Zenverge's ZN Series: Real-Life Results Support The Theories

Plenty of companies pitch me on new products, armed with nothing but a pile of PowerPoint foils. Inevitably, many of the promised chips never appear, which is why I make it a rule to not bother telling all of you about any IC that doesn’t have accompanying pricing and sample-and-production availability statistics. And that’s why I didn’t write up Zenverge’s ZN100 and ZN200 audio-plus-video-plus-DRM transcoders last November, when the company first gave me an over-dinner presentation on them.

Many of the other promised chips arrive late, accompanied by performance, power consumption and other specifications that drastically undershoot the supplier’s preliminary promises. And that’s why I was so pleasantly surprised when Zenverge invited me to meet with them again a month ago, this time at their Cupertino facilities. Here’s a block diagram of the company’s ZN200 ($50 in sample quantities), which claims to transcode MPEG-2 and various MPEG-4 video flavors (including Part 10, aka AVC, aka H.264) in high-definition at up to 4x real-time speeds:

The ZN200 also promises to downscale high-definition video to ‘portable device resolutions’ at up to 40x real-time rates. An accompanying sonic processor handles converting between AAC, AC-3 (aka Dolby Digital), MP3 and other audio formats, and the ZN200 also internally converts between various DRM standards. The lower- (albeit unannounced-) priced ZN100 has half the video processing performance of its ZN200 sibling.

The value of transcoding is something I’ve written about for many years; witness my Stream Machines writeup from December 1999, my iCompression piece from May 2000, Vweb’s processors from October 2001 and before, or ViXS’ chips from October 2004. Ironically Zenverge’s co-founders, President/CEO Amir Mobini and COO/CTO Tony Masterson, come from iCompression, which was subsequently acquired by GlobespanVirata and later sold to Conexant. And here’s how Masterson summarizes, in a technical paper he authored, the value of chips such as those his company provides:

1. Fix the format - Terrestrial HD content in the U.S. is broadcast as MPEG-2. However, portable video players and cell phones only support formats such as H.264, so transcribing MPEG-2 into H.264 is required. Cell phones lack adequate processing power. For this reason, even if the original content is in the H.264 format, transcribing is still typically needed to convert from high or main profile to baseline profile H.264 that cell phones typically support.
2. Fix the resolution - Due to the size of the screen on a cell phone, the resolution of HD content must be transcaled from either 1920×1080 or 1280×720 into the proper size for the cell phone which might be 320×240.
3. Fix the frame rate - Due to the power and size constraints on cell phones, they typically can only play content at 15 fps. For this reason, the 30 fps HDTV content must be converted to 15 fps. This requires sophisticated conversion between the different types of compressed pictures including I, P, and B pictures.
4. Fix the size - Cell phones don’t have the capacity to store HD content, which can require 9GBs to store just 1 hour of content. Transrating in combination with transcribing and transcaling will dramatically reduce the size of the content. This whole process can include lengthening the GOP structure of the compressed sequence, increasing it from 15 to up to 300.
5. Do it fast – Nobody wants to wait around while their content is being converted. Ease of use requires that this conversion happen much faster than real time. Ideally 4 to 40 times faster than real time.

To that latter point, while the ability to transcode a single high-def stream at 4x real-time rates may be overkill for your particular application, consider that Zenverge’s packet processing architecture alternatively allows the chips to concurrently handle multiple streams with little to no efficiency loss. Such an approach might, for example, enable a set-top box to record one or a few channels while playing back another in a time-shifted manner…

…which leads me to the demos I saw a month ago, based on a ZN200 mounted on a PCI Express x1 add-in card, and encompassing video encoded at various bitrates, resolutions and formats. Granted, they didn’t hit the exact speeds forecasted by Zenverge’s materials, but they came impressively close, and company officials were quick to point out that the chip was running function-focused (i.e. non-performance-optimized) code that was also full of speed-strapping debug routines. And the ZN200’s ability to substantially reduce bitrate (translation: required storage capacity) while retaining reasonable image quality was quite impressive.

So what’s next for the company? I know, because I’ve got access to confidential materials, but I can’t tell you yet Moore’s Law-fueled integration is always a good bet, however; revisit the above block diagram and let me know in the comments what currently-outside-the-IC functions you think Zenverge’s next product iteration will absorb.