Read Report / Proposal template text version

Unclassified Copy No. ........

Man Portable UWB Monitor and Locator Feasibility Study

Produced for: Radiocommunications Agency

Report No: 72/03/R/208/U October 2003 ­ Issue 2 Against ITT: AY4519

Unclassified

Unclassified

Man Portable UWB Monitor and Locator Feasibility Study

Equipment Concept Picture

Report No: 72/03/R/208/U October 2003 ­ Issue 2 Produced for: Radiocommunications Agency Against ITT: AY4519

Author(s): Peter Hulbert Gavin Owen

Approved By Malcolm Streeton .................................................... Principal Group Manager

Authorised By David Smith ....................................................... Wireless Business Unit Director

Roke Manor Research Limited Roke Manor, Romsey, Hampshire, SO51 0ZN, UK Tel: +44 (0)1794 833000 Fax: +44 (0)1794 833433 Web: http://www.roke.co.uk Email: [email protected] Approved to BS EN ISO 9001 (incl. TickIT), Reg. No Q05609

This is an unpublished work the copyright in which vests in Roke Manor Research Limited. All rights reserved. The information contained herein is the property of Roke Manor Research Limited and is supplied without liability for errors or omissions. No part may be reproduced, disclosed or used except as authorised by contract or other written permission. The copyright and the foregoing restriction on reproduction, disclosure and use extend to all media in which the information may be embodied.

72/03/R/208/U

Page 1 of 69

Unclassified

Unclassified

DISTRIBUTION LIST

Copy No. Radiocommunications Agency Radiocommunications Agency Radiocommunications Agency Radiocommunications Agency Roke Manor Research Ltd Roke Manor Research Ltd 1 2 3 4 5 Master

Full Report Mike Lipscomb Bernard Bond Ron Stanley Barry Goodyear Malcolm Streeton Project File

DOCUMENT HISTORY

Issue no Issue 1 Issue 2

Date

th 25 July 2003

Comment First Issue of document (Interim Report 1) Final Report

21 October 2003

st

72/03/R/208/U

Page 2 of 69

Unclassified

Unclassified

Man Portable UWB Monitor and Locator Feasibility Study

Report No: 72/03/R/208/U October 2003 ­ Issue 2 Produced for: Radiocommunications Agency Against ITT: AY4519

SUMMARY

Roke Manor Research Ltd. (RMR) has performed a study for the Radiocommunications Agency (RA) to investigate the feasibility of man portable equipment for locating and monitoring UWB devices. The study focuses on the UWB standard being produced by the IEEE under the IEEE 802.15.3a task group, although a less in-depth investigation into the more general case of UWB signals is also included. This study has been performed in parallel with the IEEE standardisation. Several companies have made proposals to the IEEE for the fundamental method for transmission. These are being reduced both by the IEEE voting process and by the companies' collaborating and merging proposals. During the study the number of proposals has reduced, but no final decision was made. An approach of looking at the broad range of proposals whilst limiting in depth investigation to those most likely to be adopted has therefore been used. At the time of completion of this study two proposals remain; a MB-OFDM based proposal led by Texas Instruments and a DS-CDMA based proposal led by XtremeSpectrum. The down selection procedure is still continuing and the next round on voting is scheduled for November 2003. For the in-depth technical analysis the study, has mainly investigated the Texas Instruments approach as this has the strongest backing from industry. The study has shown that it is feasible to develop man portable equipment to detect 802.15.3a based transmission. A number of location techniques have been shown to be feasible, but the recommendation is to use direction finding with a two-horn antenna, as this provides the best performance complexity tradeoff. The equipment is expected to consist of hand held dual horn antenna with each horn having cross section of about 5" square, and a processing unit. A major size-determining factor is the power consumption of around 40 W and consequently the batteries needed. For the more general case of UWB signal detection the techniques that could be employed and what they can provide has been investigated. However the requirements and techniques need further investigation before detailed implementation complexity can be analysed. For 802.15.3a it is now necessary to wait until the key parameters of the standard are decided. The implementation approach can then be updated (if necessary). The next stage of equipment development can then begin, with algorithm design and simulation, and detailed circuit design.

Roke Manor Research Limited Roke Manor, Romsey, Hampshire, SO51 0ZN, UK Tel: +44 (0)1794 833000 Fax: +44 (0)1794 833433 Web: http://www.roke.co.uk Email: [email protected] Approved to BS EN ISO 9001 (incl. TickIT), Reg. No Q05609

This is an unpublished work the copyright in which vests in Roke Manor Research Limited. All rights reserved. The information contained herein is the property of Roke Manor Research Limited and is supplied without liability for errors or omissions. No part may be reproduced, disclosed or used except as authorised by contract or other written permission. The copyright and the foregoing restriction on reproduction, disclosure and use extend to all media in which the information may be embodied.

72/03/R/208/U

Page 3 of 69

Unclassified

Unclassified

CONTENTS

1 2 3 4 INTRODUCTION ...................................................................................................................................9 REFERENCES ....................................................................................................................................11 GLOSSARY.........................................................................................................................................11 REGULATIONS AND STANDARDS...................................................................................................12 4.1 4.2 4.3 4.4 5 FEDERAL COMMUNICATIONS COMMISSION .....................................................................................12 EUROPEAN TELECOMMUNICATIONS STANDARDS INSTITUTE (ETSI) ..................................................12 INSTITUTE OF ELECTRICAL AND ELECTRONIC ENGINEERS (IEEE).....................................................13 POWER SPECTRAL DENSITY .........................................................................................................14

IEEE 802.15.3......................................................................................................................................15 5.1 5.2 5.3 5.4 5.5 5.6 COMMUNICATIONS RESEARCH LABORATORY..................................................................................16 PARTHUS CEVA INC .....................................................................................................................17 STMICROELECTRONICS................................................................................................................18 TEXAS INSTRUMENTS ...................................................................................................................19 UNIVERSITY OF MINNESOTA ..........................................................................................................20 XTREMESPECTRUM ......................................................................................................................21

6

SIGNAL DETECTION..........................................................................................................................22 6.1 6.2 6.3 6.4 6.5 6.6 PROTOCOL ISSUES.......................................................................................................................22 RF ISSUES ..................................................................................................................................23 DIGITISATION ...............................................................................................................................23 SIGNAL PROCESSING ...................................................................................................................24 FORWARD ERROR CORRECTION (FEC) .........................................................................................24 GENERAL ALGORITHMS ................................................................................................................25

7

BASIC DIRECTION FINDING TECHNIQUES FOR UWB...................................................................27 7.1 7.2 7.3 7.4 7.5 7.6 SINGLE DIRECTIONAL ANTENNA ....................................................................................................27 PHASED ARRAY ANTENNA ............................................................................................................28 DIRECTION FINDING CONFIGURATION .............................................................................................34 PATH LOSS CALCULATIONS ..........................................................................................................35 COMPLEXITY OF THE METHODS .....................................................................................................38 INTERIM CONCLUSIONS.................................................................................................................41

8

IMPLEMENTATION COMPLEXITY ....................................................................................................42 8.1 TEXAS INSTRUMENTS ...................................................................................................................43 8.1.1 Receiver Performance Requirements.........................................................................43 Noise Figure ........................................................................................................43 Front-end AGC ....................................................................................................43 Dynamic Range & ADC performance ..................................................................43 Oscillator Performance (stability and control range) ............................................43 Implementation Loss Factors...............................................................................43 8.1.1.1 8.1.1.2 8.1.1.3 8.1.1.4 8.1.1.5

72/03/R/208/U

Page 5 of 69

Unclassified

Unclassified

8.1.1.6 8.1.2 8.1.2.1 8.1.2.2 8.1.2.3 8.1.2.4 8.1.3 8.1.3.1 8.1.3.2 8.1.3.3 8.1.3.4 8.1.4 8.2 9

Sensitivity ............................................................................................................44 Evaluation of Technology ....................................................................................45 Level Plan ............................................................................................................47 Power Budget ......................................................................................................48 Thermal Management..........................................................................................49 Sampling Techniques ..........................................................................................49 Candidate Algorithms for Operation in a Low SNR Environment.........................49 Search Efficiency and Time-to-Detect Performance ............................................49 Algorithm Complexity...........................................................................................49

Hardware Implementation............................................................................................44

Detection algorithms....................................................................................................49

Detector performance and operational factors..........................................................49

XTREMESPECTRUM, MOTOROLA AND PARTHUS CEVA INC. MERGED ................................................50

DETECTING OF NON STANDARD UWB SIGNALS..........................................................................52

10 EQUIPMENT CONCEPT .....................................................................................................................56 11 CONCLUSIONS ..................................................................................................................................57

APPENDIX A A.1 A.2 A.3 A.4 A.5

TEXAS INSTRUMENTS DETECTOR DETAIL ..............................................................59

QUANTISATION ............................................................................................................................59 SYNCHRONISATION ......................................................................................................................59 OFDM DEMODULATOR ................................................................................................................61 FREQUENCY HOPPING IMPLICATIONS .............................................................................................67 OVERALL IMPLEMENTATION LOSS .................................................................................................67

72/03/R/208/U

Page 6 of 69

Unclassified

Unclassified

FIGURES

FIGURE 1: IEEE 802.15.X STRUCTURE...................................................................................................13 FIGURE 2: IEEE 802.15.3 STANDARD PLAN ..........................................................................................14 FIGURE 3: POWER DENSITY SPECTRUM FOR UWB ............................................................................14 FIGURE 4: VIVALDI ANTENNA.................................................................................................................29 FIGURE 5: END ON VIEW OF CROSS POLAR ARRAYS ........................................................................30 FIGURE 6: EFFECT OF WIDE BANDWIDTH ON AMBIGUITIES ­ PHASED ARRAY.............................31 FIGURE 7: RESPONSE OF 6 ELEMENT ARRAY TO 3.1-4.6 GHZ SIGNALS .........................................32 FIGURE 8: EFFECT OF WIDE BANDWIDTH ON AMBIGUITIES ­ DF STRUCTURE .............................34 FIGURE 9: NORMALISED RANGE V UPPER FREQUENCY LIMIT.........................................................36 FIGURE 10: BLOCK DIAGRAM FOR DIRECTIONAL ANTENNA CASE .................................................38 FIGURE 11: BLOCK DIAGRAM FOR DF ANTENNA CASE.....................................................................39 FIGURE 12: BLOCK DIAGRAM FOR 6-ELEMENT BEAM FORMING PHASED ARRAY........................40 FIGURE 13: BLOCK DIAGRAM FOR SWEEPING PHASED ARRAY ......................................................41 FIGURE 14: SIMPLIFIED TOP LEVEL BLOCK DIAGRAM.......................................................................45 FIGURE 15: SIMPLIFIED DIAGRAM OF THE BASEBAND FUNCTIONALITY ........................................46 FIGURE 16: LEVEL PLAN .........................................................................................................................47 FIGURE 17: LOW, HIGH & DUPLEX FREQUENCY BANDS FOR XTREMESPECTRUM .......................50 FIGURE 18: JOINT FREQUENCY BAND FOR XTREMESPECTRUM .....................................................51 FIGURE 19: TOP LEVEL BLOCK DIAGRAM FOR MERGED PROPOSAL #2 FORMAT RECEIVER .....52 FIGURE 20: AMBIGUOUS SIGNALS ........................................................................................................54 FIGURE 21: EQUIPMENT CONCEPT .......................................................................................................56 FIGURE 22: ARTIST IMPRESSION OF THE DETECTOR ........................................................................57 FIGURE 23: QUANTISER IMPLEMENTATION LOSS ..............................................................................59 FIGURE 24: HIERARCHICAL MATCHED FILTER....................................................................................60 FIGURE 25: OFDM RECEIVER .................................................................................................................63 FIGURE 26: PHASE SHIFTER LOOK UP TABLE ....................................................................................64 FIGURE 27: LOOP FILTER........................................................................................................................65

72/03/R/208/U

Page 7 of 69

Unclassified

Unclassified

TABLES

TABLE 1: COMMUNICATIONS RESEARCH LABORATORY ..................................................................16 TABLE 2: PARTHUS CEVA INC. ..............................................................................................................17 TABLE 3: STMICROELECTRONICS.........................................................................................................18 TABLE 4: TEXAS INSTRUMENTS ............................................................................................................19 TABLE 5: UNIVERSITY OF MINNESOTA .................................................................................................20 TABLE 6: XTREMESPECTRUM................................................................................................................21 TABLE 7: REQUIRED EB/NO....................................................................................................................37 TABLE 8: IMPLEMENTATION LOSS FACTORS......................................................................................44 TABLE 9: ADC DEVICES AVAILABLE .....................................................................................................46 TABLE 10: ESTIMATED POWER BUDGET..............................................................................................48

72/03/R/208/U

Page 8 of 69

Unclassified

Unclassified

1

INTRODUCTION

Roke Manor Research Ltd. (RMR) has performed a study for the Radiocommunications Agency (RA) under contract Reference AY4519 to investigate the feasibility of locating and monitoring UWB devices. The results of this study are recorded in this report. The RA has a requirement to be able to detect and investigate equipment that is transmitting outside of the regulations. This is normally performed by simple spectrum measurements. However, because UWB systems re-use the same spectrum as other systems, utilising a wide bandwidth/low spectral density signal, the same approach cannot be used. This study therefore investigates how to detect UWB signals and the feasibility of implementation in hand portable equipment that may be used, for example, inside buildings. Currently only the US FCC has licensed UWB operation, although a similar proposal is under discussion in Europe. Therefore for the purpose of this study the FCC definition of UWB is adopted (see section 4.1). There are still many different types of UWB radio systems covering communications, imaging and locations, and many different types of UWB modulation. UWB operates in an unlicensed band and only compliance with a simple transmission mask rule is required. However, the most widespread use of UWB is expected to be for equipment meeting the IEEE 802.15.3 standard for Wireless Personal Area Networks (WPAN) currently under development, and hence this is the main focus of this study. A brief investigation of more general UWB equipment is also made. This study has been performed in parallel with the IEEE standardisation. Around 30 proposals were made to IEEE on options for the fundamental method for transmission. These are being reduced both by the IEEE voting process and by the companies' collaborating and merging proposal. At the time of completion of this study two proposals remain; a MB-OFDM based proposal led by Texas Instruments and a DS-CDMA based proposal led by XtremeSpectrum. For a proposal to be accepted it must achieve 75% of the votes. At the last IEEE meeting in September 2003, the TI proposal failed in the final vote, achieving only 60% confidence. The down selection procedure is still continuing and the next round on voting is scheduled for November 2003. An approach of looking at the broad range of proposals whilst limiting in-depth investigation to those most likely to be adopted has therefore been used. For the in-depth technical analysis the study has mainly investigated the Texas Instruments approach as this has the strongest backing from industry. The study has shown that it is feasible to develop man portable equipment to detect 802.15.3a based transmission. A number of location techniques have been shown to be feasible, but the recommendation is to use direction finding with a two-horn antenna as this provides the best performance complexity trade-off. The equipment is expected to consist of hand held dual horn antenna with a cross section of about 5" square, and a processing unit. A major size-determining factor is the power consumption of around 40 W and consequently the batteries needed.

