Power Consumption by Wireless Communication. Lin Zhong ELEC518, Spring 2011

Transcription

1 Power Consumption by Wireless Communication Lin Zhong ELEC518, Spring 2011

2 Power consumption (SMT5600) Cellular network, 17, 1% Flight mode: Sleep, 3, 0% Lighting: Keyboard, 73, 3% Lighting: Display I, 148, 5% Lighting: Display II, 61, 2% LCD, 13, 0% Compute, 370, 13% Speaker, 45, 2% Bluetooth, 440, 16% GPRS, 1600, 58% 2

3 Power consumption (T-Mobile) Bluetooth Wi-Fi Cellular 1000 Power (mw) Transmission Connected Transmission Connected Transmission Connected Paging Discoverable Speaker Keyboard lighting LCD lighting LCD Computing IDLE-Flight mode 3

4 Power consumption (Contd.) Theoretical limits Receiving energy per bit > N * N: Noise spectral power level Wideband communication P TX P RX *d a P RX Distance: d Propagation constant: a ( ) 4

5 Power consumption (Contd.) What increases power consumption Government regulation (FCC) Available spectrum band (Higher band, higher power) Limited bandwidth Limited transmission power Noise and reliability Higher capacity Multiple access (CDMA, TDMA etc.) Security Addressability (TCP/IP) More 5

6 Wireless system architecture Network protocol stack Hardware implementation Application Transport Host computer Network Data link Physical Baseband RF front ends Network interface 6

7 Power consumption (Contd.) Low-noise amplifier LNA Antenna interface Local Oscillator (LO) Intermediate Frequency (IF) signal processing IF/Baseband Conversion Baseband processor PA Power amplifier Physical Layer MAC Layer & above >60% non-display power consumed in RF RF technologies improve much slower than IC 7

8 Power consumption (Contd.) 1% 6% Components Power (mw) 8% 18% 67% PA FS Mixer Power amplifier (PA) Frequency synthesizer (VCO/FS) Mixer 30.3 LNA 20 Baseband Amplifier 5 Source: Li et al,

9 Low-noise amplifier (LNA) Bandwidth (same as the signal) Gain (~20dB) Linearity (IP3) Noise figure (1dB) Power consumption

10 Circuit power optimization Major power consumers Huge dynamic range 10 5 Low-noise amplifier High duty cycle LNA Antenna interface Local Oscillator (LO) Almost always on Intermediate Frequency (IF) signal processing IF/Baseband Conversion Baseband processor PA Power amplifier Physical Layer MAC Layer & above High power consumption 10

11 Circuit power optimization (Contd.) Reduce supply voltage Negatively impact amplifier linearity Higher integration CMOS RF SoC and SiP integration Power-saving modes 11

12 Circuit power optimization (Contd.) Power-saving modes Complete power off (Circuit wake-up latency + network association latency) on the order of seconds Different power-saving modes Less power saving but short wake-up latency 12

13 Power-saving modes Radio Deep Sleep Wake-up latency on the order of micro seconds Low-noise amplifier LNA Antenna interface Local Oscillator (LO) Intermediate Frequency (IF) signal processing IF/Baseband Conversion Baseband processor PA Power amplifier Physical Layer MAC Layer & above 13

14 Power-saving modes (Contd.) Sleep Mode Low-noise amplifier LNA Wake-up latency on the order of milliseconds Low-rate clock with saved network association information Antenna interface Local Oscillator (LO) Intermediate Frequency (IF) signal processing IF/Baseband Conversion Baseband processor PA Power amplifier Physical Layer MAC Layer & above 14

15 Network power optimization Use power-saving modes Example: wireless LAN (WiFi) Infrastructure mode: Access points and mobile nodes Example: Cellular networks 15

