Ten Years of Wireless Sensor Network MACs:Aspects of time, frequency, and energy

Anish Arora

Wenjie Zeng

Jing Li

2

What is Medium Access Control (MAC)

• MAC provides two major functions Addressing Channel access – the focus of this tutorial

• Requirements for multiple access protocols Reliability – high delivery probability Throughput capacity – high delivery rate Latency – timely delivery Fairness – fair chance to deliver for

everyone Energy Efficiency – low cost per delivery

• Critical for battery-powered sensor platforms

3

Under the Hood – The Physical Layer Model

• When is a reception successful? Physical Model: when Signal to Interference Ratio (SINR) is

greater than some threshold β

Protocol Model: when the other senders are far away

• Terminals A and B send, C “captures” B Capacity is underestimated under protocol model

A B C

NPP

PSINRBA

BC

BCAC DD )1(

- Received signal strength from node i N – Background noise

iP

- Distance between node i and j Δ - Some positive value

ijD

4

• Low radio sensitivity + high attenuation

Challenges for Wireless Sensor MACs

Wired Medium: Interference Range = Carrier Sensing Range = Communication RangeWireless Medium: Interference Range ≠ Carrier Sensing Range ≠ Communication Range

BA C BA C D

• A sends to B, C cannot receive A • C wants to send to B, C senses a

“free” medium (CS fails)• Collision at B, A cannot receive the

collision (CD fails)• A is “hidden” for C

Hidden Terminal Problem Exposed Terminal Problem

• B sends to A, C wants to send to D• C has to wait, CS signals a medium in

use• Since A is outside the radio range of

C waiting is not necessary• C is “exposed” to B

5

Challenges for Wireless Sensor MACs (cont.)

• Global information sharing is difficult Time synchronization, time/frequency schedules

• The network is mobile in nature The wireless medium is unreliable - link mobility Node mobility Robustness of MACs matters

• Energy constraint High bandwidth, low latency -> high energy consumption Reliability -> retransmissions -> high energy consumption

6

Contention Based Medium Access

• Carrier Sense Multiple Access (CSMA)

• Transmit when you feel like transmitting Problem: Receiver must be awake as well

• Minimize collision – backoff scheme is critical Initial backoff – delay before putting the bits into the air

- Avoid collision in the first place Congestion backoff – delay after the medium is detected busy

- Avoid repeated collisions

• Retransmission needed when backoff fails Backoff scheme may make multiple attempts in itself

7

Contention Based Medium Access (cont.)

• A cookbook for a backoff scheme A backoff probability distribution p(n): the probability to tx in slot n

- Uniform backoff probability (CW is the contention window):

p(i) = p(j), for any slot i and slot j- Non-uniform backoff probability

The minimum and maximum backoff window

• Example 1: Binary Exponential Backoff (BEB) Uniform backoff probability Starts with some minimum backoff window Backoff window doubles upon each failure until a max value is reached

• Example 2: Sift Exponential backoff probability Fixed backoff window More suitable for WSNs: receiver’s radio consumes energy throughout

the backoff period

8

A Classic Contention MAC: Slotted Aloha

• We assume that the stations are perfectly synchronous

• In each time slot each station transmits with probability p

• In Slotted Aloha, a station can transmit successfully with probability at least 1/e, or about 36% of the time

-

-

-

-

- -

-

11

1

!2

1

Pr[Station 1 succeeds] (1 )Pr[any Station succeeds]

maximize : (1 ) (1 ) 0 1

1 1then, (1 )

n

n

n

P p pP nP

dPP n p pn pndp

Pn e

9

Unslotted (Pure) Aloha

• Unslotted Aloha: simpler, no (potentially costly!) synchronization

• However, collision probability increases. Why?

• There is a factor-2-handicap of Unslotted vs. Slotted

10

Aloha Robustness

• Round robin has a problem when a new station joins. In contrast, Aloha is quite robust

• Example: If the actual number of stations is twice as high as expected,there is still a successful transmission with probability 30%. If it is onlyhalf, 27% of the slots are used successfully. So nodesjust need a good estimateof the number of nodes intheir neighborhood

11

Scheduling Based Medium Access

• Schedule transmissions across Time: Time Division Multiple Access (TDMA)

- Use a “pre-computed” schedule to transmit messages- Distributed, adaptive solutions are difficult

Frequency: Frequency Division Multiple Access (FDMA)- Choose different frequencies for interfering neighbors- Permanent (radio broadcast), slow hopping (GSM), fast hopping

Space: Space Division Multiple Access (SDMA)- Segment space into sectors, use directed antennas- Reuse frequencies in different cells

Code: Code Division Multiple Access (CDMA)- People talk simultaneously in different languages

13

• Share a common sense of time? Synchronous vs. Asynchronous

Design Space for MAC Protocols

S R

It’s about slot 10 at R, and he’s supposed to be awake

Synchronous

S R

I’d better made sure that R is awake when I talk to him

Asynchronous

14

Design Space for MAC Protocols (cont.)

