Getting the picture? Streaming video to a mobile has long been a dream, but even without 3G it can be done - Video Apps

Getting the picture? Streaming video to a mobile has long been a dream, but even without 3G it can be done - Video Apps
Wireless Business & Technology, Oct, 2003 by Chetan Sharma

One of the aims of 3G mobile communications systems is to provide enhanced multimedia services, like video streaming and MMS, to the user. The global economic slowdown has forced most of the mobile wireless carriers to delay deployment of costly 3G-network infrastructure, and optimize the existing digital infrastructure to initiate promising multimedia services. Offering of multimedia services with current communications infrastructures (referred to as 2.5G/2.75G) is a technological challenge because of lower bit rates, high bit error rates, unpredictable delay, QoS issues, and limited device capabilities.

2.5G/2.75G networks are being implemented worldwide at a rapid pace, despite the economic downturn. These networks employ GPRS (derived from GSM), CDMA 1x, and EDGE access technologies. GPRS is an end-to-end mobile packet radio communications system that makes use of the same architecture as GSM. Its features such as multislotting and native IP support allow it to be used for delay-sensitive data applications like video streaming.

But the inherent characteristics of the wireless channels are still there and hence streaming video system designs should not only focus on the source coding, but should also incorporate channel characteristics to provide the best quality to the end user. EDGE is considered to offer higher channel data rates than GPRS, but EDGE is not widely deployed. CDMA-based networks utilizing CDMA 1x technology offer higher channel data rates, and are being implemented widely in the U.S., Canada, Korea, Japan, India, and many other countries.

Bandwidth Constraints of 2.5G/2.75G Networks

While carriers are making tremendous progress in enhancing their networks for efficient access to multimedia content, there are some issues with the current 2.5G (GPRS) and 2.75G (CDMA 1x, EDGE) networks. GPRS currently delivers only 10-20Kbps on average, while CDMA 1x averages between 20-40Kbps bandwidth to handsets. EDGE is claimed to deliver an even higher average data rate (~64Kb/s), but the actual throughput, and extent of coverage, have not been verified. During peak hours, the data traffic performance can further degrade, as voice traffic still takes priority.

On top of that, applications and services have to contend with latency in the network, which can impact synchronous applications like video streaming. Though we are moving to "always-on packet" networks, the connection can still be choppy, meaning one could hit dead zones and the coverage might drop for a few seconds before the network picks it up again.

Diversity in handsets and their capabilities can have a major impact on user experience as well. Multimedia content, like video, needs to adapt not only to the network and bandwidth conditions but also the device capabilities and limitations, to effectively push content whether it is streaming video or site browsing.

Delivering high-quality video over 2.5G wireless networks is a daunting task. Customers are not going to flock toward multimedia services until they can get "clear, uninterrupted" video, even if it's a 60-second or 2-minute clip. This is why several streaming service providers have focused on delivering high-quality "streaming audio" over 2.5G, and are preparing for full-scale multimedia services when 3G finally arrives.

Video Streaming over 2.5G Networks

Streaming codecs operate at much lower bandwidths (lower data rates) than the encoded data rate of the video clip. Buffering is used at the handset to accommodate the downloading of video frames at lower data rates. Streaming circumvents the need to download the entire video clip before it can be decoded and played, and offers near real-time playing of content. However, there is always a price to pay.

Frames will need to be buffered before they are decoded and played. If the encoded data rate is high to begin with, the buffering requirements are high. Further, the frame rate of the decoded and displayed video cannot be improved. There are two downloadable decoders (players) that claim to stream video clips over GPRS networks at 9.6Kb/s (according to Steve Wallage, "Media on the Future," 2003).

All the streamed clips over 2.5G/2.75G networks are 30-60 seconds in length at the most. Streamed video can be delivered directly via a server or in the MMS mode. SingTel Optus of Australia offers streaming video service of 30-second clips of news, sports, and movie reviews. Emblaze is another company that offers streaming video capability, but the video quality is marginal. Further, the processing load on the device's main processor tends to slow down the displayed content. In a nutshell, the "video reality" cannot be replicated.

Catch-22 Situation for 2.5G/2.75G Carriers

Wireless carriers know that delivery of multimedia services can be an important revenue source. They are also aware that a threshold quality level of content of a minimum duration (2 minutes), and at minimum cost, is important to drive user demand. Carriers also want to ensure that video streaming or MMS can be accomplished with lower data rates (10 Kb/s, 20Kb/s, 40Kb/s). Surely, it is desirable to stream at 15fps with minimum buffer requirements, utilizing as low a channel bandwidth as possible. This is not practical with inefficient video codecs. Further complicating the issue is the need to maintain a low processing requirement and low battery power consumption in the handset.

From cost and time-to-market points of view, using and enhancing the existing 2.5G/2.75G infrastructure to deliver multimedia services makes sense. However, the available DCT-based codec technology (MPEG-4, for example) has inherent limitations regarding performance, processing requirements, and content display (frames/sec, duration, etc.). Some of these deficiencies can be tolerated by going to 3G.

As mentioned earlier, 3G deployments are proceeding at a snail's pace, and 2.5G/2.75G network deployments are taking place at a rapid pace. So, this is posing a Catch-22 situation to service providers. Under these circumstances, relying entirely on MPEG-4 technology to help achieve success in the wireless multimedia marketplace with a 2.5G/2.75G infrastructure is a daunting task. However, there are solutions available, if you accept the fact that it is not necessary to depend on MPEG-4 technology for delivering multimedia services over 2.5G/2.75G networks. In fact, it may be less efficient and of marginal use to use MPEG-4 to offer an evolving set of multimedia services using MPEG-4.