16 infrastructure mode Mobile node sniffs based on a Listen Interval Listen Interval is multiple of the beacon period Beacon period: typically 100ms During a Listen Interval Access point buffers data for mobile node sends out a traffic indication map (TIM), announcing buffered data, every beacon period Mobile node stays in power-saving mode After a Listen Interval Mobile node checks TIM to see whether it gets buffered data If so, send PS-Poll asking for data 16

17 Buffering/sniffing in Gast, Wireless Network: The Definitive Guide /Bluetooth uses similar power-saving protocols: Hold and Sniff modes 17

18 Cellular networks Discontinuous transmission (DTX) Discontinuous reception (DRX)

19 Wireless energy cost Connection Establishment Maintenance Transfer data Transmit vs. receive 19

20 Energy per bit transfer Oppermann et al., IEEE Comm. Mag

21 Wasteful wireless communication Time Micro power management Space Directional communication Spectrum Efficiency-driven cognitive radio 21

22 Space waste Omni transmission èhuge power by power amplifier (PA) 22

23 Time waste Data Size (Byte) Network Bandwidth Under-Utilization Modest data rate required by applications IE ~ 1Mbps, MSN video call ~ 3Mbps Bandwidth limit of wired link 6Mbps DSL at home Time (s) ) (% e tim sy u b in a ls rv te in le Id User1 User2 User3 User4 Time 23 Energy 23

24 Spectrum waste 24

25 Observed from an g user Energy per bit Distribution of observed g throughput 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 Throughout (bps) 25

26 Temporal waste 1 Radio Activity Time(s) 90% of time & 80% of energy spent in idle listening Four g laptop users, one week 26

27 Fundamental problem with CSMA CSMA: Carrier Sense Multiple Access Clients compete for air time Incoming packets are unpredictable 27

28 Fundamental problem with CSMA 28

29 Micro power management (µpm) Sleep during idle listening Wake up in time to catch retransmission Monitor the traffic not to abuse it ~30% power reduction No observed quality degradation J. Liu and L. Zhong, "Micro power management of active interfaces," in Proc. MobiSys

30 Directional waste Ongoing project with Ashutosh Sabharwal

31 Directional waste

32 Two ways to realize directionality Passive directional antennas Low cost fixed beam patterns Desclos, Mahe, Reed, 2001 Digital beamforming Flexible beam patterns High cost 32 Phased-array antenna system from Fidelity Comtech

33 Challenge I: Rotation!!! Solution: Don t get rid of the omni directional antennas Use multiple directional antennas But can we select the right antenna in time? 33

34 Challenge II: Multipath fading 34

35 Challenge III Can we do it without changing the infrastructure? 35

36 Characterizing smartphone rotation How much do they rotate? How fast do they rotate? 11 HTC G1 users, each one week Log accelerometer and compass readings 100Hz when wireless in use 36

37 Device orientation described by three Euler angles θ and φ based on tri-axis accelerometer ψ based on tri-axis compass and θ and φ 37

38 Rotation is not that much <120 per second ms 1s 10s θ ms 1s 10s φ ms 1s 10s ψ PDF 0.2 PDF 0.2 PDF Rotational speed( /s) Rotational speed( /s) Rotational speed( /s) 38

39 Directionality indoor 5 dbi 8 dbi 39

40 8dBi antenna 5dBi antenna

41 Measurement setup RSSI measured at both ends Data packets ACK packets 41

42 Directional channel still reciprocal RSS(dBm) NLOS ind. / 5dBi antenna Dir-Client Dir-AP Omni-Client Omni-AP Direction( ) 42

43 Directional beats omni close to half of the time 30 5dBi 25 total time(%) [0,0.1) [0.1,1) [1,10) [10,inf) superiority intervals(s) Field collected rotation traces replayed 43

44 RSS is predictable (to about 100ms) 100 Zero order First order 5dBi Error(dB) ms 100ms 1s 10s Prediction Intervals(s) 44

