Distributed Resource Allocation in Communication Networks: From - - PowerPoint PPT Presentation
Distributed Resource Allocation in Communication Networks: From Competition to Cooperation Jianwei Huang Princeton University Seminar at Electrical Engineering Department Columbia University J. Huang (Princeton University) Distributed
Distributed Resource Allocation
1 / 60
◮ Princeton University: Mung Chiang, Vincent Poor, Robert Calderbank,
◮ Northwestern University: Michael Honig, Randall Berry, Aggelos
◮ Columbia University: Xiaodong Wang, Kai Yang ◮ California Institute of Technology: Kevin Tang ◮ Boise State University: Zhu Han ◮ University of Toronto (Canada): Wei Yu ◮ K.U.Leuven (Belgium): Marc Moonen ◮ Yonsei University (Korea): Jang-Won Lee ◮ HKUST (Hong Kong): Daniel Palomar
◮ AT&T: Russ Bellford ◮ Fraser Research Lab: Alexander Fraser ◮ Motorola: Rajeev Agrawal, Zhu Li ◮ Hamilton Institute (Ireland): Vijay Subramanian ◮ Marvell (Hong Kong): Raphael Cendrillon
Distributed Resource Allocation
2 / 60
SGSN Internal Network Inter net ISP
◮ Internet: carrying data, voice, video, ... ◮ Broadband access networks: DSL, cable, WiMAX, ... ◮ Wireless networks: cellular, Wi-Fi, bluetooth, ...
Distributed Resource Allocation
3 / 60
◮ Efficient utilization of the network resource ◮ Fair opportunities for network access
◮ Physically distributed: self-interest, difficult to control centrally ◮ Performance coupling: shared resource, mutual interactions
Distributed Resource Allocation
4 / 60
UNITED STATES
THE RADIO SPECTRUM
NON-GOVERNMENT EXCLUSIVE GOVERNMENT/NON-GOVERNMENT SHARED GOVERNMENT EXCLUSIVE RADIO SERVICES COLOR LEGEND ACTIVITY CODE NOT ALLOCATED RADIONAVIGATION FIXED MARITIME MOBILE FIXED MARITIME MOBILE FIXED MARITIME MOBILE Radiolocation RADIONAVIGATION FIXED MARITIME MOBILE Radiolocation FIXED MARITIME MOBILE FIXED MARITIME MOBILE AERONAUTICAL RADIONAVIGATION AERONAUTICAL RADIONAVIGATION Aeronautical Mobile Maritime Radionavigation (Radio Beacons) MARITIME RADIONAVIGATION (RADIO BEACONS) Aeronautical Radionavigation (Radio Beacons) 3 9 14 19.95 20.05 30 30 59 61 70 90 110 130 160 190 200 275 285 300 3 kHz 300 kHz 300 kHz 3 MHz 3 MHz 30 MHz 30 MHz 300 MHz 3 GHz 300 GHz 300 MHz 3 GHz 30 GHz Aeronautical Radionavigation (Radio Beacons) MARITIME RADIONAVIGATION (RADIO BEACONS) Aeronautical Mobile Maritime Radionavigation (Radio Beacons) AERONAUTICAL RADIONAVIGATION (RADIO BEACONS) AERONAUTICAL RADIONAVIGATION (RADIO BEACONS) Aeronautical Mobile Aeronautical Mobile RADIONAVIGATION AERONAUTICAL RADIONAVIGATION MARITIME MOBILE Aeronautical Radionavigation MOBILE (DISTRESS AND CALLING) MARITIME MOBILE MARITIME MOBILE (SHIPS ONLY) MOBILE AERONAUTICAL RADIONAVIGATION (RADIO BEACONS) AERONAUTICAL RADIONAVIGATION (RADIO BEACONS) BROADCASTING (AM RADIO) MARITIME MOBILE (TELEPHONY) MARITIME MOBILE (TELEPHONY) MOBILE (DISTRESS AND CALLING) MARITIME MOBILE LAND MOBILE MOBILE FIXED STANDARD FREQ. AND TIME SIGNAL (2500kHz) STANDARD FREQ. AND TIME SIGNAL Space Research MARITIME MOBILE LAND MOBILE MOBILE FIXED AERONAUTICAL MOBILE (R) STANDARD FREQ. AERONAUTICAL MOBILE (R) AERONAUTICAL MOBILE (OR) AERONAUTICAL MOBILE (R) FIXED MOBILE** Radio- location FIXED MOBILE* AMATEUR FIXED FIXED FIXED FIXED FIXED MARITIME MOBILE MOBILE* MOBILE* MOBILE STANDARD FREQ. AND TIME SIGNAL (5000 KHZ) AERONAUTICAL MOBILE (R) AERONAUTICAL