1. Research: from academia to industry
Jinliang Huang
AAA

2. Jinliang Huang Academia Industry Industry What is research like M.Sc.
1999
B.S.
Zhejiang University
2003
M.Sc.
KTH
2005
Ph.D.
2009
Ericsson
2014
Huawei
Academia
Industry
Industry
What is
research like

3. Content Introduction to Huawei Sweden Algorithm Group
Research in industry
Examples
Link adaptation
Channel estimation
MMSE Equalization
CoMP (Coordinated Multiple-Point) reception

4. The Algorithm Group Huawei Sweden
Gunnar Peters
Technical lead
5G RRM
RRM architecture
Machine Learning
(Pablo Soldati, KTH)
4G RRM
Coordinated scheduling
Small packet optimization
(Xiaojia Lu, CWC Oulo)
5G Baseband Receiver
L1 algorithms
Mac / Phy codesign
(Jinliang Huang, KTH)

5. Research in industry Performance VS. Complexity Implementaion friendly
Optimal algorithm/performance is always too complicated
Implementaion friendly
SW: low computational efficiency
HW: low flexibility
Priori information of parameter
Statistics assumptions
Stability
Stable performance/complexity
Academia
Industry

6. Interview
Q: What do you think is the biggest difference between academia and industry?
”Complexity is an important factor that always needs to be considered in industry.”
Ahmet

7. Link adaptation in Uplink
How to approach Shannon capacity?
MCS selection
Receiver design
Have we achieved the Shannon capacity?
No!
Why is link adaptation difficult?
Actual operating area

8. Link adaptation in Uplink
Ideal link adaptation
Doppler spread = 20Hz
Noise limited
Time of channel estimation
Time of transmission
Procedure:
Channel est/pred
Post equalizer SINR estimation
MCS selection
AWGN
Constraints
Doppler spread
Frequency selective scheduling
Resource for channel pred.
SINR estimation error
Quantization in MCS
10%
Post eq. SINR

9. Link adaptation in Uplink
Ideal link adaptation cont’d
Interference limited, MMSE equalizer
Constraints
Unknown interference
Known interference: complexity
SINR estimation is sensitive to phase error
Solution: outer loop link adaptaion

10. Receiver design What are we interested in today? Channel estimation
Equalization
Demapper +decoder

11. Channel Estimation
Pilot allocation

12. Channel Estimation
Frequency domain processing

13. Channel Estimation MMSE-based method[1]: Long term estimation
[1] J.-J. van de Beek, O. Edfors, M. Sandell, S. Wilson, P. Borjesson, "On channel estimation in ofdm systems", vol. 2, pp , jul

14. Channel Estimation ZF + Transformed domain + windowing IFFT

15. Channel estimation Time domain processing Wiener filter Doppler spread
data symbols
Wiener filter
Estimation
error
Doppler spread
Limited data in frequency
Limited data in time

16. MMSE equalization Multi-antenna combining Receiving BS Serving UE #1
Neighboring UE #3
Neighboring UE #2
Multi-antenna combining
signal
Interference
Coordinated cells
Non-coordinated cells
Information exchange
More channel estimation
MIMO processing
No information exchange
Less channel estimation
SIMO processing

17. Is MU-MIMO really good in UL?
Total capacity
UE #2 performance
MLD is not possible
Multi-user interference due to linear equalizer
Receiving BS
MU-MIMO
SIMO
Serving UE #1
Serving UE #2
Serving UE #3
Coverage issue
Celll average performance

18. CoMP reception To improve the coverage MMSE equalizer interference
noise
To improve the coverage
MMSE equalizer
interference

19. CoMP reception In ideal situation The more, the better! Post SINR:
In case
The more, the better!

20. CoMP reception In real situation Constraint:
Unbalanced
SNR
In real situation
Constraint:
joint detection is dependent on channel estimation

21. Key Takeaways Find out the physical constraints
Be careful with the assumptions
Know the tradeoff between performance and complexity

22. Any questions? Jinliang.huang@huawei.com Skalholtsgatan 9,
Kista, Sweden

23. AAA