Vous êtes sur la page 1sur 22

Femtocell Synchronisation and Location

A Femto Topic Brief

Published 08 May 2012

Small Cell Forum Ltd PO Box 23 GL11 5WA United Kingdom

Tel: +44 (0)845 644 5823 Fax: +44 (0)845 644 5824 Email: info@smallcellforum.org website: www.smallcellforum.org

Small Cell Forum Ltd (formerly Femto Forum Ltd). Registered in the UK no. 6295097

The Small Cell Forum, formerly known as the Femto Forum, supports the wide-scale adoption of small cells. Small cells are low-power wireless access points that operate in licensed spectrum, are operator-managed and feature edge-based intelligence. They provide improved cellular coverage, capacity and applications for homes and enterprises as well as metropolitan and rural public spaces. They include technologies variously described as femtocells, picocells, microcells and metrocells. The Small Cell Forum is a not-for-profit, international membership organisation, with membership open to providers of small cell technology and to operators with spectrum licences for providing mobile services. The Forum has 137 members including 63 operators representing more than 1.99 billion mobile subscribers 33 per cent of the global total as well as telecoms hardware and software vendors, content providers and innovative start-ups. The Forum has three main aims: To promote adoption of small cells by making available information to the industry and the general public; To promote the rapid creation of appropriate open standards and interoperability for small cells; To encourage the development of an active ecosystem of small cell providers to deliver ongoing innovation of commercially and technically efficient solutions.

The Forum is technology agnostic and independent. It is not a standards setting body, but works with standards organisations and regulators worldwide to provide an aggregated view of the small cell market.

If you would like more information about the Small Cell Forum or would like to be included on our mailing list, please contact: Email info@smallcellforum.org Post Small Cell Forum, PO Box 23, GL11 5WA UK Member Services Lynne Price-Walker lynne@smallcellforum.org For a full list of members and further information visit our website www.smallcellforum.org

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

Contents
1. 2. 3. 4. 5. 6. 7. 8.
Tables
Table 2-1 Table 3-1 Table 4-1 Table 5-2 Multipath Characterization ............................................................... 3 Scenario characterisation ................................................................ 9 Network Scenarios of AGPS-PTP Hybrid ........................................... 13 Availability of Time Synch from Macro Networks ............................... 16

Introduction .................................................................... 1 Assisted GNSS ................................................................. 1 Network-based Synchronization (NTP, PTP IEEE1588, NTR) ................................................................................ 8 Hybrid AGPS-Packet Timing ........................................... 12 Cellular Network Listen.................................................. 15 Hybrid AGPS-Femto Sniff ............................................... 17 Implementation Considerations ..................................... 17 Conclusions and Future Work ........................................ 17

Figures
Figure 2-1 Figure 2-2 Figure 2-3 Figure 2-4 Figure 2-5 Figure 2-6 Figure 2-7 Figure 2-8 Figure 2-9 Figure 3-1 Figure 3-2 Figure 3-3 Figure 3-4 Figure 4-1 Figure 4-2 Figure 4-3 Pseudorange Calculation ................................................................. 2 Figure Title Signal Propagation Paths to Indoor Antennas ..................... 3 Statistical Signal Level Models.......................................................... 4 Fix Probability in a Rural Standalone House on Flat Ground .................. 5 Fix Probability in a Rural Brick Apartment on Flat Ground .................... 5 Fix Probability in a Rural Steel & Concrete Office on Flat Ground ........... 6 Fix Probability in an Urban Steel and Concrete Office with Tinted Windows ....................................................................................... 6 Fix Probability in a Dense Urban Concrete Apartment .......................... 7 Fix Probability in a Dense Urban Steel and Concrete Office with Higher Attenuation Tinted Windows ............................................................ 7 Plot of individual time offset measurements over 83 minutes via a typical home DSL broadband connection ........................................... 9 Typical NTP Frequency Convergence over ADSL ................................ 10 Typical NTP Frequency Convergence over Cable ............................... 11 Typical NTP Frequency Convergence over a Private LAN Connection .... 11 AGPS-PTP network and clusters ..................................................... 12 Servers for 100K femtocells in AGNSS scenarios............................... 14 Required number of Servers Compared to PTP standalone ................. 15