• Who controls the coordination? Sender-centric vs. Receiver-centric

S R

Sender-centric and asynchronous

S:I’m sending R: z..S:I’m sending R: zz..S:I’m sending R: zzz..S:I’m sending R:roger that

S R

Receiver-centric and asynchronous

S:I’m waiting. R: z..S:I’m waiting.. R: zz..S:I’m waiting… R: zzz..S:finally… R:send now

15

Sender-Centric and Asynchronous MACs:B-MAC and X-MAC

• Low Power Listening (LPL) Introduced by B-MAC Periodically wake up shortly to detect incoming traffic Senders transmit preambles as long as the sleep interval before

sending the actual data to ensure the receiver is listening for the packet

• X-MAC improves upon B-MAC Embed node ID into preambles

- Avoid overhearing- Send only the 802.15.4 header to reduce receiver detection time

Packetized preambles- Receiver acknowledges preamble packets destined to itself- Allow early preamble termination

16

Sender-Centric and Asynchronous MACs:BoX-MAC

• Improved upon B-MAC and X-MAC• Key observation

In X-MAC, receiver detection time is coupled with length of preamble packets

• Decouple by detecting presence of energy in the channel• Use data packet as preambles

Save energy in handshaking

17

Sender-Centric and Synchronous MACs:S-MAC

• All nodes choose and broadcast awake schedules• Synchronized sleep/wake within cluster

Boarder nodes• Low channel utilization b/c of control overhead and

introduced cluster contention• Use RTS/CTS to resolve contention

Allow senders go back to sleep if another sender wins the medium

Schedule 1 Schedule 2

Schedule 1+2

increased latency

18

• Network wide synchronized channel polling• Short tones at polling time to signal pending traffic• Geared towards ultra-low duty cycle

Short polling instead of longer listen period Adapt listen interval dependent on the current network load

Sender-Centric and Synchronous MACs:SCP-MAC

SCP-MAC

S-MAC

19

Receiver-Centric and Asynchronous MAC:RI-MAC

• Flip the LPL operations

• Senders listen for receiver preambles if they have pending data Allow short preambles from the receiver

• Sender long preamble becomes long listening Better channel utilization

B-MAC RI-MACSender TX preambles Listen for preamblesReceiver Listen for preamble TX preamble to initiate TX

20

Receiver-Centric and Synchronous MAC:O-MAC

• Neighborhood synchronization vs. network synchronization Larger memory footprint vs. longer convergence time

• Each node wakes up at pseudo-random slots• Compute neighbor’s schedule using its broadcasted seed• Approximate TDMA at low cost

Ideally, the wakeup frequency should minimizecontention per wakeup

• Slots need not be alignedFrame

Slot Slot

Frame

SlotSlotSlot Slot Slot Slot

Listen

Transmit

Sleep

Frame

Slot Slot

Frame

SlotSlotSlot Slot Slot Slot

Sender

Receiver

21

A Brief Summary of Exemplar MACs

Name Synchrony Centricity PreamblesS-MAC/SCP-MAC synchronous sender yesB-MAC/BoX-MAC asynchronous sender yes(B-MAC)

no(BoX-MAC)RI-MAC asynchronous receiver yesOMAC synchronous receiver no

22

Applications Define the “Best” MAC

• Synchronous MACs are more efficient if control overhead can be mitigated, e.g. in static networks Synchronization can be complicated in mobile network

- Node mobility due to movement- Link mobility due to changing link quality

• Asynchronous MACs are more robust

• Over a wide range of traffic loads and network sizes, receiver-centric and synchronous MACs are most energy efficient

23

Challenges of Multi-channel Protocols

• Coordination of communicators Compared with single channel protocols, needs agreement on

both channel and time

• Restricted ratio interfaces Most hardware platforms for WSN do not support multiple

interfaces Channel switch time is not negligible

• Restricted energy supply Control overhead has to be reduced for energy efficiency

24

Main Categories of Multi-channel MACs

• Static scheme Different channels are chosen for 2-hop neighbors to avoid

potential interferences

• Dynamic scheme Negotiate which channel to use per packet

• Load balancing scheme Splitting traffic loads evenly over available channels

• Probabilistic scheme Probabilistically allocates a fraction of nodes into the next

channel when the current channel becomes overloaded

25

Static Channel Assignment

• MMSN (Infocom’06): Frequency assignment

- every node is assigned a channel for data reception such that most of two-hop neighbors do not communicate on the same channel.