MOBILE (OR) STANDARD FREQ. Space Research MOBILE** AERONAUTICAL MOBILE (R) AERONAUTICAL MOBILE (OR) FIXED MOBILE* BROADCASTING MARITIME MOBILE AERONAUTICAL MOBILE (R) AERONAUTICAL MOBILE (OR) FIXED Mobile AMATEUR SATELLITE AMATEUR AMATEUR FIXED Mobile MARITIME MOBILE MARITIME MOBILE AERONAUTICAL MOBILE (R) AERONAUTICAL MOBILE (OR) FIXED BROADCASTING FIXED STANDARD FREQ. AND TIME SIGNAL (10,000 kHz) STANDARD FREQ. Space Research AERONAUTICAL MOBILE (R) AMATEUR FIXED Mobile* AERONAUTICAL MOBILE (R) AERONAUTICAL MOBILE (OR) FIXED FIXED BROADCASTING MARITIME MOBILE AERONAUTICAL MOBILE (R) AERONAUTICAL MOBILE (OR) RADIO ASTRONOMY Mobile* AMATEUR BROADCASTING AMATEUR AMATEUR SATELLITE Mobile* FIXED BROADCASTING STANDARD FREQ. AND TIME SIGNAL (15,000 kHz) STANDARD FREQ. Space Research FIXED AERONAUTICAL MOBILE (OR) MARITIME MOBILE AERONAUTICAL MOBILE (OR) AERONAUTICAL MOBILE (R) FIXED FIXED BROADCASTING STANDARD FREQ. Space Research FIXED MARITIME MOBILE Mobile FIXED AMATEUR AMATEUR SATELLITE BROADCASTING FIXED AERONAUTICAL MOBILE (R) MARITIME MOBILE FIXED FIXED FIXED Mobile* MOBILE** FIXED STANDARD FREQ. AND TIME SIGNAL (25,000 kHz) STANDARD FREQ. Space Research LAND MOBILE MARITIME MOBILE LAND MOBILE MOBILE** RADIO ASTRONOMY BROADCASTING MARITIME MOBILE LAND MOBILE FIXED MOBILE** FIXED MOBILE** MOBILE FIXED FIXED FIXED FIXED FIXED LAND MOBILE MOBILE** AMATEUR AMATEUR SATELLITE MOBILE LAND MOBILE MOBILE MOBILE FIXED FIXED MOBILE MOBILE FIXED FIXED LAND MOBILE LAND MOBILE LAND MOBILE LAND MOBILE Radio Astronomy RADIO ASTRONOMY LAND MOBILE FIXED FIXED MOBILE MOBILE MOBILE LAND MOBILE FIXED LAND MOBILE FIXED FIXED MOBILE MOBILE LAND MOBILE AMATEUR BROADCASTING (TV CHANNELS 2-4) FIXED MOBILE FIXED MOBILE FIXED MOBILE FIXED MOBILE AERONAUTICAL RADIONAVIGATION BROADCASTING (TV CHANNELS 5-6) BROADCASTING (FM RADIO) AERONAUTICAL RADIONAVIGATION AERONAUTICAL MOBILE (R) AERONAUTICAL MOBILE AERONAUTICAL MOBILE AERONAUTICAL MOBILE (R) AERONAUTICAL MOBILE (R) AERONAUTICAL MOBILE (R) MOBILE FIXED AMATEUR BROADCASTING (TV CHANNELS 7-13) MOBILE FIXED MOBILE FIXED MOBILE SATELLITE FIXED MOBILE SATELLITE MOBILE FIXED MOBILE SATELLITE MOBILE FIXED MOBILE AERONAUTICAL RADIONAVIGATIONALLOCATIONS FREQUENCY
BROADCASTING FIXED MOBILE* BROADCASTING FIXED BROADCASTING FIXED Mobile FIXED BROADCASTING BROADCASTING FIXED FIXED BROADCASTING FIXED BROADCASTING FIXED BROADCASTING FIXED BROADCASTING FIXED BROADCASTING FIXED BROADCASTING FIXED FIXED FIXED FIXED FIXED FIXED LAND MOBILE FIXED AERONAUTICAL MOBILE (R) AMATEUR SATELLITE AMATEUR MOBILE SATELLITE (E-S) F I X E D F i x e d M o b i l e R a d i o - l o c a t i o n F I X E D M O B I L E LAND MOBILE MARITIME MOBILE FIXED LAND MOBILE FIXED LAND MOBILE RADIONAV-SATELLITE FIXED MOBILE FIXED LAND MOBILEDistributed Resource Allocation
5 / 60
Distributed Resource Allocation
6 / 60
Distributed Resource Allocation
7 / 60
◮ Distributed ◮ Low complexity ◮ Local computation ◮ Limited or no message passing
Distributed Resource Allocation
8 / 60
◮ Optimization theory ◮ Game theory ◮ Distributed computation and control ◮ Microeconomics
◮ Realistic network deployments and tests ◮ Engineering problem structure
Distributed Resource Allocation
9 / 60
Distributed Resource Allocation
10 / 60
Distributed Resource Allocation
11 / 60
Distributed Resource Allocation
12 / 60
