ProBeam: A Prac,cal Mul,cell Beamforming System for Small- cell Networks

Transcription

1 ProBeam: A Prac,cal Mul,cell Beamforming System for Small- cell Networks Jongwon Yoon Karthik Sundaresan Mohammad Khojastepour U. Wisconsin- Madison NEC Labs NEC Labs Sampath Rangarajan NEC Labs Suman Banerjee U. Wisconsin- Madison ProBeam / MobiHoc

2 Mo,va,on for small- cells Prolifera,on of handheld devices Traffic des,ned to mobile devices rapidly increases Cisco Visual Networking Index: Small- cells Global Mobile to scale Data Traffic network Forecast Update, capacity ProBeam / MobiHoc

3 Why small- cell? Macrocell Short range, low cost and low power AP Reuse macro spectrum Small- cell Benefits Ø Coverage (higher quality like) Ø Spectral efficiency Ø Lower power ProBeam / MobiHoc

4 Challenges in small- cell networks c1 Cell1 c2 Cell2 Dense and un- coordinated deployment Ø High interference - > degraded performance ProBeam / MobiHoc

5 Solu,on1: Resource isola,on c1 Cell1 c2 Cell2 Resource alloca7on for cell1 Resource alloca7on for cell2 Freq. Time Time or Frequency resource isolation Freq. Time ProBeam / MobiHoc

6 Solu,on2: Spa,al reuse Beamforming c1 Cell1 c2 Cell2 The abililty to transmit energy in specific direc,ons Weigh,ng the signals transmieed from an antenna array Freq. Time Space (spatial reuse) Freq. Time ProBeam / MobiHoc

7 Prior works Time or Frequency resource isolation Efficient resource management scheme FERMI [MobiCom 11] RADION [MobiHoc 12] Space (spatial reuse) Beamforming DIRC [SigComm 09] Speed [MobiCom 10] Not consider client associa7on Per- pkt based conflict graph Reuse schedule determina7on Not practical in OFDMA ProBeam / MobiHoc

8 Goal Our focus leverage beamforming to maximize spa,al reuse with small- cells Prac,cal constraints (beam paeerns) change at coarse,me scales (not per- frame due to overhead, complexity and feasibility) depend on the set of client associated with the cell (influences SNR/interference differently to clients) depend on the throughput received - - >,ed to scheduling and fairness model assumed ProBeam / MobiHoc

9 Outline Challenges in small- cell networks Mo,va,on Coordinated beamforming Joint client associa,on ProBeam SINR es,ma,on Joint beam selec,on and client associa,on Evalua,on and conclusion ProBeam / MobiHoc

10 Mo,va,on1: Coordinated beamforming Benefits Increase link capacity via improved SNR Increase network capacity via reduced interference - > higher spa,al reuse ProBeam / MobiHoc

11 Observa,on1 fixed C1 BS1 C2 BS2 cycle through Two cell network Aggregated Throughput (Mbps) BS2 Beam Pattern BS1 BS2 Interference C1 achieves projected highest by throughput neighboring when BS BS2 depends employs on the 9, beam lowest selection when 16 th ProBeam / MobiHoc

12 Observa,on2 Aggregated Throughput (Mbps) Max. aggregated Max. C BS2 Beam Pattern BS1 BS2 Aggregated Throughput (Mbps) Max. aggregated Max. C1 BS1 BS2 BS BS2, BS3 Beam Pattern Pattern maximizing local throughput differs from Well-coordinated beamforming increases 40% the one for aggregated throughput ProBeam / MobiHoc

13 Mo,va,on2: Joint client associa,on How to associate clients? SNR- based C2 C1 BS1 C3 BS2 Flexible C2 C1 BS1 C3 BS2 ProBeam / MobiHoc

14 Throughput gain Throughput (Mbps) Topology SNR-based Flexible Jointly As high address as 40% client throughput association gain, with 25% beam on average selection ProBeam / MobiHoc

15 Outline Challenges in small- cell networks Mo,va,on Coordinated beamforming Joint client associa,on ProBeam SINR es,ma,on Joint beam selec,on and client associa,on Evalua,on and conclusion ProBeam / MobiHoc

