To understand the basic operation of a CCD image sensor it is important to determine the factors that limit its performance. The following list describes some of these factors:
a) Resolution
Spatial resolution is defined as the ability to discriminate between closely spaced points in the image. Any image can be Fourier analysed and the spatial frequency of its Fourier components is normally expressed in line pairs per mm. The spacing of the electrodes may also be expressed as a spatial frequency/o (elements per mm) and the spatial frequency / of the image may be normalised as the ratio///û. The resolution is limited by the Nyquist limit, defined as the spacing of the resolution elements to a normalised spatial frequency of^%=0.5. Resolution can be analysed in terms of the Modulation Transfer Function (MTF) of the output. MTF is represented by a plot of normalised amplitude versus spatial frequency out to the Nyquist limit. Methods for computing MTF have been developed in the past for photographic imaging systems. Techniques specially adapted for solid state digital imaging systems have also been developed (Campana, 1977; Foreman and Dereniak, 1984; Boberg, 1994). Qualitatively, the greater the area under the MTF curve the better the fidelity of the reproduced image (Jaff, 1972). Collet (1985) described several ways of improving resolution.
b) Sensitivity and dynamic range
The ideal CCD image sensor has a high sensitivity and a wide dynamic range. The sensor should be able to operate at low levels of illumination and have a sufficiently wide dynamic range to accommodate large variations in illumination. Device sensitivity is a function of both noise performance and responsivity. The dynamic range of a CCD is determined at the upper end by the maximum charge handling capability and at the lower end by the various noise sources present.
c) Responsitvity
The spectral responsivity depends on the photoelement structure (photodiode or MOS capacitor) and on the type of silicon substract used. Responsivity can be characterised in terms of radiometric units. The number of charge carriers , Ng,
Chapter 3. Data Acquisition
HSAT
collected at each element during the integration period T, is given by ; Ns = q
where H is the image radiance, S is the effective responsivity, A is the geometric area of each element and q is the charge of the electron.
e) Charge Transfer Efficiency
The basic limitation on the performance of a CCD is the efficiency with which a charge packet can be transferred from one potential well to the next. Ideally any signal charge packet should remain into the potential well until the voltages are changed and then to transfer instantaneously and totally to the next potential well. In practice this does not happen. Copper et al. (1976) described a way of cancelling unwanted residuals, that consists of coding the signal before entry to the CCD and decoding the output with a well chosen code sequence.
f) Noise performance
i) Shot (photonic) noise
The emission of photons from any source is a random process. The photogenerated carriers in a potential well during a certain period of time is a random variable, the standard deviation of which represents the photonic noise. This noise is a fundamental limitation in all imaging applications since it is a property of the light itself and not of the sensor. This limitation is more relevant under conditions of low scene contrast.
ii) Thermal generation or deark current noise
As described in section 3.2.3. electrons or holes receive energy from photons which makes them able to transit from one state of energy to another. However, it is possible for the electrons or holes to make the same transition by absorbing thermal energy. This process provides an undesired source of charge carriers that can not be distinguished from the photogenerated carriers. This competition between optical and thermal generation limits the usefulness of image sensors. Dark current is highly temperature dependent, the cooling of an array is thus a possibility for decreasing the dark current to a suitable value. Declerck et al.
Hi) Bias charge noise
In certain sensors an additional bias charge is introduced to improve the charge transfer efficiency. This bias charge can be either electrical or optical (shot noise). This bias charge is also called fat zero noise.
iv) Blooming and smear
In case of a particularly bright spot in an image, more charges may be generated than the respective well can hold. When such optical overload occurs the excess carriers spread sideways into adjacent picture elements to an extent proportional to the magnitude of the overload. The main problem is to confine or dispose of the excess carriers produced by a local overload so that the rest of the image is not degraded. Beynon and Lamb (1980) described three techniques to provide anti-blooming: accumulation mode, diffused drain with control gate and diffused drain with implanted barrier.
Smear is similar to blooming. It can occur before the blooming correction fails and it is difficult to correct. Smear increases with overexposure but it is not limited to overload of illumination, it may contribute to loss of contrast at any light level. Image smear is a fundamental limitation of CCD image sensors. Berger et al. (1984) provided the formulae used to quantify the smear and blooming errors.
vj Tailing
At the illumination level where blooming is not yet introduced white spots may be deformed in the horizontal direction and only to the right. This means that after the signal reaches a peak white level it goes down to the black background with a certain delay. This deformation depending on the illumination can almost double the horizontal diameter of the spot. Tailing is substantially different from blooming or smearing, which tend to change the form of the object in all directions with accentuation of the vertical (Dahler, 1987).