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

1. Introduction
Synchronization and location determination are critical functions for successful operation of femtocells. While seemingly distinct topics, they are often discussed together since the techniques used to accomplish one may often also be used to accomplish the other. The requirements of the standards for syntonization accuracy vary from 250 ppb for a WCDMA femtocell to 50 ppb for CDMA2000, TD-SCDMA and LTE. The requirements for time synchronization accuracy vary from no requirement for FDD WCDMA and FDD LTE to 1.5 s for TD-LTE and TD-SCDMA. Location requirements vary depending on the jurisdiction, the applications envisaged for the femtocell and the operators commercial strategy. The requirement for location to support the E-911 regulations in USA is 50m with 50% probability and 150m with 95% probability. Beyond these hard requirements there are operator-specific specifications and desires. Location may be required in order to confirm that the femtocell is being used within the operators licensed region. It may be used to ensure that commercial restrictions on the use of the femtocell are complied with. Location may be needed to allow time synchronization to be achieved with the required accuracy via macrocell sniffing. Location may also be needed to support locations specific or location based functions. If the femtocell is sold via a retail outlet independently of the operator then location may be completely unknown by the network a-priori. In this case, it may be desirable for the femtocell to determine its own location. Where the operator supplies the femtocell, the nominal location may be determined from the subscribers address. However, it may be desirable or necessary for the femtocell to independently confirm the location. The femtocell synchronization and location application is distinctly different from that of larger wireless base stations in a number of important respects. Firstly, the femtocell is self-contained and operating indoors with no possibility of using antennas external to the building. Indeed, it is distinctly undesirable to use antennas external to the femtocell enclosure. Secondly, the back-haul is generally a tunnel through the open internet, often connected via DSL. Thirdly, the economics are distinctly different as the entire femtocell cost needs to be extremely low with figures as low as US$50 being projected for mass market deployment. Offset against these differences, a single femtocell serves a very small number of customers and hence the probability of failure can be allowed to be higher than in the case of larger Node B devices. These differences all have serious implications for the design of synchronization and location solutions for femtocells. In fact, they render the synchronization problem for femtocells very distinct from that of any other telecoms synchronization application. For these reasons, the topic deserves a detailed study within Small Cell Forum. Recognition of that led to the publication of a topic brief within WG1 and subsequently to the formation of a group within WG2 to study the topic in greater depth and to produce this white paper. This document is a summary of an intense study reported in a companion White Paper using available data. For each relevant technology, several scenarios were defined and characterized. The performance of that technology within each scenario was then analysed based on that characterization. Of course, there are varying performance capabilities represented within a given technology and the study has attempted to take that into account. Indeed it attempts to provide sufficient data to allow readers of this document to define benchmarks to be met by technology or component suppliers. Beyond performance, the study also attempts to define and quantify other selection parameters that may be used to select technologies. The following sections of this document consider each of the technologies concerned in turn. Firstly a description is provided followed by a summary of a discussion and analysis of each technology. Various environmental scenarios relevant to the femtocell application are introduced and the performance achievable in each scenario is summarised and discussed along with a selection of other factors to facilitate comparison. The final section provides notes on implementation considerations.

2. Assisted GNSS
2.1 Description

The study focuses on GPS, the most mature and widely used of several Global Navigation Satellite Systems (GNSS), which also include GLONASS and Galileo.

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

GPS receivers routinely produce location and precise time but can also produce syntonised frequency. GPS operates through multilateration using signals concurrently received from multiple satellites. The receiver extracts the times of transmission of the satellite signals from the data in the navigation messages modulated onto the satellite signals and measures the times of receipt and the frequencies based on its own clock. It determines the locations of the satellites at the times of transmission from orbital data provided in the navigation messages and computes the errors in the assumed location, time and frequency. By correcting these errors it establishes and maintains precise absolute time, frequency and location.

Figure 2-1

Pseudorange Calculation

Assisted GNSS operates by supplying the key navigation message data to receivers via a separate route. That allows the receivers to operate with signals much weaker than the data extraction thresholds of the receivers. It also avoids the delays associated with extracting the navigation messages. High sensitivity is achieved by using very long correlation times. This is conventionally done using a combination of coherent and non-coherent processing but the sensitivity achievable in this way is limited. The sensitivity can be further extended by using data wiping which requires a-priori knowledge of the data bits and precise synchronization with them. However, that approach is clearly not applicable when we are trying to use the GPS receiver to obtain precise synchronization. Other techniques used in AGPS positioning and navigation involve unresolved millisecond ambiguities that are not acceptable in a timing receiver. Others require one satellite signal above the data extraction threshold and knowledge of location to within 75km to permit unambiguous synchronization to all of the weak satellite signals used. There are non-linear signal processing techniques that are employed to achieve high sensitivity without any of the above restrictions. Hence, some AGPS timing solutions provide very high acquisition sensitivity (eg 160 dBm) but require at least one much stronger signal (eg -145 dBm) and location to within 75 km while others provide slightly poorer acquisition sensitivity (eg -157 dBm) without those limitations. The phenomena limiting the effectiveness of AGNSS solutions indoors include the following: 1. 2. 3. 4. 5. Attenuation of the signals by roofs (10dB-20dB), floors (10dB), walls (5dB 10dB) and even by tinted windows (10dB), Multipath fading effects which limit signal availability, Pseudorange errors resulting from path length increases caused by reflections both inside the building and, more importantly, from nearby structures as illustrated in Figure 2-2, Cross-correlation effects which limit the dynamic range of signals with which the receiver can work concurrently, Presence of CW interference sources which limits the availability of signals even in receivers with sophisticated CWI mitigation features.

