1. tseng:1 Power Saving and Power Management in WiFi and Bluetooth Networks Prof. Yu-Chee Tseng Dept. of Comp. Sci. & Infor. Eng. National Chiao-Tung University ( 交通大學 資訊工程系 曾煜棋 )

2. yctseng: 2 Outline Power control:  S.-L. Wu, Y.-C. Tseng, and J.-P. Sheu, "Intelligent Medium Access for Mobile Ad Hoc Networks with Busy Tones and Power Control", IEEE Journal on Selected Areas in Communications, 18(9):1647-1657, Sep. 2000. Power management:  Y.-C. Tseng, C.-S. Hsu, and T.-Y. Hsieh, "Power-Saving Protocols for IEEE 802.11-Based Multi-Hop Ad Hoc Networks", Computer Networks, Elsevier Science Pub., Vol. 43, No. 3, Oct. 2003, pp. 317-337. WiFi vs Bluetooth:  T.-Y. Lin and Y.-C. Tseng, "An Adaptive Sniff Scheduling Scheme for Power Saving in Bluetooth", IEEE Wireless Communications, Vol. 9, No. 6, Dec. 2002, pp. 92-103.

3. tseng:3 Introduction: Basic Concept

4. 4 Introduction Battery is a limited resource in any portable device.  becoming a very hot topic is wireless communication Power-related issues:  PHY: transmission power control  MAC: power mode management  Network Layer: power-aware routing

5. 5 Transmission Power Control tuning transmission energy for higher channel reuse example:  A is sending to B (based on IEEE 802.11)  Can (C, D) and (E, F) join? No! Yes!

6. 6 Power Mode Management doze mode vs. active mode example:  A is sending to B (based on 802.11)  Does C need to stay awake?

7. 7 Power-Aware Routing routing in an ad hoc network with energy-saving in mind Example: in an ad hoc network +–+– +–+– +–+– +–+– +–+– +–+– SRC N1 N2 DES T N4 N3

8. wu-ICCCN99:8 WirelessNet Tseng Intelligent Medium Access for Mobile Ad Hoc Networks with Busy Tones and Power Control S.-L. Wu, Y.-C. Tseng, and J.-P. Sheu, IEEE J. of Selected Areas on Communications (JSAC)

9. wu-ICCCN99:9 WirelessNet Tseng Abstract n A New MAC Protocol ubased on RTS/CTS uwith Busy Tones uwith Power Control

10. wu-ICCCN99:10 WirelessNet Tseng Power Control n Use an appropriate power level to transmit packets. uto increase the possibility of channel reuse uto increase channel utilization n Example: u(a) without power control: åthe transmissions from C to D and from E to F are prohibited. u(b) with power control: åall these can coexist.

11. wu-ICCCN99:11 WirelessNet Tseng How to Tune Power Levels n Assumptions: uA mobile host can choose on what power level to transmit a packet. uOn receiving a packet, the physical layer can offer the MAC layer the power level on which the packet was received. n Suppose P t and P r are the power levels a packet is sent and received, respectively.  = carrier wavelength un = path loss coefficient (typically 2 ~ 6) ud = distance between sender and receiver ug t and g r : antenna gains at the sender and receiver sides, respectively

12. wu-ICCCN99:12 WirelessNet Tseng n Note: during a short period, the values of n and d can be treated as a constant. This makes power control possible. n Let P min be the minimum power level to decode a packet. uSuppose X sends an RTS to Y with power P t. uIf Y wants to reply a CTS to X with a power level P CTS, such that X receives the packet at the smallest power level P min, then we have: n Dividing the above formulas, we have:

13. wu-ICCCN99:13 WirelessNet Tseng General Rules in This Paper n Busy Tone (BT) uSenders should send BTt, but gauge any BTr. uReceivers should send BTr, but gauge any BTt. n General Rules: uData packet and BTt: transmitted with power control. uCTS and BTr: transmitted at the normal (largest) power. uRTS: at a power level based on how strong the BTr are around the requesting host. n Channel Model: BTt BTr control channel data channel frequency

