Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
1/15
Seunghak Lee
School of Electronics and Computer Engineering, Hanyang University17 Haengdang-Dong, Seongdong-Gu, Seoul 133-791, Korea
Tel : 82-2-2299-6267, Fax : 82-2-2299-6263E-mail : [email protected]
http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
Implementation of software-based
2X2 MIMO LTE base station
system using GPU
2/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
Contents
• Introduction of GPU– Structure of GPU
• Proposed System– 2x2 SM MIMO LTE system– Implementation of Scrambling– Implementation of Channel Estimation– Implemented System
• Performance Evaluation– Computation Time
• Conclusion
3/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
1. Introduction of GPU
4/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
• Graphics Processing Unit ( GPU )• Single Instruction Multiple Data( SIMD )• The GPU is massively parallel processor • Kernel
Structure of CUDA Memory
Block(1,0)
Computation grid
Block(0,1)
Block(1,1)
Block(2,1)
Block(0,0)
Block(1,0)
Block(2,0)
Thread(0,3)
Thread(1,3)
Thread(2,3)
Thread(3,3)
Thread(0,2)
Thread(1,2)
Thread(2,2)
Thread(3,2)
Thread(0,1)
Thread(1,1)
Thread(2,1)
Thread(3,1)
Thread(0,0)
Thread(1,0)
Thread(2,0)
Thread(3,0)
Kernel
SIMD processor architecture
DATA ALU
DATA ALU
DATA ALU
DATA ALU
DATA ALU
INSTRUCTION
DATA
DATA
DATA
DATA
DATA
Introduction of GPU – Structure of GPU
5/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
2. Proposed System
6/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
Proposed System – 2x2 SM MIMO LTE system
• Block diagram of 2x2 SM MIMO LTE Downlink system
• The parallelization of signal processing algorithms appropriately for the SIMD architecture - To use GPU and CPU resources effectively
7/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
Proposed System – Implementation of Scrambling
2mod)()()(~ icibib
• Bit Scrambling – To avoid burst error that can happen during the procedure of data transmission
• Transmit bit :
• Scrambling bit :
31-bit long Pseudo-random sequence
• 10 threads for the 10 sub-frames such that 10 operations can be performed with a single instruction.
)(ib)(ic
8/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
Proposed System – Implementation of Channel Estimation
Mapping of downlink reference signals
• 2D linear interpolation using 8 reference symbols existing in 2 resource blocks
• Linear interpolation is performed along the time axis using the reference signals (3rd , 9th / 6th , 12th)
• The frequency axis using the channel estimation obtained by the time-axis interpolation
• 500 parallel operations which consists of 10 CUDA Blocks along the horizontal axis and 50 CUDA Blocks along the vertical axis.
Resource Block1 Resource Block2
Time
Frequency
9/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
Proposed System – Implemented System
2x2 SM MIMO LTE system
GPU: Modem
CPU:MediaPlayer
GPU: Modem
CPU:Streaming
Server
RFTransceiver
RFTransceiver
Terminal Computer Base-station Computer
10/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
3. Performance Evaluation
11/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
Performance Analysis – Computation Time
Transmitter Receiver
• Profiler provided by NVIDIA
12/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
Performance Analysis – Computation Time
• Specifications of LTE : 1 frame = 10ms
• Enough for the real time processing.
GPU Processing time for 1 frame
GPU Processing time
Transmitter 4199.332 µs
Receiver 7617.342 µs
13/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
5. Conclusion
14/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi
Conclusion
• GPU can be a solution because of its powerful computation capacity
• Also with the GPU, wireless communication systems can be implemented effectively for SDR.
• GPU can operate all waveform effectively.
LTE
CDMA
MobileWiMAX
GPU
15/15http://dsplab.hanyang.ac.kr, Tel : 82-2-2299-6267 Fax : 82-2-2299-6263 All rights reserved by Prof. S. Choi