David E. Culler University of California, Berkeley

June 2008 WEI Short Course L7-link 1

Wireless Embedded Inter-Networking

Foundations Ubiquitous Sensor Networks

Low-Power Reliable Wireless Link Layer

David E. CullerUniversity of California, Berkeley


TranTranTranTran

Building A Low-Power Link

• Media Management Control – not MAC

NetNetNetNet

PhyPhyPhyPhy

LinkLinkLinkLink

Media Management ControlMedia Management ControlMedia Management ControlMedia Management Control

Remote MediaRemote MediaRemote MediaRemote Media Link StatsLink StatsLink StatsLink StatsNeighbor Table

Addr Period Phase Pending RSSI PRR

Local MediaLocal MediaLocal MediaLocal Media

Sample Period

Sample Phase

DatDataa

DatDataa AckAckAckAck

Ad

ap

tati

on

Ad

ap

tati

on

Ad

ap

tati

on

Ad

ap

tati

on


IEEE 802.15.4 Frame Format


Elements of Robust Communication

• Application: feasible workload– Packet rates, pattern, timing

• Network: finding and using good paths– Topology discovery and route selection– Route cost determination, selection– Forwarding

• Link: Framing, Media Management Protocol– On to receive during transmission– Frame structure, error detection, acknowledgement– Avoiding contention (MAC, CCA, Hidden Terminal)– Link quality estimation

• Physical: Signal to Noise Ratio– Device Transmission Power / Receive Sensitivity– Antenna design and orientation, obstructions, attenuation– Receive signal vs interference, noise, multipath– Modulation, channel coding


In a nutshell

TX

RX


Why Multihop Routing?

• Power!– Power to transmit distance D grows as D3 or worse

– Power to route distance D grows linearly

• Bandwidth (spatial multiplexing)– With n nodes in a single cell, each gets at most 1/n bandwidth

– Many small cells => many simultaneous transmissions.

• Reliability– Individual links experience interference, obstacles, and

multipath effects

– Even short-range “wireless wires” require human nurturing

» IRDA, Bluetooth, WiFi, Cell Phone

– Provides spatial diversity and receiver diversity

» rather than antenna diversity

– Protocol level reliability


Properties of a good Link layer

• It works!

• It supports the network layer above it.

• It allows low-power operation

• It allows highly reliable communication

• It allows for low latency

• It allows bursts

• It avoids contention

• It is simple, robust and flexible


The “Idle Listening” Problem

• The power consumption of “short range” (i.e., low-power) wireless communications devices is roughly the same whether the radio is transmitting, receiving, or simply ON, “listening” for potential reception

– includes IEEE 802.15.4, Zwave, Bluetooth, and the many variants

– WiFi too!

– Circuit power dominated by core, rather than large amplifiers

• Radio must be ON (listening) in order receive anything.

– Transmission is infrequent. Reception α Transmit x Density

– Listening (potentially) happens all the time

Total energy consumption dominated by idle listening


MAC Caution

• The idle listen problem is often associated with Media Access Control (MAC) protocols,

– TDMA, CSMA, …

• but MACs provide arbitration among multiple transmitters attempting to utilize a shared medium simultaneously.

– Reduce Contention and associated loss.– May involve scheduling (TDMA) or transmission detection (CSMA)

• The problem here is the opposite.– Most of the time, nothing is transmitting.– Avoid listening when there is nothing to hear.– Scheduling and detection are involved, but to determine when to turn

on receiver, rather than when to turn off transmission.

• You need Media Access Control, but that is not enough


What we mean by “Low Power”

• 2 AA => 1.5 amp hours (~4 watt hours)

• Cell => 1 amp hour (3.5 watt hours)

Cell: 500 -1000 mW => few hours active

WiFi: 300 - 500 mW => several hours

GPS: 50 – 100 mW => couple days

WSN: 50 mW active, 20 uW passive

450 uW => one year

45 uW => ~10 years

Ave Power = fact * Pact + fsleep * Psleep + fwaking * Pwaking

* System design

* Leakage (~RAM)

* Nobody fools mother nature


Where the energy goes

• Sleep– 7 uA for TI MSP

• Sensing

• Transmitting results

• Management Traffic

• Routing Structure Maintenance– only parent tracking for leaf

•Listening• Forwarding

– non-leaf

• Overhearing packets destined for others


Communication Power Consumption

Sleep~10 uA

Transmit~20 mA x 1-5 ms[20 - 100 uAs]

