In search of the right processing power for High Definition

By Tsahi Levent-Levi

Categories: Technology

July 10th, 2008

Everything today is High Definition (HD). Whether you’re talking about televisions, projectors, video conferencing or even voice calls – the latest and greatest is high-definition capable. That said, high definition is really complex to deal with, not just in terms of bandwidth requirements, but in terms of processing power – especially if we’re dealing with visual communications.

The challenge

Have you ever tried watching a movie encoded in high definition format on your laptop? Assuming you have a single core CPU in your laptop, the results would be less than satisfactory. The reason is there is simply not enough “juice” in your processor to decode the HD stream and display it. Add to that the need to encode in real-time (necessary for two-way communication) and you’re really stuck.

To sum it up, it is basically an issue of scale:

• High definition video requires processing 9-20 times the amount of data required in the past. According to Moore’s law, such a giant leap in processing power will take at least 4 years to become a reality.

• Video telephony requires not only decoding but also encoding high-definition. Encoding and decoding must be done in parallel and in real-time.
• Encoding takes about 2-3 times more processing power than decoding power. This means visual communications are about four times harder to achieve than HDTV, for instance (which requires “just” decoding).

This amount of additional processing required for HD is a difficult idea to absorb, and therefore there are several different approaches on how to obtain information.

Solutions

• Software codec
  Use a normal, general purpose CPU, and write a “pure software codec” on top of it. Although this is the most flexible solution, it is irrelevant when considering today’s general-purpose CPUs (found in most computers). The latest and greatest laptops, dual cores inside, are still unable to encode and decode a 720p stream at the same time. On the other hand, this solution is the simplest to manage – single operating system, one CPU.
• Hardware codec
  In this case, the codec is packaged in a piece of hardware (designated chip, for instance) connected to the main CPU. Although the efficiency of this solution is the best, flexibility (changing the codec to accommodate different use-cases), improving it, adding your own IP to it, or simply fixing bugs is very limited.
  Most of today’s high definition video conferencing endpoints are based on hardware codecs.
• DSP-based codec
  This kind of codec runs on a general purpose DSP (Digital Signal Processor), which is connected to a host CPU that runs the application. Writing an optimized codec on a DSP is harder than developing a software codec, but most DSPs offered today have special instruction sets that manage the resource-intensive aspects of the codec very well.
  Most of today’s video conferencing bridges are based on DSP-based codecs.
• Acceleration
  There is another option, a hybrid of the above solutions, which is becoming very popular – splitting the codec between a general purpose processor (CPU or DSP) and a dedicated hardware accelerator. A hardware accelerator is a piece of hardware that is optimized for low-level resource-intensive functions of which codecs are comprised. The codec therefore works in a distributed mode – the processor implements the codec “logic” and framework, while the accelerator takes care of the more low-level “hard” tasks. This solution retains a lot of the flexibility lost in a full hardware codec, but requires delicate and complex synchronization between the various components.

There are many solutions to the problem of developing a high definition video conferencing product, but still no magic silver bullet has been found – there’s a lot of hard work involved. And it all begins with selecting from one of the solutions above, the one that best fits your needs and capabilities.