Your browser is not supported. Please update it.

19 July 2005

Minerva – the goddess of wisdom? or a wise old IPTVer

Companies like Minerva Networks, a specialist IPTV middleware company with 45 US IPTV installations over the past 5 years, are threatened by Microsoft’s move into the business. We asked CEO Mauro Bonomi about the state of IPTV and in particular the numbing effect of having Microsoft re-invent the business and immediately grab all the big Telcos as potential customers.

Bonomi was fresh this week from signing a deal with Eagle Broadband and France Telecom satellite subsidiary GlobeCast to bring up to 200 channels of IPTV right out of the sky via satellite, eliminating the need for a head end (see separate story).

Will IPTV always stretch to 200 TV channels, we asked? Surely this high number was only required to drive up the number of advertising slots, and given that most viewers watch perhaps 15 or so channels and store them on their ‘favorites’ lists, along with the rise of DVRs and VoD, which usually don’t carry advertising, this would surely fall.

‘There will always be a need for live TV feeds, although the number of events that really need to be shown live is limited. But perhaps we (IPTV suppliers) need to advertise that we can do the same as cable and more. So there is a transition period when IPTV will need to be able to handle 200 channels. Perhaps it will come down after that. On Demand choice is going up exponentially and this is pulled from a server not pushed over a multicast line, so the number of real time channels will one day decrease substantially,’ agreed Bonomi.

Bonomi believes that IPTV will become the dominant form of TV globally, due to its flexibility, depth of choice and the use of converged services. He gives the famous example of incoming IP phone callers being shown on the TV screen, and use of the TV remote to push the call to either an answer phone or to pause the TV to answer the call.

But he knows that his business currently only survives in opportunities where Microsoft either fails, or which are beneath its notice. Bonomi bemoaned the non-standard approach that latecomer Microsoft was bringing to IPTV. ‘Microsoft’s IPTV system does not support IGMP (the Internet Group Management Protocol) and it doesn’t use multicast to support channel change, so Microsoft has to use a proprietary multicast protocol. It handles all channel changes using servers.

‘There are two approaches possible in IPTV. You can offer the core functions of IPTV first, or you can try to WOW everyone. Microsoft chose to WOW everyone. It uses a mosaic EPG so you can see thumbnail videos of multiple channels. The mosaic can be tailored to individuals and it can show whatever you choose, perhaps the programs that right now are hot, viewed by the highest number of viewers currently watching them.

‘This mosaic is not difficult to do, but is requires more bandwidth, more CPU power and multiple decodes. Effectively a more expensive set top.

‘One of the other key features has been fast channel change.’ Of course in a world where the number of channels reduces, rather that rises, as more On Demand content becomes available, this should become less relevant.

‘Our channel change speed is the same as cable today, roughly a single second. We can reduce that further, but we will not do this Microsoft’s way. Microsoft jumps out of a simulcast (multicast) and instead of waiting perhaps half a second for the new channel stream to present the set top with an I-Frame, a Microsoft server Unicasts the right I-Frame and some older I-Frames, directly to the set top, which then switches to a new simulcast. This is what is supposed to lead to the instantaneous channel change. But it is expensive in terms of bandwidth and an operator needs extra servers to support that feature, and it is unlikely to scale.

‘There are other ways of going about that problem and our challenge in the next 12 months is to add the same capabilities that Microsoft has ‘ faster channel changes, VoIP, a network DVR, more creative advertising opportunities ‘ to our Core IPTV, which already works, but to do it in a less hardware and network intensive way.’

The biggest gamble, according to Bonomi, is in the way Microsoft deals with speeding up channel change. The latency in channel change can be anything between 1 second and 2 seconds in IPTV or cable systems, and this is made up of a number of components.

The middleware responds to channel change requests from the remote control of the user instantaneously, so there is usually no delay in the channel change due to middleware.

In traditional IPTV, as in cable, the three most important factors in channel change time are the tuning time, the delay waiting for a reference I-frame, and the time waiting for the video buffer to fill.

The tuning time is all about leaving one multicast stream and joining another. A set top cannot accept content from a new multicast stream until it has told the upstream components to stop sending the last stream and waited long enough for the network to respond to this. Otherwise the set top risks combining two streams and getting more data than its last mile bandwidth can handle.

Added to this is the time it takes to acquire the video, audio and encryption stream identifiers. Then there are other dependencies. A video stream cannot even begin to start trying to render a picture until it has one full I-frame or reference frame, and preferably more.

The I-Frame is the reference frame and is mostly a bit map with some compression, but not much, carried out by grouping identical pixels.
I-Frames are usually either 10 or 15 frames apart.

The more frames you use between I frames, the more efficiency you have in terms of bandwidth. However you sacrifice some fidelity and it takes longer to change channels, waiting for an I-frame to come around.

This wait for an I-Frame takes from a quarter to a half second. Finally it takes much of the rest of the time to fill a set top’s buffer.

If you ask any of the traditional suppliers like Minerva how they might go about improving all of that, they probably won’t tell you, wanting to keep their secrets from the competition, but they might suggest that there are a number of sensible options.

The middleware could carry video identifiers in its channel map, the space between I-Frames can be less and an I-Frame can be displayed statically until video is ready to render and while the buffer fills.

All of these approaches minimize the impact on the network, and they are a combination of small improvements from the middleware, encoders, and set-top box.

Microsoft’s approach has been to put the weight of expectation back on the network. When a remote control is pressed, a short Unicast of the new channel selected is started at a remote server, who’s job is just to phase in new channels. If the server responds instantaneously, it first sends the most recent I-frame. But actually it is difficult to resolve a fully accurate screen from just one I-frame, so the last handful are sent in a burst, without the frames in between. Some of Microsoft’s competitors say that this burst will be about 35 Mbps for around half a second, if the network will sustain that.

If the network won’t sustain that, the data gets there when it gets there but it slows down the channel change, which defeats the entire object of the exercise. Also the operators can choose to work from less I-frames or just one.

This now gives the set top enough information to begin showing all the subsequent frames, and it can phase in the new multicast stream in the full knowledge that it has the data it needs to show the next image.

However this only works IF the server responds instantaneously. Now we are back to the latencies that are both in the network to and from the server and inside the server itself. Assuming it has to order this I-Frame data from disk, there are virtually no savings. But if it is in memory already, it is likely that this will result in faster channel switching times.

Minerva, and the entire existing world of IPTV thinks of this as sacrilege, and points out that there is the potential for disaster, when a channel switching server is oversubscribed and needs to carry out multiple disk reads in succession, building a queue of channel switch requests.