Multimode Terminal - Synchronisation

1
November 2005 Synchronisation for Multimode
Terminals Prof. Steve McLaughlin University of
Bristol Dr Chris Williams University of Bristol

2
Overview

• Motivation
• Channels
• Review of OFDM synchronisation
• Robust timing synchronisation in multipath and
  single frequency networks
• Performance of timing estimators
• Timing variance reduction
• Multiple antennas for synchronisation
• Increased mobility results
• Conclusions and future directions

3
Motivation

• Synchronisation for OFDM (multimode)
• Enhance mobility
• Particularly for current broadcast standards
• Robust in multipath environments
• and single frequency networks
• Signals from different transmitters arrive in
  clusters
• Efficient data transmission
• Reduction of the required guard time for OFDM
• Focus on processing that is common to the
  different standards
• and make it more efficient / less complex

4
The Channel
5
The Channel - in time

• Two classes of channel
• Single transmitter
• Multiple (on channel) transmitter single
  frequency network for broadcast (OFDM)
• Model SFN with independent multipath clusters,
  with relative delay and power as parameters
• For SFN effective delay spread a function of
  transmitter spacing as well as the environment
• Potentially, long effective delay spreads a
  problem
• Clustering also appears in the spatial domain

6
Cluster Statistics

• Experimental evidence for multipath clustering,
  even with single transmitter
• But typically less than 3 or 4 clusters
• Urban SIMO trials in Bristol
• Some clustering evident
• Can this be exploited?
• Multipath clusters may not be separable in time

7
Spatial Characteristics

• Evenly select 12 channels from one measurement
  run
• Search for 1,2 or 3 beams to find maximum
  energy collected related to beam width (5º grid)
• e.g. 2 beams of 90 degrees loses less than 1dB
• Limited loss for coarser search grid
• Doppler spread characteristics related to cluster
  parameters

1 beam 2 beams 3 beams
8
OFDM Synchronisation
9
Timing Synchronisation

• Positive timing error introduces ISI and ICI, so
  much less tolerance.
• Pre-FFT coarse timing correction
• Timing offset induces phase offset given by
  f2pkD/N (k-carrier index, D-time offset, N-FFT
  size).
• Negative timing error tolerable up to Nyquist
  limit (density of pilots, 1 in 12).

10
Frequency Synchronisation

• Introduces ICI
• More critical for OFDM
• Integer and fractional parts
• Pre-FFT correction for fractional part (NDA)
• Post-FFT correction for integer part (NDA/DA)

11
Pre-FFT Synchronisation

• Use structural features
• Guard band (frequency)
• Guard interval/cyclic prefix (CP)
• Imposed structure (repeated symbol), e.g. WLAN

12
Pre-FFT Synchronisation Methods

• Beek
• ML in AWGN
• Correlation between repeated cyclic prefix
• Time and frequency estimate
• Simplify energy correction term

13
Basic Correlation Technique

• MLF derived for AWGN channel by Beek
• DVB-T N2048, G512
• Focus on timing estimate error

Tg
AWGN
Multipath A0.4
Multipath A1.5
14
Derivative Based Methods

• Other techniques too dependent on the actual
  channel characteristics
• Derivative of MLE is maximum near first peak, and
  has edge shortly after (or negative going zero
  crossing of 2nd-Derivative)

Derivative
Linear projection (opt)
Smoothed derivative
15
Timing Estimate Performance
16
Simulation Parameters

• DVB-T System, 2k mode
• Pilot structure coding (RS convolutional)
• Short cyclic prefix
• 64 samples (1/32 useful symbol)
• Model (LOS) proposed by Bug
• Less impact of the equaliser
• Two multipath clusters (2 SFN Tx)
• Estimate filters 15pt median, 16 pt averaging
  FIR
• No rules based processing
• See deliverables and ICR for NLOS short CP, and
  long CP results

17
Performance Eb/N0

• SFN power0dB, SFN delay31 samples

18
Performance - SFN delay

• SFN relative power 0dB, Eb/N020dB

Maximum multipath delay exceeds guard interval
19
System Level Performance

• Run performance simulations
• Equaliser can have a large impact on the results

20
Benefits

• For the same CP length, longer multipath delay
  spreads can be tolerated without the system
  becoming synchronisation limited.
• In broadcast scenarios, this would allow
  transmitters to be place further apart, reducing
  infrastructure costs or giving more flexibility
  in transmitter positioning.
• For new air interface designs, a shorter CP may
  be used from the view of synchronisation, hence
  improving spectrum efficiency.
• Derivative method is applicable to repeated
  symbol preambles for summing over half the
  preamble length, and for OFDMA.

