nology (ﬁeld method). This method, in which each video image is composed of

two ﬁelds, was originally used in television technology. The speed at which this

happens is determined by the video standard. In Europe, the so called CCIR stan-

dard is normally applied. Under the CCIR standard, 25 whole images per second

are produced from 50 ﬁelds. In the USA, the RS170 and the EIA standards, under

which 30 whole images are produced per second from 60 ﬁelds, are commonly used.

These standards are used almost exclusively in television and video devices that are

currently in general use.

There are different ways of producing ﬁelds in a sensor. In the ﬁeld integra-

tion mode, the two ﬁelds prescribed under the CCIR or RS170 video standard are

recorded on the sensor at completely different time points. In each ﬁeld, the charges

are shifted from two superimposed pixels into a cell of the vertical shift register, i.e.,

they are in effect added up. As a result, the brightness is almost doubled. The posi-

tion of the two pixels changes from ﬁeld to ﬁeld with a pixel offset in the vertical

direction. The problem that arises as a result is that if an object is in motion, it is

photographed at two different time points. If the ﬁelds are merged into a full image,

the so-called comb effect is produced.

In the frame integrated mode, the two ﬁelds are also photographed at different

time points, but in contrast to ﬁeld integration mode, the integration times overlap.

For each ﬁeld, only the charges of one pixel are shifted to a cell of the vertical shift

the vertical direction, which means that the integration sections intersect each other

over a certain length of time. Hence, a “standing” image can be generated by using a

ﬂash which is triggered precisely at the overlap. Sensors using the progressive scan

technology produce an image consisting of a complete, full image, rather than of

two ﬁelds. Typical image formats are VGA resolution (640 × 480 pixels) or super-

VGA resolution (1,280 × 1,024 pixels), where the video signal is usually put out in

the non-interlaced format. Normal television and video devices cannot handle this

format, but it offers enormous advantages for image processing (compatible with

PC technology). It is no longer necessary to trigger a ﬂash at the right instant, s ince

all pixels of the sensor are exposed simultaneously.

The two-channel progressive scan is an enhancement in which the CCD sensor

operates with two horizontal shift registers. In this way, the sensor contents can be

read out simultaneously via two video channels. To do this, all odd lines are read out

via video channel 1 and all even lines of a full picture, via video channel 2. In the

next full image, all even lines are transferred via video channel 1 and all odd lines,

via video channel 2, etc. As a result, the sensor can now operate at double speed

(for example, 50 or 60 full images per second), and, moreover, a normal interlaced

signal, which is compatible with the normal video standard, is applied at each video

channel.

196 4 Structure of a Digital Airborne Camera

4.4.3 Properties and Parameters

A multitude of physical quantities is required for deﬁning and assessing CCD

sensors and cameras. The principal ones are described below.

4.4.3.1 Signal-to-Noise Ratio

The signal-to-noise ratio (SNR) in a CCD sensor or in a CCD camera can be

expressed in simpliﬁed form as a ratio of signal electrons to noise electrons:

SNR = n

signal

noise

. (4.4-1)

where SNR is the signal-to-noise ratio, n

signal

the number of signal electrons and

noise

the number of noise electrons. The number of signal electrons, which depends

on the luminance or on the impinging photons, can be stated as follows:

signal

= (/h ·v) · t ·AP ·η. (4.4-2)

where Φ is the intensity in [W/m

], h ·ν the photon energy [Ws], t the exposure time

[s], AP the pixel area [m

] and η the quantum efﬁciency.

4.4.3.2 Noise Sources

Noise is deﬁned as a signal of limited power,the random (statistical) properties of

which are known. A distinction is made between noise sources in time and space.

Noise in time can be minimised, but not eliminated. Noise in time typical of CCDs

includes shot noise, reset noise, output ampliﬁer noise and dark current noise. By

contrast, spatially determined noise can be eliminated to a large extent through

suitable correction algorithms, characteristic ones being PRNU and DSNU.

