1
Some Aspects on Hybrid Wideband Transceiver Design for mmWave Communication Systems Marcin Iwanow 12 , Nikola Vuˇ ci´ c 1 , Mario H. Casta˜ neda 1 , Jian Luo 1 , Wen Xu, 1 and Wolfgang Utschick 2 1 Huawei Technologies Duesseldorf GmbH, Munich, Germany 2 Associate Institute for Signal Processing, Technische Universit¨ at M¨ unchen, Germany Introduction mmWave frequency ranges are considered to be a key enabler for extremely high throughput in 5G Principal issues to solve on the PHY layer: transmit/receive strategy, channel estimation/tracking, multi-user communication, waveform design, etc. Our focus: Broadband transceiver design in the hybrid beamforming architecture Problem Description Assuming perfect CSI, define the transmit and receive strategies for a multitap channel Main issues: For practical solutions, the analog beamformers P k ,A as well as combiners G k ,A are constant across the subcarriers Computational complexity of solutions Relevant Work in the Literature Narrowband solutions Exploiting channel spatial sparsity a Exploiting estimated channel subspaces b Multicarrier solutions Codebook-based approach with suboptimal solutions more efficient than the exhaustive search. c Sequential searching algorithm with limited feedback assumption. d Solution exploiting (slowly varying) second order statistics. e a O. El Ayach et al. “Spatially Sparse Precoding in Millimeter Wave MIMO Systems” b H. Ghauch et al. “Subspace Estimation and Decomposition for Hybrid Analog-Digital Millimetre-Wave MIMO systems” c C. Kim et al. “Multi-beam transmission diversity with hybrid beamforming for MIMO-OFDM systems” d A. Alkhateeb et al. “Frequency Selective Hybrid Precoding for Limited Feedback Millimeter Wave Systems” e A. Adhikary et al. “Joint Spatial Division and Multiplexing for mm-Wave Channels” System Model Digital Tx Precoder OFDM Processing: IFFT,CP insertion OFDM Processing: IFFT,CP insertion . . . RF chain # 1 . . . RF chain # N RF tx Analog RF Tx Precoding Channel H[d] Analog RF Rx Equalizer RF chain # 1 . . . RF chain # N RF rx OFDM Processing: CP deletion,FFT . . . OFDM Processing: CP deletion,FFT Digital Rx Equalizer . . . s 1 s Nsubc . . . N tx Tx antennas . . . N rx Rx antennas . . . ˜ s 1 ˜ s Nsubc H k = 1 N subc D-1 d =0 H [d ] exp j 2πk N subc d ˜ s k = G H k ,D G H k ,A H k P k ,A P k ,D s k + G H k ,D G H k ,A η k Channel Model H [d ]= β N rx N tx L N cl l =1 N l path r =1 α r ,l p(dT s - τ l - τ r )a rx (θ r ,l )a H tx (φ r ,l ) Assumptions: The number of paths is significantly lower than for the sub-6GHz frequency band. The paths propagate in space and time clusters. Proposed Solution Based on the perfect (or estimated) CSI, design optimal linear precoders and combiners Decompose the optimal precoding/combining matrices jointly for a set of subcarriers Ω, such that the analog components are equal for all the subcarriers in the set For the decomposition, use the (heuristic) objective of minimizing the Frobenius norm of the error matrix (Brief) Description of the Algorithms BCD: Optimize in each iteration either the digital or analog part. Assume the other one fixed In each step, project the analog matrix to the nearest (in Euclidean sense) matrix of the required set OMP: In each of the N RF steps, choose a vector from the codebook that has the maximal projection on the remains of the matrix being decomposed. In each step, project out the selected vector BCD Decomposition, Measured Channels -30 -25 -20 -15 -10 -5 0 5 10 0 2 4 6 8 10 12 14 16 18 20 SNR [dB] Rate [bit/c.u.] Full digital Hybrid str., separate analog part subc. Hybrid str., |Ω| =1 Hybrid str., |Ω| =3 Hybrid str., |Ω| = 10 Hybrid str., |Ω| = N subc Formal Description P Ω,A , P Ω,D = arg min P Ω,A ,P Ω,D P Ω - P Ω,A P Ω,D F s.t. P Ω,A P Ω,D 2 F ≤|Ω| P tx N subc , P Ω,A ∈P A where P Ω = P ω 1 ,..., P ω |Ω| C Nrx ×(|Ω|Ns ) , P Ω,D = P ω 1 ,D ,..., P ω |Ω| ,D C N RF tx ×(|Ω|Ns ) , P Ω,A C Ntx ×N RF tx Ω ⊂{1,..., N subc } ω l Ω, l ∈{1,..., |Ω|} BCD Decomposition, Statistical Channel Model -30 -25 -20 -15 -10 -5 0 5 10 0 2 4 6 8 10 12 14 16 18 20 SNR [dB] Rate [bit/c.u.] Full digital Hybrid str., separate analog part subc. Hybrid str., |Ω| =1 Hybrid str., |Ω| =3 Hybrid str., |Ω| = N subc OMP Decomposition, Measured Channels -30 -25 -20 -15 -10 -5 0 5 10 0 2 4 6 8 10 12 14 16 18 20 SNR [dB] Rate [bit/c.u.] Full digital Hybrid str., separate analog part subc. Hybrid str., |Ω| =1 Hybrid str., |Ω| =3 Hybrid str., |Ω| = N subc Solutions Considered strategies: Full digital setup. Linear precoding with P k and linear combining with G k for each subcarrier k Full hybrid transceiver. Linear precoding with P k ,A P k ,D and linear combining with G k ,A G k ,D H at each subcarrier. Full hybrid transceiver. Linear precoding with P Ω,A P k ,D and linear combining with G Ω,A G k ,D H at each subcarrier. For solving the optimization problem, we use either the Orthogonal Matching Pursuit (OMP) or the Block Coordinate Descent (BCD) algorithm OMP Decomposition, Statistical Channel Model -30 -25 -20 -15 -10 -5 0 5 10 0 2 4 6 8 10 12 14 16 18 20 SNR [dB] Rate [bit/c.u.] Full digital Hybrid str., separate analog part subc. Hybrid str., |Ω| =1 Hybrid str., |Ω| =3 Hybrid str., |Ω| = N subc Discussion A practical solution for wideband linear precoding/combining has been presented A reduced set of subcarriers can be used for designing the analog transmit/receive strategy reduction of complexity The choice of the decomposition algorithm is remarkably important Future work should consider channel estimation Associate Institute for Signal Processing Technische Universit¨ at M¨ unchen