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

Direct through roof, walls & windows


Attenuated No added path length Dominant for suburban residences Attenuation is relatively slight in standalone houses but high in apartments

Reflected internally
Attenuated Small added path length Dominant for large apartment blocks

Reflected externally
Attenuated Large added path length Occasional but more common in inner city apartment blocks

Figure 2-2

Figure Title Signal Propagation Paths to Indoor Antennas

Scenario

Maximum Fade Variation (dB) 2 3 3 6 10

Typical Maximum Path Extension (m) 2 10 20 100 100

Standalone House Brick Apartment Office Building Dense Urban Apartment Dense Urban Office
Table 2-1

Typical Maximum Timing Error (ns) 15 40 60 300 300

Multipath Characterization

2.2

Scenario Characterization

The scenarios we have chosen for full characterization are: Rural standalone house in flat terrain, Rural brick apartment building in flat terrain, Rural steel and concrete office building with tinted windows in flat terrain, Urban steel and concrete apartment building with tinted windows Dense urban concrete apartment, Dense urban steel and concrete office building with higher attenuation tinted windows.

Data on L-band satellite to indoors attenuation is very sparse but some studies of attenuation or of indoor GPS signal levels have been reported in the literature [1] [2] [3]. Using the data from these sources we have constructed probabilistic models for indoor GPS signal level for various scenarios and then interpolated and extrapolated based on the field experience of the authors. While the source data and experience relied upon is believed to provide a good guide it is recommended that these models be rigorously field tested and refined as necessary. Small Cell Forum WG2 has indicated an interest in undertaking such testing using the direct resources of the members. The methodology employed in constructing the statistical signal level models is detailed in the main body of this report. The result is the model illustrated in Figure 2-3. Note that each curve represents the probability of the signal from any given satellite above the horizon at that location having a level above the horizontal ordinate. These curves can be used to estimate the availability of location fixes and timing fixes based on the sensitivity characteristics of a given GPS receiver.

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

Table 2-1 gives the typical maximum fade variation, path extension and timing error to be expected as a result of multipath in each scenario.

100.00% 90.00% 80.00% 70.00% 60.00% 50.00% 40.00% 30.00% 20.00% 10.00% -170 -165 -160 -155 -150 -145 -140 0.00% -135 -130

Rural Flat Standalone House Rural Flat office Building Dense Urban Concrete Apartment
Figure 2-3 Statistical Signal Level Models

Rural Flat Brick Apartment Typical Urban Office Building Severe Dense Urban Office Building

2.3

Comparison Measures

Figure 2-4, Figure 2-5, Figure 2-6, Figure 2-7, Figure 2-8 and Figure 2-9 illustrate the Fix Probability in each of the six chosen scenarios as a function of the Initial Acquisition Sensitivity (IAS) and Ultimate Acquisition Sensitivity (UAS) of the GPS receiver. IAS is defined as the sensitivity achievable with no a-priori synchronization to the satellite signals and UAS is that achievable with a-priori synchronization to the satellite signals. It would be prudent to derate the receiver performance by 2 to 3 dB before consulting these charts. On that basis, AGPS fix availability varies from below 20% to about 87% in a dense urban office building with high attenuation tinted windows depending on the AGPS solution employed. It is much higher in the other scenarios. AGPS fix acquisition time indoors varies from a few tens of seconds to twenty or thirty minutes deep indoors where time-varying fading is common. In some locations the receiver may have to wait for the satellites to move into a more favourable geometry before a fix will be possible. Table 1 is a good guide to the accuracy achievable from a single fix. However, as a guide, in steel and concrete multi-storey buildings the horizontal accuracy of an individual fix will typically be limited to the footprint of the building. Vertical accuracy will be worse. However, positioning performance can be improved by averaging and most timing receivers offer the feature of self-surveying. Using 24-hour self-surveying, accuracy of better than 20m is achievable with about 95% certainty. AGPS timing accuracy indoors typically ranges from 30ns to 300ns depending on the multipath scenario. In the worst conditions, timing errors of up to 1 s are possible but extremely rare. Frequency accuracy from AGPS receivers either directly or by oscillator disciplining using the 1PPS timing pulse is typically better than 5ppb with occasional excursions to around 10ppb.