Toggle snooping- When a node has a unicast packet to send, the node snoops on

two frequencies alternatingly: fself and fdest

Cons: - High overhead of distance-2 coloring & toggle snooping- Regardless of real traffic conditions

26

Dynamic Channel Assignment

• TMMAC (ICC’07) A TDMA-based multi-channel MAC protocol Consists of a 2-dimensional ATIM window wherein each node

negotiates time slot and channel it will use before transmission

ATIM window is updated and checked at each node• Cons:

Message overhead of explicit negotiation Energy overhead of continuous listening for channel updates

27

Load Balancing Channel Assignment

• Y-MAC (IPSN’08) Extended from Crankshaft protocol, TDMA-based Every receiver wakes up at its non-overlap slot within each

frame on home channel If more packets need to be received, the receiver hops to the

next channel for reception• Cons:

per slot channel switching Instantaneous channel

condition is not considered

28

Probabilistic Channel Assignment

• A practical multi-channel MAC (IPSN’08) Starts in home channel Feedback control scheme for channel expansion & shrinking

- Switch probability is a function of success transmission rate & sender/receiver role

Periodic channel status updates • Cons:

Channel update & control overhead Instantaneous channel condition is not considered

29

Throughput Capacity of Wireless Networks

• Definition: for each source-destination pair, the average number of bits delivered successfully from a source to a destination

• The problem: how does throughput capacity change with an increasing number of nodes in an area of fixed size

30

Network Model

• Network of n nodes i.i.d in a unit area region An intuitive observation – the worst-case throughput capacity

λ(n) is

This occurs when the n nodes form a clique network

• Static network, independent and random destination• Each node can transmit at W bps• Packets sent from node to node in multihop fashion• Transmissions are slotted• Several nodes may transmit simultaneously• Use the protocol model

)()(n

Wn

31

Results for Always On Network

• The throughput capacity λ(n) is

• The result carry over to the physical model• Dividing W into multiple channels does not change result

• Implications: Network aimed at smaller scale of users are better When nodes communicate only nearby nodes at a distance of

less than O(1/sqrt(n)) away, a constant bit rate is feasible per source-destination pair- E.g. smart home networks

)log

()(nn

Wn

32

Results for Duty Cycled Network

• If every node runs at a duty cycle ψ, the throughput capacity λ(n) is

• Throughput capacity scales better than linearly in ψ (note that 0≤ ψ ≤ 1)

• Throughput capacity scales better than the reciprocal of

)()(n

Wn

nn log

33

Energy Efficiency of Wireless Networks

• Definition: the ratio from the energy expense in receiving unique and useful data packets to the total energy expense

• To optimize energy efficiency Maximize throughput capacity Minimize total energy consumption

• For full-duty-cycled networks, maximizing energy efficiency is equivalent to maximizing throughput capacity

34

Results for Energy Efficiency in MAC

• Deriving the exact form of energy efficiency as a function of network size n turns out to be difficult

• Given a network size, optimal efficiency is obtained by searching through different duty cycles and traffic loads

• Numerical and experimental results are obtained

35

References

• B-MAC: Versatile Low Power Media Access for Wireless Sensor Networks, Polastre, J. and Hill, J. and Culler, D., SenSys 04’

• X-MAC: X-MAC: A short preamble mac protocol for duty-cycled wireless sensor networks, Buettner et.al., SenSys 06’

• BoX-MAC: BoX-MACs: Exploiting physical and link layer boundaries in low-power networking, Moss, D. and Levis, P., Technical Report SING-08-00, Stanford University, 08’

• S-MAC: An Energy-Efficient MAC Protocol for Wireless Sensor Networks, Ye, W. and Heidemann, J. and Estrin, D., Infocom 02’

• SCP-MAC: Ultra-low duty cycle MAC with scheduled channel polling, Ye, W. and Silva, F. and Heidemann, J., SenSys 06’

• RI-MAC: RI-MAC: A receiver-initiated asynchronous duty cycle MAC protocol for dynamic traffic loads in wireless sensor networks, Sun, Y. and Gurewitz, O. and Johnson, D.B., SenSys 08’

• O-MAC: O-MAC: A Receiver Centric Power Management Protocol, Hui Cao, Kenneth W. Parker, Anish Arora, ICNP 06’

36

Place Holder Slide

The previous page is the last slide. This page can be removed.

