Eleftherios Kofidis - Channel Estimation in Filter Bank-based Multicarrier Systems: Fundamentals and Recent Advances

Transcript

1. Channel Estimation in
   Filter Bank-based Multicarrier Systems:
   Fundamentals and Recent Advances
   Eleftherios Kofidis
   Computer Technology Institute, Greece
   University of Piraeus, Greece
   

   View Slide

2. 31 Aug. 2015 CentraleSupelec, Rennes 2
   Future mobile networks – Vision and
   needs
    High
    data rate
    reliability
    QoS
   in demanding transmission scenarios
    Increased flexibility
    Efficient use of fragmented spectrum
    Robustness to asynchronism
    Co-existence of different systems (HetNets)
    …
   

   View Slide

3. 31 Aug. 2015 CentraleSupelec, Rennes 3
   Toward a new PHY – Modulation
    Is OFDM an adequate solution?
    Poor spectral containment
    Bandwidth/power inefficiency
    Challenging synch in multi-access
    Sensitivity to severe dispersions
    …
    FBMC: an attractive alternative
    Good spectral (/time) containment
    High spectral (/power) efficiency
    Flexibility (e.g., for multi-mode comms)
    Relaxed synch requirements
    Able to cope with severe multipath (e.g., large cells) and
   high mobility
    …
   

   View Slide

4. 31 Aug. 2015 CentraleSupelec, Rennes 4
   FBMC research and applications
   Filter bank-based multi-carrier modulation:
   • FBMC/OQAM
   • FMT
   • GFDM
   • UFMC
   • …
   

   View Slide

5. 31 Aug. 2015 CentraleSupelec, Rennes 5
   FBMC research and applications
   Filter bank-based multi-carrier modulation:
   • FBMC/OQAM
   • FMT
   • GFDM
   • UFMC
   • …
   • Max. spectral efficiency
   • Time-freq. localization
   • Robust to lack of synch
   • But: Intrinsic interference
   

   View Slide

6. 31 Aug. 2015 CentraleSupelec, Rennes 6
   FBMC/OQAM challenges - Solutions
    Intrinsic ISI/ICI
    Frequency / time selective subchannels
    Challenges in Channel Estimation (CE)
    Classical assumption: channel of low freq./time
   selectivity  CE analogous (similar) to OFDM
    Preamble/pilots design for increased accuracy
    However: in many realistic scenarios  Severe
   performance error floors  outperformed by
   OFDM at higher SNRs
    More recently: CE training and techniques for
   demanding channels
   

   View Slide

7. 31 Aug. 2015 CentraleSupelec, Rennes 7
   Outline
    Fundamentals of FBMC/OQAM
    System model
    Intrinsic interference effect
    FBMC/OQAM CE fundamentals
    Preamble-based
    Pilot-based
    Preamble-based CE
    Low frequency selective channels
    Highly frequency selective channels
    Simulation examples
    Additional results - on-going/future work
   

   View Slide

8. Fundamentals of FBMC/OQAM
   31 Aug. 2015 CentraleSupelec, Rennes 8
   

   View Slide

9. 31 Aug. 2015 CentraleSupelec, Rennes 9
   FBMC/OQAM vs. OFDM/QAM
    
   1 2
   F
    
   complex QAM
   real
   imaginary
   F=1/T: sub-carrier spacing
   T: OFDM/QAM symbol
   duration
   T-F density:
   OFDM/QAM (without CP): 1/(TF)=1
   OFDM/OQAM:
   Spectral efficiency (e.g., (O)QPSK):
   OFDM/QAM (without CP): 2/(TF)=2
   OFDM/OQAM:
    
   1 2
   F
    
    
   1/ 2
   F
    
   Phase space
   

   View Slide

10. 31 Aug. 2015 CentraleSupelec, Rennes 10
    Offset-QAM Modulation (staggering)
    Re
    Im
    2
    2 z-1
    +
    d2k,n
    c2k,m
    Im
    Re
    2
    2 z-1
    +
    d2k+1,n
    c2k+1,m
    even sub-carriers
    odd sub-carriers
    

    View Slide

11. 31 Aug. 2015 CentraleSupelec, Rennes 11
    FBMC/OQAM Transmitter
    IFFT
    2
    0
    ( )
    A z
    2
    1
    ( )
    A z
    2
    1
    ( )
    M
    A z
    
    2
    M
    
    2
    M
    
    2
    M
    
    
    
    1
    
    z
    1
    
    z
    
     
     
    
    0,n
    
    
    0,n
    
    
    1,n
    
    
    1,
    M n
    
    
    
    1,
    M n
    
    
    
    1,n
    
    0,n
    d
    1,n
    d
    1,
    M n
    d
    
    C2R
    C2R
    C2R
    OQAM modulation Transform block
    Polyphase
    filtering
    P/S
    conversion
    SFB:
    P. Siohan et al., “Analysis and design of OFDM/OQAM systems based on filterbank theory,”
    IEEE Trans. SP, May 2002.
    