21
Improving Performance
22
Reducing Estimate Variance

• Some estimates have large error
• particularly for short cyclic prefix
• Have used longer median and FIR filters
• Possible to use knowledge of the correlation and
  derivative peaks to bound the estimates
• Correlation peak within CP
• 1. Start of symbol before peak
• 2. Start of symbol after peak position minus CP
• 3. Start of symbol after derivative peak

23
Simulation Parameters

• Bug UN2 (NLOS) channel, DVB-T 2k mode
• Estimate filters (per symbol)
• Short 5pt median, 8pt FIR
• Long 15pt median, 16pt FIR
• Approach with estimate outside bounds
• Hard limit
• Replace previous pre-filter estimate
• Replace previous post-filter estimate

24
Application of Rules
All 3 rules are used consistently
Hard limit
No Rules
Replace estimate with previous pre-filter value
Replace estimate with previous post-filter value
25
Rules SFN delay

• Suppression of variance increase when multipath
  delay exceeds CP length
• Little loss in performance with short filter

26
Frequency Estimation and Mobility
27
Mobility Limitations

• In environments with multipath clusters spatially
  separated, it may be possible to increase
  mobility by synchronisation to each cluster
• Time frequency

28
Channel Considerations

• On the premise spatial clusters exist
• Each cluster will have Doppler offset
• Doppler spread proportional to angular spread
• Greater cluster angular separation in this model
  implies larger offset differences ( v.v.)
• No great angular discrimination required (3-4
  antennas OK)
• Assumptions weak with local scattering but
  unlikely to be travelling fast
• Clusters may have time separation, but not always
• For different transmitters (SFN), each will have
  independent carrier offset
• Spatial discrimination seems the best way forward

29
The Process

• Separate signal into clusters
• Estimate frequency ( time) offset for each
  cluster
• Correct each for frequency offset
• Combine (weighted?) pass to FFT
• Timing correction before or after combination?
• Before N estimates, each signal individually
  corrected
• Potential to reduce delay spread easier
  equalisation or reduced CP length, etc.
• After Combine N estimates, from previous
  discussion need to choose the earliest one (if
  branch power exceeds a threshold)

30
Multiple Antenna Processing

• How to separate clusters?
• Could do DoA estimation then signal separation
• For small terminals may make more sense to have
  directional elements (on 4/6 edges) process
  each non-adaptively
• More antennas (directional) more Doppler spread
  reduction (but beware!)
• In a multimode terminal use MIMO capability (same
  frequency band?), even if not MIMO processing
• With MIMO/diversity processing can still do
  frequency/time correction prior to FFT (1 for
  each channel)

31
The Model

• AGC on each antenna (equal SNR)
• Common timing correction (earliest)
• Power weighted signal combining
• Sectored antennas
• Antennas are co-located, so limited additional
  diversity gain

32
Performance for different Doppler

• Opposed signals can be separated, allowing higher
  Doppler shifts
• 2 equal power clusters, angles 20?, -160?,
  angular spread 45?, Bug UN2 (Eb/N016dB)

33
Conclusions

• Method for improved timing synchronisation
  (patent application filed)
• New derivative based method outperforms the peak
  detection method
• System performance limited by equaliser
• Rules processing reduces variance and may allow
  shorter estimation filters
• Proposed to use multiple antennas to improve
  mobility
• Benefits demonstrated
• Performance degraded when a cluster is split
  between branches
• Controlling the directionality and beam width
  would help
• Real channels to be investigated

34
Further Reading

• D-WE2.1.1/2.3.1 Architectures, Link Enhancement
  and Synchronisation Techniques for Multimode
  Baseband Terminals
• Part 2 on fundamentals of synchronisation, and
  review of synchronisation for OFDM
• D-WE2.3.4 Synchronisation for multimode terminals
• Comparison of pre-FFT synchronisation methods,
  including first derivative method
• ICR-WE2.3.1 Enhancements to Synchronisation for
  OFDM
• Rules-based enhancements, and second derivative
  comparison
• Robust OFDM timing synchronisation in multipath
  channels, submitted to IEEE Trans. Veh. Tech
• Robust OFDM timing synchronisation, Elect.
  Letts., Vol. 41, No. 13, pp. 751-752, 23 June
  2005
• Synchronisation in a receiver , patent
  application GB0419399.1

35
Thank you ! For further information
please contact Dr Chris Williams E-mail
chris.williams_at_bristol.ac.uk Tel [phone]
5049
36
The Model

• Channel paths processed separately
• Each cluster has mean angular offset spread
• Each cluster has Bug power delay profile
• Angular distribution of paths is Laplacian
• Path angles change randomly (av. Once every 20
  sym)
• Each path has classical Doppler spectrum
• Doppler spread is proportional to angular spread
  (all paths)
• Doppler offset scaled according to angle of
  arrival (ea. Path)