Advertisement

Blog

Integrated CMOS Image Sensor Architecture: The Analog Camera

How does a camera see the image it is about to capture? If it's a digital camera, the microcontroller “sees” the image ultimately and works its magic by recording and manipulating the analog image detected by the CMOS image sensor.

The camera works like our eyes, which convert the light from the subject (image sensor) to electronic signals via the optic nerve, which ultimately reach the brain (processor) where they are decoded. The lens focuses the light from the image and transmits this to the retina at the back of the eye. The lens of the eye can fine tune the focus — muscles stretch the lens, changing its shape and changing the focus. The iris widens and contracts to control the amount of light coming into the eye.

Figure 1

The eye captures a viewed image that the camera tries to emulate. (Source: University of Alabama, College of Engineering)

The eye captures a viewed image that the camera tries to emulate.
(Source: University of Alabama, College of Engineering)

A recent paper from the IEEE Journal of Solid-State Circuits in January 2013, “CMOS Image Sensor With Per-Column ΣΔ ADC and Programmable Compressed Sensing,” describes a CMOS image sensor, some delta-sigma ADCs in the analog section, and random pixel processing in the digital realm. All of this can be done in CMOS, and a high level of integration can be achieved easily with added system functions.

The principle upon which all signal processing acquisition systems are based is the Shannon-Nyquist Theorem: A band-limited signal can be effectively recovered via sample sequences as long as the sampling rate is more than two times the bandwidth of the original signal.

Figure 2

An image is captured through a lens and into an image sensor much as the eye does naturally. In this figure, a camera employs a technique called random phase-shift mask technique. This implements CS.
(Source: CMOS Image Sensor With Per-Column ΣΔ ADC and Programmable Compressed Sensing)

An image is captured through a lens and into an image sensor much as the eye does naturally. In this figure, a camera employs a technique called random phase-shift mask technique. This implements CS.
(Source: CMOS Image Sensor With Per-Column ΣΔ ADC and Programmable Compressed Sensing)

The compressed sensing (CS) technique says that fewer samples can be taken than the Nyquist theorem suggests if they are at an information rate lower than the Nyquist rate.

The sensor has a compressed sensing mode with various sensing ratios of 1/4, 1/8, and 1/16 at 480, 960, and 1,620 frames per second (fps), respectively, as well as no compression at 120 fps. The remarkable fact about this architecture is that it is done with only 1.8 percent die area overhead.

Figure 3

A microphotograph of the image sensor IC shows the high level of integration into a relatively small die area.(Source: CMOS Image Sensor With Per-Column ΣΔ ADC and Programmable Compressed Sensing)

A microphotograph of the image sensor IC shows the high level of integration into a relatively small die area.
(Source: CMOS Image Sensor With Per-Column ΣΔ ADC and Programmable Compressed Sensing)

The image reconstruction with this technique shows minimum image quality degradation as compared to normal image capture and much higher image quality than downsampling. Downsizing usually improves noise by reducing the image to a lower resolution.

Power management
Most CMOS image sensors are used in mobile devices nowadays, and you want your battery to last a long time between charges. The IEEE paper cited above states that the biggest contributor to power consumption in a CMOS image sensor with column-parallel ADCs is the ADC conversion and then the readout stage. Therefore, power consumption in CMOS image sensors increases linearly with resolution and frame rate.

Figure 4

Energy consumption per frame, structural similarity (SSIM), and peak signal-to-noise ratio (PSNR) versus compression ratio CR.
(Source: CMOS Image Sensor With Per-Column ΣΔ ADC and Programmable Compressed Sensing)

Energy consumption per frame, structural similarity (SSIM), and peak signal-to-noise ratio (PSNR)
versus compression ratio CR.
(Source: CMOS Image Sensor With Per-Column ΣΔ ADC and Programmable Compressed Sensing)

While the readout rate power consumption is reduced by on-chip image compression, this does not improve the ADC power consumption during conversion.

Compressed sensing (CS)
CS is a relatively new sampling technique that reduces the number of captured measurements so that fewer ADC measurements are needed. All this is done without degrading the recovered signal. The IEEE paper discusses an improved CS architecture that starts with the premise that if multiple pixel values are sequentially applied to the delta-sigma ADC, and you follow up with decimating the entire output sequence, the ADC output now represents a quantized version of the sum or average of the pixel values digitized.

Maintaining high SNR is critical to help reduce the image quality degradation caused by image compression.

The delta-sigma ADC
As an alternative to the conventional, higher-order noise shaping in traditional delta-sigma ADCs, the algorithmic scheme samples the modulation residue iteratively by feeding back the integrator output to the input, according to the IEEE paper, “Delta-Sigma Algorithmic Analog-to-Digital Conversion.” This technique combines the advantages of delta-sigma ADC conversion with algorithmic (cyclic) ADC conversion in one architecture, as discussed in the paper.

Figure 5

First-order algorithmic delta sigma ADC with its timing diagram.
(Source: CMOS Image Sensor With Per-Column ΣΔ ADC and Programmable Compressed Sensing)

First-order algorithmic delta sigma ADC with its timing diagram.
(Source: CMOS Image Sensor With Per-Column ΣΔ ADC and Programmable Compressed Sensing)

The algorithmic delta-sigma ADC architecture does not suffer from idle tones, since it operates in the incremental mode in which the ADC modulator and decimation filter are reset before each ADC conversion cycle.

Summary
This technique demonstrated a significant reduction in power consumption and/or increase in frame rate using this architecture. With image recovery complexity off-chip, this allows a smaller, lower-power solution to be integrated into the front-end camera system and imager.

Related posts:

8 comments on “Integrated CMOS Image Sensor Architecture: The Analog Camera

  1. David Maciel Silva
    July 31, 2013

    Congratulations for the article Steve,

    As an example, Samsung makes the sale of these sensors for different applications.

    http://www.samsung.com/global/business/semiconductor/product/cmos-imaging/catalogue

  2. Steve Taranovich
    July 31, 2013

    Thanks for sharing Maciel

  3. Davidled
    August 1, 2013

    Camera focusing technology is mainly affecting the image resolution. They might use auto-focus function to capture the image with data transmission rate. High speed camera is very useful for real-time tracking applications. I think that data transmission rate relies on the delta modulator partially as shown in Fig 5.

  4. fasmicro
    August 6, 2013

    It would be hugely valuable if he can share the circuit of the imager. I know that Eric Fossum's API CMOS imager is the industry standard. It is used in most of the latest imagers today. Ridiculously simple with 5 transistors, it continues to reshape the industry. We have the ADC, can we see the sensor circuit?

  5. fasmicro
    August 6, 2013

    Compressive sensing is certainly going to become a huge area as data explodes and companies delay expansion. I see a future where companies that figure how to pipe more data over smaller BWs will win this competitive market. This is the game and the future and there is no going back. Compressive sensing is doing a lot in video transmission.

  6. SunitaT
    August 20, 2013

    A CMOS imaging chip is a sort of active pixel sensor prepared using the CMOS semiconductor procedure. Additional circuitry next to each photo sensor alters the light energy to a voltage. Extra circuitry on the chip may be involved to convert the voltage to digital data.

  7. Brad_Albing
    August 23, 2013

    @DaeJ – do you know how they do autofocusing?

  8. Davidled
    August 24, 2013

    Lens is adjusted automatically to obtain focus on the subject. There is sensor for autofocus which measures the relative focus at the focus point with sharp edge resolution to make sure that lens focus on subject, not background.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.