Docstoc

stanford

Document Sample
stanford Powered By Docstoc
					    KU Leuven Department of Electrical Engineering                                      1/6/2002




            Reduced Complexity MUD and
           Multi-user Power Allocation in DSL

                          Raphael Cendrillon and Marc Moonen
                    SISTA Research Group, Electrical Engineering Dept.
                                 K.U. Leuven, Belgium

                In co-operation with Katleen Van Acker and Piet Vandaele
                                 Alcatel Antwerp, Belgium



Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        1
    KU Leuven Department of Electrical Engineering                                      1/6/2002




                                           Contents

    • VDSL
          – Reduced Complexity MUD (Partial Co-ordination)
          – Power Allocation in Partially Co-ordinated Environments
    • ADSL
          – Per-tone Equalisation (PT-EQ)
          – Per-tone Echo Cancellation (PT-EC)
          – Other PT-EQ work:
                • Windowing
                • Doubletalk Suppresion
                • Adaptive Implementations
    • Open Questions and Current Work



Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        2
    KU Leuven Department of Electrical Engineering                                      1/6/2002




                                   VDSL Research


    •   Reduced complexity MUD in VDSL
          – 90% performance gains of conventional MUD (e.g. vectoring)
          – 38% complexity
    •   Multi-user Power Allocation in Partially Co-ordinated Scenarios
          –   Reduced complexity MUD uses a form of ‘Partial Co-ordination’
          –   Gives best performance when combined with Multi-user Power Allocation
          –   Power Allocation must perform well with or without MUD
          –   That is: in mixtures of Interference Channels (IC) and
              Multi-Access Channels (MAC)




Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        3
    KU Leuven Department of Electrical Engineering                                      1/6/2002




        Power Allocation and Partial Co-ordination

                 Co-ordinated
                         RXs




               Un co-ordinated
                          RXs




Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        4
    KU Leuven Department of Electrical Engineering                                                            1/6/2002




        Power Allocation and Partial Co-ordination

                 Co-ordinated
                         RXs




               Un co-ordinated
                          RXs


  MAC
        – Optimal power allocation known (David Tse)
        – Simultaneous (Iterative) Vector Waterfilling
                                                                                                       
                                                                                                  
                         Si ( f )  1                              1                             
                                           i      
                                     i h H (k ) Z(k ) 
                                                          j i h j (k )h H (k )S j (k ) h i (k ) 
                                                                           j            1
                                                                                                   


Reduced Complexity MUD and Multi-user Power Allocation in DSL                         R. Cendrillon and M. Moonen        5
    KU Leuven Department of Electrical Engineering                                                          1/6/2002




        Power Allocation and Partial Co-ordination

                 Co-ordinated
                         RXs




               Unco-ordinated
                         RXs


  Interference Channel
        – Good distributed solution proposed (Wei Yu)
        – Iterative Scalar Waterfilling (Optimal in Scalar MAC)
                                                                                         
                                                                                    
                                               1  i (k )   j i hij (k ) S j (k ) 
                                                    2                      2

                                  Si ( f )   
                                              i            hii (k )
                                                                      2
                                                                                     
                                                                                    


Reduced Complexity MUD and Multi-user Power Allocation in DSL                       R. Cendrillon and M. Moonen        6
    KU Leuven Department of Electrical Engineering                                      1/6/2002




        Power Allocation and Partial Co-ordination
    •   Iterative Scalar Waterfilling
          – Good solution in Interference Channel
          – Quite sub-optimal when co-ordination used at RX (e.g. MUD)

                                                                               MAC
                                                                               Rate Region




                                                                                 It. Scalar
                                                                                 Waterfilling




Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        7
    KU Leuven Department of Electrical Engineering                                      1/6/2002




        Power Allocation and Partial Co-ordination
    •   Need good performance in both Interference and MACs
    •   Alternative Scheme proposed
    •   Simple (no constrained optimisations)
    •   Requires centralized control of in-domain lines
    •   Autonomous control not required in our application
    •   Partial co-ordination arises from
         – reduced complexity MUD
         – not unbundling




Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        8
    KU Leuven Department of Electrical Engineering                                      1/6/2002




                        Simulation Environment
    •   Similar to Wei Yu’s paper
    •   Empirical TFs from RLCG model
    •   0.4 mm lines (AWG 26)
    •   4 users @ 1000 ft, 4 users @ 3000 ft
    •   Coding gain 3 dB
    •   998 FDD bandplan
    •   Alien Crosstalk: Noise A model from T1 standards
    •   Total power constraint: 11.5 dBm




Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        9
    KU Leuven Department of Electrical Engineering                                           1/6/2002




                             Interference Channel
                                                                                    Proposed
                                                                                    Scheme




                                                      It. Scalar
                                                      Waterfilling
                            Eq. FEXT




           Similar performance

Reduced Complexity MUD and Multi-user Power Allocation in DSL        R. Cendrillon and M. Moonen        10
    KU Leuven Department of Electrical Engineering                                      1/6/2002




                            Multi-access Channel
                                                                                  Proposed
                                                                                  Scheme




                                                                                  It. Scalar
                                                                                  Waterfilling




    •   It. Waterfilling takes interference into account (unnecessary)
    •   Far-end users forced into high frequency band
    •   Significant performance degradation

Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        11
    KU Leuven Department of Electrical Engineering                                       1/6/2002




                      Reduced Complexity MUD


                                                                                   Standard MUD
                                                                                   (Linear MMSE)


                                             Reduced Complexity MUD
                                            (38% Standard Complexity)