I

I

Time

Time

Listen~20 mA

Receive~20 mA x 2-6 ms


Radio

• Trade-offs: – resilience | performance => slow wake up

– Wakeup vs interface level

– Ability to optimize vs dedicated support

* Polastre, Culler, BMAC, Sensys 2005


Brief ComparisonVendor TI Crossbow MeshNetics Ember DustNetworksModel CC2430 IRIS M2110CA ZigBit ZDM-A1281 EM250 SmartMesh-XT M2030Type System-on-Chip OEM module OEM module System-on-Chip OEM module

Linkhttp://www.ti.com/lit/gpn/cc2430

http://www.xbow.com/Products/productdetails.aspx?sid=264

http://www.meshnetics.com/zigbee-products/

http://www.ember.com/pdf/EM250/120-0082-000H_EM250_Datasheet.pdf

http://www.dustnetworks.com/docs/M2030.pdf

Micro Controller8051 core integrated in CC2430 Atmega 1281 Atmega1281

Integrated 16-bt XAP2b MCU Integrated

Program Memory (KB) 32/64/128 128 128 128SRAM (KB) 8 8 8 5Nominal Voltage (V) 3 3 3 1.8 3MCU Active current (mA) 9.5 8 14 8.5MCU Active RX current (mA) 26.7 24 19 35.5 22MCU Active TX current (mA) 26.9 (at 0 dBm) 25 (at 3 dBm) 18 (at 0 dBm) 35.5 (at 0 dBm) 20 (at -2 dBm)Numer of power saving modes 3 1 1 1 2

Power mode 1 (uA) 190 8 (sleep mode) 6 (power save mode) 1 (deep sleep)51 (low power networking)

Power mode 2 (uA) 0.5 10 (sleep)Power mode 3 (uA) 0.3Number of timers 4 6 6 3

Granularity of timers one general 16-bit timer four general 16-bit timers four general 16-bit timers two 16-bit general timertwo general 8-bit timers two general 8-bit timers two general 8-bit timers one 16-bit sleep timerone MAC timer

ADCprecision (bits) 12 10 10 12# channels 8 8 4 4DMA Yes Yes

Radio CC2430 AT86RF230 AT86RF230 EM250 IntegratedData rate (kbps) 250 250 250 250 250Receiver Sensitivity (dBm) -92 -101 -101 -97 -90Current at RX (mA) 17.2 16 16 27Current at TX at 0dBm (mA) 17.4 (at 0 dBm) 24.3 (at 0 dBm)Current at minimum TX (mA) 18.3 (at -25.2 dBm) 10 (at -17 dBm) 10 (at -17 dBm) 19.5 (at -32 dBm)Current at maximum TX (mA) 32.4 (at 0.6 dBm) 17 (at 3 dBm) 17 (at 3 dBm) 27 (at 3 dBm)Any link or signal indicator RSSI, LQI RSSI, LQI RSSI, LQI RSSI, LQINon standard modeHow high in the stack 802.15.4/ZigBee 802.15.4/XMesh 802.15.4/ZigBee 802.15.4/ZigBee 802.15.4/TSMP


Power States at Node Level

Sleep WakeUP Work Sleep WakeUP Work

Active Active

Telos: Enabling Ultra-Low Power Wireless Research, Polastre, Szewczyk, Culler, IPSN/SPOTS 2005

http://www.cs.berkeley.edu/~culler/papers/spots05-telos.pdf

http://www.cs.berkeley.edu/~culler/papers/spots05-telos.pdf


3 Basic Solution Techniques

• Listen during scheduled time slots– Arrange a schedule of communication opportunities– Maintain coordinated clocks and schedule– Listen during specific “slots”– Many variants:

» Aloha, Token-Ring, TDMA, Beacons, Bluetooth piconets, …» S-MAC, T-MAC, PEDAMACS, TSMP, FPS, …

• Sampled Listening– Listen for very short intervals to detect eminent transmissions– On detection, listen actively to receive– DARPA packet radio, LPL, BMAC, XMAC, …– Maintain “always on” illusion, Robust

• Listen after send (with powered infrastructure)– Generally, device is not listening and will not receive.– After it transmits to a receptive device, it listens for a time window– Many variants: 802.11 AMAT, Key fobs, remote modems, …

• Many hybrid optimizations possible


Communication Power – Passive Vigilance

• Listen just when there is something to hear …

Sleep~10 uA

Transmit~20 mA x 1-5 ms[20 - 100 uAs]

I

I

Time

Time

Listen~20 mA

Receive~20 mA x 2-6 ms


Approaches

• Powered Router / Duty cycle Leaf

• Coordinator / Beacon

• Network schedule

• Preamble Sampling

• Slotted Preamble Sampling

• Quasi-scheduled


Powered Router + Leaf

• Algorithm– Leaf nodes never forward, turn off when ever they like

» Typically a minimum frequency

– Forwarding nodes never turn off

– Special case transmit to leaf

» Buffer and wait till hear

• Usage– Ember, Sensicast, Millennial, Figure 8

» All started with fancier schemes and gave up


Scheduled Listen

• Pave Psleep + Plisten Tlisten / Tschedule + Pxmit Txmit / Txmit-interval

+ Pclock-sync-ave + Pdiscover-ave

• Communication to maintain synch. sets lower bound.