Some Aspects on Hybrid Wideband Transceiver Design for ......A. Alkhateeb et al. “Frequency Selective Hybrid Precoding for Limited Feedback Millimeter Wave Systems” e A. Adhikary

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

  • Some Aspects on Hybrid Wideband Transceiver Design formmWave Communication Systems

    Marcin Iwanow12, Nikola Vučić1, Mario H. Castañeda1, Jian Luo1, Wen Xu,1 and Wolfgang Utschick21Huawei Technologies Duesseldorf GmbH, Munich, Germany

    2Associate Institute for Signal Processing, Technische Universität München, Germany

    Introduction

    mmWave frequency ranges are considered tobe a key enabler for extremely highthroughput in 5GPrincipal issues to solve on the PHY layer:transmit/receive strategy, channelestimation/tracking, multi-usercommunication, waveform design, etc.Our focus: Broadband transceiver design inthe hybrid beamforming architecture

    Problem Description

    Assuming perfect CSI, define the transmitand receive strategies for a multitap channel

    Main issues:For practical solutions, the analogbeamformers Pk,A as well as combiners Gk,Aare constant across the subcarriersComputational complexity of solutions

    Relevant Work in the Literature

    Narrowband solutionsExploiting channel spatial sparsity aExploiting estimated channel subspaces b

    Multicarrier solutionsCodebook-based approach with suboptimalsolutions more efficient than the exhaustivesearch.cSequential searching algorithm with limitedfeedback assumption. dSolution exploiting (slowly varying) secondorder statistics. e

    aO. El Ayach et al. “Spatially Sparse Precoding in Millimeter Wave MIMO Systems”bH. Ghauch et al. “Subspace Estimation and Decomposition for Hybrid Analog-Digital Millimetre-Wave MIMO systems”cC. Kim et al. “Multi-beam transmission diversity with hybrid beamforming for MIMO-OFDM systems”dA. Alkhateeb et al. “Frequency Selective Hybrid Precoding for Limited Feedback Millimeter Wave Systems”eA. Adhikary et al. “Joint Spatial Division and Multiplexing for mm-Wave Channels”

    System Model

    DigitalTx

    Precoder

    OFDM Processing:IFFT,CP insertion

    OFDM Processing:IFFT,CP insertion

    ...

    RF chain# 1

    ...

    RF chain# NRFtx

    AnalogRFTx

    Precoding

    ChannelH [d]

    AnalogRFRx

    Equalizer

    RF chain# 1

    ...

    RF chain# NRFrx

    OFDM Processing:CP deletion,FFT

    ...

    OFDM Processing:CP deletion,FFT

    DigitalRx

    Equalizer

    ...

    s1

    sNsubc

    ...

    Ntx Tx antennas

    ...

    Nrx Rx antennas

    ...

    s̃1

    s̃Nsubc

    Hk =1√

    Nsubc

    D−1∑d=0

    H[d ] exp( j2πk

    Nsubcd)

    s̃k = GHk,DGHk,AHkPk,APk,Dsk + GHk,DGHk,Aηk

    Channel Model

    H[d ] = β

    √NrxNtx

    L

    Ncl∑l=1

    N lpath∑r=1

    αr ,lp(dTs − τl − τr )arx(θr ,l )aHtx(φr ,l )

    Assumptions:The number of paths is significantly lowerthan for the sub-6GHz frequency band.The paths propagate in space and timeclusters.