37

Low Power Listening (B-MAC)

• Nodes wake up for a short period and check for channel activity. Return to sleep if no activity detected.

• If a sender wants to transmit a message, it sends a long preamble to make sure that the receiver is listening for the packet. preamble has the size of a sleep interval

• Very robust No synchronization required Instant recovery after channel disruption

preamble data

listen

channel sniff

38

Low Power Listening (B-MAC)

• Problem: All nodes in the vicinity of a sender wake-up and wait for the packet. Solution 1: Send wake-up packets instead of preamble, wake-

up packets tell when data is starting so that receiver can go back to sleep as soon as it received one wake-up packet.

Solution 2: Just send data several times such that receiver can tune in at any time and get tail of data first, then head.

• Communication costs are mostly paid by the sender. The preamble length can be much longer than the actual

data length.

• Idea: Learn wake-up schedules from neighboring nodes. Start sending preamble just before intended receiver wakes

up. WiseMAC

overhearing problem

encode wake-up pattern in ACK

message

39

Sensor MAC (S-MAC)

• Coarse-grained TDMA-like sleep/awake cycles.

• All nodes choose and announce awake schedules. synchronize to awake schedules of neighboring nodes.

• Uses RTS/CTS to resolve contention during listen intervals. And allows interfering nodes to go to sleep during data

exchange.

listen sleep sleeplisten

frame

time

increased latency

40

Sensor MAC (S-MAC)

• Problem: Nodes may have to follow multiple schedules to avoid network partition.

Schedule 1 Schedule 2

Schedule 1+2

• A fixed sleep/awake ratio is not always optimal.- Variable load in the network.

• Idea: Adapt listen interval dependent on the current network load.- T-MAC

41

Hybrid Protocols

• Protocols may use information from upper layers to further improve their performance. Information about neighborhood Routing policies

• Minimize costly overhearing of neighboring nodes Inform them to change their channel sniff patterns

• Use randomization to resolve schedule collisions

schedule collision

optimization for WiseMAC

like in Dozer

42

Motivation – Hidden Terminal Problem

• A sends to B, C cannot receive A • C wants to send to B, C senses a “free” medium (CS fails)• Collision at B, A cannot receive the collision (CD fails)• A is “hidden” for C

BA C

Interference Range ≠ Carrier Sensing Range

43

Motivation – Exposed Terminal Problem

• B sends to A, C wants to send to D• C has to wait, CS signals a medium in use• Since A is outside the radio range of C waiting is not

necessary• C is “exposed” to B

BA C D

44

Adaptive Slotted Aloha

• Idea: Change the access probability with the number of stations• How can we estimate the current number of stations in the system?• Assume that stations can distinguish whether 0, 1, or more than 1

stations transmit in a time slot. • Idea:

If you see that nobody transmits, increase p. If you see that more than one transmits, decrease p.

• Model: Number of stations that want to transmit: n. Estimate of n: Transmission probability: p = 1/ Arrival rate (new stations that want to transmit): λ (with λ < 1/e).

n̂n̂

45

Adaptive Slotted Aloha Q&A

Q: What if we do not know , or is changing?A: Use = 1/e, and the algorithm still works.

Q: How do newly arriving stations know ?A: We send with each transmission; new stations do not send before

successfully receiving the first transmission.

Q: What if stations are not synchronized?A: Aloha (non-slotted) is twice as bad.

Q: Can stations really listen to all time slots (save energy by turning off)? Can stations really distinguish between 0, 1, and ¸2 sender?

A: Maybe. One can use systems that only rely on acknowledgements.

n̂n̂

46

Backoff Protocols

• Backoff protocols rely on acknowledgements only.

• Binary exponential backoff If a packet has collided k times, we set p = 2-k

Or alternatively: wait from random number of slots in [1..2k]

• It has been shown that binary exponential backoff is not stable for any arrival rate λ > 0 (if there are infinitely many potential stations)

• Interestingly when there are only finite stations, binary exponential backoff becomes unstable with λ > 0.568; Polynomial backoff however, remains stable for any λ < 1.