◮ Licensed bands: IEEE 802.22 (Television band) ◮ Unlicensed bands: IEEE 802.19
◮ Emergency networks ◮ Military networks
Distributed Resource Allocation
13 / 60
Distributed Resource Allocation
14 / 60
◮ Maximum allowed interference measured at a measurement point ◮ Equivalent to a total received power constraint
Distributed Resource Allocation
14 / 60
Distributed Resource Allocation
15 / 60
3 1
2
3 2
1
◮ Transmission power: pn ◮ Channel gain: hn ◮ Signal-to-interference plus noise ratio (SINR):
B
Distributed Resource Allocation
16 / 60
n
U (SINR )
n
SINRn
Distributed Resource Allocation
17 / 60
◮ Maximizing total weighted rate
Distributed Resource Allocation
18 / 60
◮ Weighted max-min fair
Distributed Resource Allocation
19 / 60
◮ SINR and utility may not be concave in power
◮ Local information: utility functions, channel gains ◮ Selfish objectives
◮ Mutual interference ◮ Shared received power at measurement point
Distributed Resource Allocation
20 / 60
◮ SINR and utility may not be concave in power
◮ Local information: utility functions, channel gains ◮ Selfish objectives
◮ Mutual interference ◮ Shared received power at measurement point
◮ Distributed in nature ◮ Capture interactions between users
Distributed Resource Allocation
20 / 60
Distributed Resource Allocation
21 / 60
◮ Highest bidder gets the good and pays the bid
Distributed Resource Allocation
21 / 60
◮ Highest bidder gets the good and pays the bid
◮ Good: resource ◮ Auctioneer (manager): representing seller of the good ◮ Bidders (users): buyers of the good
Distributed Resource Allocation
21 / 60
◮ Highest bidder gets the good and pays the bid
◮ Good: resource ◮ Auctioneer (manager): representing seller of the good ◮ Bidders (users): buyers of the good
◮ Bids: what the bidders submit to the auctioneer ◮ Allocation: how auctioneer allocates the good to the bidders ◮ Payments: how the bidders pay the auctioneer
Distributed Resource Allocation
21 / 60
◮ Highest bidder gets the good and pays the bid
◮ Good: resource ◮ Auctioneer (manager): representing seller of the good ◮ Bidders (users): buyers of the good
◮ Bids: what the bidders submit to the auctioneer ◮ Allocation: how auctioneer allocates the good to the bidders ◮ Payments: how the bidders pay the auctioneer
◮ Indivisible auction ◮ Divisible auction: suitable for communication resource allocation
Distributed Resource Allocation
21 / 60
Distributed Resource Allocation
22 / 60
Distributed Resource Allocation
22 / 60
Distributed Resource Allocation
22 / 60
Distributed Resource Allocation
22 / 60
Distributed Resource Allocation
23 / 60
Distributed Resource Allocation
23 / 60
◮ Weighted proportional allocation rule ◮ Positive reserve bid makes sure that the values of bids count, not just
Distributed Resource Allocation
23 / 60
◮ User-centric payment ◮ Proportional to user’s achieved QoS (SINR) ◮ Leads to fair allocation
◮ Network-centric payment ◮ Proportional to the allocated resource (power) ◮ Leads to efficient allocation
Distributed Resource Allocation
24 / 60
◮ User’s payoff (benefit) = utility - payment ◮ Both utility and payment depend on bn and b−n (bm, ∀m = n)
Distributed Resource Allocation
25 / 60
◮ User’s payoff (benefit) = utility - payment ◮ Both utility and payment depend on bn and b−n (bm, ∀m = n)
◮ Best response: Bn(b−n)
bn [Un(SINRn(bn; b−n)) − Cn (π, bn; b−n)]
Distributed Resource Allocation