16 Problem Large- scale enterprise/outdoor deployment of small- cells Problem CC 1. Selec,ng beam paeerns for each small- cell to increase the spa,al reuse and improve the link quality 2. Op,mizing the client associa,on with given small- cells Hardness Selec,ng beam paeern: O(k n k: # of beam panerns ) n: # of small- cells Finding op,mal associa,on: O(n m ) m: # of clients Finding both beam paeern for each cell and client associa,ons are not easy Decouple beam selection from client association ProBeam / MobiHoc

17 ProBeam Large- scale enterprise/outdoor deployment of small- cells ProBeam central controller (CC) Interference es,ma,on for beamforming from each BS Joint beam selec,on and client associa,on Scheduling CC ProBeam / MobiHoc

18 Key component1: SINR es,ma,on SINR measurement: O(k n ) required ProBeam: SINR es,ma,on Measures SNR, INR from all cells, O(kn) SINR i, j = k i SNR i, j INR k, j +1 (INR +1 INR) SINR(dB) = SNR(dB) INR(dB) k: # of beam panerns n: # of small- cells SINR can be es7mated from its SNR from the desired BS and aggregated INR from all interfering BSs ProBeam / MobiHoc

19 SINR es,ma,on SINR es7ma7on verifica7on 95% of es7ma7on 1dB error Signal Strength (db) SNR(bs) SNR(intf) INR(est) offset Intf(A) Intf(B) Intf(A+B) CDF SINR(meas) - SINR(est) (db) SINR i, j,π (db) = SNR i, j,π (i) (db) 10log 10 ( SNR k, j,π (k ) ) + β(db) k i ProBeam / MobiHoc

20 Key component2: CABS algorithm Joint client associa,on and beam selec,on Input: client s SNR measurements Output: beam selec,on and client associa,on for each cell; change the beam paeern for each NP-hard client; find problem the associa,on with given beam paeerns Maximize CABS u,lity is in a every ½-approximation,me epoch; algorithm (π *, X * under proportional fairness ) = argmax π,x x j,i U(t π j,i ) s.t However, its performance is within 96% x j,i of 1, j K j K i S i S upper bound If u,lity is not increasing; skip this clients π (i) X = {x j,i } K: set of clients, S: set of small- cell BSs *propor,onal fairness (avg. throughput) U(t π j,i ) = log(t π j,i ) ProBeam / MobiHoc

21 Outline Challenges in small- cell networks Mo,va,on Coordinated beamforming Joint client associa,on ProBeam SINR es,ma,on Joint beam selec,on and client associa,on Evalua,on and conclusion ProBeam / MobiHoc

22 Implementa,on Implemented on PicoChip WiMAX testbed Each BS has an eight element phased array antenna One set of small- cell network GPS module Small- cell BS High gain panern Low side- lobe Smart antenna Client ProBeam / MobiHoc

23 Testbed 4 Small- cells with direc,onal antenna 16 direc,onal paeerns + 1 omni paeern 30 Client loca,ons 4x17=68 measurements at each loca,on ProBeam / MobiHoc

24 Throughput gain - Decoupled: find best beam panern with given associa7on - CABS- all: associate all clients - UB- beam: exhaus7ve beam search O(k n ) Throughput (Mbps) Decoupled CABS-all within 96% CABS UB-beam Throughput (Mbps) Decoupled CABS-all CABS UB-beam 115% gain Number of Clients Number of Base-Stations - CABS Removing performance bottleneck is within clients 96% from of scheduling upper bound - Provides in an a epoch net gain improves of 115% the over spatial decoupled reuse approach ProBeam / MobiHoc

25 Fairness Load balancing: fairness index among the number of clients associated with each BS BS Load Balancing Decoupled CABS-all CABS UB-beam CABS is close to Number of Clients ProBeam / MobiHoc

26 U,lity / Fairness (associa,on) U,lity: captures both throughput and fairness Utility Decoupled CABS-all CABS UB-beam very close to UB w/ low complexity 46% higher than CABS- all Client Fraction UB-beam CABS CABS: only small frac7ons (10-20%) are bypassed UB- beam: more aggressive, but marginal gain Number of Clients Number of Clients Maximizing reuse gains without compromising on fairness ProBeam / MobiHoc

27 Conclusion Decoupling beamforming from client scheduling is necessary Jointly address client associa,on with beamforming Implemented on a real WiMAX testbed with commercial, off- the- shelf clients Applicable to LTE and LTE- A ProBeam / MobiHoc

28 Thank you! Ques,ons? ProBeam / MobiHoc