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

100.00%

100.00% 100.00% 100.00% 99.99% 99.90% 99.79% 99.70% 99.58% 99.50% 99.30% -142 99.38% -140

-156

-154

-152

-150

-148

-146

-144

Initial Acquisition Sensitivity (dBm) UAS = -156dBm UAS = -159dBm


Figure 2-4

UAS = -157dBm UAS = -160dBm

UAS = -158dBm

Fix Probability in a Rural Standalone House on Flat Ground

100.00%

100.00% 99.97% 99.99% 99.96%

100.00% 99.50% 99.00% 98.68% 98.50% 98.00% -14298.04% -140

99.32%

-156

-154

-152

-150

-148

-146

-144

Initial Acquisition Sensitivity (dBm) UAS = -156dBm UAS = -159dBm


Figure 2-5

UAS = -157dBm UAS = -160dBm

UAS = -158dBm

Fix Probability in a Rural Brick Apartment on Flat Ground

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

100.00% 95.00% 90.00% 85.00% 80.00% 75.00% -156 -154 -152 -150 -148 -146 -144 70.00% -142 -140

Initial Acquisition Sensitivity (dBm) UAS = -156dBm UAS = -159dBm


Figure 2-6

UAS = -157dBm UAS = -160dBm

UAS = -158dBm

Fix Probability in a Rural Steel & Concrete Office on Flat Ground

100.00% 95.00% 90.00% 85.00% 80.00% 75.00% 70.00% 65.00% -156 -154 -152 -150 -148 -146 -144 60.00% -142 -140

UAS = -156dBm UAS = -159dBm


Figure 2-7

UAS = -157dBm UAS = -160dBm

UAS = -158dBm

Fix Probability in an Urban Steel and Concrete Office with Tinted Windows

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

100.00%

100.00% 99.92% 99.96% 99.88%

100.00% 99.00% 98.00% 97.38% 97.00%

98.63%

-156

-154

-152

-150

-148

-146

-144

Initial Acquisition Sensitivity (dBm) UAS = -156dBm UAS = -159dBm


Figure 2-8

96.00% -142 -140 96.13%

UAS = -157dBm UAS = -160dBm

UAS = -158dBm

Fix Probability in a Dense Urban Concrete Apartment

100.00% 80.00% 60.00% 40.00% 20.00% 0.00% -142 -140

-156

-154

-152

-150

-148

-146

-144

Initial Acquisition Sensitivity (dBm) UAS = -156dBm UAS = -157dBm UAS = -158dBm UAS = -159dBm UAS = -160dBm

Figure 2-9 Fix Probability in a Dense Urban Steel and Concrete Office with Higher Attenuation Tinted Windows A package of assistance data for 12 satellites but not including the optional almanac data can be packaged into a binary payload of around 1052 bytes. The maximum bandwidth requirement is under 0.15 bytes per second. A pull assistance scheme in which the AGPS receiver requests only the assistance it needs when it needs it, minimizes the assistance volume that needs to be supported by the servers. To spread the load at ephemeris changeover times or, for example, after power blackouts, over a reasonable time frame it is desirable for the requests to be randomised in time. This can be done by the AGPS receiver or by the femtocell main processor. AGPS can be used to disciplines the master oscillator directly with that same master oscillator clocking the AGPS receiver itself. In addition, AGPS can be used to generate adaptive temperature compensation tables for use in significantly extending the holdover of relatively low cost oscillators with inherent stability of only around 1PPM. Holdover as good as 50ppb over 24 hours and 3 s over 30 minutes has been demonstrated using such oscillators. However, this does require careful selection of oscillators taking account of the rate of change of frequency with temperature and hysteresis effects. In addition, a start-up frequency requirement of around 250ppb or better demands pre-loading of temperature tables in production. A typical GPS receiver solution would involve between 10 and 30 passive components, possibly a Low Noise Amplifier chip, a SAW filter and an oscillator (which may be the master oscillator of the femtocell) in addition to the GPS chip or, in the case of a software GPS solution, a GPS front-end chip. Module solutions
Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

have the advantage that they come fully tested from the factory and are very simple to integrate both in design and in production. Total cost for the Bill Of Materials in 100KU volumes could vary from $15 to $20 for a module solution down to $6 to $10 for a chip level solution down to $? to $? for a software solution. However, other costs and offsetting cost savings should also be taken into account such as royalty charges and the absorption of other functions including the master oscillator and disciplining functions (where applicable) and synchronization switching functions (where applicable).

