ist Javier R. Fonollosa An overview The I-METRA …ipd481/Papers varios/IMETRAabr02.pdfThe I-METRA...

The I-METRA ProjectAn overview

Javier R. Fonollosa

http://www.ist-imetra.org

2

10/04/2002

I-METRA Presentation at the AA Cluster meeting

2www.ist-imetra.org

Outline• Objectives and general description of I-METRA• Adaptive Modulation Schemes for MIMO HSDPA

and HSUPA• Link and System level simulations:

• Standardised MIMO channel model• Preliminary results on G parameter distribution

• Adaptive Space-Time Processing:• Orthogonal Space-Time Coding and Beamforming• BLAST and linear dispersion codes• Turbo Space-Time Coding

10/04/2002


3www.ist-imetra.org

The I-METRA project•I-METRA builds on the legacy of METRA with theobjective of incorporating analysis, development andsimulation of adaptive transmission technologies for multiple-input, multiple-output (MIMO) systems

•The main emphasis is given to incorporating re-configurability capabilities in order to allow the radio network,including terminal and base stations, to adjust automatically totraffic, user and channel requirements

•The project will investigate the potential implications of thesetechnologies into Systems beyond 3G.

10/04/2002


4www.ist-imetra.org

Objectives• Quantification of the performance improvement provided by adaptive

MIMO technologies in terms of increased data rates and spectrumefficiency

• Evaluation of the ability to adjust automatically to different scenariirequirements set up by the traffic, user or channel varying conditions

• Implications of the proposed techniques in terms of technicalcomplexity of mobile and base station software and hardwarearchitecture

• Compatibility of proposed techniques with current 3GPP specificationsas well as the feasibility of incorporating them in future releases

• Impact of the MIMO configurations at the mobile terminal and basestation in the performance of transmission technologies that areessential in Systems beyond 3G (HSDPA and HSUPA)

10/04/2002


5www.ist-imetra.org

Adaptive MIMO for HSDPA andHSUPA

• The concept of HSDPA has been recently defined in 3GPP for UMTS• HSDPA considers enhancements that can be applied to UTRA to

provide very high-speed downlink packet access by means of a high-speed downlink shared channel (HS-DSCH)

• I-METRA project is focused on MIMO antenna processing techniquesfor HSDPA which are related to the evolution of other procedures,such as Adaptive Modulation and Coding (AMC) or Hybrid AutomaticRepeat on Request (H-ARQ), in order to include adaptive MIMOtechniques in the HS-DSCH structure

• HSUPA has received little attention in standardisation so far and ischaracterised by distinctive requirements

10/04/2002


6www.ist-imetra.org

Transport Channel Coding Structure for HS-DSCH

Code Block Segmentation

CRC attachment

Channel Coding

H-ARQ

Physical Channel Segmentation

Interleaving

Physical Channel Mapping

Spreading

Modulation

RV parameter

...

...

...

...

10/04/2002


7www.ist-imetra.org

Link and System LevelSimulation

• Link-level simulations alone are not enough to conclude about theperformance of a HSDPA system

• MIMO HSDPA simulations in I-METRA will be performed in twostages: link-level simulations and system-level simulations. Severallook-up tables obtained in the link-level simulation stage should feedthe system-level simulations

• Look-up tables relating FER values to a given metric, will becomputed for different Modulation and Coding (MC) schemes anddifferent MIMO techniques

• The MIMO singularity on top of the HSDPA adds some specificrequirements to both the link- and system-level simulator, namely:channel modelling, metric definition and HS-DSCH structure

10/04/2002


8www.ist-imetra.org

Channel Modelling (I)• The I-METRA project will proceed with the MIMO channel model

already developed in METRA• This model was filed in 3GPP in Feb. 2001 and after few months,

major companies such as Nokia, Lucent, Siemens and Ericssonendorsed the stochastic philosophy of the METRA's MIMO model

• This model has the structure of a tapped delay line and its taps arematrices whose size depends on the number of active elements at thetransmitting and receiving ends. This model appears as a naturalextension of well-accepted ITU profiles

• This model is of stochastic nature. It manages to embed the fullcorrelation information of the channel into two correlation matricesdefined independently at both ends. A simple Kronecker product isperformed to combine these matrices so as to achieve the fullcharacterisation of the correlation properties of a given MIMO channel

10/04/2002


9www.ist-imetra.org

Channel Modelling (II)• The model in itself is able to address a wide variety of simulation

environments.• It can model the time dispersion, the fading and the spatial

properties of MIMO channels, using a reduced set of parameters,namely:

• Power Delay Profile (PDP)• Power Angular Spectrum (PAS)• Angle of Arrival (AoA)• Angle of Departure (AoD)• Azimuth Spread (AS)• Power Doppler Spectrum (PDoS).

