Read noise versus shot noise – what is the difference and when does it matter?

When determining the best metrology camera, one of the first considerations is to determine the dominant noise source.  This can help you prioritize the most important camera parameters, such as sensitivity or dynamic range. In this post we will consider temporal noise sources that vary with time including shot noise and read noise. With this, it is also helpful to consider the difference between signal to noise ratio (SNR) and dynamic range.

What is read noise?

Read noise is a combination of noise from the pixel and from the ADC. The Read Noise (RN) of the sensor is the equivalent noise level (in electrons RMS) at the output of the camera in the dark and at zero integration time. Note that the build up is different for a CMOS sensor and a CCD sensor. The ADC with CCD image sensors is done outside the sensor and the ADCs for a CMOS image sensor are in each pixel.

Read noise basically determines the contrast resolution that the camera is able to achieve. The lower the read noise level, the lower the minimum number of signal electrons that can be detected. A lower read noise means you can see smaller changes in signal amplitude, thus detect details with smaller contrast differences. Read noise is also important in combination with expressing the sensitivity of a camera. Low read noise means that you can see small contrast changes, which is typically present in scenes taken in low light conditions. A lower RN therefore results in a more sensitive sensor.

Of course when talking about sensitivity, QE plays an important role as well as the pixel sensitive area and the use of microlenses. QE (Quantum Efficiency) is a measure of how efficiently the sensor converts light (photons) to charge (electrons). The more electrons in a pixel during the integration period, the higher the output level of the sensor, so the more sensitive the sensor is for that specific wavelength of the light. A QE of 1 means that every photon generates (in average) one electron. Normally the QE is less than 1 (or 100%). These two combined gives the overall sensitivity of the sensor as QE/Read Noise, or the minimum amount of light you can see. So read noise is one of the parameters to mind when you want to observe ‘dark scenes’.  Read here for more information.

What is shot noise?

Shot noise originates from the discrete nature of electrons. Shot noise is caused by the arrival process of light photons on the sensor. Consider the following example: imagine standing at an overpass above a highway and counting the amount cars passing by in one minute. The next minute, and the next, and the amount counted is probably not the same. The resulting measurement varies from minute to minute, following a Poisson distribution.  In the electron domain this is similar: the standard deviation of the amount of captured electrons in a pixel is the square root of the mean signal level.

For an increasing amount of electrons, the shot noise increases as well. Noise however should not be considered on its own. In applications, the noise has to be seen relative to the signal. In practice, the Signal to Noise Ratio (SNR) is used. The SNR for higher signal levels is dominated by shot noise.

The formula you see typically is:

SNR = 20*LOG(SQRT(FWC)).

Where FWC is Full Well Capacity.

Sensor specifications typically use the maximum FWC number that can be realized in a pixel. In order to realize this FWC, the sensor is operated in saturation mode.  So in metrology systems when saturation behavior needs to be avoided, it is relevant to use the maximum FWC number at which the sensor still has a linear response (e.g. an increase in light of 2x causes an increase in video signal of 2x from the sensor).

So when you want to observe contrast changes in a bright scene, you require a camera with an image sensor that has a higher FWC. In bright scenes the contrast detecting ability is dominated by the shot noise of the camera, basically the FWC.  Some functionality such as binning and averaging can be done inside the camera to further reduce shot noise.

What is SNR?

Since we referenced SNR is Signal-to-Noise-Ratio, let’s go into some more details.

The formula for SNR is listed above. It is expressed in decibels (dB). The higher the number, the better the ability to detect contrast changes in bright scenes.

What is DNR?

Dynamic Range (DNR) and the signal to noise ratio (SNR) are sometimes considered interchangeable for CCD and CMOS image sensors and cameras, but they are not the same. While SNR basically expresses that ability to see contrast details in bright parts of the image, DNR indicates the total contrast resolution of the sensor (from dark to white).

With industrial cameras the formula used is:

DNR = 20*LOG(FWC/Read Noise).

It is expressed in dB as well.  FWC and Read noise are expressed in electrons.

It is a measure for the contrast range that can be detected. The larger the number, the
more grey levels that can be captured, thus the higher the contrast detecting resolution of the camera.

Dynamic range can be measured and calculated using the photon transfer curve if desired.  For more information on the Photon Transfer Curve, click here.

Dynamic range provides a much more useful indication (compared to SNR) regarding the ability of the camera to provide the desired image details.  When comparing dynamic range values from different cameras, be sure to verify they were measured under the same conditions.

Conclusion

The system is said to be light limited when, while not yet reaching saturation, the exposure of the sensor cannot be increased. Large full well is not of the essence here.  Other performance parameters such as read noise, QE (sensitivity), and fixed pattern noise behavior are more important.

The system is said to be shot noise limited when there is an abundance of light and a large full well is desired to optimize it. The absolute noise level in electrons will be higher for the shot noise than for the read noise. If more light can be captured (for example by increasing the integration time), the better the SNR becomes according to the equation. It pays off here to use a camera with a larger full well.

For more information about the different noise sources and the differences with CCD and CMOS image sensors, see additional information from Albert Theuwissen of Harvest Imaging: http://harvestimaging.com/blog/?p=1049

Related Posts:

What can you do with 4 electrons of read noise now from a CMOS industrial camera – half of that of a CCD?

How to Select the Best Industrial Camera – Are you shot noise limited?

Dynamic Range DNR and Signal to Noise Ratio SNR for CCD and CMOS image sensors