14. wu-ICCCN99:14 WirelessNet Tseng Illustrative Example (I) n A is sending to B. uA’s data packet and BTt at the minimal level (yellow circle). uB’s BTr at the largest level (white circle). n C intends to send to D. uC hears no BTr. uD hears not BTt. uSo the transmission can be granted (pink circle). A D C B

15. wu-ICCCN99:15 WirelessNet Tseng Illustrative Example (II) n Now we moe C into A’s circle. n A is sending to B. uA’s data packet and BTt at the minimal level (yellow circle). uB’s BTr at the largest level (white circle). n C intends to send to D. uC hears no BTr. uD hears no BTt. uSo the transmission can be granted (pink circle). A D C B

16. wu-ICCCN99:16 WirelessNet Tseng Illustrative Example (III) n Next we move D into A’s circle. n A is sending to B. uA’s data packet and BTt at the minimal level (yellow circle). uB’s BTr at the largest level (white circle). n C intends to send to D. uC hears no BTr. uD hears A’s BTt. uSo the transmission can NOT be granted (pink circle). A D C B

17. wu-ICCCN99:17 WirelessNet Tseng Illustrative Example (IV) n A is sending to B. uA’s data packet and BTt at the minimal level (yellow circle). uB’s BTr at the largest level (white circle). n C intends to send to D. uC hears A’s BTt and B’s BTr. uD hears no BTt. uThe transmission can be granted if C controls its transmission power (pink circle). A D C B

18. wu-ICCCN99:18 WirelessNet Tseng Illustrative Example (V) n A is sending to B. uA’s data packet and BTt at the minimal level (yellow circle). uB’s BTr at the largest level (white circle). n C intends to send to D. uC hears A’s BTt and B’s BTr. uD hears no BTt. uThe transmission can be granted if C controls its transmission power (pink circle). A D C B

19. wu-ICCCN99:19 WirelessNet Tseng Many Transmission Pairs with Power Control and Busy Tones A F E D C B BTt and DATA: yellow circles BTr: white circles

20. wu-ICCCN99:20 WirelessNet Tseng The Protocol n P max : the maximum transmission power n P min : the minimum power to distinguish a signal from a noise n P noise : the maximum power at which an antenna will regard a signal as a noise uP min - P noise should be a very small value n Basic “Power” Rules: uData packet and BTt: transmitted with power control. uCTS and BTr: transmitted at the largest power P max. uRTS: at a power level based on how strong the BTr are around the requesting host.

21. wu-ICCCN99:21 WirelessNet Tseng Detailed Protocol n On a host X intending to send a RTS to Y, uX senses any receive busy tone BTr around it uX sends a RTS on the control channel at power level P x : åIf there is no BTr, let P x = P max. åO/w, let Pr be the power level of BTr that has the highest power among all heard BTr’s. 4The RTS should not go beyond the nearest host that is currently receiving a data packet. 4P max is used because BTr is always transmitted at the maximal power.

22. wu-ICCCN99:22 WirelessNet Tseng n On Y receiving X’s RTS, uY senses any transmit busy tone BTt around it. åIf there is any BTt, then Y ignores this RTS. åO/w, Y does the following: 4reply with a CTS at the maximum power Pmax 4turn on its receive busy tone BTr at the maximum power Pmax n On X receiving Y’s CTS, uX transmits its data packet at power P x. uX turns on its transmit busy tone BTt at power P x. åP r is the power level at which X receives Y’s CTS. P x is the minimal possible power level to ensure that Y can correctly receive the data packet.

23. wu-ICCCN99:23 WirelessNet Tseng Many Transmission Pairs with Power Control and Busy Tones F E D C G H BTr BTt A B RTS CTS

24. wu-ICCCN99:24 WirelessNet Tseng Analysis n Scenario: uA is currently sending to B. uAnother pair, C and D, is intending to communicate. n Goal: We want to find out the probability that C can send to D. n Through complicated calculus, we find that …

25. wu-ICCCN99:25 WirelessNet Tseng When BC < r max n INTC(Ra, Rb, AB) = the intersection of the circles centered at a and b uRa = radius of the circle centered at a uRb = radius of the circle centered at b uAB = distance of a and b n The probability that C can send to D when A is sending to B: ui.e., the coverage of Rc excluding the coverage of Ra uFig. 6

26. wu-ICCCN99:26 WirelessNet Tseng

27. wu-ICCCN99:27 WirelessNet Tseng cont... n Integrating over  = 0.. 2 , and then over CB = 0.. r max n Integrating over AB = 0.. r max, we have the final result n On the contrary, the DBTMA has probability of 0.