Roke Manor Research Limited Roke Manor, Romsey, Hampshire, SO51 0ZN, UK Tel: +44 (0)1794 833000 Fax: +44 (0)1794 833433 Web: http://www.roke.co.uk Email: [email protected] Approved to BS EN ISO 9001 (incl. TickIT), Reg. No Q05609

This is an unpublished work the copyright in which vests in Roke Manor Research Limited. All rights reserved. The information contained herein is the property of Roke Manor Research Limited and is supplied without liability for errors or omissions. No part may be reproduced, disclosed or used except as authorised by contract or other written permission. The copyright and the foregoing restriction on reproduction, disclosure and use extend to all media in which the information may be embodied.

72/03/R/208/U

Page 9 of 69

Unclassified

Unclassified

For the more general case of UWB signal detection, the techniques that could be employed and what they can provide has been investigated, however the requirements and techniques need further investigation before detailed implementation complexity can be analysed. For 802.15.3a it is necessary to now wait until the key parameters of the standard are decided, or a dominant proprietary system enters the market. Roke Manor Research can then update the implementation approach. In addition a detailed requirements specification should be produced, with inputs from staff that will use the equipment to ensure operational requirements are fully included. The next stage of equipment development can then begin, with algorithm design and simulation, and then detailed circuit design. Following reference and glossary sections the report is structured as follows:

· · · · · · · ·

Section 4 provides an overview of the regulations and standards applicable to UWB. Section 5 describes the transmission regulations and the options for the signal format currently proposed within the IEEE 802.15.3 Task Group. Section 6 presents an overall method of detecting/receiving these formats. Section 7 describes the methods developed in this study to locate equipment for each of the signal types, including the use of fixed or phased array directional antenna. Section 8 then addresses the method and feasibility of implementation. Section 9 provides a brief examination of the more general case of non-standard UWB equipment. Section 10 shows what the detector equipment is likely to look like when developed Finally Section 11 provides a conclusion, and recommendation for further research and development in this area.

Additional detail on implementation is included in an Appendix A.

72/03/R/208/U

Page 10 of 69

Unclassified

Unclassified

2

[Ref 1]

REFERENCES

IEEE 802.15 Working Group for WPAN http://grouper.ieee.org/groups/802/15/ IEEE 802.15 WPAN High Rate Alternative PHY Task Group 3a (TG3a) http://www.ieee802.org/15/pub/TG3a.html IEEE 802.15 Task Group 3a call for Proposals http://grouper.ieee.org/groups/802/15/pub/TG3a_CFP.html Multi-band OFDM Proposal for IEEE 802.15 TG3a Texas Instruments, IEEE 802.15-03/267r5, July 2003 BBC Resources, "The Effects of Frequency Errors in OFDM", Report No. BBC RD 1995/15 J G Proakis, "Digital Communications", 2nd Edition, McGraw Hill, 1989

[Ref 2]

[Ref 3]

[Ref 4]

[Ref 5] [Ref 6]

3

GLOSSARY

The following lists abbreviation definitions used within this report. ADC AGC BPSK CDMA COFDM CRL dB DF DFD DS DSP EIRP ETSI FCC FEC FFT FH FIR FPGA ICI ISI IEEE Analogue to Digital Converter Automatic Gain Control Bi-Phase Shift Keying Code Division Multiple Access Coded OFDM Communications Research Laboratory Decibel Direction Finding/Finder Device For Detection Direct Sequence Digital Signal Processor Effective Radiated Power Referenced to Isotropic European Telecommunications Standards Institute Federal Communications Commission Forward Error Correction Fast Fourier Transform Frequency Hopping Finite Impulse Response Field Programmable Gate Array Inter Channel Interference Inter Symbol Interference Institute of Electrical and Electronic Engineers LNA MAC M-BOK Mbps NF NCO OFDM PHY PPM PSK QPSK RA RF RMR RS SNR TI UWB VCO VCXO WPAN Low Noise Amplifier Multiply Accumulate Multiple Bi-Orthogonal Keying Mega Bits Per Second Noise Figure Numerically Controlled Oscillator Orthogonal Frequency Division Multiplex Physical Layer Pulse Position Modulation Phase Shift Keying Quadrature Phase Shift Keying Radiocommunications Agency Radio Frequency Roke Manor Research Limited Reed-Solomon Forward Error Correction Signal to Noise Ratio Texas Instruments Ultra Wideband Voltage Controlled Oscillator Voltage Controlled Crystal Oscillator Wireless Personal Area Network

72/03/R/208/U

Page 11 of 69

Unclassified

Unclassified

4

REGULATIONS AND STANDARDS

An Ultra Wideband (UWB) signal allows the transmission of data with low spectral density by spreading the signal over a very wide bandwidth. The original approach was based on the transmission of very short RF pulses. However, now many different methods have been proposed. Wideband signals can allow very high bit rates; provide very fine range resolution and precision distance and/or positioning measurement capabilities. The Federal Communications Commission (FCC) has made the UWB spectrum available in the USA and it is expected to be legal for use in other countries in due course.

4.1

FEDERAL COMMUNICATIONS COMMISSION

The FCC first defined the main UWB spectrum for communications to be 3.1 GHz to 10.6 GHz. There is an additional low band, and higher bands for non-communications applications. A minimum spectral width of 500 MHz, or 25% of the centre frequency is specified (hence 500 MHz applies for the 3.1 GHz to 10.6 GHz band). In February 2003, the FCC eased some UWB restrictions to support the use of UWB technology in the emergency services. This is stated within the FCC Memorandum Opinion & Order (MO&O) and Further Notice of Proposed Rule. In addition the FCC confirmed the regulations it passed in February 2002 and made some minor changes to rules relating to through-wall imaging and ground penetrating radar (GPR) systems. In the latest FCC rules define UWB as "any radio technology having a spectrum that occupies a bandwidth greater than 20% of the centre frequency or a bandwidth of at least 500 MHz". The main band remains, as 3.1 GHz to 10.6 GHz for communications devices and within this range the transmission power is limited to ­41.3 dBm/MHz. Two spectral masks with different out of band limits have been specified, one for indoor and one for outdoor use. In summary the main FCC regulations for UWB communications are currently: 1. 2. 3. An UWB signal has a fractional bandwidth of 20% or 500 MHz at the ­10dB point. Maximum power spectral density is -41.3 dBm/MHz Frequency range is 3.1 GHz to 10.6 GHz

4.2

EUROPEAN TELECOMMUNICATIONS STANDARDS INSTITUTE (ETSI)

ETSI has set up an ERM-TG31A Group with the task of investigating and developing generic and/or specific ETSI radio standards for Short Range Devices using UWB technology under the direction of TC-ERM. TG31A will also identify spectrum requirements to ensure spectrum efficiency and compatibility with other radio services

72/03/R/208/U

Page 12 of 69

Unclassified

Unclassified

The Generic Standards being produced and due to be published in November 2003 are:

· ·

EN 302 065 (2 parts), Communications & Measurements between 3.1 to 10.6GHz using UWB. A draft version is available. EN 302 066 (2 parts), Radio Location Services between 100MHz to 6GHz is being worked on

In addition CEPT SE24 (Spectrum Engineering) are performing a UWB sharing study.

4.3

INSTITUTE OF ELECTRICAL AND ELECTRONIC ENGINEERS (IEEE)

The IEEE has a number of Task Groups focussing on the IEEE 802.15.x standards as illustrated in Figure 1. TG3a is working to define a higher speed PHY enhancement amendment to IEEE 802.15.3 for applications, which involve imaging and multimedia. The higher data rate categories are: >110 Mbps at 10m, >200 Mbps at 4m and >480 Mbps. UWB is seen as the approach, but concerns over frequency allocation and interference have been raised. A number of companies have submitted proposals for this PHY and TG3a are in the process of down selecting them. A plan of the TG3a's activities is shown in Figure 2.

TG2

Task Group 2 Coexistence

TG1

Task Group 1 WPAN/ BluetoothTM

TG3

Task Group 3 WPAN High Data Rate

IEEE 802.15.x

TG4

Interest Group

4a Alt PHY

Task Group 4 WPAN Low Data Rate

TG3a

Task Group 3a WPAN Higher Data Rate Alt PHY

Figure 1: IEEE 802.15.x Structure

72/03/R/208/U

Page 13 of 69

Unclassified

Unclassified

Session #22

·Complete TRD ·Selection Criteria ·Download Procedure

Session #23

·Proposals Heard ·Criteria Amended

Session #24

·2nd Round of Proposals ·Technical Review

Session #25

·Final Round of Presentations ·Voting ·Start Draft

Session #26

·Draft Improved

Session #27

·Complete Draft ·Approval for Letter Ballot

Jan 03

Mar 03

May 03

Jul 03

Sept 03

Nov 03

Figure 2: IEEE 802.15.3 Standard Plan

4.4

POWER SPECTRAL DENSITY

The masks for the UWB spectrum are given in Figure 3 for the various standards bodies.

UWB Mask

-30 UWB EIRP Emission Levels (dBm) -35 -40 -45 -50 -55 -60 -65 -70 -75 -80 0.1 1 Frequency (GHz) 10 100

FCC Indoor FCC Part 15

ETSI Indoor Singapore UFZ Trial

Figure 3: Power Density Spectrum for UWB

72/03/R/208/U

Page 14 of 69

Unclassified

Unclassified

5

IEEE 802.15.3

The IEEE 802.15 High Rate Alternative PHY Task Group (TG3a) for Wireless Personal Area Networks (WPANs) is working to provide a higher speed PHY enhancement amendment to IEEE 802.15.3 for applications, which involve imaging and multimedia. A large number of proposals (about 30) were submitted to TG3a. During the process of collaboration and th down selecting, this was reduced to only seven by 17 July 2003 (i.e. during this phase of the study). It is these seven that were initially investigated in this study. Although by the end of this study most of these proposals were eliminated, the conclusions are included as these may still be employed as proprietary systems. Each is introduced and described with the main issues for receiving it. The contenders considered are:

· · · · · · ·

Communications Research Laboratory (Single & Multi-band) Oki Electric Industry Co Ltd ParthusCeva STMicroelectronics (Single band) Texas Instruments (Multi-band) University of Minnesota (Multi-carrier) XtremeSpectrum (Multi-band)

Each of these proposals, with the exception of the Oki proposal have been summarised in the following sub-sections to provide a comparison of the claimed performance. The Oki proposal has not been included because it was based on high directivity ceiling mounted antennas. The kind of detection equipment envisaged here would not be appropriate for detection for such a system and it is unlikely that it would be subject to the kind of illegal usage envisaged. The figures given in these sections have been taken from the submitted proposals and have not been validated by RMR.

72/03/R/208/U

Page 15 of 69

Unclassified

Unclassified

5.1

COMMUNICATIONS RESEARCH LABORATORY

This proposal is based on a highly flexible modulation format referred to as `soft spectrum'. This proposal would present some problems for the UWB monitor equipment design as its proponent's stated objectives are to give spectrum freedom, waveform freedom and system freedom. The proposal is essentially a collection of variables in terms of occupied bands, number of bands and pulse shapes. It is not clear from the proposal how terminals would `discover' one another and agree a set of parameters for operation. It is also uncertain whether the standard would be proposed to be interoperable between different users' equipment and/or between different generations of equipment. CRL have formed a consortium, which includes Advantest Corporation, Anritsu Corporation, CASIO Computer Co., Ltd., Fuji Electric Co., Ltd., Fujitsu Limited, Furukawa Electric Co., Ltd., Hitachi Cable, Ltd., Hitachi Communication Technologies, Ltd., Hitachi Kokusai Electric Inc., Matsushita Electric Works, Ltd., Matsushita Electric Industrial Co., Ltd. and Meiji University. Proposal Reference # Date of Proposal Modulation Method IEEE 802.15-03/097r3

th 9 May 2003

Soft-Spectrum / Free-verse pulse shaping / Geometrical pulse / QPSK & BPSK

FEC Coding Scheme (at 480Mbps data Rate) Parallel Detection Need Link Margin at 1m (at 480Mbps data Rate) Modes of Operation Maximum Simultaneous Piconets Bands

Viterbi coding: Length, K = 7 and Rate, R = 3/4

Unknown Not specified. However, 224.3Mbps and 4m gives 9.9dB 5 Unknown Required Single or Multi-band Table 1: Communications Research Laboratory Optional Future

72/03/R/208/U

Page 16 of 69

Unclassified

Unclassified

5.2

PARTHUS CEVA INC

The physical layer is based on a single band transmission. Bit rates are available from 60 Mbps in powers of two multiples up to 960 Mbps. The occupied bandwidth is 3.8 to 7.7 GHz for all bit rates so it is assumed although not stated that the modulation is based on narrow pulses, which occupy the whole band regardless of pulse rate. The minimum pulse rate is 480 Mpps (Mega pulses per second). The maximum pulse rate is specified as 7700 Mpps. It is not clear how this pulse rate can be reconciled with the specified occupied bandwidth but this may use something like offset QPSK. The modulation/spreading is bi-orthogonal based on a 32-chip length codeword. Because every codeword can be transmitted either inverted or non-inverted this provides 64 possible states allowing 6 bits to be encoded by every 32 chip symbol. Proposal Reference # Date of Proposal Modulation Method IEEE 802.15-03/123r3

th 5 May 2003

DSSS Coding Scheme ­ biorthogonal coding Ternary spread spectrum

FEC Coding Scheme (at 480Mbps data Rate) Parallel Detection Need Link Margin at 1m (at 480Mbps data Rate) Modes of Operation Maximum Simultaneous Piconets Bands

Rate, R = 4/6 16 state Convolutional coding No 18.8dB

1 Unknown Required 3.8GHz to 7.7GHz Table 2: Parthus Ceva Inc. Optional Future

72/03/R/208/U

Page 17 of 69

Unclassified

Unclassified

5.3

STMICROELECTRONICS

The modulation is based on Pulse Position Modulation (PPM) combined with polarity signalling. For this proposal as well as for the CRL proposal the pulse shape is somewhat arbitrary. As for that proposal this means that developing monitoring equipment could prove problematic. Proposal Reference # IEEE 802.15-03/139r3 IEEE 802.15-03/140r0 (3rd March 2003) Date of Proposal Modulation Method FEC Coding Scheme (at 480Mbps data Rate) Parallel Detection Need Link Margin at 1m (at 480Mbps data Rate) Modes of Operation Maximum Simultaneous Piconets Bands Single band, 3 to 7 GHz (BW limited due to present technology) Single band, 3 to 10 GHz Table 3: STMicroelectronics 1 Unknown Required Optional Future

th 9 May 2003

Pulse Position Modulation / Polarity Turbo Codes Parallel Concatenation of Convolutional Codes (PCCC) Rate, R = 7/8 No 9.1dB

