Delivering Synchronization across the Packet Network

 

Phil Tolcher - Semtech

 

With the powerful driver from the Mobile Wireless Operator to reduce OPEX and increase ARPU, the decisions on CAPEX investment on equipment infrastructure are hugely important and the challenge to reduce development, installation and maintenance costs has never been so great.

 

When considering a typical Third Generation (3G) mobile network, a significant portion of the operating cost, in some installations up to 50%, can be attributed to the radio access network and the recurring costs of the backhaul between the Base Station and the BSC/MSC, leased on a monthly basis from the local carrier as well as remote monitoring and management of equipment.

 

For the next generation mobile network, the “Holy Grail” is quickly evolving towards a Packet-Everywhere architecture. Just like the SDH/SONET networks, the need to synchronize the network is mandated and needs careful consideration. In wireless networks, the synchronization needs are driven more by the air interface than by the services being carried, and so tight synchronization is definitely still needed. Indeed, wireless network have some of the strictest requirements for synchronization, and they are not at all easily met.

 

Synchronization within a base station UTRAN mandates a common timing reference amongst adjacent Node B’s and is defined to ensure that Node B’s are traceable to the RNC and to minimise cross interference between adjacent Node B’s. The synchronization method depends on whether a Frequency Division Duplex (FDD) or Time Division Duplex (TDD) mode is being deployed within the UTRAN. 

 

The delivery of synchronization today typically uses either a leased line (E1 or DS1), a re-timed clock from a SDH/SONET multiplexer or GPS. Synchronising a base station to a common clock is usually attempted, in GSM and WCDMA systems at least, using the frequency of the landline signal. Although a landline clock is usually traceable to an atomic clock, and therefore has excellent long-term accuracy, the amount of noise allowed on a land line makes most of them unsuitable for this purpose. Either special synchronization links have to be assigned, at greater expense, or noise filters have to be used but these can introduce their own problems. CDMA2000 systems have mandated that the air interface carries GPS time. Because the cable delay of a landline is not known, the CDMA2000 base stations cannot be given GPS from a common point in the network but must acquire it individually, and so base stations are usually equipped with a GPS receiver, at great installation and maintenance expense.  

 

A more ideal synchronisation scheme would control the air interface by providing accurate frequency and phase whilst also providing an accurate copy of GPS time, all without requiring specially-conditioned lines or individual GPS receivers.

 

As the links from the BSC/MSC to the base stations start to migrate to packets, it becomes possible to adopt a fundamentally different synchronisation scheme which closely approaches the ideal scheme outlined above. It provides equivalent performance to the best of today’s methods but avoids many of their pitfalls whilst being robust and extremely cost-effective. In 2002, the IEEE ratified IEEE 1588, which is a new standard for distributing a high precision time base around a network. In 2003, this was adopted by the IEC and given the number IEC 61588. Although it was originally targeted towards relatively compact LANs, it was seen as being potentially useful in wider networks, particularly in the connections to wireless networks. In 2004, the IEEE opened a PAR to enhance IEEE 1588 to suit this larger audience. The benefit of this new standard is that it will allow a time base to be generated in remote equipment without relying on quiet transmission links. It can operate with reasonable network delays and packet jitter without deviating away from the synchronization requirements. The common clock can either be owned by the land network operator or the wireless network operator. Because it builds a time base, which can be GPS-locked, it provides frequency, phase and time accuracy, and it suits both FDD and TDD systems. To get it to work, especially in the presence of typical levels of packet delay and jitter, requires some careful design of the slave clock. Semtech have employed their extensive experience of conventional designs for SDH and PDH equipment to this problem and have prototyped an IEEE 1588 master slave pair  with the right levels of performance. Semtech are also key contributors to the IEEE 1588 PAR, with the aim of making sure the new version of the standard has the right enhancements to suit this, and similar, applications.

 

The basic operation of IEEE 1588 is similar to that of NTP (IETF RFC 2030) in that master and slave clocks exchange timestamp values at a relatively slow rate, and the slave uses these timestamp values to align its timebase to that of the master. The difference between IEEE 1588 and NTP is that IEEE 1588 uses hardware-assisted timestamping which avoids the considerably-variable latency suffered by software methods and this leads to much-improved accuracy.

 

In the Semtech patent pending technology, strong filtering of perceived packet offsets finds the average path delay and maintains the stability of the slave’s time base even when packet jitter is high. Strong filtering, though, particularly with the low rate of transmission of IEEE 1588 timing packets, has to work over long time constants and the local oscillator in the slave clock must have suitable stability to support this. Since there is no getting away from the holdover requirement, a decent OCXO has still to be provided in each base station, and this can be at the heart of the IEEE 1588 slave clock. Because the OCXO will have good short- and medium-term stability, the slow rate of IEEE 1588 packets does not limit the performance which can be obtained for this application. The basic performance seen on the prototypes is more than adequate for wireless base station requirements.

 

The original version of IEEE 1588 did not have the right features to provide sufficient resiliency for this application and this is being addressed in the PAR. Things like redundant network clocks and paths will be provided, so that the wireless base station can continue uninterrupted by a network failure.

 

This new synchronisation scheme offers a way to simplify and cost-reduce the synchronisation of a wireless network. Being able to work independently of the land network operator’s clock removes the need for expensive conditioned land lines, whilst the ability to transport GPS time from an RNC to its base stations means dedicated GPS receivers are no longer needed at each base station, thereby saving considerable installation and maintenance costs. GPS reception has always been subject to good antenna positioning and is vulnerable to heavy weather. Worse, availability of GPS signals has always been at the behest of the US Navy, which owns and operates the GPS system and they have the ability to make all CDMA2000 systems, and other GPS-based systems, inoperable. This potential weakness has been feared for some time in these applications. Being able to take GPS (directly, or via UTC) from a protected central point means the CDMA2000 networks can continue even during GPS outages.

 

Semtech is not a provider of backhaul services and quoting a “guarantee” of savings would be unwise. Semtech has however discussed the potential savings with a large and well respected service provider, and under conditions of anonymity, we can offer some “numbers” using the Pound Sterling.

 

Traditional access pricing model;

 

Assumed 100Mbps of an STM1 is used at RNC and 3xE1 at each Node B  (giving 17 Node Bs in RNS).

Assumed distance between RNC and provider network POP to be 5 miles, and average distance between provider network POP and Node Bs to be 5 miles.

 

Then, yearly rental costs:

 

RNC-end:        £100,295 + £18,000                              = £118,295

Node B ends:  100M/E1 x (£4098 + £1000 + £1192)         = £314,500

Total annual cost:                                                     = £432,795

 

Simple Ethernet backhaul pricing model;

 

Assumed 100Mbps Ethernet at RNC:                                       = £14,000

Assumed 6Mbps Ethernet at each of 17 Node Bs:  17 x £2600     = £44,200

Total annual costs:                                                             = £58,200

                                                         

Potential ANNUAL SAVING :  £432,795 - £58,200 = £374,595.

 

While this is an example and cannot be promised, certainly not by Semtech, it provides a glimpse into the new paradigm of cost structure for mobile wireless network backhaul services.

 

 

To learn more about a new timing delivery paradigm for next generation IP/packet networks, please contact Phil Tolcher (ptolcher@semtech.com).