28. wu-ICCCN99:28 WirelessNet Tseng When r max < BC < 3r max n Main difference: C’s RTS will be sent with max. power. n The probability that C can send to D when A is sending to B: uSee Fig. 7: åAt point C1, node C can always send. åAt point C2, node C can’t send if D is in A’s range.

29. wu-ICCCN99:29 WirelessNet Tseng

30. wu-ICCCN99:30 WirelessNet Tseng cont... n Integrating over  = 0.. 2 , and then over CB = r max..3r max n Integrating over AB = 0.. r max, we have the final result

31. wu-ICCCN99:31 WirelessNet Tseng cont. n On the contrary, the DBTMA has a success probability of X change to r max

32. wu-ICCCN99:32 WirelessNet Tseng Discrete Power Control n The levels of power provided by hardware may not be infinitely tunable. uWe may have a discrete number of power levels. n Theorem: uGiven a fixed integer k, evenly spreading the k power levels will be the best choice. uI.e., (1/k)*P max, (2/k)*P max, (3/k)*P max, …, (k/k)*P max.

33. wu-ICCCN99:33 WirelessNet Tseng Simulation Parameters n Simulation parameters uphysical area = 8km  8km umax transmission distance (rmax) = 0.5 or 1.0 km unumber of mobile hosts = 600 uSpeed of mobile hosts 0 or 125 km/hr. ulength of control packet = 100 bits ulink speed = 1 Mbps utransmission bit error rate = 10-5/bit

34. wu-ICCCN99:34 WirelessNet Tseng Simulation Results: Channel utilization vs. traffic load (a) r max = 0.5 km(b) r max = 1.0 km

35. wu-ICCCN99:35 WirelessNet Tseng Channel utilization vs. data packet length at various traffic loads

36. wu-ICCCN99:36 WirelessNet Tseng Channel Utilization vs. Number of Power Levels n r max = 1 km; arrival rate = 200 or 400 packets/ms; packet length = 1 or 2 Kbits uSo 4 to 6 levels will be sufficient.

37. wu-ICCCN99:37 WirelessNet Tseng Channel Utilization vs. Traffic Load n mobility = 0 km/hr and 125 km/hr n The transmission distance r max = 1.0 km

38. wu-ICCCN99:38 WirelessNet Tseng Short Conclusion n a new MAC protocol upower control on top of RTS/CTS and busy tones n Channel utilization can be significantly increased because the severity of signal overlapping is reduced.

39. 39 Power Mode Management in IEEE 802.11 Y.-C. Tseng, C.-S. Hsu, and T.-Y. Hsieh, "Power-Saving Protocols for IEEE 802.11- Based Multi-Hop Ad Hoc Networks", Computer Networks, Elsevier Science Pub., Vol. 43, No. 3, Oct. 2003, pp. 317-337 (also in INFOCOM).

40. 40 Power Consumption IEEE 802.11 power model  transmit: 1400 mW  receive: 1000 mW  idle: 830 mW  sleep: 130 mW

41. 41 Power Mode Management Power modes in IEEE 802.11  PS and ACTIVE Problem Spectrum:  infrastructure  ad hoc network (MANET)  single-hop  multi-hop ad hoc networks

42. 42 Infrastructure Mode two power modes: active and power- saving (PS)

43. 43 Ad Hoc Mode (Single-Hop) PS hosts also wake up periodically.  interval = ATIM (Ad hoc) window Beacon Interval ATIM Window Host A Host B Beacon BT A =2, BT B =5 power saving state Beacon ATIM ACK active state data frame ACK

