Part of the Oxford Instruments Group

Low Noise Mode in the Sona 4.2B-6 back-illuminated sCMOS Camera

For most applications the 16-bit high dynamic range mode of the Sona 4.2B-6 will be the most flexible, and therefore the most useful mode for most imaging applications as it combines speed, the widest dynamic range and a low noise floor. But for when the highest sensitivity and lowest possible noise floor is required, the low noise mode function of the Sona 4.2B-6 can be utilized. This low noise mode uses a two times correlated multisampling (2-CMS) approach so that a low noise floor can be achieved while still maintaining high frame rates. Therefore, select the low noise mode for imaging the weakest signals, and for example, live cell imaging applications that seek to reduce illumination intensities and use short exposure times to minimize adverse effects on cell biology.

How does the 2-CMS low noise mode work?

The Components of Readout Noise

During the normal sensor readout process, noise is generated by the sensor itself termed “read noise”. Although the amount of read noise in modern sCMOS sensor designs is very low it does set a limit on the noise floor and ultimately impacts the sensitivity of the camera i.e. the ability of the camera to detect very low signal levels against the noise component.

Figure 1: The read noise of scientific cameras is very low. However, when the signal level and the signal to noise ratio are reduced (right) then the read noise of the camera imposes a limit to the sensitivity.

The read noise itself comes from two main sources: from the amplifiers that increase the low-level signal so it can be further processed which is termed “amplifier noise”, and from each time charge is cleared from the pixel, which is termed “reset noise”. In the case of sCMOS cameras, this read noise is expressed as a distribution as a median or rms figure since each pixel effectively has its own circuit and thus noise value. This is discussed in the technical note, understanding read noise in sCMOS.

How can we overcome Read Noise and Improve the Signal to Noise Ratio?

There are several practical ways to improve the signal to noise ratio, two common examples of which are:

  • Increase the exposure time - collect a larger signal and therefore improve the signal to noise ratio. But this will not only reduce frame rates, but also mean increased illumination times which will have future impact on fluorophores and cells.
  • Pixel binning – commonly used for CCD sensors to sum the output of pixels in e.g. 2x2 or 4x4 array before readout, to give a higher signal to noise at the expense of spatial resolution. However, sCMOS sensors have a different sensor architecture so that it is not possible to sum pixels before readout and thus not gaining the benefit of reduced read noise.

Another strategy is to tackle the source of the read noise itself. The reset noise component is commonly reduced in digital circuits (including sCMOS cameras) using correlated double sampling (CDS), leaving the amplifier noise as the main component. The amplifier noise could be reduced by taking multiple samples of the signal in a process called correlated multi-sampling (CMS). The potential downside of this approach is that with multiple samples being taken through a single read channel the frame rate of the sensor could be reduced. While this may work for fixed cells it would not be suitable for many live cell imaging applications.

Maintaining speeds at low noise using 2-CMS noise reduction

Under normal operation, sCMOS sensors with dual channel amplifier circuits the upper and lower readout channels act in a low gain (for high level signals), and a high gain (for low level signals) channel format. These can be used in a single channel format or combined to give a full image data range (Figure 2). Combing the 2 gain channels effectively is difficult, but used in this way, sCMOS cameras can provide a much higher dynamic range capability when compared to CCD and EMCCD cameras. From a practical microscopy point of view it means more samples are in range more often with less adjustments and reimaging due to “oversaturated” or “too dim” samples. For Andor Sona cameras they exploit an exclusive technology termed “Extended Dynamic Range” to achieve a more linear and thus wider effective dynamic range than other cameras employing the same sensor.

Figure 2: An sCMOS sensor with a dual channel amplifier, such as the Sona shown here, can use 2 gain channels to reconstruct a 16-bit image from the information sampled using the 2 channels. This provides an exceptional dynamic range.

The GS2020BSI sensor used in the Sona 4.2B-6 (and the Marana 4.2B-6) also allows a further configuration of the 2 gain channels. In the configuration used for 2-CMS, both readout channels act as “high gain” channels, shown in Figure 3. By taking a sample from each channel at the same time, the read noise level can be cut by a factor of Ö2. This equates to a reduction in the normal read noise of the GS2020BSI sensor from ~1.6e- [1.8e-rms] to ~1.2e- [1.3e- rms].

Figure 3: A simplified representation of how the Gain Channels are configured for 16-bit and for 2-CMS mode in the GS2020BSI Sensor.

