Presentation@ICC15:
To discover the function splitting plan with best computation-fronthauling tradeoff, we formulate the problem as a graph clustering problem and develop a genetic algorithm to solve it.
Paper: Graph-based Framework for Flexible Baseband Function Splitting and Placement in C-RAN
Boost Fertility New Invention Ups Success Rates.pdf
Graph-based Framework for Function Splitting in CRAN
1. Graph-based Framework for
Flexible Baseband Function Splitting
Jingchu LIU*, Sheng ZHOU*, Jie GONG*, Zhisheng NIU*, and Shugong XU+
Network Integration for Ubiquitous Linkage and Broadband (NiuLab)
Tsinghua University, Beijing, China *
and
Intel Labs, Beijing, China +
2. Contents
• Fronthaul Bottleneck of C-RAN
• Baseband Compression via Function Splitting
• Graph-based Framework for BB function splitting
• Customized Genetic Algorithm
• Simulation Results
• Future Work
2
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
3. Contents
• Fronthaul Bottleneck of C-RAN
• Baseband Compression via Function Splitting
• Graph-based Framework for BB function splitting
• Customized Genetic Algorithm
• Simulation Results
• Future Work
3
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
4. FH Bottleneck of C-RAN
• C-RAN and Fronthaul (FH)[1]
• C-RAN centralizes BB processing from radio site
• FH carries BB samples (I/Q data) btw. RRH and BBU
• Reduce cost, ease implementation of cooperative technology, flexibility
• FH Bottleneck[2]
• Bandwidth: 10Gbps/ant. for typical LTE UL/DL (100Mbps backhaul)
• Delay: 3ms maximum processing latency for LTE HARQ -> maximum
500us fronthaul transmission dealy
• Synchronization: ~1ppb frequency accuracy, ~1us phase difference
• C-RAN is sound, but FH is (too) expensive
4
FH = Fronthaul BB = Baseband
RRH = Remote Radio Head BBU = Baseband Unit
[1] CMRI. C-RAN: The road towards green RAN
[2] CPRI Specification, “interface specification,” 2013.
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
5. Contents
• Fronthaul Bottleneck of C-RAN
• Baseband Compression via Function Splitting
• Graph-based Framework for BB function splitting
• Customized Genetic Algorithm
• Simulation Results
• Future Work
5
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
6. BB Compression via Function Splitting
• Transparent BB compression[3]
• Time-domain or frequency domain
• Compress at RRH(BBU), reconstruct at BBU(RRH)
• Transparent: no changes needed for BBU/RRH
6
[3] Samardzija, “Compressed Transport of Baseband Signals in Radio Access Networks.,“ TWC 2012
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
7. BB Compression via Function Splitting
• Transparent BB compression
• Around 3x compression at the cost of distortion (EVM<1.5%)
• Enough?
7
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
8. BB Compression via Function Splitting
• Transparent BB compression
• Compression via function splitting[4][5]
• Preprocessing at remote site to reduce FH bandwidth
• … “move” some BB processing from data center back to radio site
8
[4] Lorca and Cucala. “Lossless compression technique for the fronthaul of LTE/LTE-advanced
cloud-RAN architectures,” in WoWMoM 2013
[5] Dotsch, "Quantitative Analysis of Split Base Station Processing and Determination of
Advantageous Architectures," Bell Labs Journal, 2013
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
9. BB Compression via Function Splitting
• Transparent BB compression
• Compression via function splitting
• Why it works?
9
200+ times
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
10. BB Compression via Function Splitting
• Transparent BB compression
• Compression via function splitting
• Why FLEXIBLE function splitting[6][7][8]
• Various implementation constraint (BW, delay)
• Continuous evolvement of PHY technology (massive MIMO)
• Unpredictable service requirements (sub-ms latency in 5G)
• Decide splitting option “on-the-fly”!
• Need a framework for deciding the “best split”
10
Research Question:
How to decide the best BB function splitting scheme?