                                 SUD




Reduced Complexity MUD and Multi-user Power Allocation in DSL    R. Cendrillon and M. Moonen        12
    KU Leuven Department of Electrical Engineering                                      1/6/2002




                      Reduced Complexity MUD


                                                                                  Standard MUD
                                                                                  (Linear MMSE)




                                 SUD




Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        13
    KU Leuven Department of Electrical Engineering                                      1/6/2002




                                   ADSL Research

    • Equalisation
          – Channel Shortening TEQ
          – Per-tone Equalisation (PT-EQ)
    • Echo Cancellation
          – Time/Frequency Echo Canceller
          – Per-tone Echo Cancellation (PT-EC)
    • Other Work
          – Adaptive PT-EQ
          – Double-talk Suppression and PT-EC
          – RFI Suppression through Windowed PT-EQ




Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        14
    KU Leuven Department of Electrical Engineering                                               1/6/2002




                            Equalisation in ADSL
    •   Conventional Scheme: Channel Shortening TEQ [Chow, Al-Dhahir, Cioffi]


                                                                      1 Tap FEQs
                      Channel Shortening                        1
                                                                          X




                                                      FFT

                                                                ...


                                                                          ...
                              TEQ
                                                                          X
                                                                N



    •   TEQ design not directly related to bit-rate maximisation
    •   Finding optimal TEQ leads to non-convex optimisation
    •   Performance sensitive to synch delay




Reduced Complexity MUD and Multi-user Power Allocation in DSL            R. Cendrillon and M. Moonen        15
    KU Leuven Department of Electrical Engineering                                        1/6/2002



                               Per-tone Equaliser
    •   Channel shortening TEQ
          – T-tap TEQ common to all tones
          – 1-tap FEQ for each tone

    •   Per-tone Equaliser (PT-EQ) [Van Acker]
          – T-tap FEQ for each tone

                                                     T Tap FEQs
                                 Sliding FFT



                                               1
                                                       FEQ
                                               ...



                                                         ...
                                                       FEQ
                                               N

    •   Exploit ‘sliding FFT’ properties  Similar complexity to conventional
        channel shortening TEQ
    •   Further complexity reduction by varying FEQ length (T) on tone-by-tone
        basis

Reduced Complexity MUD and Multi-user Power Allocation in DSL     R. Cendrillon and M. Moonen        16
    KU Leuven Department of Electrical Engineering                                                 1/6/2002



                                            Per-tone Equaliser
    •   Equaliser design directly related to bitrate optimisation                         64 tap   PTEQ
    •   40% gain in US bitrate (on average) over conv. TEQ                                64 tap   TEQ
                                                                                          16 tap   PTEQ
    •   Reduced sensitivity to synch delay                                                16 tap   TEQ

                              1.2

                              1.0
           Data-rate (Mbps)




                              0.8

                              0.6

                              0.4

                              0.2

                               0 -10   -5   0   5   10   15       20   25     30     35      40
                                                    Synch Delay
    •   Currently at implementation phase for next gen. of Alcatel ADSL modems


Reduced Complexity MUD and Multi-user Power Allocation in DSL          R. Cendrillon and M. Moonen            17
    KU Leuven Department of Electrical Engineering                                                              1/6/2002



                                              Echo Cancellation
    •   Time/Freq Domain Echo Cancellation [Ho, Cioffi]
          –   EC split between time and freq
          –   Efficient Implementation
    •   Per-tone Echo Cancellation (PT-EC) [Van Acker]
          –   Extension of Per-Tone concept to Echo Cancellation
          –   Similar complexity to Time/Freq EC
          –   Now EC design directly related to bit-rate optimisation
                               IFFT
                       . . .




                                                 Sliding FFT
                                                      . . .
                                PT-EC                                      P tap Echo Cancellers
                                                    FEC


                                                                  FEC
                                                                                                             Hybrid
                                                                                  PT-EQ
                                                -         . . .




                                                                                               Sliding FFT
                                                     +                          FEQ
                                                                        . . .




                                                                                       . . .
                                      . . .




                                                             -
                                                                  +             FEQ

Reduced Complexity MUD and Multi-user Power Allocation in DSL                         R. Cendrillon and M. Moonen          18
    KU Leuven Department of Electrical Engineering                                      1/6/2002




                                        Other Work
    • Adaptive Implementation of PT-EQ [Ysebaert]
    • Double-talk Suppression and PT-EC [Ysebaert]
          – Double talk decreases convergence speed in EC adaptation
          – Far-end signal can be removed prior to adaptation
          – Integrated into PT-EC structure
    • RFI Suppression through Windowed PT-EQ




Reduced Complexity MUD and Multi-user Power Allocation in DSL   R. Cendrillon and M. Moonen        19
    KU Leuven Department of Electrical Engineering                                              1/6/2002




              Open Questions and Current Work
    •   Quantify sub-optimality of different power allocation schemes in IC
    •   Completely autonomous power allocation in IC
          – Near-far Scenario: It. Waterfilling gives large gains over Conv. PBO
          – Still need to determine total power alloc. for each user
          – One suggestion: Independent 3rd party
          – May be prickly politically
          – Also possible to vary total power constraint in autonomous way
            e.g. based on line length
          – Similarly, could apply masks to w.f. spectrum
                • Mask varies with line length
                • Short lines: Tight masks prevent excessive loading into low freqs
                • Long lines: Loose masks allow far-end users to fully exploit waterfilling gains


    •   Bonded lines (Full Vectoring), Unbundled crosstalk cancellation, Adaptive
        considerations, Channel ID and tracking


Reduced Complexity MUD and Multi-user Power Allocation in DSL          R. Cendrillon and M. Moonen         20

				
DOCUMENT INFO