New Ways to Manage IP Streams Offer Bandwidth Gains on HFC

Amit Eshet, senior director, media processing, BigBand Networks

Amit Eshet, senior director, media processing, BigBand Networks

March 25, 2011 – BigBand Networks last month demonstrated a new approach to minimizing the bandwidth impact of IP video streams that goes well beyond adaptive streaming to help lower MSOs’ costs of introducing IPTV services.

With IP stream counts escalating as video-capable devices proliferate, the providers of access bandwidth, including telcos and mobile operators as well as cable companies, need new ways to minimize the impact, says Amit Eshet, BigBand’s senior director of media processing. “Adaptive streaming (AS) is a good option today, but it’s not good enough for future needs,” Eshet says.

IP data of all descriptions is delivered in unicast mode over the Internet, including video. AS, which adjusts the bit rate of a video stream accessed by a user based on device requirements and how much bandwidth is available, has come into play as a way to maintain a steady video stream as bandwidth fluctuates without causing buffering delays or undue losses in quality.

But AS is designed for unmanaged networks. With managed networks like cable’s in play to deliver IP video, new options are available, including multicast, the IP equivalent of broadcast, as well as unicast.

For operators, Eshet says, the solution entails making the network smart enough to leverage AS to greater advantage while also exploiting the bandwidth efficiencies associated with multicast, especially as the latter can be applied in conjunction with multiplexing the IP streams over QAM (quadrature amplitude modulation) channels together with legacy MPEG-TS (transport streams).

“The spectrum dedicated to the delivery of video to set-top boxes can also be used for the delivery of video to IP devices in a ‘bandwidth-free’ manner that doesn’t degrade the legacy service or require massive capex investments up front,” he says. “The net benefit to the cable operator is a smoother and more cost effective migration to IPTV.”

Eshet makes clear the solution demonstrated at CableLabs’ Winter Conference in Atlanta is not yet a product. “We’re leveraging a technology concept that is part of BigBand’s advanced video processing capabilities, but it’s not a product today,” he says. “We’re working with customers to shape the solution to their needs, and then we will productize it.”

BigBand, of course, is not alone in putting new ideas on the table to help cable operators avoid big capex outlays for network upgrades as they move to IPTV services. While BigBand’s solution depends on the techniques it has developed for edge-based video processing that bypass the DOCSIS CMTS (cable modem termination system), CMTS vendors report they are likewise devising new approaches to managing IP streams that go beyond the capabilities built into the DOCSIS 3.0 specifications.

Indeed, Eshet acknowledges that some of the concepts posited by BigBand are doable through video processing positioned in the CMTS. But he cites additional efficiencies having to do with allowing IP streams to share MPEG transport channels. This can’t be done on the integrated CMTS platform but could be done with modular CMTS architecture, which leverages edge QAMs to handle modulation of the DOCSIS streams.

“Our point is that when you use intelligence at the edge QAM, you gain two things,” he says. One is the ability to apply video processing techniques that are used on legacy streams to the IP streams. The other entails leveraging existing QAM channels for video rather than allocating more QAMs for CMTS-delivered IP video. This avoids having to make sacrifices, such as cutting back analog channels or squeezing more digital channels into QAMs, to free up bandwidth for the CMTS, he says.

Advocates of the IP video-over-CMTS approach argue there are greater bandwidth savings to be found through delivering all IP-based services – voice, high-speed data and video – over bonded channels using the power of statistical multiplexing to get the most out of the allocated bandwidth. And they say this approach allows operators to make the transition to all-IP operations as the end game with a single point of management over all streams.

BigBand, in the CableLabs demo, showed how its approach could net at least 30 percent savings in bandwidth. Eshet says other techniques not part of the demo but part of BigBand’s evolving product strategy could add another ten to 20 percent to the savings.

The demo focused on how network intelligence can be applied to gain efficiency on unicast streams that employ AS. One benefit of the BigBand approach is to prevent IP device clients from demanding more bandwidth than they really need for a given session, which is often the case when the client-based native HTTP (Hypertext Transfer Protocol) AS process is in play.

This happens because the “chunks” or stream segments transmitted to a given client every few seconds depend on how much bandwidth the client tells the server it has available rather than on what the client actually needs. Thus, an AS rate plan that has a large-screen HD resolution such as 780p as its maximum level will provide bandwidth suited to that resolution to a device that has access to that much bandwidth even if the device is a handheld where 780p is overkill.