10/04/2002


10www.ist-imetra.org

Channel Metric• For a given data rate, the user performance in a conventional

HSDPA channel, usually stated in terms of FER, is reliablypredicted using an estimate of the C/I

• It is not clear which parameter is the C/I counterpart in a MIMOHSDPA channel

• In order to deal with an HSDPA system able to adapt to channelpropagation conditions, it is mandatory to define a metricfunction that maps the channel estimation to the FER on thatchannel. It is expected that this metric will be different fordifferent MIMO techniques

• Additionally, this metric should be simple enough to betransmitted to Node B within the uplink PDCCH-HS using up to6 bits every TTI (2 ms).

10/04/2002



Modulation

Spreading

How do MIMO systems modify the HS-DSCH structure?

CRC attachment

Channel Coding

H-ARQ


Interleaving



10/04/2002



Modulation

Spreading


CRC attachment

Channel Coding

H-ARQ


Interleaving



Modulation

Spreading

Code reuse and DSTTD

10/04/2002



Modulation

Spreading


CRC attachment

Channel Coding

H-ARQ


Interleaving



Modulation

Spreading

Code reuse and DSTTD

Code Block SegmentationPARC

Modulation

Spreading

Channel Coding

H-ARQ


Interleaving



10/04/2002



Code reuse MIMO transmission

CodingInterleaving

Mapping

DEMUX ...

Spreading Code 1

Spreading Code 2

Spreading Code 10

ScramblingCode

ScramblingCode

Highspeeddatastream

• Equal transmission rates on each antenna• Coding followed by demultiplexing• Channel code spans space and time dimensions

10/04/2002



CodingInterleaving

Mapping

DEMUX

...

Spreading Code 1

Spreading Code 2

Spreading Code 10

ScramblingCode

ScramblingCode

CodingInterleaving

Mapping ...

...

...

Highspeeddatastream

PARC MIMO Transmission• Distinct transmission rates on each antenna• Demultiplexing followed by coding• Channel code spans time dimension only

10/04/2002



System Level Simulations (I)• System-level simulations are mandatory in order to include system

attributes such as Fast Cell Selection, H-ARQ or Node B scheduling• System level simulations will take a similar format to that proposed

in 3GPP for the evaluation of MIMO techniques for HSDPA.• These will take as input a range of link-level results that will be used

in a look-up table - relevant factors may include modulation scheme,coding, redundancy, ratio of in-cell/out-of-cell interference (G-parameter).

• The referencing into the look-up table will depend on the chosenMIMO technique and the associated metric.

• The G-parameter is important as the interference situation willdetermine the modulation and coding (for example) chosen by theNode B

10/04/2002



System Level Simulations (II)

0.00

5.00

10.00

15.00

20.00

25.00

-9 -6 -3 0 3 6 9 12 15 18 21 24 27 30 33 36

G-parameter (dB)

% C

alls Macro

Micro

10/04/2002



Adaptive Space-TimeProcessing

• Orthogonal Space-Time Coding and Beamforming withlimited state information at the transmit side (CSIT)

• BLAST and linear dispersion codes (CSIR only)

• Turbo Space-Time Coding (CSIR only)

10/04/2002



Combining OSTBC and BF (I)• Traditional approach to the design of space-time block codes is based

on the minimisation of an upper bound to the pairwise codeword errorprobability:

• [Jöngren] proposes to introduce the available state information andminimise

( ) ( )

σ−≤→ 2

2

4,exp

21 lk

lkdP CCCC

( ) ( ) ( ) ( )( )Hlklklk

lklkl CCCCCCA

CCACC −−==→ ,,

,det1

( ) ( ) ( ) ( )( )hCCAIhCCCChhCC lkNH

lklk

lk ddP ,,4

,exp21ˆ, 2

2

2

⊗=

σ−≤→

10/04/2002



Combining OSTBC and BF (II)• It is assumed that the transmitter has some information on how good

the channel estimation is. It therefore uses two different kinds ofinformation:

– The actual channel estimation itself:– A measure of the goodness of the approximation:

( ) ( ) ( )

( ) ( )( ) 1ˆ2

ˆ1

ˆ11

ˆˆ

,4

1,

,detlog,,

−

−−−

+⊗σ

=

−=

hhh

hhhhhhhhhh

RCCAICCΨ

CCΨmRCCΨRmCC

lkNlk

lklkH

lkl

hhhhh mR ˆˆ ,h

Assumedknown to thetransmitter!!!