Need for a More Efficient Video Encoder

More efficient encoders are needed to deliver high-quality video clips over 2.5G/2.75G wireless networks. Packet Video is quoted recently in the trade press (Steve Wallage, Media on the Future, 2003) as saying that their codecs can deliver video at 5-6 frames/s at 30Kb/s, falling to 2-3 frames/sec at 9.6Kb/s. This is the streamed data rate; the raw encoded data rate is much higher. According to QUALCOMM, a typical QCIF image at 15 frames/s will require 96-128 Kb/s, with MPEG-4 codecs.

Clearly, this level of performance is not acceptable to enable video streaming and MMS services to truly take off in 2.5G/2.75G wireless networks. As we mentioned earlier, it is necessary to increase the clip duration from 2 to 5 minutes. Perhaps we will someday be able to play a complete movie on a handset without recharging the battery.

Many leading market research firms are forecasting a big market for wireless multimedia services. Strategy Analytics (2003) estimates that wireless streamed media services to handheld devices will generate revenues of around $5.7 billion by 2008. This does not include other related multimedia services (i.e., MMS messaging of still images, graphics, and text).

Over the last few years, several alternative encoding formats have been launched to deliver multimedia services over the Internet. They are not compatible, but are all based on the DCT encoding technology that has serious limitations in high compression environments. This battle with multiple proprietary compression standards has been carried over to the wireless world. Many proprietary solutions have been introduced in the marketplace. Thin Multimedia, Office Noa (Nancy Technology), and Sindhara SuperMedia are companies with proprietary encoding technologies. In addition to providing streaming solutions, these companies offer proprietary algorithms to push content and small footprint Java applets to decode at the wireless devices. With this software-based solution, content delivery over wireless is made relatively easy. But the question remains: Which of these solutions use less bandwidth and processing/power resources, and offer good video quality?

Codecs Based on Wavelet Technology

It has been well documented in the literature that wavelet encoding is superior to the DCT-method of encoding that's at the heart of MPEG-4 codecs. Wavelet codecs are more efficient and robust, and degrade gracefully, in the hostile transmission environment of wireless. All the focus to date has been on using wavelet encoding of still images. JPEG-2000 is one example where wavelet transforms are used to process still images.

Spatially and temporally correlated data such as image and video frame data can be efficiently compressed by decorrelating the pixel data using one of several transformation techniques available. Unlike classical Fourier-based transforms (like DCT) used in MPEG-4, wavelets employ multi-resolution analysis (a comparatively new technique), perfected in recent years.

Wavelets decorrelate the data in such a way as to preserve vital spatial information about the entire frame, which is lost if using DCT-like transforms. Wavelet encoding leads to higher fidelity reconstruction of the video sequence; all other factors remaining invariant. Current video coding standards (H.263, MPEG-4) employ block-based discrete cosine transform (DCT). At very low bit rates, DCT based codecs suffer from blocking artifacts and "mosquito noise."

Improving on the wavelet transform and encoding schemes, to enhance intraframe encoding efficiency, and conduct processing-intensive Motion Estimation/Motion Compensation (ME/MC) in a way that takes advantage of the intrinsic features of wavelets, rather than depending on the spatial correlation of frames to derive motion vectors and frame-difference information, will result in a very efficient encoder. Wavelet codecs have arrived, and should be considered for wireless multimedia applications.

Sindhara's High-Efficiency Wavelet Codecs

While there are several wavelet codecs on the market, solutions from Sindhara are superior in terms of quality and resources. A well-implemented wavelet codec (such as QwikStream from Sindhara) can provide streamed video over wireless connections from as low as 10Kbps, thus enabling deployment of streaming solutions in present-day GPRS and CDMA infrastructures. Part of the value of such codecs is their ability to dynamically adjust the stream to allow for prevailing network conditions.

The scalability feature allows the content to be created only once and distributed to a wide range of users connected through different bit-rate channels.The error-resilient tools of such codecs incorporate features that allow the reconstruction of the entire frame even if 75% of the frame data is lost or corrupted. Other error-resilient tools include error detection, correction, and concealment features that make it more robust to channel noise (bit errors) as well as packet loss.

Network-aware middleware can be used to adapt to the changing network conditions and deliver streaming video in the best possible way with consistent quality, under any network conditions. Thus, frequent disconnections and broken streams are never experienced. The delivery and playback mechanism may also incorporate features to minimize the buffering and network delay, ensuring smooth playback. It can also incorporate a mechanism for prioritization of media streams using network-specific protocols to increase the overall throughput and minimize delays. With the right tools, a QCIF (176x144) video clip at 15 frames/sec can be encoded and transmitted at less than 64Kb/s (including audio). This compares favorably against the capability of MPEG-4 video codecs that require around 128Kb/s. It is not necessary to play the numbers game. Plainly, well-designed wavelet codecs are indeed superior to DCT codecs.

Conclusion

Mobile wireless carriers will focus on achieving maximum use of their current 2.5G network infrastructure, with minimal enhancements. Wireless multimedia has been identified as a major source of new revenues. Video streaming is indeed a lucrative service. Any technology that enhances the capability of the 2.5G networks to deliver high-quality video streaming and MMS services will be in great demand. Companies like Sindhara offer multimedia solutions based on proprietary implementations of wavelet technology, offer high-performance delivery of text, still image, and video over 2.5G/2.75G wireless networks.

Chetan Sharma is a wireless industry consultant focused on enterprise and multimedia applications.

He is co-author of Wireless Data Services: Technologies, Business Models, and Global Markets to be published in 4Q03 by Cambridge University Press.