72/03/R/208/U

Page 18 of 69

Unclassified

Unclassified

5.4

TEXAS INSTRUMENTS

A merged proposal has been formulated by Texas Instruments in partnership with Staccato Communications, Focus Enhancements, Intel, Femto Devices, Institute for Infocomm Research, General Atomic, Sony, Time Domain, Philips, Samsung Advanced Institute of Technology, Panasonic, Mitsubishi, Samsung Electronics and Wisair. This is based on frequency hopped COFDM with guard periods long enough to protect against most anticipated delays spreads. Proposal Reference # Date of Proposal Modulation Method FEC Coding Scheme (at 480Mbps data rate) Parallel Detection Need Link Margin at 1m (at 480Mbps data rate) Modes of Operation Maximum Simultaneous Piconets Bands No 18.2dB in Mode 1 17.5dB in Mode 2 2 3 13-bands with a bandwidth of 528MHz. Centre Frequencies for each band: Group A: Band #1 Centre Frequency at 3432MHz Band #2 Centre Frequency at 3960MHz Band #3 Centre Frequency at 4488MHz Group B: Band #4 Centre Frequency at 5016MHz Band #5 Centre Frequency at 5808MHz Group C: Band #6 Centre Frequency at 6336MHz Band #7 Centre Frequency at 6864MHz Band #8 Centre Frequency at 7392MHz Band #9 Centre Frequency at 7920MHz Group D: Band #10 Centre Frequency at 8448MHz Band #11 Centre Frequency at 8976MHz Band #12 Centre Frequency at 9504MHz Band #13 Centre Frequency at 10032MHz Table 4: Texas Instruments Required Optional Future IEEE 802.15-03/267r1

th 14 July 2003

OFDM / QPSK / Frequency Hopped Viterbi Coding: Length, K = 7 and Coding Rate, R = 3/4

For devices with improved performance

72/03/R/208/U

Page 19 of 69

Unclassified

Unclassified

5.5

UNIVERSITY OF MINNESOTA

The university is proposing a Multi-carrier Fast Frequency Hopping UWB-OFDM system. Unlike the TI proposal, the hopping is not between OFDM symbols but within them. Proposal Reference # Date of Proposal Modulation Method FEC Coding Scheme (at 480Mbps data Rate) Parallel Detection Need Link Margin at 1m (at 480Mbps data Rate) Modes of Operation Maximum Simultaneous Piconets Bands Unknown Unknown Multi-carrier Bandwidth of >500MHz Two options that avoid UNII band: Low Band: 3.1GHz to 4.75GHz Whole Band: 3.15GHz to 9.9GHz and miss out 4.8GHz to 5.9GHz Table 5: University of Minnesota Required Optional Future No Unknown IEEE 802.15-03/147r3

th 10 May 2003

Fast Frequency Hopping / OFDM / QPSK Convolutional Channel Coding rate, R = 3/4

72/03/R/208/U

Page 20 of 69

Unclassified

Unclassified

5.6

XTREMESPECTRUM

XtremeSpectrum will be cooperating with Motorola. Currently they have an UWB chipset named Trinity, which is a 3-band solution, split into 2 bands named Low and High and uses a Multiband CDMA PSK approach. Proposal Reference # Date of Proposal Modulation Method FEC Coding Scheme (at 480Mbps data Rate) Parallel Detection Need Link Margin at 1m (at 480Mbps data Rate) Modes of Operation Maximum Simultaneous Piconets Bands Unknown Not stated. However, 600Mbps at 4m is 1.7dB (high band) 3 Spectral Modes (Low, High and Multi ­ band) Unknown Multi-band Low Band: 3.1GHz to 5.15GHz High Band: 5.825GHz to 10.6GHz Multi-band: 3.1 to 5.15GHz and 5.825 to 10.6GHz Required Optional Future IEEE 802.15-03/153r5 May 2003 DS-CDMA / PSK / M-BOK Reed-Solomon, RS (225,223) with byte Convolutional interleaver

Table 6: XtremeSpectrum

72/03/R/208/U

Page 21 of 69

Unclassified

Unclassified

6

SIGNAL DETECTION

Although there are many proposals for 802.15.3a they have many features in common. Specifically, most of them involve the detection of multiple bands. In some cases, there is some modulation involved in the choice of band so it is necessary to receive all bands in parallel. This section and the following section on location can therefore examine the approach in general, making reference to particular aspects of the options where necessary. For signal detection the overall sensitivity needs to be maximised, and in general better than normal 802.15.3a devices. It is therefore necessary to maximise the signal that can be used.

6.1

PROTOCOL ISSUES

The detector needs all or most of the capabilities of a standard 802.15.3 terminal. There are potentially two modes, which may be applied for detection... Piconet participation ­ This mode provides an attractive method of signal detection. In this case the detector interacts with the device for detection (DFD) in order to get it to transmit. Clearly in order to elicit responses from the DFD the detector must be able to transmit to it. If the DFD may be illegal because of increased transmitter power then it may be necessary for the detector also to have an optional increased transmit power capability to provide reciprocal paths. The advantages of this mode of operation are:o o The DFD may be triggered to transmit at will. All parameters of operation are available and exchanged. By participating in the protocols the detector will be explicitly signalled with the signalling bit rate, the used band, the FEC mode employed, etc.

The problem with the participation mode is that it may not be possible if the piconet is set up with a closed set of addresses for operation. It may be possible to perform some limited participation in which the detector and the DFD exchange initial information and rapidly conclude that they are not part of the same piconet. This exchange may be adequate for the detector to identify the DFD and measure its power. Observer ­ In this mode the detector has no transmitter and must rely on observing any transmissions made by the DFD. Such a mode is viable provided that salient physical layer parameters are transmitted regularly in a way that is independent of the history of the exchange. Thus, many, if not all of the transmissions must be prefaced with a component that can be detected and decoded without knowledge of any data-exchange-specific physical parameters. This should be possible for even highly adaptive systems. For example, a terminal might listen to its environment to detect bands of other systems (e.g. 802.11a) which must not be interfered with. Such a terminal might adapt its transmission to avoid transmitting into those bands where necessary. However, it would not be safe for that terminal to assume that other terminals would recognise the same condition and act in the same way since those other terminals could be out of

72/03/R/208/U

Page 22 of 69

Unclassified

Unclassified

range of the other systems. Thus it would always be necessary to signal any band avoidance conditions explicitly. There would be a case for making the equipment capable of operating in either mode as required in order to maximise flexibility.

6.2

RF ISSUES

The detector ideally should operate with superior sensitivity compared with the normal terminal. There are few degrees of freedom available for achieving this. One, however, is to provide improved noise figure in the receiver. This can be achieved by increasing the gain of the low noise amplifier. Improved large signal handling capability may be required in order to maintain the necessary dynamic range. In single chip solutions for 802.15.3 there may be problems with interference from the digits to the RF. In the RA detector there is more scope for avoiding such problems since the RF and digital parts of the receiver can be separated and screened as necessary. For most of the proposals, some form of variable gain amplifier will be needed to cope with the dynamic range of the signals. This will operate in conjunction with an automatic gain control (AGC) sub-system to keep the level of the signal into the digitiser within the optimum range. Some of the proposals involved multiple local oscillators; others (the largest consortium proposal) have a single fast hopping local oscillator, which may be tuned by switching a limited set of mixing operations. Such operations will produce a number of spurious responses. However, these should be low enough not to cause difficulties. If the final selection has been made at the time of developing the detector and if it appears that non-compliant alternatives are unlikely then such an approach could be applied. If not, a more generalised approach would be recommended. It is likely that any physical layer proposal involving multiple bands will lend itself either to simplified generation of a fast hopping oscillator or of parallel multiple oscillators. For multiple parallel band solutions, the down-selection of proposals is leading to a small number of proposals with only a few bands. Thus, providing multiple local oscillators should not be problematic. These could be implemented as separate phase lock loops sharing a common reference. Because of the wide bandwidth being considered, full flexibility will not be possible in that we cannot have all voltage controlled oscillators (VCO) covering the whole band. Thus we will need a number of more limited range VCOs. However, we could arrange for the ranges of the VCOs to overlap in such a way that any proposal could be accommodated by suitable assignment of VCOs to bands. An alternative would be to generate one or more combs of frequencies and select the bands of interest by filtering. This approach may be simpler although it is clearly less flexible given the fixed filters that would be needed. Additionally, there would be significant spurious responses from the finite attenuation of the unwanted comb components.

6.3

DIGITISATION

Every proposal will require the conversion of the signal from analogue to digital at some point. The receiver architecture is not mandated in any of the physical layer proposals but their credibility has to be supported by presentation of viable receiver options. Many proposals suggest the use of extremely highspeed (even as high as 20 Gsps) analogue to digital converters (ADCs). This capability is afforded within a low cost mass-market device by the use of very limited precision conversion; sometimes as low as 1 bit (in

72/03/R/208/U

Page 23 of 69

Unclassified

Unclassified

this case the ADC is nothing more than a single comparator). In some cases 3 level ( 1.5 bits) digitisation is employed. This may be intended to work in conjunction with ternary modulation. Inevitably, limited precision conversion will led to some loss of sensitivity. In a conventional spread spectrum system, for example, 1 bit precision leads to a loss of about 2 dB sensitivity against noise. Thus, for RA's detector it would be desirable to use a minimum of 4 bits of precision, which for most modulation formats is adequate to bring losses down to a fraction of a decibel. In fact packaged 6 or 8 bit ADC devices with a sampling frequency higher than 1 GHz are available.

6.4

SIGNAL PROCESSING

Given the high sampling rates typically required, the signal processing will clearly also need to be very high speed, certainly higher than can be achieved with general purpose digital signal processing (DSP). The obvious contender for the processing is therefore the use of field programmable gate arrays (FPGAs). These will not provide the same cost/performance/power consumption characteristics as the full custom ASICs that will be developed for 802.15.3. However, it should be possible to implement the required functions within parameters that are acceptable for the RA detector. Where speed considerations are crucial, it may be necessary to apply some parallelism for certain functions. This might apply, for example, in the Fast Fourier Transform (FFT) required for the multi-band OFDM physical layer proposal. It may be appropriate also to incorporate a DSP to operate in conjunction with the FPGA. This can work on some of the slower functions.

6.5

FORWARD ERROR CORRECTION (FEC)

It is not clear at this stage whether FEC decoding will be a requirement for the detector. If the detector is operating as an `observer' then it will not be able to participate in any automatic repeat request (ARQ) protocols so may be unable to decode complete messages. The requirements of the detector are to detect the offending UWB device (possibly in the presence of other legitimate UWB transmitters). There is no explicit requirement to decode the data transmitted except insofar as this assists the primary requirement. FEC decoding might provide some useful information...

·

Where addresses are error protected it may be necessary to decode these. This could make it possible to distinguish members of a legitimate piconet from the offending device. Whether this is a real requirement in practice is debatable. Assuming that the operator has a directional antenna he/she may be able to `home-in' on the DFD and locate it to a small area (see Section 7). There could be only a limited number of devices within that area so separating them to identify the DFD may be straightforward. With weak signals, FEC provides coding gain which could make measurement of the received signal power more accurate. If the data is decoded then (after performing a checksum to test for validity) it can be re-encoded to create a reference for synchronous measurement of the received signal level. Typically this may provide a few (up to about 3) decibels improvement in sensitivity. Again it is questionable whether such improvement is enough to justify the additional complexity.

·

Thus at this stage there is doubt whether there is adequate benefit from implementing FEC decoding to justify its inclusion in the detector.

72/03/R/208/U

Page 24 of 69

Unclassified

Unclassified

6.6

GENERAL ALGORITHMS

In order to develop the detector it will be necessary to implement a number of receiver algorithms within the signal processing. These will include... 1. AGC ­ This could be implemented entirely in the analogue circuitry but provides best accuracy and flexibility if the measurements and control signals are performed/generated digitally. This approach would, of course, require a digital to analogue converter (DAC) to provide the control signal in a suitable form to drive the analogue variable gain amplifier. However, the sample rate for this could be very much lower than for the ADC since the bandwidth of the AGC loop will generally be low. Synchronisation ­ The timing of the signal must be determined. Every physical layer proposal will contain a transmission burst format description. The burst format will include synchronisation bursts, which consist of known data. In some cases the sync data may be one of a set of possible code words so that the receiver is addressed by the sync word. This would be problematic for the detector since this would never be addressed explicitly. Thus, if this approach is adopted, the receiver would need some form of parallel correlation. The synchronisation is needed to identify the signal as complying with the 802.15.3 standard, and to obtain its timing. Depending on the physical layer proposal, means for detecting multipath components may be provided as part of the sync word or the latter may be divided into two components, one for initial sync, the other for multipath sounding. Also, depending on the format of the physical layer proposal, it may be necessary to lock the frequency of the receiver clock to that of the received signal. Conventional approaches for doing this involve pulling the frequency of the reference voltage controlled crystal oscillator (VCXO) in the receiver to achieve frequency lock. However, this may not be possible because the dynamics of the oscillator may limit its rate of tuning unacceptably. For this reason, digital methods of frequency tracking may be required. Resolving of the multipath components may be necessary, for example, to extract the timings and weights to be applied to a `Rake' receiver in which the signal energy from multiple components are summed to improve the sensitivity. Inevitably, only a proportion of the total energy can be collected, as some of the components are too weak to be accurately measured for the purposes of generating combining weights. If there are a large number of such paths (as may be the case for a UWB signal) these can sum to a significant fraction of the total. Not all of the proposals are amenable to use of a Rake receiver. Those with a high symbol rate may require an equaliser. For this case, the channel must still be estimated in order to obtain the necessary parameters for setting up an equaliser. The OFDM based concepts do not require equalisation and all that is needed is the complex weight for each of the tones across the multiplex. This is simpler than for the other approaches. In one sense, as soon as synchronisation to a single path has been achieved then the signal has been detected. However, if only a fraction of the energy has been detected then it may not be possible to demonstrate that a power limit has been exceeded. Thus, measurement of the powers of a significant fraction of the paths is necessary, for a pulse-based approach, to obtain a realistic reading. This requirement should be mitigated to some degree if a horn antenna is used since this should exclude multipath energy arriving from angles widely separated from the line joining the DFD to the detector. Thus the longer (and typically the weaker) paths should be rejected.

2.

72/03/R/208/U

Page 25 of 69

Unclassified

Unclassified

Although the total energy is reduced through the rejection of multipath components outside of the main beam of the antenna this is more than offset by the gain of the antenna relative to isotropic. 3. Carrier Phase Tracking ­ Early UWB proposals were based on wideband pulses which did not have a carrier and therefore, for which carrier phase tracking was not applicable. Most of the remaining proposals have one or more bands that are narrow enough that carrier phase is a meaningful concept. Carrier phase can be estimated based on known data within the signal. This process may use the sync preamble to establish an initial phase. Additional tracking may be applied during the burst, based either on embedded pilot symbols or on decision directed estimation based on the received data signal.