• The codes are then designed to minimise:

10/04/2002



Combining OSTBC and BF (III)• If, in addition, we force with OSTBC, we can obtain

the “optimum” beamformer for this situation.kk CWC = kC

CK WSymbols

OrthogonalSpace-Time

Block Coding

Beamformingmatrix

(one beamformerfor code branch)

10/04/2002



Asymptotic situations• Perfect channel state information:

• No channel state information:

0ˆ →hhhR

01ˆ →−hhhR

Traditional beamforming

MMIW 1→

Pure space-time code

[ ]00wW �M→

Left singular vector of thechannel matrix

CK wSymbols


Block Coding

Beamforming vector(only one code

branch is processed)

CKSymbols


Block Coding

10/04/2002



Simulations

2TX, 1RX 4TX, 4RX

10/04/2002



Advantages BF+OSTBC• The architecture provides the optimum performance with respect to

both OSTBC and conventional beamforming:

BER

SNR

ConventionalBeamformer

OSTBC

ProposedBeamformer

• Simple processing at the user equipment (OSTBC)

10/04/2002



Disadvantages BF+OSTBC• Necessity of MIMO channel feedback: vector quantisation (In current

HSDPA specifications only 2 bits/slot are available )• Perfect knowledge of the estimation statistics is assumed and, even so,

mild gains are obtained. Behaviour in a practical situation?• Relatively high computational requirements at Node B. A new

beamformer must be calculated at each channel update, and a complexoptimisation procedure must be carried out for non-diagonal covariancematrices.

• A new beamformer matrix must be calculated every time thetransmission rate is changed. The architecture not very suitable fromthe re-configurability point of view.

10/04/2002



Linear Dispersion Codes (I)• Each symbol is multiplied by a transmit matrix over a particular space-

time support. For the real-valued (BPSK) case:

∑=

=Q

kkks

1

CX

CQ

Symbols

Linear DispersionBlock Coding

S/P

C1

...

• This architecture subsumesOSTBC and BLAST as specialcases:

Space-timecoding matrix

−=

=0110

1001

21 CC

=

=

=

=1000

0100

0010

0001

4321 CCCC

(2 x 2) Alamouti code

(2 x 2) V-BLAST

10/04/2002



Linear Dispersion Codes (II)• The channel knowledge is only assumed at the receiver. The code

matrices are designed as follows:

( ) ( )

ρ+ TqqNTHC

CCM

ETq

HHIdetlog1max

averaged over all possible realisations of the channel.

• Depending the number of signals sent (Q), the block length (T), andthe size of the constellation of the symbols (r bits/symbol), the overalltransmit rate will be

rTQR 2log=

• The receiver can be almost as simple as in OSTBC. Linear receiversare viable.

10/04/2002



Implementation LD Codes

• Ease of rate re-configurability in aMIMO architecture.

• Given the possibility of linearreceivers, the channelperformance metric seems easilycomputable.

• The architecture admits thepossibility of more receive thantransmit antennas.

ADVANTAGES DISADVANTAGES

• No channel state information isused by the transmitter.

• The codes are optimised forparticular channel statistics.

10/04/2002



Turbo Codes: Encoder

• Standard Turbo Codes employ two parallel Recursive SystematicConvolutional (RSC) Codes concatenated via an interleaver.

• The systematic bit and the two parity bits are rearranged in a serialorder.

RSC

RSCp

P/S

Systematic bit

Parity bits

10/04/2002



Turbo Codes: Decoder• Two soft output decoders serially concatenated via interleavers• Each decoder takes three inputs: a received systematic bit, a received

parity bit and soft a-priori information from the other encoder• 5-10 iterations of decoding are used• On the last iteration, a hard decision is applied on the soft outputs

SoftDecoder p

p -1SoftDecoder

p

10/04/2002



Space Time Turbo Codes

• Space Time Codes (STC) are codes designed for Multiple Inputs andMultiple Outputs (MIMO) systems

• Space Time Turbo Codes (STTC) are STC codes which use theconcept of iterative turbo decoding

10/04/2002



Simple STTC Scheme

• A possible simple scheme of STTC is using the standard Turboarchitecture, but simultaneously transmitting the bits through a fewtransmit antennas. Thus, achieving both coding diversity, and transmitdiversity

RSC

RSCp

10/04/2002



General STTC Scheme

• In a more general STTC scheme, the transmitter consists of a generalencoder for coding diversity, serially followed by a general STC fortransmit diversity.

• The receiver has a joint decoding and demodulating algorithm basedon the Turbo principle.

pChannelEncoder

STC

10/04/2002



Hochwald & Brink’s Scheme

10/04/2002



• Interleaving on bit-level (~1.5 dB gain compared to symbol level)• Only one component encoder connected to outputs at given time• If puncturing is not applied, bandwidth efficiency is halved

Tujkovic’s Scheme: Encoder

(Recursive)STTC

RecursiveSTTC

Bit-level interleaver

Punctureand/or

multiplex

10/04/2002



Tujkovic’s Scheme: Decoder

symbolMAP 1

sym/bits

inter-leaver

bit/sym

symbolMAP 2

bit/sym

inter-leaver

sym/bits

+

-

+

-

ist Javier R. Fonollosa An overview The I-METRA …ipd481/Papers varios/IMETRAabr02.pdfThe I-METRA...

Documents

Transcript of ist Javier R. Fonollosa An overview The I-METRA …ipd481/Papers varios/IMETRAabr02.pdfThe I-METRA...