44. 44 Problem Statement (Multi-Hop MANET) Clock Synchronization:  a difficult job due to communication delays and mobility Neighbor Discovery:  by inhibiting other's beacons, hosts may not be aware of others ’ existence Network Partitioning:  with unsynchronized ATIM windows, hosts with different wakeup times may become partitioned networks

45. 45 Network-Partitioning Example Host A Host B AB C DE F DE F Host C Host D Host E Host F ╳ ╳ ATIM window ╳ ╳ Network Partition

46. 46 What Do We Need? PS protocols for multi-hop ad hoc networks  Fully distributed  No need of clock synchronization (i.e., asynchronous PS)  Always able to go to sleep mode, if desired

47. 47 Features of Our Design Guaranteed Overlapping Awake Intervals:  two PS hosts ’ wake-up patterns always overlap  no matter how much time their clocks drift Wake-up Prediction:  with beacons, derive other PS host's wake-up pattern based on their time difference

48. 48 Beacon Int. (BI) Act. Win. (AW) Structure of a Beacon Interval BI: beacon interval (to send beacons) AW: active window  BW: beacon window  MW: MTIM window (for receiving MTIM)  listening period: to monitor the environment BW MW listening

49. 49 Three Protocols Based on the above structure, we propose three protocols  Dominating-Awake-Interval  Periodical-Fully-Awake-Interval  Quorum-Based

50. 50 P1: Dominating-Awake-Interval intuition: impose a PS host to stay awake sufficiently long “ dominating-awake ” property Host A Host B Beacon Interval ╳╳

51. 51 Problem:  only dectectable in ONE direction Adjustment:  odd beacon interval:  Active Window = BW + MW + listening  even beacon interval:  Active Window = listening + MW + BW Host A Host B Odd Beacon IntervalEven Beacon Interval Odd Beacon IntervalEven Beacon Interval B ╳ B BB ╳ M M M M

52. 52 Host A Host B odd beacon interval Beacon window MTIM Window Active window odd beacon interval even beacon interval Unicast Example MTIMData ACK

53. 53 Characteristics dominating awake  wake-up ratio < 1/2 sensibility  A PS host can receive a neighbor ’ s beacon once every two beacon intervals.  suitable for highly mobile environment

54. 54 P2: Periodical-Fully-Awake-Interval Basic Idea:  In every T intervals, stay awake in one full interval.  wake-up ratio  1/T  compared to 1/2 of protocol 1 Two types of beacon intervals:  Low-power interval  Fully-awake interval (in every T intervals)

55. 55 Example (T = 3) Host A Host B T(=3) Beacon Intervals Fully-awake Low-power ╳╳╳ ╳╳╳ T: Interval between the fully awake periods A PS host can receive its neighbor’s beacon frame in every T = 3 beacon intervals

56. 56 Definitions of Intervals Low-power interval:  active window + doze window  AW = BW + MW  i.e., listening period = 0 Fully-awake interval:  no doze window  i.e., AW = BI  very energy-consuming, so only appears once every T beacon intervals

57. 57 P3: Quorum-Based Quorum Sets:  Two quorum sets always have nonempty intersection.  (used here to guarantee detectability) A matrix example: n n c1c1 c2c2 r1r1 r2r2 Host A’s quorum intervals Host B’s quorum intervals Non-quorum intervals intersection

58. 58 Example (2D matrix quorum) 0123 4567 891011 12131415 0123 4567 891011 12131415 Host A’s quorum intervals Host B’s quorum intervals Non-quorum intervals Host A’ quorum intervals 1514131211109876543210 31302928272625242322212019181716 Group 1 Group 2 Host B’s quorum intervals 1514131211109876543210 31302928272625242322212019181716 Group 1 Group 2 Overlapping intervals

59. 59 Overlapping Property Overlap no matter how clocks drift demo... 0123 4567 891011 12131415 0123 4567 891011 12131415 14131211109876543210 1514131211109876543210 Host A’s quorum intervals Host B’s quorum intervals

60. 60 Quorum and Non-quorum Intervals Quorum interval:  AW = BI (i.e., fully awake) Non-quorum interval:  no beacon, only MTIM window  AW < BI  BW = 0, AW = MW Beacon window MTIM Window Active window Beacon Interval Quorum Interval Non-quorum Interval

61. 61 Summary Protocol Numbers of beacons per interval Active ratioNeighbor sensitivity Dominating11/2+BW/BIBI Periodical11/TT*BI/2 Quorum(2n-1)/n 2 (n 2 /4) * BI BI: length of a beacon interval AW: length of an active window BW: length of a beacon window MW: length of an MTIM window T: interval between the fully awake periods n: length of the square

62. 62 Summary Identify the problems of PS mode in IEEE 802.11 in multi-hop ad hoc networks.  clock drifting, network-partitioning Propose several PS protocols Connecting this problem to quorum issue in distributed systems.