72/03/R/208/U

Page 26 of 69

Unclassified

Unclassified

7

BASIC DIRECTION FINDING TECHNIQUES FOR UWB

A number of challenges exist when trying to locate a source of UWB interference. This section suggests the approach to be taken when carrying out detection using man-portable equipment. Characteristics of the signal, worst-case environments with high densities of UWB devices and requirements for the detection equipment (including basic direction finding capabilities, receiver sensitivity, SNR considerations and reasonable detection range) are covered. Central to the identification of interfering and unlawful UWB devices is the ability to carry out direction finding of the source and discriminate it against any other legal devices in the vicinity. Two approaches that have been investigated are the use of a high-gain, directional wideband antenna and the possibilities offered by the use of phased array techniques. In addition, the use of a two antenna direction finding (DF) approach has been explored. Polarisation is an important issue for the detector. It would be attractive if vertical polarisation could be assumed for the DFD since this could allow relatively omni-directional radiation in azimuth. However, this cannot be relied upon, particularly with a device that may not be in use in accordance with the regulations. Thus the antenna must operate successfully with a signal on any polarisation. This will be considered in each of the antenna options. An attractive option would be to provide circularly polarised antenna(s) which can handle any angle of polarisation. For selecting between antenna concepts we need to work with a common set of assumptions in order to make fair comparisons. There will be a practical limit to the physical size of the antenna/antenna array determined by the practicality of its being carried and used by an operator. However, the shape of the antenna is likely to be different for the different cases. At this stage it seems appropriate to apply a constraint to the total antenna area facing the field. The depth of the antenna is likely to be comparable for different antenna types so this is not considered. For the purposes of this study a figure of about 25 square 2 1 inches (160 cm ) has been chosen . This is somewhat arbitrary but should be reasonably close to a practical figure. If it needs revision the figures for each approach should scale roughly proportionally so that similar conclusions should be reached. Let us consider the issue of determining the angle of arrival of a signal at a detector. A number of generic techniques are possible, notably...

7.1

SINGLE DIRECTIONAL ANTENNA

At this point we consider the use of directional antennas in the detector. There are three benefits to this...

· · ·

Increased antenna gain leads to improved range Sweeping for the signal source helps determine its location The narrow beam will, to some extent, reject interference arriving from other directions and hence improve the detectability of the DFD

1

Imperial units have been chosen as more intuitive when considering practical size

72/03/R/208/U

Page 27 of 69

Unclassified

Unclassified

Unknown polarisation can be handled either by implementing polarisation diversity or a circularly polarised antenna. A horn antenna with square cross section can be implemented to take feeds from both vertical and horizontal polarisations. This can provide for diversity based on selection, combining or circular polarisation. There will be a small loss (up to 3 dB) some of the time from the latter but its convenience and simplicity are likely to be overriding considerations. We wish to maximise the antenna aperture given our size constraint to optimise directivity and gain. Given the need for equal gain in both polarisations this suggests a square cross section, which our size constraint suggests, should be 5" square at the opening. If we consider the lower part of the band then the useful centre frequency would be around 4 GHz. A horn of these dimensions would have a gain of around 14.6 dB. This gain would rise to 19.4 dB at 7 GHz. Assuming vertical polarisations the antenna beam width in elevation is 30° and in azimuth is 41° for 4 GHz. The figures for 7 GHz are 17° and 24° for elevation and azimuth respectively. For this approach the directivity of the antenna is used to provide angular discrimination. The antenna pointing angle is physically swept until maximum signal level is obtained. The advantages of this method are relative simplicity ­ only one receiver is required, along with high antenna gain, giving improved range. One disadvantage is that the angular discrimination is relatively poor, relying on finding the peak of a comparatively smooth function. There may be a number of UWB emitters all complying with the 802.15.3 standard. It will be necessary to distinguish these. Ideally, as any emitters are detected, all of their addresses should be obtained. The signal strength for each emitter should be associated with its address. As angular sweeping is performed to determine the direction of arrival for the emitters, it should be possible to find the peak for more than one device. Problems will arise when the average time between transmissions (particularly for the DFD) from the emitters becomes larger than the order of a second. In this case it would be necessary (in the case of manual sweeping) to move the antenna in small angles and wait for the signal to appear in each direction. The equipment could be arranged to remember the signal levels in each direction and provide a `back-count' to the strongest seen. This could be a little awkward for the operator.

7.2

PHASED ARRAY ANTENNA

Phased arrays for UWB differ from those of narrow-band signals because the phase change due to time separation of signals from different antenna elements varies greatly over the signal bandwidth (although the individual bands of some of the multi-band approaches could be treated in some ways like narrowband signals). Each element of a UWB antenna array will receive the signal independently. A variable delay-weighting network can be used to combine them. When the delay distribution is matched to the direction of arrival of the pulse then the signals received in each element will sum directly. In principle, this will provide a gain equal to the number of elements. In practice some weighting factor must be applied according to the pattern of the individual elements and this will reduce the gain somewhat. Thus the system can scan the delay pattern to effectively move the beam direction. For multiple band systems these effects are manifest in related phase shifts for each band. The correct common broadband delay will match the phase errors for all bands. The greatest sensitivity to delay will apply for the highest frequency band because the wavelength is smallest. Thus the angular resolution is related to the highest frequency rather than the bandwidth of the signal.

72/03/R/208/U

Page 28 of 69

Unclassified

Unclassified

Consider the minimum and maximum allowable distance between elements. Adjacent elements of the array need to have spacing great enough that small angular displacements separate the times of arrival of the signal by delays that are resolvable given the signal highest frequency. This suggests a spacing of a quarter wavelength at the highest frequency. For practical wideband antennas the physical size is dominated by the wavelength at the minimum frequency. This means that the minimum element spacing is likely to be based on the lowest frequency. Consider the Vivaldi antenna, which is a tapered stripline wideband antenna, providing about 8 dB of gain, see Figure 4.

Figure 4: Vivaldi Antenna The width of the antenna is nominally equal to a half wavelength at the lowest frequency. However, because the stripline is implemented on a high dielectric substrate the velocity of light in the material is reduced so the wavelength is also reduced. Velocity factors of 60 or even 50% can be obtained. In practice, because the wave is partly inside and partly outside the antenna, a factor of around 80% is more realistic. Thus the minimum antenna height is around 0.8, where is the wavelength at the minimum frequency. If the antennas were laid side by side with no gap then this would give a spacing of 0.8. In the same way as for the basic directional antenna there are two ways that polarisation uncertainties can be handled. One is to provide antennas with circular polarisation. With these a signal on either polarisation would be detected with only 3 dB attenuation. The alternative would be to have a pair of cross-polar antennas and use diversity combining. At this stage we assume that circular polarisation would be satisfactory. Either approach would require physically crossed antennas. These cross-polar antennas could be implemented as either vertical/horizontal or as ±45°. The latter has several advantages. Firstly, there is better symmetry. Secondly, for a horizontal array, as will be required for searching in azimuth, the antenna elements can be brought closer together. Thirdly, the maximum coupling between antenna elements will be lower. The array options are illustrated in Figure 5.

72/03/R/208/U

Page 29 of 69

Unclassified

Unclassified

a) Vertical / Horizontal

a) Plus/ Minus 45°

Figure 5: End on View of Cross Polar Arrays We can see that 5 elements of the ±45° antennas can be included in less space than 4 elements of the vertical/horizontal array. In fact this allows a horizontal spacing down to 0.8 /2 or 0.57 at the lowest frequency. Although inevitably there will be some coupling between the antennas this will be at an acceptable level. To get some idea of scale we note that the lowest frequency for UWB is 3.1 GHz. This corresponds to a wavelength of 9.7 cm. Thus the elements could be spaced at 0.57 x 9.7 = 5.5 cm. For a flat array it should be practicable to have up to 30 cm width or about 6 elements. This would give an overall structure size of 5.5 cm by 33 cm or 28.1 square inches, roughly consistent with our area constraint. Given 8 dB gain for each element this would provide an overall gain of 8 + 10 x log10(6) = 15.8 dB. The directional antenna had a gain of 14.6 dB. Normalising for the small difference in aperture size for the two cases, the difference is only 0.7 dB. In practice, given the imperfections in the array antenna the gains may be even closer. The spacing of adjacent antenna elements is still too great to avoid pointing ambiguities. For example, a pointing angle of +61° gives a phase shift at 3.1 GHz between adjacent elements of 180° (0.57 x 360° x sin(61°) = 180°). A pointing angle of -61° gives a phase shift between adjacent elements of -180°. These cannot be resolved. The effect is illustrated in Figure 6 for a 6 element array.

72/03/R/208/U

Page 30 of 69

Unclassified

Unclassified

0 -2 -4 -6 Antenna Gain (dB) -8 -10 -12 -14 -16 -18 -20 -90 -75 -60 -45 -30 -15 0 15 Pointing Angle (°) 30 45 60 75 90 3.1 GHz 4.1 GHz 5.1 GHz

Figure 6: Effect of Wide Bandwidth on Ambiguities ­ Phased Array A signal enters the array at an angle of ­61°. The figure plots the response of the receiver to this signal as a function of pointing angle. The signal bandwidth ranges from 3.1 GHz up to the frequency labelled on each curve. Every curve shows a peak response at a pointing angle of ­61° as expected. However, they also show responses around other pointing angles. The curve labelled 3.1 GHz corresponds to a narrowband signal at 3.1 GHz (bandwidth 3.1 GHz to 3.1 GHz). For this curve it can be seen that there is a response at +61° that is equal in strength to the response at ­61° which therefore cannot be resolved from it. This confirms the earlier simple analysis. However, as the bandwidth increases we plot the average of powers received over the bandwidth. Because the position of the ambiguous response alters with frequency we see a flattening in the overall average. Thus, for the curve labelled 4.1 GHz (corresponding to a bandwidth from 3.1 to 4.1 GHz) the peak of the ambiguous response is reduced to ­5 dB. For the unlabelled curves the upper frequency increases in steps of 1 GHz. Thus, the curves illustrate how the ambiguity can be resolved by the wideband signal. Figure 7 shows the responses to a wideband signal from 3.1 to 4.6 GHz for input angles ranging from ­60° to +60° in 15° steps.

72/03/R/208/U

Page 31 of 69

Unclassified

Unclassified

0 -2 -4 -6 Antenna Gain (dB) -8 -10 -12 -14 -16 -18 -20 -90 -75 -60 -45 -30 -15 0 15 Pointing Angle (°) 30 45 60 75 90

Figure 7: Response of 6 Element Array to 3.1-4.6 GHz Signals It can be seen that the unwanted responses are 6 dB down which should be enough reduction to avoid ambiguous detection. The individual wanted peaks are relatively narrow, allowing good angular resolution. The above discussion covers the fundamentals of creating a phased array antenna and shows that this can be done with practical size and provide unambiguously, a relatively good angular resolution for an ideal signal. We now consider the usage of such an array. There are two ways of using an antenna array to find a signal... 1. Sweep the Pointing Angle - Sweep the pointing angle over the range of angles by ramping the relative delays to each element. For a 30 cm array sweeping over ±60° the maximum delay is 0.3 x sin(60°)/c = 0.87 ns where c is the speed of light. The whole range of pointing angles is swept and the signal strength for each angle noted. The pointing angle providing the greatest signal strength is then the best estimate of the actual angle of arrival of the signal. Form Beam onto Signal - The alternative approach is to adjust the weights from all of the elements to maximal ratio combine the signals as received from each element. Once this is done, the weights are examined to obtain an estimate of the signal angle of arrival.

2.

There are pros and cons to both approaches. The first approach has the advantage of relative simplicity. The combiner weights are precisely set so there is no loss when the beam points towards the signal source. However, compromises are needed in determining the rate of sweep. On one hand the sweeping must not be so rapid that there is inadequate time to measure the signal strength with reasonable accuracy on a given pointing angle. On the other hand if the sweeping is too slow then the signal may be missed if the beam is pointing away from the signal source during the transmission period. This approach is analogous to the use of a manually steered directional antenna. The advantages over manual steering are that the steering can be arbitrarily fast and precise logging of direction is possible. In principle, multiple

72/03/R/208/U

Page 32 of 69

Unclassified

Unclassified

combining networks and receivers could be put behind the antenna array allowing multiple parallel beams. In practice this would almost certainly be prohibitively costly. The beam forming approach has the advantage that a beam can be formed almost instantly using the sync information from a transmission burst. Thus, there is no concern about the problem of missing the signal arising from the antenna pointing in the wrong direction at the time of the burst. The disadvantage is that the signal on each element individually must be strong enough to permit detection and estimation of strength and time of arrival to allow near optimal combining. Thus there will be a signal to noise penalty. Effectively, much of the range advantage obtained from combining across elements is lost. There are more fundamental issues that will affect either approach. Because of the very wide bandwidth of the signals for detection they will be resolvable into a large number of multipath components. In general there will be a rich multipath environment involving possibly as many as dozens of paths. Because all but one of these paths is non line of sight, their angles of arrival will make them, to greater or lesser degrees, unsuitable for locating the transmitter. The extent of this problem will depend on the locations of the reflectors. If the detector is at some distance from the emitter and if most of the reflectors are in the vicinity of the emitter than the estimated angle should be in the rough direction of the emitter. On the other hand if there are significant reflectors surrounding the detector then initial direction estimates may be totally erroneous. The degree to which non line of sight paths can be rejected in the receiver will depend on the structure of the signal finally selected for 802.15.3a. Let us consider two extremes. In principle classical UWB signals based on very wideband pulses would allow direction finding based on selecting only the earliest arriving pulse. If the line of sight path is obstructed by, for example, a wall then it may not necessarily be the strongest path. This could make its identification problematic. Given that the signal is usually fragmented into many components, there will generally be a significant loss in received signal strength from using only the line of sight path. Nevertheless, with relatively high gain antennas it may be possible to achieve some useful performance. Other UWB proposals are based on signals with narrower short-term bandwidth. In particular the TI proposal uses 528 MHz bandwidth OFDM signals. The individual tones of the OFDM multiplex are spaced at 4.125 MHz intervals. Thus the fine structure of the signal is not wideband. However, the received signal is the convolution of the channel impulse response with the transmitted signal. If the complex weights of the received signal on each tone can be estimated, as will be necessary to demodulate the signal then their inverse Fourier transform will yield an estimate of the channel impulse response. Given a bandwidth of around 500 MHz, the resolution will be order 2 ns. However, the signal is frequency hopped over a number of bands (usually 3). It should be possible to combine complex weight estimates from multiple bands to obtain higher resolution estimates of the channel impulse response. In order to do this it will be necessary to align phases at band edges. This can be done by estimating the phase profiles of adjacent bands and matching them together. Some gain matching may also be necessary to compensate for different RF gain paths. With a 3 adjacent band signal it may be possible to achieve resolution down to 0.66 ns. The resolution defines the minimum time separation between multipath components necessary to allow them to be recognised as separate. Although the resolution is limited to 0.66 ns, the accuracy can be markedly better than this for a resolved discrete path. Improved accuracy can be obtained either by zero-stuffing the frequency estimates or by pattern matching the impulse response. Either use of the antenna array may be applied for both types of signal.

