Serving 22 Users in Real-Time with a 128-Antenna Massive MIMO Testbed
1. IEEE SiPS
27th October 2016, Dallas
Serving 22 Users in Real-Time with a 128-
Antenna Massive MIMO Testbed
Paul Harris
Siming Zhang, Wael Boukley Hasan, Steffen Malkowsky, Joao Vieira, Siming Zhang, Mark Beach, Liang Liu, Evangelos Mellios, Andrew
Nix, Simon Armour, Angela Doufexi, Karl Nieman, Nikhil Kundargi
Communication Systems and Networks Group
University of Bristol, Bristol, UK
http://www.bristol.ac.uk/engineering/research/csn/
2. IEEE SiPS
27th October 2016, Dallas
Summary
• System Overview
• Measurement Setup
• Experimental Results
• Conclusions
• Ongoing Work
2
3. IEEE SiPS
27th October 2016, Dallas
The Massive MIMO Concept
3
Ultimate Spatial
Resolution
• Increased spectral efficiency and network capacity
• Accurate spatial multiplexing
Time
Space
Uplink Downlink
Uplink
Uplink
Uplink
Downlink
Downlink
Downlink
Cellular View
4. IEEE SiPS
27th October 2016, Dallas
NI Based ‘BIO’ Massive MIMO test-bed
4
• 128 Programmable Radio Heads (4
racks of 32 radios)
• ‘TD-LTE’ like PHY (20 MHz BW)
• 1.2 – 6.0GHz Carrier (3.51GHz used)
• Centralised MMSE, ZF and MRC/MRT
MIMO Processing
• Supports up to 12 User Clients (Full
FPGA Processing)
• 24 user clients (decimated processing)
5. IEEE SiPS
27th October 2016, Dallas
Functional Overview
5
Distributed FPGA Processing with PCIe links Compact Computer
6. IEEE SiPS
27th October 2016, Dallas
Linear Decoding/Precoding
6
• MGS Full QR
Decomposition
• Partial parallel systolic
array
• One detection matrix
per 12 subcarrier
resource block
7. IEEE SiPS
27th October 2016, Dallas
MIMO Processor
• Wide Data Path 128 x 12 Linear Detector
• Computes 128 x 12 by 128 x 1 matrix vector
multiply in 160 ns
• 24 Million times per second
7
𝒚𝑾 𝑴𝑴𝑺𝑬
12 x 128
128 x 1
× = = 𝒖
12 x 1
32 x 1 (4)12 x 32 (4)
𝑾 𝑴𝑴𝑺𝑬 𝟎
𝑾 𝑴𝑴𝑺𝑬 𝟏
𝑾 𝑴𝑴𝑺𝑬 𝟐
𝑾 𝑴𝑴𝑺𝑬 𝟑
𝒚 𝟎
𝒚 𝟏
𝒚 𝟐
𝒚 𝟑
11. IEEE SiPS
27th October 2016, Dallas
CDF Plots of SVS
11
Scenario 1-3 in ascending order of LOS distance. 200ms capture interval for 3 minutes. Averaged across frequency.
Exploitation
of azimuth
spread Closest
scenario is
the worst for
32 elements
12. IEEE SiPS
27th October 2016, Dallas
𝑯𝑯 𝑯
for 12 users with scaled N
12
Scenario 2 (12.5m Straight Line). 200ms capture interval for 3 minutes. Averaged across frequency and time.
13. IEEE SiPS
27th October 2016, Dallas
Real-Time Channel Information
13
Eigen
Structure
Individual Spatial Stream Rx Magnitude
Power Delay profiles
Frequency Domain profiles
Fading over the
array caused by
stairwell
17. IEEE SiPS
27th October 2016, Dallas
22 Streams of 256-QAM
• With the same frame structure as before this equates
to 145.6 bits/s/Hz (uncoded sum rate of 2.915 Gbps)
17
User
Inactive
User
Inactive
131 bits/s/Hz
144 bits/s/Hz
145.7 bits/s/Hz
18. IEEE SiPS
27th October 2016, Dallas
Conclusions
• Average ratio of composite channel gain (eigenvalue) to
inter-user correlation observed to be 10 dB or more for a
ratio of up to 6:1 basestation antennas to users
• Azimuth dominated array configurations could improve
close range LOS performance
18
20. IEEE SiPS
27th October 2016, Dallas
Acknowledgements and Thanks to…
• Post Graduate Students: Wael Boukley Hasan, Siming Zhang, Henry
Brice & Benny Chitambira
• Academic Colleagues & post graduates at Lund University: Steffen
Malkowsky, Joao Vieira, Liang Liu, Ove Edfurs & Fredrik Tufvesson
• Academic Colleagues at Bristol: Mark Beach, Andrew Nix, Evangelos
Mellios, Angela Doufexi and Simon Armour
• NI Staff: Karl Nieman, Nikhil Kundargi, Ian Wong, Leif Johansson &
James Kimery
20
21. IEEE SiPS
27th October 2016, Dallas
Thank You
Any questions?
Communication Systems and Networks Group
University of Bristol, Bristol, UK
http://www.bristol.ac.uk/engineering/research/csn/