    View Slide

12. 31 Aug. 2015 CentraleSupelec, Rennes 12
    FBMC/OQAM Receiver
    2
    0
    ( )
    B z
    2
    1
    ( )
    B z
    2
    1
    ( )
    M
    B z
    
    FFT
    1
    
    z
    1
    
    z
    2
    M
    
    2
    M
    
    2
    M
    
      
    
    
    
    Subchannel
    processing
    Subchannel
    processing
    Subchannel
    processing
    *
    0,n
    
    *
    1,n
    
    *
    1,
    M n
    
    
    
    *
    0,n
    
    
    *
    1,n
    
    Re
    
    *
    1,
    M n
    
    
    0,n
    d
    1,n
    d
    1,
    M n
    d
    
    Re
    Re
     
    R2C
    R2C
    R2C
    S/P
    conversion
    Polyphase
    filtering
    Transform block OQAM demodulation
    AFB:
    

    View Slide

13. System model
    31 Aug. 2015 CentraleSupelec, Rennes 13
    

    View Slide

14. 31 Aug. 2015 CentraleSupelec, Rennes 14
    System model (1)
     M: #subcarriers
     K: overlapping factor
     g: prototype filter (length )
    C2R SFB h + AFB
    Intrinsic interference:
    

    View Slide

15. 13 March 2014 Patras (ENDECON) 15
    Intrinsic interference in FBMC/OQAM
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    ,
    1
    ,
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    0
    ,
    0
    1
    ,
    0
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    n
    M
    n
    M
    n
    M
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    n
    n
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    
    
    
    
    
    
    

    View Slide

16. 13 March 2014 Patras (ENDECON) 16
    Intrinsic interference in FBMC/OQAM
    With good TF localization,
    contributions to intrinsic
    interference only come from the
    first-order neighboring TF
    points
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    ,
    1
    ,
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    0
    ,
    0
    1
    ,
    0
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    n
    M
    n
    M
    n
    M
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    n
    n
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    
    
    
    
    
    
    

    View Slide

17. 13 March 2014 Patras (ENDECON) 17
    Intrinsic interference in FBMC/OQAM
    With good TF localization,
    contributions to intrinsic
    interference only come from the
    first-order neighboring TF
    points
    
    
    
    
    
    
    
     
    
    
     
    
      
     
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    ,
    1
    ,
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    0
    ,
    0
    1
    ,
    0
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    n
    M
    n
    M
    n
    M
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    n
    n
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    
    
    
    
    
    
    

    View Slide

18. 13 March 2014 Patras (ENDECON) 18
    Example – “PHYDYAS filter”
    FBMC/OQAM TMUX transfer function (interference function):
    ( - Even k
    - after “de-phasing” ( ) to bring into the form
    - before that: green real, brown  imaginary  OQAM ! )
    time
    freq.
    n-4 n-3 n-2 n-1 n n+1 n+2 n+3 n+4
    k-1 j0.005 -j 0.043 j0.125 -j0.206 j0.239 -j 0.206 j0.125 -j0.043 j0.005
    k 0 j0.067 0 j0.5644 1 -j0.5644 0 -j0.067 0
    k+1 -j0.005 -j0.043 -j0.125 -j 0.206 - j0.239 -j0.206 -j0.125 -j 0.043 -j0.005
     
    *
    ,
    k n
    k n
    j
      
      , ,
    ,
    k n k n
    d ju k
     
    • N. J. Fliege, “DFT polyphase transmultiplexer filter banks with effective reconstruction,” EUSIPCO 1992.
    • C. S. Lee and K. Y. Yoo, “Polyphase filter-based OFDM transmission system,” VTC-2004 (Fall).
    