If we consider a single sample of the signal as per normal operation the signal and noise components are expressed as follows:

For 2-CMS mode that takes 2 samples of the signal at the same time point using both gain channels this becomes:

Figure 4: The Low noise mode of the Sona 4.2B-6 can be demonstrated to show an improvement at low light levels. In this example, an OpticalZilla light rig has been used to show a comparison between HDR mode (left) and Low noise mode (right). Sub-window shows the relative Signal and noise components of the image for both modes set to the same scale.

What are the benefits and tradeoffs to using 2-CMS?

Because the 2-CMS method uses the 2 separate channels to sample the signal at the same time, the frame rate of the sensor does not suffer the same impact on frame rates compared to the CMS approach. Since the 2 gain channels are now being combined as a 12-bit output rather than used to cover a 16-bit range the dynamic range will be reduced to that of the single gain channel (Figure 5). Therefore, it is important to maximize this so the mode can be as effective in practice as possible.

Figure 5: By effectively using 2 high gain channels to perform 2-CMS the read noise is reduced at the expense of dynamic range (above right).

GS2020BSI Sensor Parameter Sensor Mode
Method 2-CMS HDR mode
Bit depth 12-bit 16-bit
Max frame rate (fps) 43 74
Normal Mode Read Noise (e- median) ~1.2e- [1.3e- rms] ~1.6e- [1.8e-rms]
Well depth (e-) 1,800 55,000

Table 1: Standard specifications of the GSENSE2020BSI sensor in 2-CMS mode compared to HDR mode configuration as specified by sensor manufacturer.

In table 1 we can see that read noise in 2-CMS mode is reduced compared to the normal 16-bit mode. While maximum frame rate is reduced, at 43 fps it is still sufficiently fast for most applications. The main compromise of running in low noise mode is therefore the reduction in well depth. This makes maximizing the effective well depth in this mode very important.

Andor has optimized the 2-CMS mode implementation to allow the full potential well depth and linearity to make this low noise mode more suitable to a wider range of samples. The following table shows the parameters used in 2-CMS mode and how they have been implemented in two camera designs that use the GSENSE2020BSI and how the well depth, linearity and dynamic range differ.

GS2020BSI Specifications Competitor Sona 4.2B-6
Well Depth (e-) 1,000 1,800
Dynamic Range 833:1 1,500:1
Linearity >99.5% >99.7%

Table 2: A comparison of the Correlated Multi Sampling low noise mode implementation in cameras using the GSENSE2020BSI sensor.

The well depth that is achieved is 1.8x higher than is reported for another camera that has implemented this mode. This allows the Sona 4.2B-6 to have a wider window of operation and function more effectively in this imaging mode. Furthermore, as for full range 16-bit image data, the quantitative accuracy is also higher (>99.7%) for the Sona 4.2B-6 which is critical for quantitative measurements such as FRET or other ratiometric datasets.

Figure 6: The low noise mode of the Sona 4.2B-6 has been optimized for best performance at low light levels. This can be demonstrated on low light samples – in this example the widely used FluoCells #1 BPAE cells (Invitrogen) were imaged with very low exposure time to provide a simple comparison. 

As outlined above, the low noise mode of the Sona 4.2B-6 is well suited to live cell imaging that seeks to reduce illumination intensities and use as short exposure times as possible to minimize adverse effects on the cell biology being studied.

How Sona provides further improvements in sensitivity

In addition to low noise using 2-CMS mode, and high quantum efficiency provided by back-illumination, the Sona camera series features a deeper cooling capacity  that is enabled by exclusive vacuum technology. This unmatched ability to cool the sensor has two important effects on camera noise:

  • Reducing dark current inherent to the camera sensor itself that increases over time – Sona has the lowest dark noise contribution of any back-illuminated camera. This helps extend the useable exposure duration for experiments that require longer exposures.
  • Reducing the number of high or non-uniform value pixels that otherwise need correction by interpolative filters found in sCMOS cameras. Interpolative filters are undesirable for application such as localisation based super-resolution.

Figure 7: Sona-4.2B-6 has excellent noise characteristics resulting in a cleaner image and less need for interpolative filters.

Sona 4.2B-6 has an exceptional noise performance achieved due to optimized implementation of the GS2020BSI sensor that results in high sensitivity, highest quantitative accuracy and the lowest possible noise. This translates into the cleanest possible image with the minimum of additional image filters being required.

Sona low noise 2-CMS Mode benefits:

  • Achieve lowest possible noise (approaching 1 e-)
  • High frame rates (over 40 fps full frame with low noise)
  • Highest possible quantitative accuracy (>99.7%)
  • Lowest possible hot pixels without interpolative filters (up to 5 times less than other Back-illuminated cameras)
  • Lowest dark current noise contribution (as low as 0.1e-/p/s)
  • Wider Dynamic Range than other GS2020BSI based cameras (x1.8)

Related assets