72/03/R/208/U

Page 33 of 69

Unclassified

Unclassified

7.3

DIRECTION FINDING CONFIGURATION

The proposal only considered the above two approaches. However, during the course of the work a third option which holds some attractions has emerged. This is an approach based on traditional direction finding (DF) technology. The concept is straightforward and represents a compromise between the two approaches described above. In this approach a pair of circularly polarised horns are arranged side by side. In order to meet our size constraint the horns both have an aperture of 12.5 square inches and are assumed to be square. Each horn receives its signal separately. Correlation is performed to determine the time of arrival difference for the signal at the two horns and the angle of arrival is inferred from this. Because of the physical size of the horns their separation is increased compared with the phased array antenna and this exacerbates the ambiguity problem described in Section 7.2. However the wideband signal mitigates this in the same way as previously. This effect is illustrated in Figure 8, which is analogous to Figure 6.

0 -2 -4 -6 Antenna Gain (dB) -8 -10 -12 -14 -16 -18 -20 -90 -75 -60 -45 -30 -15 0 15 Pointing Angle (°) 30 45 60 75 90 3.1 GHz 4.1 GHz

Figure 8: Effect of Wide Bandwidth on Ambiguities ­ DF Structure Each horn has a beam width in azimuth of about 60°. Thus the pointing angle can only usefully be swept over ±30°. It would therefore be necessary for the operator to move the antenna in steps of 60° to cover a full 360° sweep. Essentially this approach combines the basic high gain properties of the horn with the sweeping capabilities of the phased array antenna. Compared with the phased array it reduces the number of receiver functions, greatly simplifying the hardware.

72/03/R/208/U

Page 34 of 69

Unclassified

Unclassified

7.4

PATH LOSS CALCULATIONS

Suppose, initially, that we have transmit and receive antennas that have 0 dB gain with respect to isotropic across the band. At any given frequency the free space path gain is given by

4d

2

where is the wavelength in metres and d is the range in metres. Alternatively the path loss may be expressed as...

c 4df

2

where c is the speed of light in m/s and f is the frequency in Hz The maximum transmitted power density is P mW/Hz. Therefore the received power at a range of d metres is

P

fU

fL

c 4df

c .df = P 4d

2

2

1 1 - f fU L

where fU and fL are the upper and lower frequency limits respectively. If we define the normalised range, dn as unity for fU = 10.6 and fL = 3.1 GHz then we have

dn =

1 1 - 3.1 fU

1 1 4.3813 - = 1.4133 - 3.1 10.6 fU

We plot the normalised range for 3.1 < fU < 10.6 GHz in Figure 9.

72/03/R/208/U

Page 35 of 69

Unclassified

Unclassified

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 3 4 5 6 7 Upper Frequency (GHz) 8 9 10 11

Normalised Range

Figure 9: Normalised Range v Upper Frequency Limit From this we see that the range for the isotropic antenna case is not a strong function of the bandwidth. Even with a bandwidth up to only 7 GHz, nearly 90% of the maximum available range is obtained. The expression for received power at a distance of 1 m for the isotropic antenna case can be arranged in the following form:Pdbm + 27.6 + 10 x log10(1/fL ­ 1/fU) where Pdbm is the transmitted power spectral density in dBm/MHz and the frequencies are also in MHz Using the FCC limit of ­41.3 dBm and the maximum frequency limits of 3.1 - 10.6 GHz we get a total received power for the isotropic antenna case of ­50.1 dBm at a range of 1 m. The use of this bandwidth maximises the total power that may be transmitted given compliance with the FCC spectral mask. However, most of the physical layer proposals are based on transmission in bands of bandwidth order 500 MHz. This is done to simplify the equipments whilst complying with the FCC requirement for 500 MHz as a minimum bandwidth where the fractional bandwidth constraint is not satisfied. Proposals are divided according to whether multiple bands are transmitted in parallel or whether the bands are used serially. For the latter case the rate of switching between bands determines whether the whole band or the partial band can be considered from the viewpoint of maximum permissible transmitted power. Most proposals aim to switch rapidly enough that the whole bandwidth can be considered in order to maximise the allowed transmitter power and therefore the range.

72/03/R/208/U

Page 36 of 69

Unclassified

Unclassified

For example, considering the multi-band OFDM proposal, in its Mode 1 the bandwidth is from 3.168 to 4.752 GHz. Applying this to the expression gives a received power at 1 m of -53.5 dBm. This is within 1 dB of the value quoted by the proposers. The 802.15.3 standard allows for a number of different bit rates, the highest being 500 Mbps. From the viewpoint of detection this is the worst case and since, in general, the detector will have no control over which bit rate is being used, this value should be used for the purposes of range computation. It is not clear what noise figure will be possible for the detector but this should be lower than for the standard equipment. Most physical layer proposals quote a noise figure in the region 6 to 7 dB so for the purposes of this study we will assume that the detector can achieve 5 dB. The effective receiver input noise level will then be...

6 10 x log10(kT)+10 x log10(500 x 10 ) + NF = -174 + 87 + 5 = -82 dBm.

The proposers quote a variety of figures for the required Eb/No as tabulated below.

Proposal Communications Research Lab Oki Parthus Ceva ST Microelectronics University of Minnesota XtremeSpectrum Texas Instruments Average

Minimum Eb/No (dB) 7.2

Implementation Loss (dB) 3

Total (dB) 10.2

Directional antennas ­ Not applicable 4.3 6.4 4 4 Not Quoted Directly 9.6 4.9 6.5 3 3 3.4 12.6 7.9 9.9 8.3 10.4

Table 7: Required Eb/No In each case the values for the highest quoted available bit rate is used. The figures are not intended as figures of merit for the different systems since many other factors contribute to their relative attractiveness but to highlight the general similarity over the different proposals. In view of the relatively small spread it was decided to take the average figure for the sum of minimum Eb/No and implementation loss of around 10 dB as the applicable protection ratio. Taking the multi-band CDMA proposal as typical for received signal power and taking the 10 dB protection ratio we have a link margin at 1 m of -53.5 - -82 ­ 10 = 18.5 dB. Given the variations in assumptions this is comparable with the link margin of 12.2 dB quoted for the multi-band CDMA proposal for a range of 2 m. In principle, for a line of sight path, the above would correspond to a range of 8 m with an isotropic antenna on the detector. With the considered antennas we have an additional link margin of around

72/03/R/208/U

Page 37 of 69

Unclassified

Unclassified

15 dB. It should be borne in mind, however, that the budgets are computed on the basis of an isotropic antenna at the emitter. Given that the FCC regulation limits the EIRP in the highest gain direction of the transmitting antenna, we would be very fortunate to experience this level of power. A more realistic level might be 3 to 6 dB down on this. If we take a figure of 6 dB then the overall excess path loss at 1m would become 18.5 ­ 6 + 15 = 27.5 dB. In principle this would led to a range of 23 m in line of sight conditions. In practice the channels become more complex with increasing range and it becomes more difficult to exploit the whole of the signal. At short range the computed path loss might be reasonably realistic. At longer range the signal will be highly fragmented into multipath components, resulting in some loss of signal, when direction finding is considered.

7.5

COMPLEXITY OF THE METHODS

In this section we consider the relative complexity of the methods for finding the direction of the DFD using either a directional antenna or a phased array. Some of the issues are common to both. In considering the relative complexities of the different approaches we defer detailed complexity analysis to the next section. The directional antenna is clearly the simplest of the options, requiring only one receiver. This one receiver detects each signal and the signal strength is logged. A block diagram of this option is shown in Figure 10.

ADC 0 90 ADC

LNA

FPGA

DAC

Figure 10: Block Diagram for Directional Antenna Case The filter constrains the signal to the RF bandwidth of interest. In the case of single band solutions it might be a matched filter to the transmitted pulse. A low noise amplifier to overcome the noise of subsequent stages follows this. A variable gain amplifier follows to implement automatic gain control (AGC). The band of interest is then down converted in quadrature according to a local oscillator that may be frequency agile. The I and Q components are digitised and fed to the FPGA for processing. A digital to analogue converter fed from the FPGA controls the gain of the variable gain amplifier to close the AGC loop. This single receiver can be built on to enable direction finding to be implemented using a phased array. The DF approach requires only two receivers to process the signal. If we use beam forming then many of the functions need to be multiplied up. Specifically, the following elements will need replication, one per antenna element:

·

RF Filter

72/03/R/208/U

Page 38 of 69

Unclassified

Unclassified

· · ·

Low Noise Amplifier Mixer Analogue to Digital Converters

A block diagram of the DF option is shown in Figure 11. This is the preferred solution for the detector unit.

ADC 0 90 ADC

LNA

FPGA

FPGA / DSP

ADC 0 90 ADC

LNA

FPGA

DAC

Figure 11: Block Diagram for DF Antenna Case The reference and local oscillators will be common across all receivers. This is necessary to achieve coherent operation across the receivers as well as achieving savings in hardware. The ADC outputs will need to be first processed independently and then the processed outputs brought together. Thus, the FPGA processing will also be required on a per-antenna-element basis. Finally, an overall processor will combine the outputs of the receivers. It can be seen that the architecture is a duplication of the elements of Figure 10. The only differences are a common local oscillator and a common control for the AGC.

72/03/R/208/U

Page 39 of 69

Unclassified

Unclassified

Obviously for a 6 element array there will be 6 receivers, leading to considerable complexity. This is illustrated in Figure 12.

ADC 0 90 ADC

Element 1

LNA

FPGA

Element 2

Element 3 FPGA / DSP Element 4

Element 5

ADC 0 90 ADC

Element 6

LNA

FPGA

DAC

Figure 12: Block Diagram for 6-Element Beam Forming Phased Array If the phased array antenna is used in conjunction with a sweeping antenna then considerable simplifications are possible. The block diagram is shown in Figure 13. In this case the combining of signals from the array elements is performed before most of the RF and digital processing. Each antenna element has an associated filter. These might be simplified or removed if the dynamic range of the following low noise amplifier is great enough. Each element is then followed by an analogue variable delay element. This can be implemented as a binary sequence of pin diode switched delay elements. The maximum delay is 0.87 ns to cover ±60°. A granularity of about 1% should be adequate for delay setting, suggesting 128 delays for a binary delay. For a regular linear array, every delay is a linear multiple of the minimum, nonzero, delay. Thus the `Logic' block can readily derive the set of delays for each block. Following the delays the signals are summed and then fed to a receiver that is identical to the back end of that shown in Figure 10.

72/03/R/208/U

Page 40 of 69

Unclassified

Unclassified

Logic

LNA

Delay

LNA

Delay

ADC LNA Delay 0 90 LNA Delay ADC FPGA

LNA

Delay

DAC

LNA

Delay

Figure 13: Block Diagram for Sweeping Phased Array

7.6

INTERIM CONCLUSIONS

This section has presented four options for locating the DFD... 1. 2. 3. 4. Single directional antenna Phased array with beam forming onto signal Phased array with sweeping of beam Two antenna direction finder

Option 1 is the simplest but it is likely that the operator will have difficulty using this when the transmissions are intermittent. Option 2 is the most comprehensive. It would allow any or all of the capabilities of the other options to be provided. Unfortunately it is also the most complex, requiring essentially six complete receivers. If the equipment were implemented as a manpack with a searching antenna `head' then there would need to be six cables feeding the signals back to the man-pack. Given that these would need to carry UWB signals this would be somewhat impractical. Moreover, the cost, complexity and power consumption of the multiple receivers could become prohibitive particularly for a portable, battery operated equipment. Option 3 would be a more practical approach than option 2, still providing some electronic beam steering but at substantially reduced complexity. The hardware within the dotted box of Figure 13 could possibly be included within the antenna head, the main challenge being to incorporate all the variable delays. Nevertheless, the capabilities provided by this option do not fully match with the

72/03/R/208/U

Page 41 of 69

Unclassified

Unclassified

requirements. In the same way as for option 1, an intermittent signal may be missed if the signal is not active when the beam is swept near to the peak response. Option 4 provides the best of both worlds. The overall complexity is modest, incorporating only two receivers, albeit almost complete receivers. The DF algorithm that is based on measuring delays between the received signals in the two antennas, in principle, allows the angle of arrival to be estimated from a single burst.

8

IMPLEMENTATION COMPLEXITY

The following sub-sections investigate in detail the implementation and complexity based on the approach identified in the previous sections. Additional detail associated with this is included in Annex A. To allow sufficient depth of investigation to determine feasibility only the Texas Instruments OFDM method has been investigated in detail. This has been selected, as it is the proposal currently receiving the strongest commercial backing and the most votes within IEEE. However, because it cannot be certain that th it will be adopted some analysis for the other remaining proposal (as of 16 Sept 2003) has also been include. The following areas are considered:

·

Receiver performance requirements o o o o o o Sensitivity Noise Figure Oscillator performance (stability and control range) Implementation Loss factors Front-end Gain and AGC Dynamic Range & ADC performance

·

Receiver architecture o o Block diagram of receiver Evaluation of current technology available for implementation

·

Detection algorithms o o o o o Sampling techniques ­ complexity and performance trade-offs Candidate algorithms for operation in a low SNR environment Search efficiency and time-to-detect performance Algorithm complexity Implementation factors such as latency, processing requirements, etc

·

Detector performance and operational factors

72/03/R/208/U

Page 42 of 69

Unclassified

Unclassified

8.1

TEXAS INSTRUMENTS

This proposal has been evaluated in detail in Appendix A, in which a detailed receiver structure is presented. This section summarises the operation of a receiver in less detail, including results from Appendix A.

8.1.1 RECEIVER PERFORMANCE REQUIREMENTS

An evaluation of the receiver sensitivity issues has already been given in Section 6.2. This section goes into more detail.

8.1.1.1

Noise Figure

The noise figure is a function of the level and gain plan along with the noise figures of the individual components. This is considered in Section 7.4 as a factor that is, to a first approximation, common to all proposals. From that section we obtain a figure of 5 dB.

8.1.1.2

Front-end AGC

The front end gain required to achieve the specified noise figure is 53 dB. The dynamic range required for the AGC is of the order 56 dB. Half of this will be implemented in RF and half after the RF down conversion.

8.1.1.3

Dynamic Range & ADC performance

The dynamic range has been specified above. This is met largely by the AGC. The requirements of the ADC are to handle the instantaneous dynamic range (peak to mean ratio) of the OFDM signal. Texas Instruments propose a 4-bit ADC. Simulations indicate that this will give an implementation loss of the order of 0.25 dB. In view of the ready availability of 6-bit ADCs it is proposed to use this number and obtain negligible implementation loss from this source.