3. Network-based Synchronization (NTP, PTP IEEE1588, NTR)


3.1 Description

PTP and NTP both rely on a short exchange of time-stamped packets to estimate the time at the client. Frequency is derived from a rate-of-change-of-time observation at the client using time measurements spread over some duration. Packet timing is widely and efficaciously used for both syntonization and synchronization in LAN networks. Extension of this approach to the internet requires the following primary problems to be addressed: Variability in time transfer latency (jitter) due to network latency created by hubs, switches, cables, and other hardware that reside between the clocks, latency associated with the processing of timing packets, time uncertainty introduced by asymmetry, and the cost of server and network load.

PTP is a natural choice for delivering synchronisation in an Enterprise or other managed network environment to a number of types of devices. A simple server-client model works well in single-building or small-site settings. PTP-enabled routers (boundary and transparent clocks) extend the reach and performance of PTP in larger deployments or across core networks. These deployments rely on investment in PTP-enabled routers throughout the path between grand master and client. Whilst this may appear to be a significant overhead when associated with a single service such as femtocells, PTP may also be available as a common service in future networks. Note that PTP synchronisation packets will not naturally traverse a NAT gateway (such as a home router) without specific configuration. Although conventionally NTP is used over the internet to synchronise clocks in IT equipment within a second or so, it can also be used to good effect to provide frequency control for femtocells given dedicated servers and specifically optimised client algorithms (client algorithms have no effect on compatibility with standard servers). (Note that public NTP servers support neither the capacity nor sufficient confidence in accuracy required for femtocells.) NTP clients suffer no difficulties with NAT traversal. However, some corporate networks restrict access to a specific list of trusted external NTP servers. Error! Reference source not found. illustrates typical Packet Delay Variation (PDV) for NTP measurements over 83 minutes. Proprietary low packet rate (5-20 polls per minute) solutions can maintain absolute frequency accuracy of the order of tens of ppb in a home environment. Access and backhaul networks introduce packet delay distortions that go beyond variable queuing delays, including but not limited to path diversity and route reconfiguration. Despite this, good solutions for NTP based syntonization do exist.

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

Figure 3-1 Plot of individual time offset measurements over 83 minutes via a typical home DSL broadband connection All that can be said of absolute time at the client is that it lies between the limits of round-trip delay measured using the packet protocol. Other than for exceptional links, however, correct absolute time is usually within a few milliseconds of the centre of this range. In the long term it seems likely that PTP boundary clocks will exist widely at local exchanges (and cable head-ends). This will allow a very much higher packet rate to be supported for each client with very much less apparent jitter. This will certainly reduce the cost of the timing solution at the client. The effect on infrastructure cost will depend on how it is shared and any additional costs of access. Where it is available, use of the Network Timing Reference is very efficient (potentially free and very fast) but relies on the femtocell including the broadband modem and the local-exchange operators cooperation. Coarse location may be derived from the IP address associated with the DSL or cable connection. This assumes that the femtocell or the management system has access to a database linking IP address to physical location.

3.2

Scenario Characterization

The scenarios chosen for full characterization are: Broadband Technology: xDSL and Cable Another 2 scenarios are recommended for future study: Fibre and GPON

The following Table 3-1 summarises the characterization of cable and DSL scenarios in comparison to a LAN connection which may be relevant to an enterprise deployment.

Cable Network connection Packet Latency min Packet Latency mean Packet Latency mean (of best 90%) Asymmetry (to timeserver- from timeserver) RMS Packet Delay Variation RMS Packet Delay Variation (of best 90%) Packet Loss
Table 3-1

ADSL Network Connection 16.235ms 17.076ms 16.662ms 1.374ms

LAN Connection 180.488s 204.875s 200.924s N/A

10.675ms 11.646ms 10.991ms 4.870ms

3.780ms 0.728ms 4.9%

3.973ms 0.462ms 0.05%

29.972s 24.025 s 0.4%

Scenario characterisation

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

3.3

Comparison Measures

Short term non-availability of fixes can be mitigated by using high stability oscillators and concurrent use of multiple time servers. A NTP timeserver can support up to 100,000 clients at 20% utilisation with 16 requests/min per client. PTP timeservers are limited to between 500 and 1000 clients. Figure 3-2, Figure 3-3 and Figure 3-4 illustrate typical NTP convergence at varying packet rates. At 16 pkts/min over ADSL it takes up to 4000 seconds to converge to within 100ppb but may take only 1000s at 48 pkts/min over cable and is virtually instantaneous over a LAN. Frequency accuracy depends on PDV, oscillator stability and the averaging period and the figures also show the performance achieved with 2 hours of averaging. As mentioned in the introduction, timing accuracy is of the order of milliseconds.