[6] P. Rost , et al., “Cloud technologies for flexible 5G Radio Access Networks”, Comm. Mag. May 2014
[7] D.Wubben , et al., Benefits and Impact of Cloud Computing on 5G Signal Processing, SPM, Nov 2014
[8] J. Liu, et al., “CONCERT: A Cloud-Based Architecture for Next-generation Cellular Systems”, IEEE WCM, Dec 2014
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
11. Contents
• Fronthaul Bottleneck of C-RAN
• Baseband Compression via Function Splitting
• Graph-based Framework for BB function splitting
• Customized Genetic Algorithm
• Simulation Results
• Future Work
11
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
12. Graph-Based Framework for BB Function Splitting
• BB processing structure and graph
• Basic processing structures are chains
• Cooperative processing structures like CoMP may form loops
12
Typical baseband processing structure of LTE BS
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
13. Graph-Based Framework for BB Function Splitting
• BB processing structure and graph
• Basic processing structures are chains
• Cooperative processing structures like CoMP may form loops
• General BB processing structures are graphs :
• Each node : BB processing function
• Each (directed) link : logical link btw. BB processing fcn.
• Node weight : computational complexity of a node
• Link weight : fronthaul BW requirement
• Sources and sinks terminate BB processing chain (path)
13
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
14. Graph-Based Framework for BB Function Splitting
• BB processing structure and graph
• Function splitting and graph clustering
• Graph-clustering: group nodes by “similarity”
• Here, places to place BB functions define cluster
• Therefore, where to place a function = which cluster to assign a node
14
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
15. Graph-Based Framework for BB Function Splitting
• BB processing structure and graph
• Function splitting and graph clustering
• Computational cost of cluster i w.r.t scheme
• Fronthauling cost between clusters i, j w.r.t scheme
• Path delay constraint
15
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
16. Graph-Based Framework for BB Function Splitting
• BB processing structure and graph
• Function splitting and graph clustering
• Graph clustering is multi-obj. combinatorial opt.
• Clustering schemes can be represented with integer valued vectors
• Overall formulation is 2-obj. opt. with delay constraint
16
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
17. Contents
• Fronthaul Bottleneck of C-RAN
• Baseband Compression via Function Splitting
• Graph-based Framework for BB function splitting
• Customized Genetic Algorithm
• Simulation Results
• Future Work
17
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
18. Customized Genetic Algorithm (GA)
• Natural encoding
• Use to represent the chromosome
18
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
19. Customized Genetic Algorithm (GA)
• Natural encoding
• Linearly combined and delay punished fitness fcn.
• Computational cost + fronthauling cost + delay penalty
19
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
20. Customized Genetic Algorithm (GA)
• Natural encoding
• Linearly combined and delay punished fitness fcn.
• Dispersive crossover
• 2 Parents generate 1 offspring
• Genes are randomly passed down from parents
20
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
21. Customized Genetic Algorithm (GA)
• Natural encoding
• Linearly combined and delay punished fitness fcn.
• Dispersive crossover
• Graph-based mutation
• ONLY mutate to connected clusters
• Graph seeding
• Some modules can not be placed arbitrarily in reality, e.g. antennas
• Seed nodes are assigned with fixed cluster indices
• DO NOT take part in crossover or mutation
21
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
22. Contents
• Fronthaul Bottleneck of C-RAN
• Baseband Compression via Function Splitting
• Graph-based Framework for BB function splitting
• Customized Genetic Algorithm
• Simulation Results
• Future Work
22
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
23. Simulation Results
• Simulation Setup
• BB processing structure, NOTE for CoMP
• Link weight (fronthauling cost)
23
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
24. Simulation Results
• Simulation Setup
• Node weights (computational cost)
24
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
25. Simulation Results
• Simulation Setup
• Exponential cost and productive delay fcn.
25
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
26. Simulation Results
• Simulation Setup
• GA parameters
26
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
27. Simulation Results
• Tradeoff between computational and fronthauling cost
27
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
28. Simulation Results
• Tradeoff between computational and fronthauling cost
28
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
29. Simulation Results
• Cooperative processing (like CoMP)
29
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
30. Simulation Results
• Delay constraint
30
Delay tolerance decrease
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
31. Contents
• Fronthaul Bottleneck of C-RAN
• Baseband Compression via Function Splitting
• Graph-based Framework for BB function splitting
• Customized Genetic Algorithm
• Simulation Results
• Future Work
31
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
32. Future Work
• Realistic simulation parameters
• Analytic derivation of optimal split schemes
• Information-theoretic interpretation of BB split
32
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting
33. Thanks
• Any problems / suggestions?
33
Jingchu Liu
Tsinghua University
Graph-based framework for flexible
baseband function splitting