Improving MIMO Sphere Detection Through Antenna Detection Order - - PowerPoint PPT Presentation

Improving MIMO Sphere Detection Through Antenna Detection Order Scheduling

Michael Wu, Chris Dick, Yang Sun, Joseph Cavallaro December 1, 2011

MIMO Detection

12/1/2011 2

 Spatial Multiplexing

 Increases throughput  Used in many wireless

standards

 MIMO detector recovers

• riginal signal

 Search problem n Hx y                                            

2 1 2 1 22 21 20 12 11 10 02 01 00 2 1

n n n x x x h h h h h h h h h y y y

H

2 2

ˆ min Rx y

n

Ω x





n Rx y                                             ˆ ˆ ˆ ˆ

2 1 2 1 22 12 11 02 01 00 2 1

n n n x x x r r r r r r y y y

Tree-Search Based Detection

 Depth-first tree search  Variable execution time  Fairly sequential, slow  Breadth-first tree search  More data parallel  Sort is the bottleneck

 Number of comparisons + Memory requirement (KM) 12/1/2011 3

SSFE Detector

12/1/2011 4

 Selective Spanning with

Fast Enumeration (SSFE)

 Data parallel deterministic

search

 First antenna level

 Enumerate all modulation

points

 Subsequent levels

 Pick the best outgoing node

for each path.

QPSK 2x2 Example

SSFE Detector

 “Sort-free” MIMO Detector

 Picking the best outgoing node does not require sort

5 12/1/2011

Schnoor-Euchner enumeration P=[s0, s1 … sk], yk+1, hk+1 Sk

round()

sk+1 ̂ 1 , , ,

Flexsphere Implementation

6 12/1/2011

• C. Dick, M.

Trajkovic, S. Denic, D. Vuletic, R. Rao, F. Harris, K. Amiri, FPGA Implementation of a Near-ML Sphere Detector for 802.16e Broadband Wireless Systems , proceedings of SDR conference, 2009

H

search

compute the optimal detection order

Hardware Resource Cost Flexsphere Design

x ˆ

64% 11% 16% 9% Preprocessing RVD/QRD Sphere Detector Soft Output Generator

Flexsphere Implementation: V-BLAST Reordering

12/1/2011 7

 Complex block: 3 matrix inverses and 3 matrix multiplies H Matrix multiply Matrix multiply QRD Back Subst. 4x4 Matrix Inverse Norm Search/ Reorder 4x4 Matrix multiply Matrix multiply QRD Back Subst. 3x3 Matrix Inverse Norm Search/ Reorder 3x3 Matrix multiply Matrix multiply QRD Back Subst. 2x2 Matrix Inverse Norm Search/ Reorder 2x2

Flexsphere Implementation

8 12/1/2011

• C. Dick, M.

Trajkovic, S. Denic, D. Vuletic, R. Rao, F. Harris, K. Amiri, FPGA Implementation of a Near-ML Sphere Detector for 802.16e Broadband Wireless Systems , proceedings of SDR conference, 2009

H

search

compute the optimal detection order

Hardware Resource Cost Flexsphere Design 64% 11% 16% 9% Preprocessing RVD/QRD Sphere Detector Soft Output Generator

x ˆ

Can we do better?

N-Way MIMO Detector

 Get rid of the

V-BLAST channel reordering block

 Duplicate search blocks depending on BER requirement.

 Add permute block which enforces a detection order  Example: N = 2, two search blocks

9 12/1/2011

H

Merge

x ˆ

X0  X1  X2 X1  X2  X0

y,

N-Way MIMO Detector: Merge Block

 Simple block, performs max-log-map computation

min min

• Soft Output

Generator Soft Output Generator +1

• 1

+1

• 1

LLR L0 L1

10 12/1/2011

BER Performance

 Rayleigh fading channel  Soft Output MIMO Detector + Rate 1/2 WiMAX LDPC

decoder

 1 outer iteration + 20 inner iteration with early termination

11 12/1/2011

BER Performance (16QAM)

12 12/1/2011

BER Performance (64QAM)

13 12/1/2011

Implementation: N-Way MIMO Detector

 Target: 83.768Mbps(WiMAX),

Virtex 5 @ 225Mhz

 T

• tal resource = N (RVD/QRD + Sphere Detector + Soft Output Generator)

+ Merge

N Slices LUTs/FFs DSP48 Block RAM 1 5,658 9,437/15,990 78 41 2 11,274 19,018/32,525 156 82 3 16,827 28,743/49,117 234 123 4 22,832 38,515/65,381 312 164 Flexsphere 15,657 29,776/45,944 237 146

14 12/1/2011

Implementation: N-Way MIMO Detector

 Target: 83.768Mbps(WiMAX), Virtex 5 @ 225Mhz

15 12/1/2011

Conclusion

 Scalable data parallel detection algorithm

 Search is cheap  Better performance/resource compare to Flexsphere

 Target for software implementation?

 Enumeration complexity doesn’t depend on modulation  Increase parallelism of the detection algorithm

Parallelism: M Parallelism: NM

16 12/1/2011