    View Slide

19. 13 March 2014 Patras (ENDECON) 19
    More examples
    IOTA filter
    Bregović-Saramäki filter
    P. Siohan and C. Roche, IEEE Trans. SP, Dec. 2000.
    M. G. Bellanger, ICASSP-2001.
    R. Bregović and T. Saramäki, IEEE Trans. SP, Aug. 2005
    PHYDYAS filter
    

    View Slide

20. 31 Aug. 2015 CentraleSupelec, Rennes 20
    System model (2)
     Common assumptions (locally freq./time-invariant channel):
    

    View Slide

21. 31 Aug. 2015 CentraleSupelec, Rennes 21
    System model (2)
     Common assumptions (locally freq./time-invariant channel):
    

    View Slide

22. 31 Aug. 2015 CentraleSupelec, Rennes 22
    System model (2)
     Common assumptions (locally freq./time-invariant channel):
    

    View Slide

23. 31 Aug. 2015 CentraleSupelec, Rennes 23
    System model (2)
     Common assumptions (locally freq./time-invariant channel):
    OFDM-like
    

    View Slide

24. 31 Aug. 2015 CentraleSupelec, Rennes 24
    System model (2)
     Common assumptions (locally freq./time-invariant channel):
    OFDM-like
    colored
    virtual Tx symbol
    (pseudo-symbol)
    

    View Slide

25. FBMC/OQAM channel estimation:
    Fundamentals
    31 Aug. 2015 CentraleSupelec, Rennes 25
    

    View Slide

26. 31 Aug. 2015 CentraleSupelec, Rennes 26
    Preamble-based channel estimation (1)
    Control / Data
    Preamble
    Frame: SFB
    non-zero part
    0 0
    prevents interference
    from previous frame
    (often unnecessary!)
    prevents interference
    from control/data
    channel time invariant
    

    View Slide

27. 31 Aug. 2015 CentraleSupelec, Rennes 27
    Preamble-based channel estimation (2)
    Control/Data
    Preamble
    Full
    (block-type):
    Control/Data
    Sparse
    (comb-type):
    0
    0
    protect from ICI
    

    View Slide

28. Scattered pilot-based channel estimation
     Help (auxiliary) pilot
    31 Aug. 2015 CentraleSupelec, Rennes 28
    J.-P. Javaudin, D. Lacroix, and A. Rouxel, VTC-2003 (Spring).
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    ,
    1
    ,
    1
    ,
    1
    ,
    1
    1
    ,
    1
    1
    ,
    0
    ,
    0
    1
    ,
    0
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    n
    M
    n
    M
    n
    M
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    k
    n
    n
    n
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    d
    
    
    
    
    
    
    

    View Slide

29. Preamble-based channel estimation
    31 Aug. 2015 CentraleSupelec, Rennes 29
    

    View Slide

30. Interference Approximation Method (IAM):
    Interference in a positive role!
     Known input  interference approximation possible 
    pseudo-pilots
     Choose input so as to maximize pseudo-pilot magnitude
     Compute channel estimate (as in OFDM):
    31 Aug. 2015 CentraleSupelec, Rennes 30
    0,0 0,1 0,2
    1,0 1,1 1,2
    2,0 2,1 2,2
    1,0 1,1 1,2
    M M M
    d d d
    d d d
    d d d
    d d d
      
    C. Lélé et al., “Channel estimation methods for preamble-based OFDM/OQAM modulations,”
    European.Trans. Telecomm., 2008.
    estimation error
    

    View Slide

31. Example: IAM-R
     Null side symbols ( base design on middle
    symbol only)
     Carefully choose signs so as to maximize
    pseudo-pilots’ magnitude
    31 Aug. 2015 CentraleSupelec, Rennes 31
    0
    1
    0
    0
    1
    0
    0
    1
    0
    0
    1
    0
    0
    1
    0
    0
    1
    0
    0
    1
    0
    0
    1
    0
    
    
    
    
    31
    
    d
    d
    
    d
    Idea:
    Example:
    M=8, OQPSK
    

    View Slide

32. More IAM variants – Using imaginary pilots
     Idea: Use Imaginary pilots to generate imaginary- or real-valued
    pseudo-pilots (of even larger magnitude)
     Not a strictly OQAM input!
    31 Aug. 2015 CentraleSupelec, Rennes 32
    • C. Lélé et al., ICC-2008.
    • J. Du and S. Signell, ICC-2009.
    • PHYDYAS deliverable D3.1
    • E. Kofidis and D. Katselis, EUSIPCO-2011.
    0
    0
    0
    0
    0
    0
    0
    0
    0
    0
    0
    0
    0
    0
    0
    0
    0
    0
    1
    1
    1
    0
    0
    0
    jd
    d
    d
    jd
    d
    d
    jd
    d
    