8.1.1.4

Oscillator Performance (stability and control range)

The reference oscillator for the receiver will be fixed in frequency since it is not possible to tune it rapidly enough to receive the data burst. Thus all-digital demodulation is assumed. The Texas Instruments proposal is based on reference oscillators with an accuracy of ±20 ppm. In order to simplify operation and improve performance, a ±5 ppm (initial accuracy and aging) oscillator appears a good compromise for the monitor. This leaves the overall frequency errors dominated by the transmitter errors without specifying difficult requirements, which yield diminishing returns.

8.1.1.5

Implementation Loss Factors

Five elements have been considered as contributors to the overall implementation loss. These are tabulated in Table 8 - see Appendix A.

72/03/R/208/U

Page 43 of 69

Unclassified

Unclassified

Factor Quantisation (6-Bits) Aliasing Channel Estimation Analogue Frequency Errors Finite Precision Arithmetic Total

Level (dB) 0.025 0.25 0.55 0.11 0.2 1.135

Table 8: Implementation Loss Factors

8.1.1.6

Sensitivity

The sensitivity can be readily computed from the implementation loss factors. This is based on a half rate code and 512 MHz transmission. It is acknowledged that the service bit rates are lower than this but this will be taken into account in TI's definition of implementation loss.

6 S = 10 log10(kT) + NF + 10 log10(512 x 10 ) + Impl + Req_EbNo = -76.8 dBm

where, NF Impl Req_EbNo is noise figure in decibels (= 5 dB) is the implementation loss (= 1.135) is the required Eb/No (= 4 dB)

8.1.2 HARDWARE IMPLEMENTATION

Following on from section 7.5 regarding the complexity of the detector Figure 14 shows a simplified toplevel block diagram. It is clear to see that there are two channels for DF with their outputs combined in the Control Processor. This will be operating at lower rate compared to the Baseband Processing FPGAs. The Control Processor will handle the Display, Keypad, User interface along with the outputs of the Baseband FPGAs. This device could be a DSP, microprocessor, or another FPGA.

72/03/R/208/U

Page 44 of 69

Unclassified

Unclassified

Down Converter

In-phase VCG ADC

A B

LNA BPF

AGC

0 90 Quadrature Phase VCG

Dual ADC A

Baseband Processing FPGA

B

ADC

Display

Frequency Hopping Synthesiser

Control Processor

(FPGA, DSP or uP)

Keypad User Interface

Down Converter

In-phase VCG ADC

A B

LNA BPF

AGC

0 90 Quadrature Phase VCG

Dual ADC A

Baseband Processing FPGA

B

ADC

Power Supply Regulation Circuitry

DAC

Figure 14: Simplified Top Level Block Diagram

8.1.2.1

Evaluation of Technology

As mentioned in the previous sections the critical components for the detector are: Analogue-to-Digital To improve the matching of the I and Q paths a dual Analogue-to-Digital is recommended. As previously mentioned this device would have the following parameters: Sample Rate IF Frequency Bandwidth Resolution 528 MSPS > 528 MHz 6 bits

All of these parameters can easily be met by currently available technology. Table 9 lists a selection of potential ADC devices that could be used. However, these devices will require quite large amounts of power and careful routing of the printed circuit board at these clock frequencies. Furthermore, the output of the ADC will need to be de-multiplexed to a data rate that the following stage FPGA is able to handle. For this detector it is likely to be limited to 264 MHz.

72/03/R/208/U

Page 45 of 69

Unclassified

Unclassified

Specification Manufacturer Part Number Sample Rate (GSPS) 2 1 1.5 0.6 2.5 Up to 6 Input Bandwidth (GHz) 3 0.7 2.2 2.2 5 3 Resolution (Bits) 10 6 8 8 6 6 ENOB (Bits) 7.8 5.7 7.5 7.6 5.2 5.5

Atmel Atmel Maxim Maxim Snowbush Microelectronics Rockwell

TS83102 AT76CL610 MAX108 MAX106 ADC8005 (CMOS Macro) RAD006

Table 9: ADC Devices Available Baseband Processing FPGA This will perform the main processing of the received signal including the FFT Demodulation, Deinterleaving and Decoding. A very simplified view of the functionality of the FPGA is given in Figure 15. A detailed breakdown of these blocks and their control mechanism is given in Appendix A.

Front End Circuitry

Control Logic

Output to Control Processor

Digitised Input from ADC

FFT

De-Interleave

Decode

Descramble

Synchronisation

Figure 15: Simplified Diagram of the Baseband Functionality Based on the analysis in Annex A; the FPGA will require as a minimum the following parameters: Clock Frequency Logic Cells/Elements Memory 264 MHz > 45,000 > 1.5 Mbits

It is envisaged that the implementation of the demodulator function within the FPGA will be challenging. The main reason for this is the clocking frequency required to perform the multiplications within the FFT block. To achieve this there are a number of multiplication techniques that could be utilised. However, there is a trade-off between resources, latency and speed. From Altera's Application Note 306

72/03/R/208/U

Page 46 of 69

Unclassified

Unclassified

"Techniques for Implementing Multipliers in Stratix, Stratix GX & Cyclone Devices" a 12-bit x 9-bit Firm Multiplier (using DSP blocks and Logic Elements) can achieve a performance of 270MHz with a latency of 2 clock cycles when implemented in an EP1S10F484C5 device. To implement the complete FFT and synchronisation, the performance of the synthesis and place and route tools will be crucial in achieve the design requirements. However the analysis shows that Altera and Xilinx FPGA devices should be able to support the main processing in a single device.

8.1.2.2

Level Plan

A spreadsheet level plan for the receiver has been produced and is given in Figure 16. The values used are based on currently available component technology demonstrating that it is possible to implement a receiver without using the latest devices. The referred noise figure from the level plan is 4.05 dB which is slightly better than the 5 dB assumed figure mentioned in section 7.4.

Level Plan

FCC in-band permitted level Signal Bandwidth Bandwidth ADC Information: No. of bits Sample Rate Voltage Full scale RMS Full scale Signal Power Signal to Quantisation Noise Ratio (SQNR) Quantisation noise power Parameter Gain Cascaded Gain Linear Gain Cascaded Linear Gain Noise Figure Linear Noise Figure Cascaded Linear Noise Figure Cascaded Noise Figure Mean Signal Level Peak Signal Level Thermal Noise Power (KTB) SNR (Mean Signal to Noise) SNR (Peak Signal to Noise) Input Referred Noise Figure, NF -41.30 dBm/MHz 5.28E+08 Hz 87.23 dBHz Constants: Boltzman's constant Temperature (K) KT Thermal Noise Floor

6.00 5.28E+08 0.80 0.28 2.04

bits Hz Vpk-pk Vrms dBm

1.38E-23 293.00 4.04E-21 -173.93

J/K K J dBm/Hz

34.87 dB -32.83 dBm Input Filter -1.00 -1.00 0.79 0.79 1.00 1.26 1.26 1.00 -83.00 -71.00 -86.71 3.71 15.71 dB LNA 20.00 19.00 100.00 79.43 3.00 2.00 2.51 4.00 -63.00 -51.00 -63.71 0.71 12.71

Pulse Power Pk-to-Mean Ratio AGC 25.00 44.00 316.23 2.51E+04 5.00 3.16 2.54 4.05 -38.00 -26.00 -38.66 0.66 12.66 Down Converter 28.00 72.00 630.96 1.58E+07 6.50 4.47 2.54 4.05 -10.00 2.00 -10.66 0.66 12.66 ADC Input

15.85 Units dB dB

dB

-82.00 -70.00 -86.71 4.71 16.71 4.05

-10.00 2.00 -10.66 0.66 12.66

dB dBm dBm dBm dB dB 0.20 Vpk-pk 0.80 Vpk-pk

Mean ADC Voltage Peak ADC Voltage

Figure 16: Level Plan

72/03/R/208/U

Page 47 of 69

Unclassified

Unclassified

8.1.2.3

Power Budget

The ADC's and the FPGA will require the largest amount of power for the detector. Table 10 estimates the power requirements for a Man Portable UWB Monitor. This level of power will restrict the operating period of the unit if it is powered via batteries. In addition the thermal issues of the unit will need to be carefully considered to dissipate the heat. Component Rx Channel #1 Dual 6-bit ADC FPGA 1.5 10 Clock frequency of 528MHz Clock frequency of 264MHz 10W when in synchronisation 3W obtaining synchronisation VGC LNA AGC Rx Channel #2 Dual 6-bit ADC FPGA 1 1 1 0 1.5 10 Clock frequency of 528MHz Clock frequency of 264MHz 10W when in synchronisation 3W obtaining synchronisation VGC LNA AGC Control Processor User Interface, Keypad and Display Oscillator DAC Miscellaneous Power Supply Regulation Total Power Consumption 1 1 1 1.5 0.2 0.2 0.2 1 6.4 38.5 Assumed to have a slow clock rate Termination circuitry, buffers, translators, etc. Assumes an efficiency of 80% This is the worst-case power. Power (W) Comments

Table 10: Estimated Power Budget

72/03/R/208/U

Page 48 of 69

Unclassified

Unclassified

8.1.2.4

Thermal Management

Careful consideration is needed for the thermal design of the receiver due to the high power levels given in 8.1.2.3. The printed circuit board is likely to be 10 to 12 layers with several layers used as thermal plane. Techniques such as thermal vias will be utilised and the use of large device packages which have good thermal properties. The use of heatsinks and forced air cool is also recommended since conducting the heat away may not be sufficient.

8.1.3 DETECTION ALGORITHMS

8.1.3.1 Sampling Techniques

This has been evaluated in Appendix A. There it is shown that sampling at 512 MHz provide satisfactory operation with an implementation loss of only 0.25 dB.

8.1.3.2

Candidate Algorithms for Operation in a Low SNR Environment

The algorithms are described in Appendix A.

8.1.3.3

Search Efficiency and Time-to-Detect Performance

The TI proposal is so arranged that detection of the sync word is not a determining factor in receiving the physical layer burst and can be detected on a single occurrence. This is considered in more detail in Appendix A. Therefore there is no significant search time required.

8.1.3.4

Algorithm Complexity

Algorithms have been developed that are efficient for the requirement. The complexity is considered in the implementation section. Latencies are not anticipated to cause difficulties at this stage

8.1.4 DETECTOR PERFORMANCE AND OPERATIONAL FACTORS

The performance of the detector will be dominated by its sensitivity. It should be able to receive individual bursts. At this stage only physical layer information is available. The ability to decode user-specific identifiers such as addresses will depend on how the higher layers are arranged. It is understood that representations may be made into the IEEE 802.15.3 to influence their development to ensure that monitoring can be satisfactorily performed.

72/03/R/208/U

Page 49 of 69

Unclassified

Unclassified

8.2

XTREMESPECTRUM, MOTOROLA AND PARTHUS CEVA INC. MERGED

The merged proposal #2 led by XtremeSpectrum in collaboration with Parthus Ceva is a Split Band DSCDMA approach. In the revised proposal, IEEE 802.15-03/153r9, there are 4 spectral modes of operation.

· · · ·

Low Band High Band Duplex-Band

3.1 to 5.15 GHz 5.825 to 10.6 GHz 3.1 to 5.15 GHz 5.825 to 10.6 GHz 3.1 to 4.9 GHz 6.3 to 8.1 GHz

2.05 GHz Bandwidth 4.775 GHz Bandwidth 2.05 GHz Bandwidth 4.775 GHz Bandwidth 1.8 GHz Bandwidth 1.8 GHz Bandwidth

28.5 to 400 Mbps 57 to 800 Mbps

Up to 1.2 Gbps

Joint-Band

Data rate not stated

These bands are illustrated in Figure 17 and Figure 18.

Low, High and Duplex Bands

1.2

Duplex Band

1 0.8 Power Levels

Low Band

0.6

High Band

0.4 0.2

0 1 2 3 4 5 6 7 8 9 10 11 12 Frequency (GHz)

Figure 17: Low, High & Duplex Frequency Bands for XtremeSpectrum

72/03/R/208/U

Page 50 of 69

Unclassified

Unclassified

Joint Band

1.2 1

0.8 Power Levels 0.6

0.4 0.2

0 1 2 3 4 5 6 7 8 9 10 11 12 Frequency (GHz)

Figure 18: Joint Frequency Band for XtremeSpectrum Currently, the technology does not provide adequate performance to digitise the full bandwidth of 7.5 GHz. The maximum speed of the current ADC devices, Table 9, is around 6 GSPS with an IF bandwidth of 3 GHz and a resolution of 6-bits. However, this is a preliminary front sheet. Implementing this high-speed sampling ADC in the receiver using sub-sampling techniques it would enable the power levels in the band to be measured. The proposal states a sampling frequency of 6.385 GHz for the Joint-band mode. Figure 19 gives a possible receiver architecture. Here, the front end section of the receiver has been simplified compared to the Texas Instrument approach as there is no need for down convert to get the I and Q components and perform frequency hopping. The band-pass filter arrangements have changed to have a switched band-pass filter instead. This will help to improve the performance for each particular mode of operation. The amplified signal is applied to the ADC running at Fs = 6 GHz. The sample and hold circuit, the low pass filter and the ADC form a sub-harmonically sampled system. The result of the sub-harmonic sampling is a signal that is an exact replica of the original signal (with its frequency compressed) providing that the signal bandwidth is below Fs/2. If the bandwidth of the signal is greater than Fs/2, aliasing occurs, but the aliased signal retains the same peak to mean ratio as the original signal.

72/03/R/208/U

Page 51 of 69

Unclassified

Unclassified

Switched BPF LNA AGC ADC Demux

X

Y

Baseband Processing FPGA Keypad

Oscillator

Control FPGA

Display User Interface

Switched BPF LNA AGC ADC Demux

X

Y

Baseband Processing FPGA

DAC

Figure 19: Top Level Block Diagram for Merged Proposal #2 Format Receiver The sampled ADC data at 6 GHz is divided down by a factor of 24 using the de-multiplexer, to allow the Baseband FPGA to process the data at 250 MHz. The captured data can then be analysed in terms of time and frequency domain. However, this will involve a complex algorithm to restructure the data due to the de-multiplexing function and operates at high clock frequencies. The advantage of the sub-sampling is signals can be captured and the true peak to mean value accurately calculated. Sub-sampling captures all spectral components over a wide frequency band. In addition Frequency Hopping and Down Conversion are not required. The disadvantage with this architecture is that the sampling rate is significantly high requiring careful design effort for layout, power supplies and thermal areas. Due to the high sampling rate there will be a huge amount of data that the Baseband Processing FPGA will have to process rapidly using parallel and pipeline functions. This will essentially result in a greater latency for the receiver and a more complex algorithm to be implemented in the FPGA compared to the Texas Instruments approach.

9

DETECTING OF NON STANDARD UWB SIGNALS

The problem of detecting unknown UWB emitters is extremely challenging. There are some key questions as to what is required. We assume that a complaint has been made about interference and that the emitter is being sought. Of course, a priori, it is not known whether the cause of interference is a UWB device ­ determining this is part of the requirement. The situation is more uncertain than this since an illegal device may share some of the properties of a UWB emitter but not all ­ it may not satisfy all or indeed any of the requirements described in section 4. Thus, for example, the signal may be radiated with less than 500 MHz bandwidth or even below 3.1 GHz. For the purposes of this section, therefore, we consider an illegal emitter as UWB if its radiated bandwidth is significantly greater than that of a single carrier of any legal terrestrial communications transmission in the same part of the band. Given signals with 20 MHz bandwidth we consider the minimum bandwidth as 100 MHz.