Figure 3-2

Typical NTP Frequency Convergence over ADSL

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

10

Figure 3-3

Typical NTP Frequency Convergence over Cable

Figure 3-4

Typical NTP Frequency Convergence over a Private LAN Connection

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

11

The data traffic generated for the examples above are 65 MByte/month (each way) for 16 pkts/min, 195 MByte for 48 pkts/min and 975 Mbytes/month for 240 pkts/min. The oscillator needs to be able to maintain the stability at the converged level for the averaging period. Timestamping of network packets is required for both NTP and PTP using a counter clocked by the frequency source to be conditioned. PTP requires timestamping of the incoming and outgoing packets whereas NTP requires timestamping of the incoming packets only. The timestamps must be made available with the packets. There is an issue with PTP in that the standards are such as to preclude access to a PTP server by multiple clients over a NATed network which effectively rules out PTP for use in residential networks. The amortised cost of timeservers for each femto is of the order of US$0.10 for NTP and US$10.00 for PTP.

4. Hybrid AGPS-Packet Timing


4.1 Description

As discussed earlier, AGPS cannot support synchronization with 100% availability indoors. On the other hand PTP and NTP may suffer high PDV and even packet extinction. Also large numbers of timing servers would be needed to support correspondingly massive numbers of femtocells. In addition, PDV can be minimized by minimizing the number of hops in the path from grandmaster to client. The AGPS-PTP concept takes account all of the above considerations. As a preference AGPS-PTP recovers timing from AGPS but when the AGPS is in holdover it switches to PTP. AGPS-PTP can use a grandmaster or it can use a neighbouring femtocell with AGPS synch as a master. This gives rise to the concept of femtocell clusters which are independent AGPS-PTP timing networks composed of at least one PTP master synchronized to GPS and multiple PTP slaves. These are defined by neighbour lists that are made available to each femtocell from an AGPS-PTP server. All femtocells in the same cluster exchange information allowing each femtocell to choose potential masters with good AGPS timing quality and in order to set up connections between masters and slaves. Each hybrid client can have multiple masters and multiple slaves at the same time.

Figure 4-1

AGPS-PTP network and clusters

Hence a cluster is able to maintain synchronization independently of the PTP grandmaster as illustrated in Figure 4-1. The AGPS-PTP hybrid server performs hybrid timing network management functions such as clustering, provision of AGPS assistance data and PTP Grandmaster functions. Clusters can be configured autonomously or with the help of the AGPS-PTP server to minimize the network distance between masters and slaves. In the AGPS-PTP concept PTP reduces the dependence on the GPS RF environment while AGPS reduces the dependence on the packet network environment by distributing the functions of PTP master. Secondly, clustering greatly reduces the number of servers or PTP Grandmasters required thereby reducing opex and
Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

12

capex relative to PTP alone. Thirdly, AGPS-PTP minimizes the number of packets required for synchronization and distributes the PTP traffic load through the network.

4.2

Scenario Characterization

The AGPS-PTP scenarios are an intersection of the AGPS and Packet Timing scenarios with private and public networks offering a major segmentation. The clusters may need to be larger if the AGPS availability is very low. In a private network PDV will generally be small as the number of hops between femtocells in the cluster will generally be 2 or less. Also IP Multicast may facilitate more distributed and self-organized clustering. In a public network the network distances may be much larger and more varied and hence dynamic PTP message rate may be needed for synchronization. It may also be that the involvement of AGPS-PTP servers will be necessary for clustering although a mixture of centralized and distributed clustering is envisaged. Table 4-1 highlights the major differences between the public and private network environments.

Private network Test environment Network type Network configuration PTP message type PTP message rate Delay asymmetry Clustering method
Table 4-1

Public network Residential (internet) xDSL, Cable (DOCSIS), xPON, Ethernet Variable configuration Unicast Dynamic Inherent asymmetry of xDSL, Cable (DOCSIS) and xPON Distributed / Centralized

Office (intranet) Ethernet Fixed configuration Mixed unicast / multicast, Or Unicast Constant Asymmetry from the traffic condition Distributed (decentralized) Server not necessary

Network Scenarios of AGPS-PTP Hybrid

4.3

Comparison Measures

A key measure of the success of the AGPS-PTP concept is the extent to which it reduces the required number of servers (PTP Grandmasters). Figure 4-2 illustrates the number of servers required per 100,000 femtocells as a function of cluster size for each of the AGPS scenarios analysed in this paper. It shows that a single server can support 100,000 femtocells and more for all but the worst class of building. In that case, the number of servers needed varies from 70 to 150 as the cluster size varies from 8 down to 2.