63. 63 Sniff Scheduling for Power Saving in Bluetooth

64. 64 Overview (cont.) Addressing  48-bit Bluetooth Device Address (BD_ADDR)  3-bit Active Member Address (AM_ADDR)  8-bit Parked Member Address (PM_ADDR) Four operational modes:  Active  Sniff  Hold  Park

65. 65 Bluetooth Networks Piconet  one master + at most 7 active slaves Scatternet  multiple piconets to form a larger network

66. Packets Exchange Scenario MASTER SLAVE 1 SLAVE 2 SLAVE 3 ACLSCO ACL

67. 67 Low-Power Sniff Mode A slave can enter the low-power sniff mode by setting a parameter  (T sniff, N sniff_attempt, D sniff )  in per slave basis

68. 68 LMP_PDUs for Sniff

69. 69 Sniff Scheduling Problem How to determine the sniff parameters?  Goal: balancing power consumption and traffic need Earlier Works  na ï vely adjust parameters in an exponential way  double/halve sniff interval or active window whenever polling fails/succeeds  The placement of active windows of multiple slaves on the time axis is not addressed.

70. 70 Design Goals consider multiple slaves together adaptively schedule sniff parameters  more accurate in determining the sniff- related parameters based on slaves ’ traffic loads include solutions of placing of active windows of sniffed slaves on the time axis

71. 71 Proposed Architecture

72. 72  T k,N k,D k : current sniff parameters for slave k.  U k : the slot utilization of slave k.  B k : the buffer backlog of slave k.  W k : a weighted value to indicate the current requirement of slave k.  B max is the maximum buffer space  S k : the desired slot occupancy of slave k, which is the expected ratio of N k / T k.  0 < δ < 1 (to tolerate some unexpected traffic) the evaluator

73. 73 Resource Pool (RP) Although time slots are an infinite sequence, we represent them as a sequence of 2-D matrices.  each matrix M is of the size 2 u × T  time slots are viewed in a “ row-major ” way The availability of M:

74. 74 RP Example M ’ s size = 2 3 × 15 = 120

75. 75 Example: to allocate a slot occupancy of 16/120 (** Note: 16/120 = 8/60)

76. 76 Example: to allocate a slot occupancy of 16/120 (** Note: 16/120 = 4/30 = 2/15)

77. 77 A Running Example 5 slaves Each slave initially has an equal occupancy of 1/5 of the matrix M. We discuss two strategies:  longest sniff interval first  shortest sniff interval first

78. 78 Scheduling Policies: Longest Sniff Interval First (LSIF) a) initial state (equal shares) b) reduce S2 to 2/60 c) reduce S3 to 3/120 d) increase S4 to 6/30

79. 79 Scheduling Policies: Shortest Sniff Interval First (SSIF) a) initial state (equal shares) b) reduce S2 to 1/30 c) reduce S3 to 1/60 d) increase S4 to 3/15

80. 80 Conclusions Proposed:  Power-saving protocols for IEEE 802.11-based multi-hop ad hoc networks  Sniff-scheduling schemes for Bluetooth-based piconets References: 1.T.-Y. Lin and Y.-C. Tseng, “ An Adaptive Sniff Scheduling Scheme for Power Saving in Bluetooth ”, IEEE Personal Communications (to appear). 2.Y.-C. Tseng, C.-S. Hsu, and T.-Y. Hsieh, “ Power-Saving Protocols for IEEE 802.11-Based Multi-Hop Ad Hoc Networks ”, IEEE INFOCOM, 2002.