Fundamentally, noise electrons in CCD images are generated by three processes,

which are connected with the release (photogeneration and thermogeneration) and

the readout process of electrons:

Photon noise is equal to the square root of the number of signal electrons

(in accordance with the laws of Poisson distribution in photogeneration). CCD

noise is the term used for noise electrons (n

CCD

) generated in CCD channels

(transfer, dark current, ﬁxed pattern noise, ...), statistically distributed around a

mean (rms). Ampliﬁer noise refers to electrons (n

AMP

) generated in the output

ampliﬁer.

Since none of the three sources is correlated with each other, the following holds:

noise

√

{

√

[/h ·v) · t ·A ·η] +n

CCD

AMP

}. (4.4-3)

where n

CCD

is the noise electrons in the CCD and n

AMP

the noise electrons in the

output ampliﬁer.

4.4 Opto-Electronic C onverters 197

If (4.4-2) and (4.4-3) are inserted in (4.4-1), then:

SNR = (/h ·v) ·t ·A ·η/

√

{

√

[(/h ·v) · t ·A ·η] +n

CCD

AMP

}. (4.4-4)

If a high light modulation occurs in the CCD, then photon noise dominates. This

can be expressed in simpliﬁed terms:

SNR ≈

√

(/h ·v) · t ·A ·η ≈

√

signal

or S/N is directly proportional to the square toot of quantum efﬁciency,

SNR ∼

√

η. (4.4-5)

For low modulation, the noise of the CCD and of the readout ampliﬁer dominates:

SNR ∼ η/

√

CCD

AMP

). (4.4-6)

For high modulation in the CCD, therefore, the signal-to-noise ratio is propor-

tional to the square root of quantum efﬁciency. For small signals, the ratio is directly

proportional to quantum efﬁciency and is determined primarily by noise in the CCD

and in the output ampliﬁer.

For example, if the CCD noise n

CCD

is 8e

–

(rms), and the ampliﬁer noise n

AMP

is 6e

–

(rms), then n

GES

√

CCD

+ n

AMP

) =

√

(82 + 62)e

–

= 10e

–

(rms).

The magnitudes of individual components of CCD-speciﬁc noise merit further

discussion. Shot noise, which is thermally generated and not correctable, has a

Poisson distribution:

shot

√

Q (4.4-7)

where Q is the generated charge quantity [e

–

Reset noise is generated in the channel resistance of the reset FET of the CCD

output ampliﬁer, commonly also referred to as kTC noise on account of (4.4-9).

Since the reset noise is constant over the clock period of a pixel, it can be corrected.

The method used for this purpose is termed correlated double sampling (CDS) (see

Section 4.5).

reset

√

4kTBR in [V] (4.4-8)

reset

√

4kTC/e in [e

−

]. (4.4-9)

where k is the Boltzmann constant [J/K],

T temperature [K]. B noise power bandwidth [Hz],

R = effective channel resistance []

C = node capacity [F] and e electron charge [1.6 10–19 C].

198 4 Structure of a Digital Airborne Camera

Output ampliﬁer noise is subdivided into white noise, also known as Johnson

noise, and ﬂicker noise, also termed 1/ f noise. White noise is generated thermally

through channel resistances of the output FET (source follower):

white

√

4kTBR

out

in [V] (4.4-10)

white

√

4kTBR

out

/CVF · Vin[e

−

] (4.4-11)

where CVF is the conversion gain factor [μV/e

–

] and V the ampliﬁcation of the

output ampliﬁer.

Flicker noise is inversely proportional to frequency. The frequency at which the

ampliﬁer is determined only by white noise is referred to as 1/f corner frequency. In

general, white noise increases and ﬂicker noise diminishes with increasing ampliﬁer

(chip) area; for this reason, in designing an ampliﬁer, the aim is to achieve an opti-

mum between geometry and typical operating frequency. Figures 4.4-13 and 4.4-14

illustrate the typical noise curves of CCD output ampliﬁers.