“Clients are greedy,” Eshet says. “They’ll try to get the highest bit rate possible with no awareness of what else is happening on the network.”

The upshot, as demonstrated at the CableLabs’ conference, is a lot of usable bandwidth gets wasted. In a setup where a household is entitled to 4 megabits-per-second service and three devices – a PC, iPad and iPod – are accessing that bandwidth simultaneously, “we showed you get only 70 percent utilization of the available bandwidth.”

Beyond wasting bandwidth on devices that don’t need what’s being sent, there are other aspects to AS the lead to this level of inefficiency. For example, AS is designed to work in pre-set bit-rate gradations, so that if an optimal bit rate isn’t available, the next level down might be lower than what’s really available.

Compounding the problem, when several devices asking for more bandwidth than is available to each are shifted to the next level down, the bandwidth freed up in that process could actually have been sufficient to provide at least one of them the bit rate it originally requested. All this is happening in multi-second intervals, which vary from one AS mode to the next, depending on which AS a particular device is compatible with.

Another problem has to do with inefficiencies resulting from the fluctuations in actual bit rates required for any given frame sequence in the video stream. A talking-heads sequence in a basketball game requires far less pictorial information to be transferred from one frame to the next than what’s needed to convey the action on the court.

These problems can all be addressed by network intelligence that looks at the whole bandwidth slice and divides it up moment by moment in whatever way is best suited to optimizing quality of experience across all devices. “We put that intelligence to use in allocating the bandwidth using several parameters, starting with making sure the amount of bandwidth used by any device is no more than is required for that platform,” Eshet says. “If an iPod and an iPad are in operation, you can get the same quality of experience on the two devices with much less bandwidth going to the iPod than you’re allocating to the iPad.”

Another technique exploits the advantages of operating in variable bit rate versus constant bit rate. The bit rate is adjusted frame by frame to deliver only what’s needed to satisfy the quality requirements for displaying whatever level of action is occurring in a given frame sequence.

By applying statistical multiplexing techniques, the bandwidth manager can then take advantage of capacity that’s freed up by those sessions where the action is diminished to make more bandwidth available to those devices that need it. “Basically we’re taking the same concept used for rate shaping of legacy video and applying it to AS,” Eshet says.

The demo showed these various techniques applied to the three session streams result in 100 percent utilization of available bandwidth, or, in other words, a net 30 percent gain in bandwidth efficiency, he notes. All of this can be done using existing AS clients and servers without introducing a proprietary AS, he adds.

“Our aim is to leverage the existing IP video infrastructure,” he says. “Instead of sending downstream IP video traffic through the CMTS, we’re suggesting it be sent directly to our MSP (Media Service Platform).”

This is BigBand’s frequency- and spectrum-agile edge-QAM device that employs the company’s Converged Video Exchange (CVEx) control plane software to manage advanced media processing for legacy broadcast, SDV and VOD as well as IP video. The modularly expandable MSP can also be configured to support local ad insertion and addressable advertising.

“On top of that, operators can employ BigBand’s edge resource and session managers to expand on the capabilities,” Eshet adds. Essentially there are three ways to use various combinations of these resources to conserve bandwidth, he says.

One is to perform all the bandwidth savings processes on the incoming AS video in conjunction with DOCSIS encapsulation at the edge, as was done in the demo. Another approach, which also requires DOCSIS encapsulation, avoids use of AS on the streams coming into the edge QAM, which allows unicast and multicast IP streams to be multiplexed with MPEG-TS over the core fiber network, and then applies AS along with the enhanced AS management techniques to the IP streams at the QAM so that they can be distributed in unicast mode to cable modems..

The third idea involves combining all the streams through the QAM but eliminating AS and DOCSIS encapsulation at the edge in order to maximize bandwidth efficiency across all streams all the way to subscribers’ premises. There the cable modem would capture the IP streams and apply AS for distribution to all devices in the home.

This last method would employ CVEx to tell the cable modem clients where to find the IP streams, much as it communicates to SDV clients to allow set-top boxes to identify SDV streams. This would require new processing and software capabilities in the cable modem. “We’re working with chip vendors and certain MSOs to define the client on the cable modem,” Eshet says.

This third approach can be combined with either of the others to create a hybrid architecture where AS is maintained end to end while the multicast and MPEG-TS streams are multiplexed together. “There are multiple ways to blend these approaches using edge network intelligence,” Eshet says. “Each customer can pick their own way.”