Scalable Systems

Managing product portfolio costs with proper applications

• By Mark Oliver
• Feb 01, 2010

Technological refinements and cost effectiveness in video surveillance systems are changing the security industry as we know it. Just about every aspect of a surveillance installation benefits from enhanced capabilities. While this provides a significant leap in feature set and functionality, it also requires equipment manufacturers to become technology experts if they are to remain competitive. Consider, for example, some of the changes taking place in the industry:

• Traditional CIF and 2 CIF video has already been replaced by D1 resolution in most new installations. Now, with megapixel image sensor technology made cost effective by high-definition consumer applications, surveillance video is beginning to migrate to high definition and beyond.
• Greater resolution requires more efficient compression. MPEG-4 is rapidly being displaced by H.264 AVC, made popular in high-definition television and Internet video. A newer, scalable codec, H.264 SVC, promises to deliver increased flexibility and will, over time, displace H.264 AVC.
• High-definition video cannot be effectively transported in analog form using traditional coaxial cable. Digitizing video feeds and leveraging IP networking technologies solves the transport problem but adds complexity to surveillance installations.
• Standard-definition monitors with the familiar analog composite inputs are being replaced by flat-panel monitors with varying digital or high-definition (VGA, DVI, HDMI) inputs.

Staying Adaptable

To address these changes, broader product portfolios are required. Additionally, adoption of the latest technology requires using the latest high-speed interfaces such as SATA, PCIe, gigabit Ethernet and the notoriously difficult DDR2/3 memory interfaces. This increased complexity leads to higher risks and longer design times for products that must compete for a share of an increasingly diverse marketplace.

Development cycles can take many years and cost hundreds of thousands of dollars. Add to this the expense of regulatory certification, firmware development, the porting of application software and the hidden costs of support, and developing a portfolio of such products can become prohibitively expensive.

Traditionally, manufacturers have countered these issues by using production-ready reference designs.

Supplied by major component manufacturers, these reference designs provide examples of how their component can be used. By solving some of the more difficult design problems up front, these designs can reduce program risk and shorten time to market. Unfortunately, they do little to address the inherent problems of managing a portfolio of disparate architectures. That is why the ideal product portfolio would be derived from a single, scalable reference design.

Products would then share a similar architecture, code base and programming interface, and application software would be common across the entire product range. This would dramatically reduce the costs associated with supporting and maintaining that portfolio.

A Single Solution

This was the overarching objective when the idea for a high-definition standalone hybrid DVR using its S6000 family processors was conceived. These processors are software configurable, so a single, scalable board design could be created to address varying product requirements. A feature set was defined that spanned the entire product space from CIF to high definition, from analog video to HDcctv, from composite output to HDMI and from MPEG-4 to H.264 SVC.

Furthermore, a design objective was that the reference design should retain a single programming interface for software compatibility and to allow derivative products to be created by doing little more than depopulating areas of the circuit board.

Surveillance DVR reference designs are characterized by their I/O interfaces, which can be a major design impediment. These interfaces have dedicated hardware to support the required standards. Analog video inputs, for example, have associated video decoders and analog- to-digital converters. High-definition video outputs might have HDMI transceivers. Each of these physical interfaces also will have an associated compute requirement. Analog inputs might require color space conversion, noise reduction filtering or de-interlacing.

Video outputs might require video scaling or graphical overlays. For a reference design to remain scalable, each of the optional hardware interfaces should be a self-contained entity, depopulated if not required but with access to compute resources if needed.

In a software-defined solution, it is the system firmware that delivers its functionality. Firmware requirements for DVR applications are dominated by video processing functionality and codecs. By designing these elements in a modular manner, the appropriate video processing algorithm or codec can be instantiated and associated with a particular stream. Video pipelines can be constructed to associate a particular stream with a particular hardware interface to provide the required system functionality.

The compute resources needed can be estimated by creating worst-case scenarios for the video pipelines. Processing elements can be selectively populated or depopulated to satisfy the compute requirements of any given application.

To satisfy the functional scalability requirement, it is critical that a highspeed, low-overhead bus be provided between processing elements. A bus of sufficiently high bandwidth means that the overall compute capability of the system can be calculated irrespective of the physical location of the compute elements themselves. For example, analog video input might be de-interlaced and filtered on one processing element and compressed on another elsewhere in the system.

Finally, to satisfy the requirement for code compatibility across a portfolio, the entire architecture should be hidden from the top-level application. Exposing a single application programming interface to a host processor provides a highly abstracted view of the underlying architecture and creates a common approach to controlling data flow within the system.

Put to the Test

The VRM6016 from Stretch is one implementation of the preceding design considerations.

The architecture comprises a series of physical hardware interfaces that can be selectively populated to meet the requirements of a particular design. Each of the hardware interfaces is associated with a compute element, which, in addition to its obligations to the hardware interface, forms part of a larger pool of compute resources. Additional compute elements can be added to the network as needed to satisfy the worst-case compute requirements anticipated for the product. A host processor—VRM6016, a PowerPC—provides an industry standard and a familiar platform on which to develop application code.

Those applications use an API to control the flow of data though the system without regard for the underlying architecture. In this way, the hardware interfaces and software features can be scaled to meet product requirements without the need to change application code.

An entry-level DVR design might contain three processing elements and perform H.264 compression on captured analog video, outputting video on a standard-definition monitor. Adding an HDMI physical interface (and its associated processing element) would give the reference design high-definition output capability without any further changes. Adding an additional processing element might give the reference design simultaneous decode capability.

It can readily be seen that, starting from a basic system, additional features such as high-definition output can be rapidly added to create a portfolio of products with a range of I/O capabilities. A rich API abstracts away the detail of the underlying architecture and allows the DVR application programmer to write code that can be scaled as features are added. Product design cycles are shortened from years to weeks, and the cost of bringing products to market is dramatically reduced.

With an increasing diversity of technologies becoming available to the surveillance market, the use of traditional approaches to designing portfolios of products to cover a solution space is becoming prohibitively expensive. A more scalable approach is called for, whereby a common architecture can be used to cost effectively derive a widely diverse set of solutions while retaining software compatibility. For the first time, highly cost-optimized designs can be developed that reuse existing software with dramatic reductions in program risk and development cost.