Fig. 4.4-13 1/f noise of the

output stage of a Kodak CCD

line (Kodak, 2001b)

Fig. 4.4-14 Output ampliﬁer

noise of a Kodak CCD line at

28 MHz (Kodak, 2001b)

4.4 Opto-Electronic C onverters 199

Various clocks are required for reading out the signal from the CCD, some

of which have a very high frequency (horizontal and reset clocks). Their gener-

ation and the current pulses produced by the relatively high capacitive loads of

the CCD registers (...pF to ...nF, depending on the dimensioning), can result in

considerable interference with the output signal. This noise, called clocking noise,

can be reduced only through a precise design and good ﬁltering of the power

supply.

Dark current noise is the result of defects or faults in portions of silicon low-

charge chips or at transitions to silicon dioxide. Additional electrons, which add

up to form the signal, are generated stochastically here in the mid-band between

the conduction band and the valence band. This is a thermal process which can be

prevented only by cooling the CCD sensor.

A distinction is made between surface dark current, which accounts for the

largest portion of the dark current and which is generated through defects in

the semiconductor and in the manufacturing process on the surface of silicon

dioxide, and bulk dark current, which is generated inside silicon dioxide by the

defects.

Dark current is affected by two typical noise magnitudes, namely dark current

non-uniformity and dark current shot noise. Whereas DSNU as spatial noise is cor-

rectable, shot noise is not. Just as in the case of photon shot noise, dark current noise

has a Poisson distribution:

dark

√

D in [e

−

] (4.4-12)

where D is the generated dark current [e

–

4.4.3.3 Dark Signal (or Dark Current)

The mean noise level (or noise ﬂoor) is relatively constant and can therefore be

corrected to a large extent. Depending on the requirement, a distinction is made

between the sole dark current correction and the additional correction of the dark

signal deviations between the pixels, the DSNU (see next section). In principle,

each CCD line or matrix has, at the start of its readout register, several representative

dark current pixels, which should be used as points of reference for the correction.

Hence, the signal level of the dark current pixels represents the bottom threshold

of the available dynamic range. Figure 4.4-15 shows the typical dark current image

of an uncooled CCD matrix: the “bright” side characterises increased temperature

caused, for example, by the output ampliﬁer(s).

Dark current is essentially dependent on the exposure time or integration time

and on chip temperature. It can be markedly minimised through cooling. The gen-

eral rule is that the dark current is halved every time the sensor is cooled by

approximately 7 K (see Fig. 4.4-16). Depending on the manufacturer, the physi-

cal speciﬁcations are given in [pA/Pixel] or [pA/cm

], or also in [e

–

/pixel · ms], but

most commonly in [μV/ms].

200 4 Structure of a Digital Airborne Camera

Fig. 4.4-15 Dark current

image of a CCD matrix

Fig. 4.4-16 Dark current in relation to the temperature of a (cooled) CCD matrix

4.4.3.4 Dark-Signal Non-Uniformity (DSNU)

DSNU expresses the non-uniformity of pixels of a CCD sensor in absolute darkness

and is a speciﬁcation of dark current. Absolute uniformity of all pixels would be

ideal, but minute differences between wafers caused through manufacture, defects

and, above all, chip temperature affect uniformity. Dark current and DSNU can be

minimized through tempering (cooling). Dark current and DSNU also diminish with

increasing readout frequency or shorter integration time t

INT

. The value is usually

given in [μV/ms] with respect to a certain integration time.

The use of a correction depends on the particular application and is normally

required only at very high signal resolutions or very long integration times and/or

4.4 Opto-Electronic C onverters 201

when sensors are not cooled. Since the signal level swing of DSNU in terms of the

dynamic scope of the CCD is very small, a correction is normally made digitally by

subtracting the pixel values of a stored dark current image from the pixel values of

acquired images.