72/03/R/208/U

Page 52 of 69

Unclassified

Unclassified

The requirement, then, is to detect and locate such an emitter. Problems in doing this include...

·

Interference from other users of the spectrum ­ in particular it may be that the signal that the complainant is having difficulty receiving may be a strong interferer to reception of the unknown UWB signal in the vicinity of the monitor. Unknown signal format.

·

Susceptibility to interference in detection is different for an unknown UWB emitter from that for a known UWB emitter. The main difference is that the signal from the known UWB emitter can be received making full use of the processing gain. Thus the applicable noise bandwidth for reception is the symbol rate. For unknown emitters the applicable bandwidth is the whole bandwidth that the signal occupies. This difference arises because, for a known UWB emitter, the signal can be match filtered and time synchronised. This is not usually possible for an unknown emitter. The only way that interference can be handled is by providing some discrimination between it and the signal. The obvious approach is to use angle of arrival. This implies the use of one or more of the techniques of Section 7. Given possibly large differences in signal strength (the interference to the signal from the UWB emitter may be stronger) it is unlikely that the use of a single directive antenna would be adequate. The UWB signal might be interfered with by a significant number of interference sources. However, given the very large bandwidth involved, these interferers might be separated in frequency from one another. Techniques for signal separation require at least as many antennas as signals to be separated. From a practical perspective it would be difficult to separate more than two signals in any given band. It is therefore proposed to consider the use of two antennas. This also would have the advantage of being compatible with the recommendations for 802.15.3a. It is proposed that the spectrum be viewed `piece-meal' ­ i.e. by looking at separate contiguous sub-bands in turn. There are several reasons for doing this...

· · ·

It allows processing to be performed at manageable rates There should be a maximum of one interferer in any given considered bandwidth It may be possible to use a bandwidth for each sub-band such that each experiences essentially flat fading.

One issue that must be resolved in this approach is determining whether we have encountered a single wideband signal or (for example) two narrower band signals. The situation is illustrated in Figure 20.

72/03/R/208/U

Page 53 of 69

Unclassified

Unclassified

A OR B C

Frequency

Figure 20: Ambiguous Signals This could be resolved by considering spectral properties of the two bands and/or angle of arrival. The spectral properties could be, for example, periodicities. However this would not be possible if the fundamental pulse repetition frequency is higher than the bandwidth of the sub band. Direction finding can initially be applied separately to each sub-band. Consistent DF readings should confirm that what is being detected is a signal rather than just noise. The principle would be to capture a period of signal in each of a number of wide bandwidths. This should be possible using the same hardware as proposed for the 802.15.3 receiver in Section 8.1.2. In this case a bandwidth of up to 500 MHz could be captured. This data would then be stored for post processing. By using this approach the sub-bands can be matched up across processing operations because the same signal is being used. It is proposed that we could consider only the bandwidth immediately surrounding the centre frequency of the victim receiver. The advantage of this approach is that we are considering the part of the spectrum of the interferer that is `doing the damage' and therefore the interferer is guaranteed to be present in that band (and, presumably, either side of it, if indeed it is a UWB signal). Suppose we down convert 500 MHz of bandwidth around the victim receiver to complex baseband. We can then capture digitised samples of this over a period of time. This captured data can be processed to examine sub-bands, which should be narrow enough to have substantially non-frequency selective fading. It is likely that sub-bands for most UWB signals will have a complex Gaussian envelope. Indeed this assumption is the basis for one of the key arguments supporting the use of UWB over spectrum that is already occupied. If the signal that the victim receiver is trying to receiver is present then this will overlap the relevant portion of the spectrum. Thus, in this part of the spectrum there will be two signals present. For a single antenna system it would not be possible to separate these signals. However, if more than one antenna is provided and if the two signals angle have different angles of arrival into the antennas then it is possible, at least in principle, to separate them. A number of sophisticated techniques are available for separating signals received over multiple antennas in this fashion. The set of methods is generically known as `Blind Signal Separation' or BSS. There are several algorithms available, all of which are based in some way on the examination of higher order statistics (HSS) of the signals as received over the multiple antennas. The most common approach is based on the assumption that the envelope statistics of the (in our case) two signals yield useful information. In fact this is true except where the envelopes of both signals are Rayleigh distributed (i.e. where the complex envelopes of both signals are complex Gaussian). In this case, discrimination is generally accepted to be impossible.

72/03/R/208/U

Page 54 of 69

Unclassified

Unclassified

As discussed earlier, a narrowband portion of a UWB signal is likely to have a complex Gaussian envelope. Thus, for the proposed technique to work we are reliant on the other signal having a noncomplex-Gaussian complex envelope. This may be applicable for some transmissions but increasingly, signal formats are becoming non-constant envelope and with the proliferation of systems based on OFDM (802.11a, DAB, DTV), many do, in fact, have a complex Gaussian complex envelope. There may be an alternative method of obtaining the required weights if the UWB signal is present outside the bandwidth of the signal, which is interfering with it. By examining parts of the UWB spectrum outside of the bandwidth of the interfering signal, the relative gains and phases of the UWB signal on the two antennas may be estimated. These may then be used to cancel the UWB signal within the bandwidth of the interfering signal to allow the same parameters to be extracted for the interfering signal. If the frequency difference is large then parameters may be extracted in more than one sub-band outside the bandwidth of the interfering signal to allow the rate of change of phase to be determined so that the phase difference between antennas at the centre of the bandwidth of the interfering signal may be extrapolated. This is unlikely to be a requirement if the need for the signal and interference to be flat fading is satisfied. In short, it would be possible to detect a sub-set of all possible unknown UWB signals. However, some signal formats would be very noise like over a portion of bandwidth narrow enough to experience flat fading.

72/03/R/208/U

Page 55 of 69

Unclassified

Unclassified

10

EQUIPMENT CONCEPT

A conceptual drawing of the detector equipment is given in Figure 21. The horns form a large proportion of the equipment size. However, the processing unit is required to accommodate internal batteries and fans for cooling the print circuit board(s). The dimensions are approximation and there is some tradeoffs between size and development cost and risk..

125.00mm

Power Level -55 dBm/MHz

100.00mm

125.00mm

125.00mm

250.00mm

125.00mm

125.00mm

250.00mm

Figure 21: Equipment Concept

72/03/R/208/U

60.00mm

190.00mm

Page 56 of 69

Unclassified

Unclassified

Figure 22: Artist Impression of the Detector

11

CONCLUSIONS

A definitive statement on feasibility of detecting signals from 802.15.3a equipment can clearly not be made until the standard has been defined. However based on the current proposal the study has shown that it is likely to be feasible. In particular for the TI led OFDM proposal (which has received the strongest support) the implementation has been examined in detail, suitable key components identified, and processing load calculated. A number of direction finding techniques have been evaluated and all shown to be feasible. From these the recommendation is to use direction finding with a two-horn antenna as this provides the best performance complexity trade-off. The provided direction information will enable an operator to quickly locate the source of an interfering signal. The equipment is expected to consist of hand held dual horn antenna with a cross section of about 5" square, and a processing unit. For the more general case of UWB signal detection the techniques that could be employed and what they can provide has been investigated, however the requirements and techniques need further investigation before detailed implementation complexity can be analysed. For 802.15.3a it is necessary to now wait until the key parameters of the standard are decided, or a dominant proprietary system enters the market. Roke Manor Research can then update the

72/03/R/208/U

Page 57 of 69

Unclassified

Unclassified

implementation approach. In addition a detailed requirements specification should be produced, with inputs from staff that will use the equipment to ensure operational requirements are fully included. The next stage of equipment development can then begin, with algorithm design and simulation, and then detailed circuit design.

72/03/R/208/U

Page 58 of 69

Unclassified

Unclassified

APPENDIX A TEXAS INSTRUMENTS DETECTOR DETAIL

The Texas Instruments proposal is based on frequency hopped OFDM. The physical layer burst consists of a header and a data part. The header part includes a sync component that is not OFDM. Thus reception of the burst comprises detection of the sync component and synchronisation to it followed by demodulation of the header and then demodulation and decoding of the received data.

A.1

QUANTISATION

A simulation has been used to determine the implementation loss for various numbers of bits of quantisation. In each case the optimum level setting has been applied. Results are shown in Figure 23.

0.4

0.35

0.3

Implementation Loss (dB)

0.25

4 Bit 6 Bit

0.2

0.15

0.1

0.05

0 0 1 2 Eb/No (dB) 3 4 5

Figure 23: Quantiser Implementation Loss This indicates that 4-bits would give a loss of 0.25 dB at an Eb/No of 3 dB (NB this corresponds to an Eb/No of 6 dB with a half rate code). Since ADCs are readily available with 6-bits quantisation it makes sense to use these to save this loss. The loss is then only 0.025 dB at 3 dB Eb/No.

A.2

SYNCHRONISATION

Although the TI proposal is based on frequency hopped OFDM, the initial parts of the physical layer burst (sync transmissions) are made using simple BPSK modulation of a carrier centred on the centre of the

72/03/R/208/U

Page 59 of 69

Unclassified

Unclassified

OFDM channel. The sync transmissions are frequency hopped as shown in Slide 23 of [Ref 4]. This shows 24 repeats of the sync transmission, which consists of a hierarchical sequence of length 128 with a cyclic repeat of 32 elements of the code. This mirrors the cyclic prefix applied to the OFDM symbols that are transmitted later. Hierarchical codes are obtained by taking repeats of a short code running at high rate and modulating all the elements of each repeat by an outer code running at a slower rate. They have the advantage that matched filtering involved significantly fewer multiplications. In fact the required number of multiply accumulates (MACs) is the sum of the lengths of the individual codes rather than the length of the overall code. For the TI proposal the inner code length is 8 and the outer code length is 16. Thus the total number of MACs is 24 rather than 128. A possible architecture for the matched filter is shown in Figure 24.

Complex Baseband

8 0 16 24 32 40 48 56 64 72 80 88 96 104 112 120

2

Figure 24: Hierarchical Matched Filter In fact there are four codes, one for each piconet. Depending on the higher layers it may be necessary for terminals to search for all codes. Alternatively, if a terminal is hard wired into a particular piconet then only that piconet's code need be looked for. In either event, the monitor equipment will need to search for all four codes since it cannot know a priori which code will be used by the DFD. In fact there are only two inner codes so this marginally reduces the number of operations from 4 x 24 = 96 to (2 x 8) + (4 x 16) = 82. The left hand filter in Figure 24 is matched to the inner code. The right hand filter is matched to the outer code. However, the used elements from the shift register in the right hand filter are separated by the length of the inner code. The filters are half complex in that there is one filter for the I channel and one for the Q channel. The energy at the output of the filter is computed by squaring the I and the Q results and adding them together. Each multiplication is made by a tap weight, which is an element of the code and which, therefore, is ±1. Thus the multipliers are implemented as conditional additions or subtractions into the accumulator.

72/03/R/208/U

Page 60 of 69

Unclassified

Unclassified

At this point we need to consider whether the received signal can be hard limited so that `sign-only' data is used. In general this leads to a loss in sensitivity of about 2 dB. There is, however, an additional issue. Hard limiting results in a fixed threshold for detection of the signal. If, on the other hand, multiple bit samples are processed then it become necessary to measure the background noise level so that the detection threshold is based on SNR and not absolute signal level. In addition the issue of AGC must be considered. At the start of the synchronisation transmissions the receiver has no knowledge of the level at which the signal will be received. By the end of the synchronisation burst the AGC must have settled to a point suitable for demodulation of OFDM with limited precision quantisation. We can assume, because of variations in antenna and LNA gains along with the inherent pathloss variation with frequency, that different AGC gain settings will be needed for each frequency in the hopping sequence. Thus the AGC will need to settle independently on each frequency. The sync burst is transmitted 8 times on each frequency. This raises the question as to which frequency the receiver should dwell in during idle mode. Because of the sequence of frequencies used for sync transmissions it is only possible to hear all transmissions if the receiver is tuned to the minimum frequency. This is sensible because this frequency should have the highest range capability. Thus the receiver will search on Band#1 until it receives a signal above threshold. The initial setting for AGC will be maximum gain although if there is excess received noise, the gain may be turned down slightly. Thus maximum sensitivity will apply for detection of the first burst. If the transmission is weak then it may not be detected on the first burst. In this th case no gain adjustment will be made. If it takes until the 7 burst before detection occurs there is no time for the AGC to settle. However this is of no consequence since the gain is already correct! On the other hand if the signal is very strong it will hard limit in the ADC, affecting only the most significant bit. Detection will be reliable but the AGC gain will be too high. Because of the limited maximum range (nominally 10 m) of the devices, the required dynamic range of the AGC is not great. The signal will be at its maximum in the near field (/2 separation = 0.03 m at around 4.5 GHz) and the absolute maximum range (in practice) is about 20 m. Thus the dynamic range is 20 log10(20/0.03) = 56 dB. A 6-bit ADC has a theoretical dynamic range in the region of 6 x 6 = 36 dB. In practice the dynamic range over which measurements of power can be made is significantly smaller than this, nearer to 24 dB. Thus, if a very strong signal is received ­ i.e. hard limiting in the ADC, the gain can be reduced in steps of 24 dB until the signal is measurable. In the worst-case 2 steps will take the signal down to 8 dB above a limit range signal (56 ­ 4 x 12 = 8). The signal can then be measured and this can be used to set the final level, for which there remain a further 5 measurements. Thus synchronisation and AGC settling can be performed reliably for a wide dynamic range of signals.

A.3

OFDM DEMODULATOR

There is an important difference between an OFDM receiver for this type of application and one for continuous OFDM transmissions (such as digital broadcast). In the latter case the receiver can lock onto the signal, taking a period of time during which reception of data is not possible. Because of this it is possible to lock the reference clock to the received signal and effectively compensate for phase and timing errors in the analogue domain. For this application all of the data must be received and the response time of the reference oscillator will be far too long to permit this to be performed by pulling its frequency. For this reason, all compensation for phase and timing errors must be performed digitally.