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

13

Servers per 100K Femtocells in AGNSS Scenarios


180 160 140 120 100 80 60 40 20 0 2 3 4 5 6 7 8 Average Cluster Size Rural standalone house on flat ground Rural brick apartment on flat ground Rural Steel & Concrete Office on Flat Ground Urban Steel and Concrete Office with Tinted Windows Dense Urban Concrete Apartment

Figure 4-2

Servers for 100K femtocells in AGNSS scenarios

Figure 4-3 compares the number of servers required for AGPS-PTP as compared to that for PTP in order to service 10% of the residences in Seoul across all housing types. For PTP 459 Grandmasters are required while for AGPS-PTP the figure varies from 52 down to 1 as the cluster size varies from 2 to 7.

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

14

(for 229K femtocells at a housing type proportion)


500 Required number of Servers 450 400 350 300 250 200 150 100 50 0 2
Figure 4-3

Servers in a Case Scenario

459

PTP standalone AGPS-PTP

52

20 3

8 4

3 5

2 6

1 7

1 8

Average cluster size

Required number of Servers Compared to PTP standalone

Results of actual testing of AGPS-PTP in a private network within an office and laboratory environment are also reported. A fairly typical private network was deployed with 8 units under test, two of which had good GPS signal strength, three of which had moderate signal strength and three of which had bad signal strength. Periods of GPS holdover were deliberately simulated and the synchronization performance of one unit without GPS was monitored. The average frequency accuracy achieved was 0.49 ppb with a standard deviation of 1.83 ppb. Changing the network distance via switching had little impact on the accuracy and only a very low PTP packet rate was required. Similar testing over a public network was conducted over 4 sites across two neighbouring cities. The mean frequency accuracy this time was -0.2ppb with standard deviation of 3.40 ppb. The problem of using PTP via Network Address Translation in a firewall is explained along with various potential solutions. There is also increased complexity in the integration of AGPS and PTP into the client. However, the infrastructure complexity is greatly reduced compared to PTP as shown earlier by the massive reduction in the number of servers required. The Total Timing Cost of AGPS-PTP is much less than that of PTP because of the dramatic reduction in the number of servers required.

5. Cellular Network Listen


5.1 Description

Femtocells can source synchronization from any macro-cell signals it is able to access. Frequency will always be better than 50 ppb in accuracy. Phase will also be to the required accuracy at any CDMA or TDD BTS but will be delayed by the time of flight to the monitoring femtocell. However, use of this mode of synchronization does require that macro-cell signals are available, which is not always the case, and that the type of cell (ie macro-cell as opposed to femtocell) can be established reliably. To overcome the time of

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

15

flight problem the locations of BTS and femtocell need to be established since this will introduce bias at the rate of 1 s per 300m.

5.2

Scenario Characterization

The scenarios that can be envisaged are essentially the type of macrocell which influences the availability of time (phase) synchronization. Table 5-1 details the conditions and required aids for time synchronization from each type of macro-cell.

Macro network WCDMA FDD LTE FDD WCDMA TDD SC-TDMA, LTE TDD GSM

Conditions Depends on optional synchronisation of the macro network (e.g. for MBMS) None Depends on optional synchronisation of the macro network (e.g. for TDOA)

Required Aids Location of femto and basestations Location of femto and basestations Location of femto and basestations (possible by TDOA if signals from 3 base-stations are available) Location of femto and basestations (possible by TDOA if signals from 3 base-stations are available)

