Part of the Oxford Instruments Group

Andor sCMOS Cameras for Adaptive Optics Wavefront Sensing

Andor sCMOS cameras represent ideal choices for Adaptive Optics Wavefront Sensing, due to the highly parallel pixel readout process that results in exceptionally fast frame rate capability, coupled with low read noise and high QE for optimal signal to noise ratio under conditions of short exposures. In this technical note, we consider three models from the Andor sCMOS range that show particular suitability to wavefront sensing: Marana 4.2B-6 (with CoaXpress interface), Zyla 4.2 PLUS (with CameraLink interface), Balor 17F (with CoaXpress interface), the key performance parameters of each model summarized in Table 1 below.

 Model Array Format Pixel Pitch (µm) Sensor Diagonal (mm) Frame Rate (16-bit, full array) Read Noise (median, e-) QE Max (%)
Marana 4.2B-6 2048 x 2048 6.5 18.8 74 1.6 95
Zyla 4.2 PLUS 2048 x 2048 6.5 18.8 100 0.9 82
Balor 17F 4128 x 4104 12 70 54 2.9 61

Table 1 – Comparing key imaging parameters of three Andor sCMOS models for wavefront sensing.

In Part 1 we will consider the potential frame rate performance in further detail, considering Region of Interest (ROI) capability. In Part 2, we will consider relative latency between the models, an important consideration for adaptive optics usage, as it dictates when images are ready in software for processing as part of the closed loop deformable mirror system.

Part 1 – sCMOS Frame Rates

Rapid frame rate performance is pivotal to wavefront sensing, with Region of Interest (ROI) sub-arrays commonly being employed to extend into hundreds of frames per second. Table 2 shows frame rates across a series of ROI array sizes for the three Andor sCMOS camera models being considered here as candidate wavefront sensors.

Key Imaging Parameters for Table 2 (where options are available):

  • Rolling Shutter exposure mode
  • Overlap (100% duty cycle) mode
  • 16-bit (full dynamic range) mode
  • ROI positioned centrally
  • CoaXpress (CXP) interface [Marana and Balor]; CameraLink (CL) interface [Zyla]
Array/ROI Marana 4.2B-6 Zyla 4.2 PLUS Balor 17F
4128 x 4104 N/A N/A 54
2048 x 2048 74 100 108
1024 x 1024 148 202 205
512 x 512 295 406 431
128 x 128 1166 1627 1684

Table 2 – Frame rates across a series of ROI array sizes for three Andor sCMOS camera models

Note that when comparing the Marana and Zyla models (each of which are 2048 x 2048 arrays), that while Zyla is capable of faster frame rates, the Zyla is not back-illuminated and achieves high QE through use of microlenses on each pixel. Marana on the other hand is back-illuminated, achieving up to 95% QE without microlenses.

Furthermore, please note that if the ROI for Zyla is not centred in the vertical direction, the frame rate will decrease (up to a factor of 2 slower), while for the Marana and Balor models the ROI can be placed anywhere with negligible frame rate reduction.

Part 2 – Comparative Latency

A key consideration for the usage of scientific imaging cameras as wavefront sensors is ‘latency’. Since wavefront sensors are part of a closed loop system in AO configurations, it is imperative that the image is quickly made available in software for real time processing, such that it can continually inform the deformable mirror system on how to reshape and flatten the incident wavefront on its way to the primary science detector.

This means that when comparing multiple cameras for potential wavefront sensing, we need to develop a clear understanding of the relative timings related to exposure, sensor readout and any image transfer overheads.

There can be some subjective variation around the definition of ‘latency’ within the timing flow of the imaging process. For the sake of standardising within the current comparative study, we will consider the total end-to-end time between the start of an exposure and the point at which the full image/ROI from that exposure is available for processing by software. We will also standardise by assuming an exposure time of 10 milliseconds (yielding 100 fps). Note however, that this 10 ms exposure corresponds to different ROI array sizes, and corresponding fields of view, across the three camera models that are being considered.

Figures 1 and 2 below show timing schematics comparing Marana 4.2B-6 to Zyla 4.2 PLUS. The key latency distinctions between the considered sCMOS camera formats is as follows:

  • Zyla must first readout the entire ROI array (10 ms) to the camera FPGA where the image is assembled, before the image is then transferred across the CameraLink interface, which itself takes a further ~ 10 ms. Since these processes operate sequentially rather than simultaneously, then the entire end-to-end process approximates to exposure (10 ms) + readout (10 ms) + data transfer over CameraLink (10 ms) = 30 ms. Note, the reason that the Zyla image must first be assembled on the FPGA is due to the more complex sensor readout, which involves reading out two halves of the array simultaneously, starting with the middle rows and moving outwards towards the top and bottom rows.
  • Marana has a more straightforward sensor readout architecture which means there is no requirement to assemble the image on the camera FPGA before it is transferred to the host PC. Instead, as soon as a pixel row is readout, it is processed by the FPGA and is immediately transferred across the CoaXpress (CXP) interface. This means that the image transfer occurs simultaneously with the image readout, rather than sequentially, thus overcoming a significant contributor to latency. The entire end-to-end process for Marana approximates to exposure (10 ms) + simultaneous readout/data transfer (10 ms) = 20 ms.
  • Balor, although not specifically represented in the shown schematics, has a similar unidirectional sensor readout architecture to Marana, although with a distinction that Balor achieves a speed efficiency boost by reading out every group of 4 rows simultaneously. As such, if an ROI array was defined for Balor that resulted in a 10 ms exposure (and corresponding 10 ms readout), then the entire end-to-end process for Balor would also approximate to exposure (10 ms) + simultaneous readout/data transfer (10 ms) = 20 ms.

As such, the latency of Marana and Balor is reduced relative to that inherent to the Zyla platform. However, bear in mind that, as shown in section 1, faster sensor frame rates are possible for Zyla 4.2 PLUS relative to Marana 4.2B-6. In selecting the most appropriate wavefront sensor for your set-up, both factors should be considered within the context of your exact experimental requirements.

Key Imaging Parameters for Figures 1 and 2 (where options available):

  • Exposure time / readout time – 10 milliseconds (ROI selected, if applicable, to achieve this)
  • Rolling Shutter exposure mode

Figure 1 – Zyla 4.2 PLUS: Timings schematic representing exposure, readout and image transfer (over CameraLink interface).

Figure 2 – Marana 4.2B-6: Timings schematic representing exposure, readout and simultaneous image transfer (over CoaXpress interface). Note, Balor is very similar.

Date: April 2020

Author: Dr Colin Coates

Category: Technical Article

Download as pdf


Related assets