4.4.3.5 Photo Response Non-Uniformity (PRNU)

Oftenalsoreferredtoasvignetting or ﬁxed pattern noise, PRNU expresses the non-

uniformity or the deviation from uniformity of photo responsivity of the pixels of

a CCD sensor under uniform exposure conditions (ﬂat ﬁeld). Absolute equality

of all pixels would be ideal here, but unfortunately not all pixels on a CCD chip

have uniform responsivity. In extreme cases, i.e., at very low pixel responsivity, we

speak of “dead” pixels. Minimal differences across a wafer arising from the man-

ufacturing process, defects and differences in ampliﬁcation are the cause of this

non-uniformity. It is important to differentiate between non-uniformity when there

is illumination (PRNU) and when there is darkness (DSNU). PRNU is expressed in

[%] relative to saturation voltage U

SAT

A pixel-related correction is required in any case in signal processing for signal

resolutions >6 bits; depending on the requirements, it must be made multiplicatively

(since an ampliﬁcation fault is involved), in analogue fashion (MDAC) or digitally.

Correction images or correction factors are produced by averaging several images

of all photosensitive pixels at about 50% U

SAT

at absolutely uniform illumination

(ﬂat ﬁeld) and stored as a correction matrix. It should be borne in mind that in this

manner the PRNU is eliminated to a very large extent, but shot noise is increased by

a factor of

√

2 (this also holds for DSNU correction).

4.4.3.6 Dynamic Range (DR)

The ratio of the maximum output signal (saturation limit of a pixel) to its photo

responsivity is known as the dynamic range. A CCD’s photo responsivity is limited

by dark current noise. As mentioned earlier, the dynamics of a CCD sensor can be

improved by cooling. The DR is expressed in [dB] and is one of the basic parameters

for assessing a sensor’s range of applications.

DR = 20 log (USAT/UDARK

rms

) (4.4-13)

or also

DR = 20 log (QSAT/QDARK) (4.4-14)

where USAT is the saturation voltage [V], UDARKrms the dark current voltage

(rms) [V], QSAT (= FW) the full well charge quantity [e

–

] and QDARK (= D)the

generated dark current [e

–

202 4 Structure of a Digital Airborne Camera

4.4.3.7 Noise Measurement (Photon Transfer Curve, PTC)

Generating a photon transfer curve is an expedient way of characterising CCD sen-

sors or CCD cameras. Apart from basic noise measurement, the photon transfer

curve also provides the full well and the transfer in electrons per A-D converter

state [e

–

/LSB]. In this way, the dynamic range can be calculated. Each dot on the

PTC represents a pixel group, illuminated by a ﬂat ﬁeld, with varying exposure time

because, although the integration time is constant, the time of light incidence on to

the sensor varies. In this manner the dark current can be subtracted as a constant

error. The noise (standard deviation) is now plotted for each pixel group with refer-

ence to the mean signal value. The result is a curve comprising three basic sections:

the ﬁrst section (ﬂat) shows the noise ﬂoor level, i.e., minimum system noise; the

second (rising) section shows the typical operating range of the sensor system; and

the third section shows the limit by pattern noise (correctable).

Figure 4.4-17 shows, for example, that (similar to the calculation in (4.4-14)) this

system provides the following dynamic range:

DR = 20 log (FW/D) = 40,000/25 = 64 dB (4.4-15)

where FW is full well [e

–

] and D noise ﬂoor or dark current [e

–

Fig. 4.4-17 PTC of a Kodak

full-frame matrix, including

board at 28 MHz (Kodak,

2001b)

4.4.3.8 Blooming and Antiblooming; Smear and Transfer Efﬁciency

All stored data should be read out in its entirety, so that no data belonging to the last

image acquired remains after reading out a CCD. But even a CCD is not an ideal

component. Blooming, transfer inefﬁciency and smear effects impair the charge

transfer and/or the image quality.

The capacity of a pixel is determined by its size. There is a linear relationship

between the number of electrons generated and the incident light. When a pixel