CDMA2000

None

Table 5-1

Availability of Time Synch from Macro Networks

5.3

Comparison Measures

For frequency control a single weak macro-cell signal is sufficient. For timing a single macro-cell signal is required provided location is known. However, for location at least 3 synchronized base stations are required along with their locations. Cellular Network Listen (CNL) can provide an important benefit in typical deployments but cannot be relied upon for universal coverage especially when time synchronization is required. CNL provides synchronization almost instantly. Location accuracy of 100-300m is possible which is poor compared to GPS but frequency and timing accuracy is quite sufficient for femtocell applications provided location is known to 300m. The complexity of CNL is primarily in the RF chain. In the simplest implementation only transmissions from the same technology and band are monitored and only during discontinuous periods while the femtocell is not carrying any traffic. Even then this may require additional circuitry to allow reception on the femtocells transmit frequency. In the most complex case one or more dedicated receivers are provided allowing reception of a range of bands and technologies. The discontinuous approach offers a significant simplification. Time synchronization requires frequent sniffs at intervals of no more than a few minutes and short enough that the interruption can be tolerated by most codecs. Some management overhead is also required to identify and verify that the signal being used for frequency control is from a sufficiently reliable source. There may also be some overhead associated with establishing the location of the base station if time synchronization is required. The costs associated with the femtocell itself include those of additional antennas, RF amplifiers and filters of the order of a few tens of cents. Future generations of transceiver IC are expected to support discontinuous CNL without overhead. Some cost may be associated with obtaining the BTS locations from other operators networks.

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

16

6. Hybrid AGPS-Femto Sniff


6.1 Description

This hybrid concept again involves using AGPS-synchronized femtocells as synchronization masters. However, in this case, neighbouring slave femtocells synchronize to the masters by using femtocell-sniffing. This could be advantageous where femtocells are deployed in close proximity so that they can form clusters. The advantages include the fact that phase biases caused by time of flight from master to slave will be small owing to their close proximity, in deployments where the macro-cells and femtocells do not share the same frequency there is no need to provide an additional RF section in the femtocell and very high synchronization availability should be achievable where femtocells are clustered. There is a complication in that it is necessary for the slaves to identify masters with good AGPS synchronization and how this can be done is not clear. The main disadvantage is the cost of incorporating both AGPS and femto sniff.

6.2

Scenario Characterization

The scenarios of interest are steel and concrete apartment or office buildings where AGPS availability may be lower. In the worst such scenarios the AGPS availability could be between 20% and 87% depending on the AGPS technology employed.

6.3

Comparison Measures

If the AGPS technology had IAS of -144dBm then the cluster size would need to be above 31 to give availability of 99.9%. However, if the AGPS technology had IAS of -155dBm then a cluster size of only 4 would give availability of 99.97%. Fix time would be very fast when neighbouring femtocells already had AGPS fixes. Location uncertainty would contribute less than 1 s to overall timing error of less than 2 s. Location error would be between 50m and 300m. The back-haul demands would be negligible as for AGPS and the oscillator demands could also be very low as for AGPS. The complexity would be as for AGPS with some additional complexity to allow reception on the femtocell transmit frequency. The cost would be slightly more than for AGPS alone.

7. Implementation Considerations
AGPS assistance can be supplied to those femtocells that need it from conventional AGPS assistance servers such as those already deployed in many wireless networks or from AGPS assistance service providers. Alternatively it could be supplied from GPS receivers at the gateways provided those gateways were within a few hundred km of the femtocells themselves. One of the factors affecting the success of AGPS integration is GPS interference from internal sources within the femtocell. Clock frequencies with harmonics close to GPS should be avoided or rise times should be limited and track lengths kept low. Patch antennas should be provided with ground planes as large as practicable. Synchronization switching is a complication requiring careful consideration when multiple synch sources are used. A common oscillator should be disciplined to maintain both frequency and time. Switching should occur on the control side of the oscillator. When switching between sources, large shifts in timing should be undertaken slowly and smoothly.

8. Conclusions and Future Work


It is not feasible to draw conclusions as to the optimal choices of synchronization or location technology for various scenarios as this would require a multi-dimensional comparison matrix. In addition, the degree of subtlety to be considered in making the choice does not lend itself to that sort of guidance. Instead we
Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

17

recommend a strand of further work to perform a number of detailed case studies aimed at providing greater guidance for specific applications of general interest. Another proposed strand of further work is to perform a systematic program of field testing aimed at confirming and refining the scenario characterization presented within this document. This could be performed by the broader membership of the Small Cell Forum using standardized equipment and methodology to be documented and supplied by the synchronization and location team.

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

18

References
1 2 3 HSGPS Signal Analysis and Performance under Various Indoor Conditions, G Lachapelle et al, ION GPS/GNSS 2003, 9-12 Sep, 2003, Portland, OR A Detailed Experimental Study of the LEO Satellite to Indoor Channel Characteristics, Zoltan Bodnar et al, International Journal of Wireless Information Networks, Vol 6, No 2, 1999 The Nordnav Indoor GNSS Reference Receiver, Alexander Mitelman et al, ION GNSS 2006, 26-29 Sep, 2006, Fort Worth, TX

Report title: Femtocell Synchronisation and Location Issue date: 08 May 2012 Version: Published

19

Vous aimerez peut-être aussi