Determining Dynamic Range of CCD Image Sensors

In previous articles, I discussed both dynamic range as a general engineering concept and dynamic range as a performance specification in imaging systems. In this article, we'll examine CCD dynamic range at the semiconductor and circuit level. We’ll consider the following question: What determines the range of luminance that a CCD and its associated signal-processing circuitry can record?

For more on the subject of CCD (charge-coupled devices), make sure to catch up on our selection of related articles: 

• Basics of image sensors
• Basics of CCDs
• CCD types (e.g., full-frame, interline-transfer, and frame-transfer)

 

Maximum Signal and Noise Floor

In a CCD, the input signal is electromagnetic radiation, and the output signal is a packet of electric charge whose quantity corresponds to the intensity of that electromagnetic radiation.

The dynamic range of the input signal is essentially infinite: luminance can vary from zero, meaning the complete absence of photons, to whatever intensity is produced by the brightest object in the universe. The dynamic range of the output signal, however, is finite.

Because of noise, zero luminance will not produce zero electric charge; because of full well capacity, luminance that increases beyond a certain point will no longer produce corresponding increases in electric charge. For that reason, the dynamic range of a CCD is the ratio of full well capacity to noise floor, i.e., the maximum output signal level that a pixel can generate divided by the signal level that will be generated even when a pixel receives no incident light.

 

 

If you’ve been following the Image Sensor Technology series, you know that the dominant noise sources in a CCD are dark noise, photon noise, and read noise. However, when calculating the dynamic range, we consider only dark noise and read noise.

I see two reasons for this: First, photon noise is not a property of the CCD or the accompanying readout circuitry—it’s inherent in the nature of light and doesn’t vary from one system to another. Second, photon noise doesn’t contribute to the minimum possible output signal level, because with zero incident light, there is also zero photon noise.

This leads us to the following formula:

\[dynamic\ range=\frac{N_{SATURATION}}{N_{NOISE}}\]

 

where \(N_{SATURATION}\) is the full well capacity—i.e., the number of electrons at which the output signal saturates—and \(N_{NOISE}\) is the sum of dark noise and read noise expressed in electrons RMS.

If you prefer to use decibels or stops instead of an ordinary ratio, we have the following:

\[dynamic\ range\ (dB)=20\log_{10}{\frac{N_{SATURATION}}{N_{NOISE}}}\]

 

\[dynamic\ range\ (stops)=\log_{2}{\frac{N_{SATURATION}}{N_{NOISE}}}\]

 

CCD Full Well Capacity

An important factor that influences full well capacity is the area of the pixel, or the area of the photodiode if only a portion of the pixel is sensitive to light. A larger photoactive element corresponds to a larger section of silicon in which electric charge can be integrated. Thus, we can extend the dynamic range by increasing the pixel size. Since this will result in decreased resolution if the total sensor area doesn’t change, we will need a larger sensor if we want to both increase dynamic range and maintain resolution.

 

Physically larger photodiodes provide more space for the accumulation of light-generated free electrons.

 

The CCDs currently on the market offer wide variation in full well capacity. For example, Oxford Instruments has devices ranging from 25,000 electrons (pixel area = 100 µm², dynamic range = 64 dB) to 510,000 electrons (pixel area = 676 µm², dynamic range = 94 dB).

 

Effective Full Well Capacity

Operating conditions can affect full well capacity. ON Semi’s KAI-2020 CCD, for example, is specified at a full well capacity of either 20,000 electrons or 40,000 electrons. The actual physical capacity is closer to 40,000 electrons, but the output amplifier can’t handle 40,000 electrons at the full readout speed of 40 MHz. Thus, when calculating dynamic range, we need to consider the effective full well capacity, rather than the electron capacity that corresponds only to the pixel’s physical characteristics.

Also, the effective full well capacity of the KAI-2020’s photodiodes depends on the applied substrate voltage. A lower substrate voltage produces larger full well capacity (and therefore larger dynamic range), but it also makes the sensor more susceptible to blooming.

 

Dynamic Range and Analog-to-Digital Conversion

Though the dynamic range of a CCD is determined by full well capacity and noise, we have to keep in mind that the camera can’t do anything with the CCD’s data until it is digitized. The CCD signal passes through the ADC on its way to the rest of the system, and consequently, we need to make sure that the analog-to-digital converter provides adequate dynamic range.

If you pay good money for a high-performance CCD with 90 dB of dynamic range and then pair it with an 8-bit ADC, the final dynamic range of the image data is the dynamic range of the ADC, with only 8 bits of resolution, is actually about 48 dB. You’ve lost 42 dB.   

Fortunately, achieving adequate dynamic range from the ADC is usually not a major challenge. Many CCDs will be in the 60–70 dB region, and you can maintain that sort of dynamic-range performance with 12 bits of ADC resolution. Analog Devices and Texas Instruments sell highly integrated CCD signal processors that support 12-bit A/D conversion.

 

Conclusion

We’ve seen that dynamic range is determined by a CCD’s full well capacity and noise characteristics. However, we need to keep in mind that dynamic range isn’t always the most critical performance metric. Dynamic range helps the camera to capture high contrast scenes, still, sometimes we might be more interested in large well capacity or low noise, rather than in combining noise performance and well capacity in a way that maximizes dynamic range.