    Proposed Solution

    Based on the perfect (or estimated) CSI,design optimal linear precoders andcombinersDecompose the optimalprecoding/combining matrices jointly for aset of subcarriers Ω, such that the analogcomponents are equal for all the subcarriersin the setFor the decomposition, use the (heuristic)objective of minimizing the Frobenius normof the error matrix

    (Brief) Description of the Algorithms

    BCD:Optimize in each iteration either the digitalor analog part. Assume the other one fixedIn each step, project the analog matrix tothe nearest (in Euclidean sense) matrix ofthe required set

    OMP:In each of the NRF steps, choose a vectorfrom the codebook that has the maximalprojection on the remains of the matrixbeing decomposed.In each step, project out the selected vector

    BCD Decomposition, Measured Channels

    −30 −25 −20 −15 −10 −5 0 5 100

    2

    4

    6

    8

    10

    12

    14

    16

    18

    20

    SNR [dB]

    Rate[bit/c.u.]

    Full digitalHybrid str., separate analog part ∀ subc.Hybrid str., |Ω| = 1Hybrid str., |Ω| = 3Hybrid str., |Ω| = 10Hybrid str., |Ω| = Nsubc

    Formal Description

    (P?Ω,A,P?Ω,D

    )= arg min

    PΩ,A,PΩ,D‖P?Ω − PΩ,APΩ,D‖F

    s.t. ‖PΩ,APΩ,D‖2F ≤ |Ω|Ptx

    Nsubc, PΩ,A ∈ PA

    where

    P?Ω =[P?ω1 , . . . ,P

    ?ω|Ω|

    ]∈ CNrx×(|Ω|Ns),

    PΩ,D =[Pω1,D, . . . ,Pω|Ω|,D

    ]∈ CNRFtx ×(|Ω|Ns),

    PΩ,A ∈ CNtx×NRFtx

    Ω ⊂ {1, . . . ,Nsubc}ωl ∈ Ω, ∀l ∈ {1, . . . , |Ω|}

    BCD Decomposition, Statistical Channel Model

    −30 −25 −20 −15 −10 −5 0 5 100

    2

    4

    6

    8

    10

    12

    14

    16

    18

    20

    SNR [dB]

    Rate[bit/c.u.]

    Full digitalHybrid str., separate analog part ∀ subc.Hybrid str., |Ω| = 1Hybrid str., |Ω| = 3Hybrid str., |Ω| = Nsubc

    −30 −25 −20 −15 −10 −5 0 5 100

    2

    4

    6

    8

    10

    12

    14

    16

    18

    20

    SNR [dB]

    Rate[bit/c.u.]

    Full digitalHybrid str., separate analog part ∀ subc.Hybrid str., |Ω| = 1Hybrid str., |Ω| = 3Hybrid str., |Ω| = Nsubc

    OMP Decomposition, Measured Channels

    −30 −25 −20 −15 −10 −5 0 5 100

    2

    4

    6

    8

    10

    12

    14

    16

    18

    20

    SNR [dB]

    Rate[bit/c.u.]

    Full digitalHybrid str., separate analog part ∀ subc.Hybrid str., |Ω| = 1Hybrid str., |Ω| = 3Hybrid str., |Ω| = Nsubc

    −30 −25 −20 −15 −10 −5 0 5 100

    2

    4

    6

    8

    10

    12

    14

    16

    18

    20

    SNR [dB]

    Rate[bit/c.u.]

    Full digitalHybrid str., separate analog part ∀ subc.Hybrid str., |Ω| = 1Hybrid str., |Ω| = 3Hybrid str., |Ω| = Nsubc

    Solutions

    Considered strategies:

    Full digital setup. Linear precoding withP?k and linear combining with G?k for eachsubcarrier kFull hybrid transceiver. Linear precodingwith P?k,AP?k,D and linear combining with(G?k,AG?k,D

    )Hat each subcarrier.

    Full hybrid transceiver. Linear precodingwith P?Ω,AP?k,D and linear combining with(G?Ω,AG?k,D

    )Hat each subcarrier.

    For solving the optimization problem, we useeither the Orthogonal Matching Pursuit (OMP)or the Block Coordinate Descent (BCD)algorithm

    OMP Decomposition, Statistical Channel Model

    −30 −25 −20 −15 −10 −5 0 5 100

    2

    4

    6

    8

    10

    12

    14

    16

    18

    20

    SNR [dB]

    Rate[bit/c.u.]

    Full digitalHybrid str., separate analog part ∀ subc.Hybrid str., |Ω| = 1Hybrid str., |Ω| = 3Hybrid str., |Ω| = Nsubc

    −30 −25 −20 −15 −10 −5 0 5 100

    2

    4

    6

    8

    10

    12

    14

    16

    18

    20

    SNR [dB]

    Rate[bit/c.u.]

    Full digitalHybrid str., separate analog part ∀ subc.Hybrid str., |Ω| = 1Hybrid str., |Ω| = 3Hybrid str., |Ω| = Nsubc

    Discussion

    A practical solution for wideband linearprecoding/combining has been presentedA reduced set of subcarriers can be used fordesigning the analog transmit/receivestrategy → reduction of complexityThe choice of the decomposition algorithmis remarkably importantFuture work should consider channelestimation

    Associate Institute forSignal Processing Technische Universität München