MIMO SVD based Algorithm Implementation Dejan Markovic, Prof' Bob Brodersen

1
MIMO SVD based Algorithm Implementation
Dejan Markovic, Prof. Bob Brodersen

• Large capacity benefit

MIMO
Tx
Rx
Rx diversity
Traditional system with 1 Tx and 1 Rx antenna
BB-equivalent channel models
y H x z
1.
y U ? V x z
SVD
2.
x' V x y' U y z' U z

• LMS based blind tracking of SVD components U,?,V
• U?,V updated every symbol period
• Periodic feedback every few hundred data symbols
• Additional blocks required at the Rx side for
  blind tracking

y' ? x' z'
Unitary X-form
3.
Need to know V at Tx and U,? at Rx

• Floating-to-Fixed point conversion
• Architectural and circuit optimization of the
  main building blocks
• Implementation in 0.1mm

Module Compiler estimates _at_ fCLK100MHz, 0.25mm
tracked
Building Blocks
inputs
output
P mW
A mm2
D ns
N
1.
5.9
0.156
7.7
C4?1 , C4?1
C1?1
Cmult_Vd_V
8
ideal
2.
0.4
0.075
2.1
C1?1 , C1?1
C1?1
Csub_d
8
3.
5.6
0.158
6.9
C1?1 , C4?1
C4?1
Cmult_S_V
8
4.
2.2
0.064
7.2
R1?1 , C4?1
C4?1
Cmult_RS_V
8
0.13mm
0.18mm
5.
1.2
0.027
2.0
C4?1 , C4?1
C4?1
Csum_V_V
8
0.25mm
6.
20.1
0.459
12.3
C4?1 , R1?1
C4?1
Cdiv_V_RS
8
Summary of the overall SVD hardware
characteristics
SVD
inputs
output
P mW
A mm2
D ns
N
64
1.74
8.2
R1?1 ,C4?1 , C4?1
C4?1
V_est
8
R1?1 , C4?1
U?_est
8
R4?4 , C4?1
8.2
210
5.54
Optimal step size depends on singular values
Weak subchannel