72/03/R/208/U

Page 61 of 69

Unclassified

Unclassified

The architecture has been generated with a view to minimising complexity. In particular, two decisions were taken... 1. Sampling is at the minimum rate of 512 MHz. The use of this frequency involves a small implementation loss due to aliasing. This is estimated at less than 0.4 dB based on a 3-pole Butterworth anti-alias filter. With careful optimisation it should be possible to bring this figure down to 0.25 dB or better. No mathematical operations are performed prior to the FFT. In a conventional all-digital receiver, phase and timing compensation would be applied close to the ADC. Phase compensation would require a complex multiplier and numerically controlled oscillator (NCO) operating at the sample frequency. Timing adjustment would require interpolators. Interpolators are implemented as finite impulse response (FIR) filters where the tap weights are varied as a function of the required timing offset. The required length of the interpolator filter is a function of the accuracy that is needed and of the factor by which the data has been oversampled. In our case, because the data has been sampled only just above the Nyquist rate, the interpolator filters would need to be long. Moreover, the tap weights of the interpolators would need to be updated regularly (probably once per OFDM symbol). This could be done either by computation, which would be intensive or by storing the interpolator tap weight sets for a full range of timing offsets. Either way, considerable complexity is involved. It is possible to perform compensation for phase and timing errors after the FFT. The only problem arises if the uncompensated frequency errors prior to the FFT are large enough to introduce significant inter channel interference (ICI) ­ i.e. interference between the different tones in the OFDM multiplex. For the proposed system the maximum permitted frequency error is ±20 ppm. For our receiver we can afford to use a more accurate reference oscillator. Accuracies as good as ±1 ppm are available from low cost oscillators. However, once ageing is considered, a figure of ±5 ppm is perhaps more realistic, leading to a maximum end-to-end error of ±25 ppm. The highest RF frequency we need to consider is the centre of the highest band: 7920 MHz. Thus the maximum frequency error is 25 x 7920 = 198 kHz. The frequency spacing is 4.125 MHz. Thus 3 6 the relative absolute maximum frequency error is 198x10 /4.125x10 = 0.048. According to [Ref 5], this will led to a worst case ICI attenuation of a little over 21 dB. Because the number of tones is modest, the system is relatively insensitive to effects due to sampling clock frequency errors and these will be negligible compared with the carrier frequency effects. -6 According to [Ref 6] a ½ rate constraint length 7 Convolutional code gives a BER of 10 (giving a better than 95% probability of error free reception for a maximum length frame containing 4095 bytes) at around 5 dB with soft decision decoding. This corresponds to a QPSK symbol SNR also equal to 5 dB, because QPSK provides 2 bits per symbol, which, in turn, provide one data bit from a half rate code. Thus the ICI is 16 dB lower than the noise under this condition. -16/10 ) = 0.11 dB. The dominant This, in turn, leads to a degradation of only 10 log10(1+10 contributions of ICI come from the tones that are closest in frequency to the victim tone. In a multipath fading channel the fading on these will tend to be correlated with the fading on the victim tone so that the ICI will tend to remain not much stronger than the figure of 21 dB below the victim tone. Thus one may expect the degradation to be even lower in the case of fading.

2.

72/03/R/208/U

Page 62 of 69

Unclassified

Unclassified

An example architecture has been developed for the OFDM receiver based on post FFT processing. This is illustrated in Figure 25. This is proprietary to RMR but is included here as a possible approach for evaluation.

Phase & Delay Compensated

2R

R

LOOP FILTER

12C

Channel Compensated

128R

12C

1C

S/P

128C

FFT

128C

128C

× e ji

128C

SELECT PILOTS

12C

REMOVE PILOT MODULATION

*

128C

R

SYMBOL 1 REGISTER

128C

DEINTERLEAVE

1R

DECODE

SYMBOL 2 REGISTER

ESTIMATE PHASE SHIFT

1C

1C

128C

DESCRAMBLE

AVERAGE IN FREQ.

R

REMOVE PILOT MODULATION

128C

R:- This block is repeated for frequency hopping

Figure 25: OFDM Receiver All data paths are illustrated with a single line. However, the lines are annotated according to the number of connections involved. In turn, connections can represent a group of links from 1 to 20 bits wide. The format of the annotation is NR or NC where is N is the number of connections and where R denotes real and C denotes complex. Thus, for example, the lines labelled 128C involve 256 connections. If these connections were individually 8 bits wide then a total of 2048 logical wires would be required for this case. In practice the connections requiring very large numbers of wires would almost certainly be implemented in a serial/parallel configuration. The circuit operates as follows. The outputs of the I and the Q ADCs (6-bits wide each) enter the receiver at the left hand side and are fed into a serial to parallel converter (S/P). This can be implemented as a block of memory where the data is written into sequentially addressed locations. It is assumed that control has been established to load the correct 128 complex samples for the first channel-estimating symbol based on correlation of the sync word. The contents of the S/P memory are sent to an FFT function. For the first channel-estimating symbol this generates symbols embodying initial estimates of the gains for all the frequencies over the channel. These are first captured into the SYMBOL 1 REGISTER and immediately copied into the SYMBOL 2 REGISTER block. When the second channel estimating symbol is received and processed by the FFT it is multiplied, term-by-term by the complex conjugate of the previous response. This is achieved by the multiplier to the right of the FFT block and the contents of the SYMBOL 2 REGISTER passed through the switch (Note the `*' on the lower input to the multiplier denoting complex

72/03/R/208/U

Page 63 of 69

Unclassified

Unclassified

conjugation). The outputs of the multiplier at this stage are exactly what are needed to give an optimal estimate of the carrier phase shift between the first and second OFDM symbols. This gives an indication of the received frequency error. The block labelled ESTIMATE PHASE SHIFT performs this estimation. Since the channel estimates from the first two pilot symbols will be used for reception over the whole of the burst we want them to be as accurate as possible. Accuracy can be improved by using both pilot symbols. However, there will be a phase shift between the two symbols according to the carrier frequency error. Thus the first symbol is rotated in phase by the lower multiplier according to the estimated phase shift to match the second symbol before the two are added together. After this, the fixed phased modulation is removed from the pilot symbols (`REMOVE PILOT MODULATION'). This is done by dividing by the known pilot symbols. Actually division is not needed and multiplication by the complex conjugate of the known pilot symbol is all that is required. This actually amounts to a possible switching between the real and imaginary components and/or a possible inversion. Then the channel estimates are averaged with a suitable filter across the frequency domain to give improved accuracy. The results are stored in the filter output and used for the remainder of the burst and fed to the multiplier at the output of the FFT via the switch, which from now on will be in the right hand position. If the local oscillator and symbol-timing clock were perfectly synchronised, the set of outputs at the point following the initial multiplier (`Channel Compensated') would be suitable for reception. However, phase and frequency tracking is needed for both the local oscillator and the symbol-timing clock and this is done by the remaining circuitry using information derived from the pilot tones. There are twelve pilot tones, with a mutual separation of 10 times the frequency spacing. These are used to provide estimates of phase and time shifts. Firstly these pilots are separated out (`SELECT PILOTS'). Then the pilot modulation is removed (`REMOVE PILOT MODULATION') in the same way as described earlier. Next, the estimates are filtered. This can be done simply using first order digital recursive filters, also known as trackers. The pilot outputs are then fed to the `LOOP FILTER' block, which provides the necessary phase shifts for the next symbol. The outputs of the LOOP FILTER block feed 128 multipliers by

× e ji . These perform a complex multiply by cos() + j.sin(). A lookup table can best implement this ­ a

suitable structure is illustrated in Figure 26.

x

LOOKUP TABLE

x.cos()

x.sin()

Figure 26: Phase Shifter Look Up Table Serial use of one of these plus an adder and subtractor will achieve the required result. The exact precision requirement would be determined by simulation but 6-bits for all inputs and outputs may be adequate, particularly if the sign of x is handled externally and if is reduced to the first quadrant. The (6+6) x (6+6)/8/1024 = 6 KB which is manageable. As an alternative, effective memory size then becomes 2 a sin/cos lookup table could be used in conjunction with a hardware multiplier.

72/03/R/208/U

Page 64 of 69

Unclassified

Unclassified

The LOOP FILTER block is illustrated in Figure 27.

Advance/Retard Serial to Parallel Capture

R:- This block is repeated for frequency hopping

Sample Clock Time Error

* pk pk -1 k =2 12

Decrement

R

ACC

R

1R

ACC

g1

1R

Increment

g2

Type II Loop Filters

63

Local Oscillator Phase Error

62

12C

From Pilot Filters

pk

k =1

12

R

ACC

1R

R

ACC

To Phase Compensation Mixer

g3

1R

-63

g4 Initialise LO Frequency Error

-64

Combine Timing and Phase Corrections

Figure 27: Loop Filter As explained earlier the loop filter needs to compute adjustments to compensate for both carrier phase and for timing errors. The block diagram shown in Figure 27 caters for both of these. Consider first, timing errors. An error in time corresponds to a phase error that is proportional to frequency. Thus a timing error should result in a phase shift between adjacent pilot tones that is constant for each adjacent tone pair. We can thus get an estimate of this phase shift by averaging over the phasors for each tone pair. If the kth * pilot is pk then the phase shift between the kth and k-1th pilot will be pk pk-1 . We can get a weighted average of the phase error by forming

72/03/R/208/U

Page 65 of 69

Unclassified

Unclassified

* pk pk -1 k =2

12

The function is shown in the top left hand of Figure 27. The output is fed to a type II loop filter to produce the current value for the required adjustment of phase between adjacent tone pairs (not adjacent pilot pairs). The output feeds into the blocks labelled `Combine Timing and Phase Errors'. Because the required correction for timing error is proportional to frequency, the phase shift for each tone is multiplied by the index corresponding to the tone (which is that tone's relative frequency). It should be noted that timing adjustments made by phase shift ramps at the output of the FFT do not correspond to absolute changes in time ­ rather they implement cyclic rotations of the processed data. Because of the cyclic prefix on the OFDM symbols, these remain equivalent as long as the absolute timing does not drift either backwards into the multipath from the previous symbol or forwards into the beginning of the next symbol. The circuit limits the shift that can be implemented by cyclic rotation to ± 0.5 samples. When the output of the phase shift accumulator (top right hand) goes outside the permitted range, one or other of the comparators' outputs will go high. This will cause a phase shift corresponding to a delay of either plus or minus one sample to be subtracted from the accumulator. At the same time the output from the relevant comparator feeds across to the serial to parallel converter (S/P ­ see Figure 25) to implement a compensating sample shift for the next OFDM symbol. The overall phase error is measured as

pk

k =1

12

Again this is processed by a type II loop filter. The first accumulator in this filter holds the effective frequency error. At the start of reception of the burst this accumulator is initialised with the output of the ESTIMATE PHASE SHIFT block (see Figure 25). The output of the overall filter is the current required overall compensating phase shift. This must be added to all of the tones. The adders in the blocks labelled `Combine Timing and Phase Errors' implement this function. It should be noted that the two elements of timing and phase error measurements are orthogonal. Thus the phase errors do not affect timing error measurements and timing errors do not affect phase error measurements. Actually there is a second order effect in that a large timing error would cause the average over pk to fall. However in steady conditions, all errors should be close to zero so this condition will not arise in practice. The output of the phase compensator block consists of phase compensated I and Q data. Because of the multiplication by the channel estimates, this data is suitably weighted for soft decision Viterbi decoding. The data is fed to a de-interleaver and thereafter is Viterbi decoded to provide the scrambled data, which is then de-scrambled to provide the required data. Note that detailed control issues have been omitted from the above description. In practice the header is separately encoded and this provides control information, such as the code rate, the data burst length and the scrambler code to allow the remainder of the burst to be decoded.

72/03/R/208/U

Page 66 of 69

Unclassified

Unclassified

A.4

FREQUENCY HOPPING IMPLICATIONS

For simplicity of explanation the above description has been given without reference to frequency hopping (FH) and its implications. When FH is included, much of the architecture remains unaltered but there will be some important changes. The channel estimates will be different for each frequency, as will the phase and timing errors. In fact the blocks in Figure 25 and Figure 27 labelled R will need to be replicated to the number of frequencies used (3 or 7). Returning to Figure 25, consider Mode 1 operation in which hopping takes place over 3 frequencies (the extension to 7 frequencies is trivial). The first pilot symbol will be received on frequency 1. This will be stored, after passing through the left hand register into the right hand register corresponding to that frequency. The operation is then repeated for the first pilot symbols on the other two frequencies. The three replicated right hand registers will then each contain the pilot symbols for their applicable frequency. The second pilot symbols are then received and processed in turn, being combined with the first pilot symbol and averaged in frequency into the applicable AVERAGE IN FREQ. register. As the receiver frequency hops thereafter the relevant AVERAGE IN FREQ. register output is connected to the switch. Much of the processing continues as normal from this point on. The pilot filters are frequency-specific and must be switched accordingly. Now refer to Figure 27. The accumulators in the loop filters must be frequency-specific. Different piconets have different time-frequency codes. Some are very simple in that the time gap between visiting the same frequency is constant (e.g. 1 2 3 1 2 3). Others are more complicated in that the time gap could be small or large (e.g. 1 1 3 3 2 2). When relating the phase shift from one symbol on a given frequency to the next, it is necessary to scale according to the time gap so that the correct interpretation of frequency is made. This should be a trivial operation. It may be desirable, although not absolutely essential, to perform related compensation tasks within the pilot filters.

A.5

OVERALL IMPLEMENTATION LOSS

The detection of the sync word is shown in [Ref 4] not to dominate the performance in terms of reception. Other sources of implementation loss considered are... · · · Quantisation ­ We have seen that with 6-bits the effect of this is negligible (0.025 dB) Aliasing ­ This will be less than 0.25 dB. Channel Estimation ­ This will depend on how much filtering is possible in the frequency domain to improve the estimate accuracy. This in turn will depend on the power delay profile of the multipath in the channel. Based on the size of the guard interval, it is reasonable to suppose that it should be possible to gain about 7 dB improvement from the frequency domain filtering. Given the two pilot symbols we should get a total of 10 dB enhancement of the channel estimate signal to noise ratio. Simple simulations indicate a loss of 0.55 dB under this condition. Frequency error effects ­ It has been shown that correcting frequency after the FFT results in loss of up to about 0.11 dB Finite precision arithmetic ­ It will be desirable to limit the precision of the arithmetic in order to have a practical implementation. The degree of implementation loss is entirely under the control

·

·

72/03/R/208/U

Page 67 of 69

Unclassified

Unclassified

of the designer. Experience suggests that a figure of 0.2 dB represents a good compromise between complexity and performance. The total implementation loss is then 0.55 + 0.25 + 0.025 + 0.11 + 0.2 = 1.135 dB. Thus the implementation loss will be of the order of 1 dB, which compares favourably with the 2.5 to 3.0 dB quoted in [Ref 4].

72/03/R/208/U

Page 68 of 69

Unclassified

Unclassified

72/03/R/208/U

Page 69 of 69

Unclassified

Unclassified

Roke Manor Research Limited Roke Manor, Romsey Hampshire, SO51 0ZN, UK Tel: +44 (0)1794 833000 Fax: +44 (0)1794 833433 Email: [email protected] Web: http://www.roke.co.uk

Unclassified

Information

Report / Proposal template

72 pages

Report File (DMCA)

Our content is added by our users. We aim to remove reported files within 1 working day. Please use this link to notify us:

Report this file as copyright or inappropriate

703261


Notice: fwrite(): send of 201 bytes failed with errno=104 Connection reset by peer in /home/readbag.com/web/sphinxapi.php on line 531