25 / 60
◮ User’s payoff (benefit) = utility - payment ◮ Both utility and payment depend on bn and b−n (bm, ∀m = n)
◮ Best response: Bn(b−n)
bn [Un(SINRn(bn; b−n)) − Cn (π, bn; b−n)]
◮ Nash Equilibrium (N.E.): b∗ = {b∗
n, ∀n}
◮ Fixed point solution of all users’ best responses
n = B
−n
◮ Stable outcome of the game
Distributed Resource Allocation
25 / 60
Distributed Resource Allocation
26 / 60
Distributed Resource Allocation
27 / 60
Distributed Resource Allocation
28 / 60
Distributed Resource Allocation
29 / 60
Distributed Resource Allocation
29 / 60
1
Distributed Resource Allocation
30 / 60
1
2
Distributed Resource Allocation
30 / 60
1
2
3
Distributed Resource Allocation
30 / 60
1
2
3
4
Distributed Resource Allocation
30 / 60
1
2
3
4
5
Distributed Resource Allocation
30 / 60
1
2
3
4
5
6
Distributed Resource Allocation
30 / 60
1
2
3
4
5
6
7
Distributed Resource Allocation
30 / 60
1
2
3
4
5
6
7
8
Distributed Resource Allocation
30 / 60
Distributed Resource Allocation
31 / 60
Distributed Resource Allocation
32 / 60
Distributed Resource Allocation
33 / 60
1
2
Distributed Resource Allocation
33 / 60
◮ Example: logarithmic utility Un(SINRn) = log(SINRn)
Distributed Resource Allocation
34 / 60
◮ Example: logarithmic utility Un(SINRn) = log(SINRn)
1
2
Distributed Resource Allocation
34 / 60
Distributed Resource Allocation
35 / 60
Distributed Resource Allocation
36 / 60
Distributed Resource Allocation
37 / 60
Distributed Resource Allocation
37 / 60
◮ Common information: P, n0 and π. ◮ Local information: Un, hn and SINR(t−1)
n
Distributed Resource Allocation
37 / 60
◮ Only limited explicit message passing: ⋆ bid (user to manager) and resource allocation (manager to user) ◮ No need to know anything about other users
Distributed Resource Allocation
38 / 60
◮ Only limited explicit message passing: ⋆ bid (user to manager) and resource allocation (manager to user) ◮ No need to know anything about other users
◮ Similar arguments also apply for the power auction ◮ Only works for specific utility functions ⋆ Examples: log(1 + SINRn) and log(SINRn)
Distributed Resource Allocation
38 / 60
Distributed Resource Allocation
39 / 60
Distributed Resource Allocation
40 / 60
5 10 15 20 16 17 18 19 20 21 22 23 24 Iterations Achieved SINR (dB) user 1 user 2 user 3
Distributed Resource Allocation
40 / 60
◮ SINR auction: fair allocation ◮ Power auction: efficient allocation ◮ Large system: two auctions are both efficient
Distributed Resource Allocation
41 / 60
Distributed Resource Allocation
42 / 60
Distributed Resource Allocation
43 / 60
Distributed Resource Allocation
43 / 60
◮ Multi-carrier transmissions: Discrete Multitone Modulation
Distributed Resource Allocation
44 / 60
Distributed Resource Allocation
45 / 60
Distributed Resource Allocation
45 / 60
Distributed Resource Allocation
45 / 60
Distributed Resource Allocation
45 / 60
◮ RT generates strong crosstalk to CO line, especially in high tones ◮ CO generates little crosstalk to RT in all tones
Distributed Resource Allocation
46 / 60
Distributed Resource Allocation
46 / 60
◮ No multi-user joint decoding
Distributed Resource Allocation
46 / 60
◮ No multi-user joint decoding
Distributed Resource Allocation
46 / 60
◮ No multi-user joint decoding
◮ Non-convexity: total weighted rate not concave in power. ◮ Physically distributed: local channel information ◮ Performance coupling: across users (interferences) and tones (power