    
    
    
    0
    0
    0
    1
    0
    0
    0
    0
    1
    0
    0
    0
    0
    1
    0
    0
    0
    0
    1
    0
    j
    j
    j
    j
    
    
    
    
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    
    
    
    
    
    
    
    
    
    
    
    
    j
    j
    j
    j
    j
    j
    j
    j
    j
    j
    j
    j
    IAM-I IAM-C E-IAM-C
    1/3 of the subcarriers:
    

    View Slide

33. Price for good performance: high PAPR!
    31 Aug. 2015 CentraleSupelec, Rennes 33
    SFB-modulated preambles (magnitudes squared)
    Sample no.
    • M=256, K=4
    • OQPSK
    Interf. from data part
    

    View Slide

34. Optimal preambles (1)
     Preamble optimization:
    Minimize MSE subject to transmit power/energy constraint
     For low frequency selective channels:
     FBMC/OQAM
     Block-type: equal pilot tones
     Comb-type: equispaced & equipowered
     OFDM/QAM (no account for CP energy):
     Block-type: DFT matrix column
     Comb-type: equispaced & equipowered
    31 Aug. 2015 CentraleSupelec, Rennes 34
    • D. Katselis et al., IEEE Trans. SP, May 2010.
    • E. Kofidis et al., Signal Processing, July 2013.
    • C. Mavrokefalidis et al., EURASIP JASP, May 2014 (for relaying networks).
    

    View Slide

35. 31 Aug. 2015 CentraleSupelec, Rennes 35
    Highly frequency selective channels
     No simplifying assumptions:
    D. Kong et al., IEEE TSP, Jan. 2014
    E. Kofidis, ICASSP-2014.
    

    View Slide

36. 31 Aug. 2015 CentraleSupelec, Rennes 36
    Optimal preambles (2)
     Optimization problem:
     Problem structure:
    E. Kofidis, ICASSP-2014
    

    View Slide

37. 31 Aug. 2015 CentraleSupelec, Rennes 37
    Optimal preambles (3)
     Block-type preamble:
     Complex-valued:
     Real-valued:
     Simple estimation procedure (for real preamble):
     Take the first terms of
     Divide them by
    

    View Slide

38. 31 Aug. 2015 CentraleSupelec, Rennes 38
    Optimal preambles (4)
     Comb-type preamble ( pilot tones):
     Equipowered and equispaced
     Estimation procedure:
     Prototype filter autocorrelation:
     Compute the “weighted” freq. response first:
     Compute the “weighted” impulse response via IFFT and
    divide by the weights to arrive at the impulse response
    estimate:
    E. Kofidis, ISCCSP-2014
    

    View Slide

39. 31 Aug. 2015 CentraleSupelec, Rennes 39
    Simulation example: Block-type
    error floor
    ?
    

    View Slide

40. 31 Aug. 2015 CentraleSupelec, Rennes 40
    Simulation example: Comb-type
    ITU-VehA channel model error floor
    

    View Slide

41. 31 Aug. 2015 CentraleSupelec, Rennes 41
    More and on-going
     Preamble-based CE:
     POP etc. [1,3]
     MIMO case [2,3,4]
     Multiuser case [7]
     Longer preambles [5,8]
     LMMSE channel estimation [10]
     Scattered pilot-based CE:
     Extend help pilot idea to highly selective channels
     Take into account
     virtual (edge) subcarriers [6]
     interference from data [6]
    1. C. Lélé et al., EW-2007.
    2. E. Kofidis and D. Katselis, ICSIPA-2011.
    3. E. Kofidis et al., Signal Process., July 2013.
    4. E. Kofidis, EW-2015.
    5. M. Newinger et al., VTC-2013 (Spring).
    6. L. Baltar et al., EUSIPCO-2014.
    7. F. Rottenberg et al., ISWCS-2015.
    8. E. Kofidis, ISWCS-2015.
    9. EMPhAtiC deliverable D3.1
    10. L. Caro et al., VTC-2015 (Spring).
    …
    

    View Slide

42. Thank you!
    Questions?
    

    View Slide