To the cellular operator backhaul is a perennial problem, a process made up of a delicate balance between Opex (rent existing E1 capacity) and Capex (build out an equivalent microwave network) and every wave of technology adopted by a cellco ‘ 3G, HSDPA, mobile multicasting ‘ threatens to continually double and quadruple the requirements for base station connectivity.
There are reports that backhaul takes up 30% of Opex and just as much of Capex for many cellular operators.
At some point in the far off future, when base stations are connected to handsets using OFDM, and speak to one another and the core network in pure IP, the vagaries of managing different flavours of circuit switched traffic will all be in the past. But we’re not there yet.
All of which attracted us to the invitation by Israeli start up Celtro to discuss why mobile TV, and especially the MBMS variant of it, represented yet another bottleneck in the design of cellular networks.
We have quoted in these pages from time to time, the concern that Unicast cellular TV services can eat all the capacity at a base station with between 4 and 8 separate users watching different programming. The provision of quality mobile TV will require QVGA screen resolution and even when compressed in H.264 that requires around 250 to 400 Kbps for each channel carried.
An MBMS TV service offering 15 TV channels, a not unlikely number, would require around 5 Mbps of backhaul capacity. So we are back to the Capex Opex argument ‘ is that another 3 E1 lines per base station costing perhaps €600 per month, or is it the sighting of a new microwave transmitter or is this the rental of a broadcast satellite delivery, along with the capex of a satellite receiver for each base station?
Even if we think that the right way to deliver ALL forms of mobile TV is through a new broadcast network and never through the cellular backhaul, the Celtro approach is still pretty interesting for cellular operators.
Faultline spoke to CEO Ron Zor and CTO Shahar Gorodeisky this week and got down and dirty about cellular backhaul.
‘How cellular operators carry out backhaul has not changed in the last 10 or 15 years, it is the constant addition of E1 lines,’ explained Zor.
‘These days about 60% of this backhaul is wireless, because it costs less to set up, and about 40% is in fixed lines. Today it is roughly 1 base station for 1,000 cellular subscribers across a cellular network, so that 10 million customers means about 10,000 bases stations. Each of these today have about 5 E1 lines to support them, and these in turn are aggregated in regions with about 40 or 50 E1s being merged onto one line and then this in turn transported to a hub.
‘On a 10,000 cell network there are around 30 switches and none of this traffic at present is IP, but multiple switched lines transported on TDM if it’s a GSM base station and on ATM if it’s UMTS 3G,’ Zor said as he warmed to the problem.
‘In about 3 years time operators will start to think about using the IP protocol to backhaul these sites, but at present it is complicated. The network itself has to be synchronous to support cell handovers and to do that it needs to accurately distribute a synchronization signal. Because of this the traffic tends to be kept separate throughout its journey through the network from traffic in other networks.’
He paints a picture whereby cellcos really need between 3 and 6 E1 lines for each of 2G, 3G and HSPA for every base station. ‘The recommended backhaul for UMTS was 6 E1 lines, but what most networks do is add two and then fight congestion issues by slowly adding more.’
What Celtro does is offer network appliances (its DynaMate range of Compact Cellular Backhaul Switches) which translate the 2G and 3G traffic (and the HSDPA and MBMS traffic if need be) into internet packets, complete with their timing data, where this is required, and then mingles all of it and multiplexes it across existing E1 lines as if they were a single pipe, to another similar appliance which puts all the signals back together again right next to the regional concentrator or hub.
In order to do this the appliances have to carry out deep packet inspection, at real time speeds ‘ anything under 20 milliseconds of delay (Celtro reckons this can be brought down to 5 milliseconds for a 3G network) ‘ and decide which data has a high priority and attach Quality of Service tags to the data it recognizes, as well preserving data that is needed to keep the service synchronous and then recreating the 2G and 3G streams at the other end.
The larger of these appliances can take as many 50 E1 lines and multiplex all their traffic, at speeds up to 2 Gbps and the important thing about this is that the different types of traffic can be multiplexed together, with statistical multiplexer gains being shared among the entire bandwidth.
Celtro gives a lot of customer endorsed statements showing how it upgraded clients to 3G or HSDPA and actually reduced the total number of E1 lines, by aggregating all the traffic on to the existing lines.
‘At Vodacom in South Africa, we installed 1,000 of our switches when they upgraded to an HSDPA network. Their problem was of course the price of the extra E1 lines, but also the fact that capacity was just not available. We put our switches into every cellsite and every hub site and they needed no additional capacity,’ he said.
One of the Celtro tricks is to split time dependent data from the non-time dependent data, flattening the best effort data out over time ‘ delaying it or offloading it to things like DSL backhaul, preserving the E1 lines for the urgent applications.
It manages all this in a box which has standard networked processors and a handful of DSPs, with most of the tricks in the software.
Like all these things this is a constantly moving target, if a new data type arrives the system treats it as high priority traffic until it has a handle, which takes a few milliseconds, on what type of data it is.
‘When you are carrying video, if you just throw some of it away or send some by a slower route, that can damage the video for long periods. Deep packet inspection allows us to isolate I- frames in a video stream and preserve them and then we can adopt a policy of throwing away (or delaying) some of the B frames and P frames, for instance every other one. That way the video still gets through, but it is only loses a 30th of a second of quality, not the I-Frame which is referenced by all subsequent frames,’ said Zor.
‘In many cases, by beginning to use our system, cellcos can then erode E1 prices and re-open negotiation for more capacity at a later point and a lower price,’ he added.
Celtro says that the network operator can also handle encrypted data streams in one of two ways, it can either make decryption keys available to the Celtro switches and they can decrypt and then inspect the packets, or the Celtro systems can manage without. ‘We can use pattern recognition and behavioural data to work out what kind of applications are running across cellular backhaul, so we don’t have to decrypt it.’
Celtro was founded in 2003 from a 3 year old research team inside ECI Telecom, and has been delivering product for three years. It finds itself today up against similar products from Cisco, Nokia Siemens, Tellabs, Alcatel Lucent and Huawei. That’s a tough group to try to keep up with, but it has customer benchmark data that claims it can squeeze as much, or more data, into backhaul lines than any of them so far. Naturally it is only a matter of time before a 100 person $30 million company becomes attractive enough for an acquisition, or it gets overtaken by the increasing efforts of these giants.
Right now Celtro is licking its lips at the prospect of all the surges in backhaul requirements due to the various version of HSDPA and the arrival this year and next of MBMS, and it hopes to be able to change the Capex Opex cellular calculation from one that weighs satellite delivery of video against more E1 lines, into one that makes better use of existing backhaul capacity for the far smaller Capex of installing its switches, which go for anywhere between $1,000 and $10,000 each, depending on capacity, something like a few $million for each network.