Distributed Resource Allocation
47 / 60
◮ Non-convexity: total weighted rate not concave in power. ◮ Physically distributed: local channel information ◮ Performance coupling: across users (interferences) and tones (power
◮ Static channels ◮ Multi-carrier transmissions ◮ Typical network topology ◮ Unique channel frequency responses with good empirical models ◮ Cannot decode explicit message passing
Distributed Resource Allocation
47 / 60
◮ Non-convexity: total weighted rate not concave in power. ◮ Physically distributed: local channel information ◮ Performance coupling: across users (interferences) and tones (power
◮ Static channels ◮ Multi-carrier transmissions ◮ Typical network topology ◮ Unique channel frequency responses with good empirical models ◮ Cannot decode explicit message passing
Distributed Resource Allocation
47 / 60
◮ IW: Iterative Water-filling [Yu, Ginis, Cioffi’02]
Distributed Resource Allocation
48 / 60
◮ IW: Iterative Water-filling [Yu, Ginis, Cioffi’02] ◮ OSB: Optimal Spectrum Balancing [Cendrillon et al.’04] ◮ ISB: Iterative Spectrum Balancing [Liu, Yu’05] [Cendrillon, Moonen’05]
Distributed Resource Allocation
48 / 60
◮ IW: Iterative Water-filling [Yu, Ginis, Cioffi’02] ◮ OSB: Optimal Spectrum Balancing [Cendrillon et al.’04] ◮ ISB: Iterative Spectrum Balancing [Liu, Yu’05] [Cendrillon, Moonen’05] ◮ ASB: Autonomous Spectrum Balancing [Huang et al.’06]
Distributed Resource Allocation
48 / 60
◮ A virtual line representative of the typical victim in the network ◮ Good choice: the longest CO line ◮ Parameters (power, noise, crosstalk) are publicly known
Distributed Resource Allocation
49 / 60
Distributed Resource Allocation
50 / 60
Distributed Resource Allocation
50 / 60
n
Distributed Resource Allocation
51 / 60
n
n
◮ Locally computable without explicit message passing
Distributed Resource Allocation
51 / 60
Distributed Resource Allocation
52 / 60
◮ Duality gap is zero (under large number of tones) ◮ Satisfied in real DSL networks (ADSL: 256 tones, VDSL: 4096 tones) ◮ Can be solved using dual decomposition
Distributed Resource Allocation
52 / 60
n
n
n,m/hk n,npk m − σk n
+
◮ Reference line is not active in high frequency tones
Distributed Resource Allocation
53 / 60
n
n
n,m/hk n,npk m − σk n
+
◮ Reference line is not active in high frequency tones
Distributed Resource Allocation
53 / 60
n
n
n,m/hk n,npk m − σk n
+
◮ Reference line is not active in high frequency tones
Distributed Resource Allocation
53 / 60
Distributed Resource Allocation
54 / 60
1
2
3
◮ Parallel updates: more realistic with cleaner proof.
Distributed Resource Allocation
55 / 60
Distributed Resource Allocation
56 / 60
Distributed Resource Allocation
56 / 60
Distributed Resource Allocation
57 / 60
Distributed Resource Allocation
58 / 60
◮ Mutual interference ◮ Integer constraints of channel utilization in OFDMA systems ◮ Discrete choices of multimedia coding and content selection
◮ Local utility information ◮ Local channel/buffer information ◮ Local multimedia content information ◮ Self interests
◮ Mutual interference ◮ Shared total received power ◮ Shared total downlink transmission power ◮ Shared relay resources or source budgets ◮ Correlation between traffic arrival and departure
Distributed Resource Allocation
59 / 60
Distributed Resource Allocation
60 / 60