45 Multi-directional antenna design (MiDAS) One RF chain, one omni antenna, multiple directional antennas Directional ant. only used for data transmit and ACK Reception Standard compliance Tradeoff between risk and benefit 45

46 Packet-based antenna selection Assess an antenna by receiving a packet with it Leveraging channel reciprocity Continuously assess the selected antenna Find out the best antenna by assessing them one by one Potential risk of missing packets Stay with omni antenna when RSS changes rapidly No change in network infrastructure 46

47 Symbol-based antenna selection Assess all antennas through a series of PHY symbols Similar to MIMO antenna selection Needs help from PHY layer Antenna training packet Regular packet SEL ACK 47

48 Trace based evaluation Rotation traces replayed on the motor RSSI traces collected for all antennas Algorithms evaluated on traces offline -45 Dir 3 RSS(dB) Omni Dir 1 Dir time(second) 48

49 An early prototype 1 omni antenna 3 directional antennas WARP Laptop Controllable motor Finalist of MobiCom 08 Best Student Demo 49

50 The busier the traffic, the better 6 5 Upper bound Symbol-based Packet-based Gain(dB) ms 100ms 1s 10s Average Packet Interval 50

51 Two 5dBi antennas enough 6 5 Upper bound Symbol-based Packet-based Gain(dB) three two-opp two-adj one Antenna Configuration 51

52 Two 5dBi antennas enough NLOS ind. / 5dBi antenna Gain(dB) Upper bound Symbol-based Packet-based 5dBi 8dBi Antenna Gain RSS(dBm) RSS(dBm) Dir-Client Dir-AP Omni-Client Omni-AP Direction( ) NLOS ind. / 8dBi antenna Dir-Client Dir-AP Omni-Client Omni-AP Direction( ) 52

53 Real-time experiments: 3dB gain -45 Omni Multi antenna Avg. RSS(dB) NLOS ind. Environment LOS ind. Packet-based antenna selection Three 5dBi antennas Continuous traffic (1400 byte packets) Field collected rotation trace 53

54 Throughput improvement Throughput(Mbps) Omni Multi antenna NLOS ind. Environment LOS ind. 54

55 SNR vs. transmission rate (802.11a) Goodput (Mbps) Mbps 9Mbps 12Mbps 18Mbps 24Mbps 36Mbps 48Mbps 54Mbps SNR (db) (D. Qiao, S. Choi, and K. Shin, 2002) 55

56 MiDAS+rate adaptation+power control Recall that RSS is quite predictable up to 100ms 200 % Goodput Gain-Upper bound Goodput Gain-MiDAS TX power reduction-upper bound TX power reduction-midas Omni SNR(dB) 56

57 Protocol waste Cellular network WLAN (Wi-Fi) Transmission efficiency Connection Availability

58 How to combine the strength of both Wi-Fi and Cellular network? Estimate Wi-Fi network condition WITHOUT powering on Wi-Fi interface 58

59 Use context to predict WiFi availability Visible cellular network towers Motion Time of the day, day of the week Statistical learning Context Wi-Fi Conditions P(WiFi Context) Ahmad Rahmati and Lin Zhong, "Context for Wireless: Context-sensitive energy-efficient wireless data transfer," in Proc. MobiSys 07. Journal version with new results to appear in IEEE TMC 59

60 Cellular network offers clues

61 Cellular network offers clues

62 We don t move that much 50% 40% 30% 20% 10% 0% moving (1, 5] (5, 10] (10, 30] (30, 60] (60, 120] (120, inf) Length of motionless period (minute) Data collected from 2 smartphone users 2006 Shoehorned smartphone with accelerometer 62

63 Our life is repetitive Probability of same Wi-Fi availability (normalized autocorreletaion) Time (days) Data collected from 11 smartphone users 63

64 WiFi availability is HIGHLY predictable Prediction accuracy of Wi-Fi availability Time (minutes) Application Mobile EKG monitoring 35% battery life improvement (12 to 17 hours) 64