Post a Comment
Feature
Video surveillance: It’s all about computing power
Prying Eyes: Explore a reference design for concentrating video from a cluster of cameras.
By Ron Wilson, Executive Editor -- EDN, 9/27/2007
|
1. The board uses four Techwell TW2815s (see details below) to gather 16 video and 16 audio channels. These four-channel devices acquire NTSC/PAL composite video and analog audio, filter it, and convert it to BT.656 digital video for subsequent processing.
2. Four Stretch S6100s (see details on the S6000 family from EDN's DSP Directory) compress the incoming video using a software-based multistandard compression engine. Each device works on four channels, with one doubling as a supervisor. The chips working together are capable of H.264 D1 resolution encoding at 30 frames/sec on each channel. Significantly, there is processing room left over for video analytics that can scan images for content, adjust cameras or compression rates, or provide raw data for downstream algorithms.
3. As is increasingly the case with advanced SOCs, the power-supply circuitry consumes nearly as much space as the computing heart of the board. The design must support the needs of the computing chips, DDR2 DRAMs, and analog front-end devices.
4. Each Stretch chip uses a pair of Micron D9GMG DDR2 DRAMs for local storage in addition to the SRAM on the S6100 dice.
Read more Prying Eyes
5. Stretch designed the board to use PCI Express as its means of communication with a host processor, which would save the compressed video locally or transmit it to a remote hub. The S6100 chip directly supports the interface.
Techwell TW2815 details
Conversion of analog NTSC or PAL video to a Bt.656-standard YCrCb digital signal used to be a board-level function in its own right. The process requires anti-aliasing and sampling the incoming video signal in a way that doesn't undermine later steps. Then the video goes through an ADC sufficiently fast and with enough dynamic range to make a high-quality digital representation possible. Both of these essentially analog jobs are non-trivial, and industry experience has shown that the design team undertaking them needs not only fine analog design skills but a detailed knowledge of what does, and does not, contribute to image quality in the composite video signal.
The rest of the job is digital signal processing. The now-digitized signal must be sent through a comb filter to isolate the chrominance and luminance portions of the signal. But it must also be clamped, gain-adjusted, and otherwise massaged to ensure that the chrominance and luminance components are freed from noise sources such as IF interference, mutual interference and signal-strength variations. If the signal being captured includes audio, an entire capture-and-conditioning chain is necessary for the audio signal as well.
That this process is understood at all is a tribute to engineering skill. But the state of the market today is that you can go out and buy a four-channel chip that performs all of these functions under the control of a host processor. So drop a chip onto your board, and you can gluelessly deliver Bt.656 digital video to your video processing subsystem from an analog input.
An instance of this ability is the Techwell 2815. The chip uses 10-bit integrated video ADCs. It provides a combination of automatic signal-conditioning features, such as white-peaking and clamping, and programmable adjustments such as hue, saturation, contrast, brightness, and contrast. The part operates at up to D1 resolution.