• Full power listen to discover and join schedule.

P

P

Time

Time

Transmit duringListen interval

Listen intervalsno transmit

Synchronizationskew

Tperiod

Tschedule


Schedule Mechanisms

• Compute schedule off-line and distribute it to the nodes– Requires some unscheduled communication mechanism to perform survey

of who-communicates-with-whom and who-interferes-with-whom, collect results, and distributed schedule.

– Easy for star and clique.

– Changing conditions, additions and deletions are problematic

• Define set of slots, advertise, resolve– Relatively easy for star, clique, or ring.

– Complicated for general mesh. Essentially constraint resolution.

– Typically, coordinator schedules for one-hop neighbors and coordinators (cluster heads) stay powered.

» As network grows, leaf-fanout trends to 1

– Embedded a “scheduling tree”

• Essentially the traffic-light coordination problem, but the roads and flows are changing.


Communication Scheduling: TDMA

• Time Division Media Access– Each node has a schedule of awake times

– Typically used in star around coordinator

» Bluetooth, ZIGBEE

» Coordinator hands out slots

– Far more difficult with multihop (mesh) networks

– Further complicated by network dynamics

» Noise, overhearing, interference


S-MAC

• Carrier Sense Media Access

• Synchronized protocol with periodic listen periods

• Integrates higher layer functionality into link protocol

– Hard to maintain set of schedules

• T-MAC [van Dam and Langendoen, Sensys 2003]

– Reduces power consumption by returning to sleep if no traffic is detected at the beginning of a listen period

Schedule 2Schedule 1

Wei Ye, USC/ISI

Node 1

Node 2

sleeplisten listen sleep

sleeplisten listen sleep

sync

sync

sync

sync

Ye, Heidemann, and Estrin, INFOCOM 2002


Flexible Power Scheduling• TDMA-like scheduling of listening slots

– Distributed allocation

– CSMA with LPL within a slot

• Node allocates – listen slots for each child

– Transmission slots to parent

– Hailing slot to hear joins

• To join listen for full cycle– Pick parent and announce self

– Get transmission slot

• CSMA to manage media– Allows slot sharing

– Little contention

• Reduces loss & overhearing

• Connectivity changes cause mgmt traffic

Holte, PhD 2004, IPSN 2004


IEEE 802.15.4 SuperFrame


Coordinator / Beacon

• Algorithm– Coordinator broadcasts periodic beacon assigning speaking/receiving slot for each node– Node listens for full beacon period to learn slot– Listens / Recv / Transmit in each slot to maintain synch and communicate

• Usage– 802.15.4 provision, baked deeply into Zigbee

» No Zigbee provider uses it, just powered routers + leaf

• Strengths– Localized timesynch, not whole network

• Weaknesses– Fragile: can’t comm till find coordinator and bind. What happens if none around. Lose

coordinator temporarily.– Adaptation

» Scan when joining network.» When to reclaim slots

– Utilization» Only fraction of channel available, even when only one node needs to burst

– Efficiency» Need to participate in beacon coordination even if nothing to send or receive

– Routing» Within cell easy, need to coordinate across multiple coordinators

• Power bounds– Beacons + time synch


TSMP

• Connectivity Survey of entire network on every channel.

• Offline computation of a communication opportunity schedule.


Scheduled Listen Tradeoffs

+ Transmission scheduling that reduces contention, reduces loss, reduces retransmission, saves energy.

– regardless of whether reception is scheduled.

+ Transmission and Reception costs predictable in steady state (synched and stable)

-- Listening Cost: Plisten Tlisten / Tperiod – Pay the cost even when nothing is received.– Tlisten must be greater than the maximum synchronization variation.

• Delay– In the worst case, delay is Tcycle per hop. Expected Tcycle/2. Aligning

slots for one direction of flow increases delay for others.– Jitter in latency is small,

» except that loss requires retransmission and whenever fanin exceeds reserved slots for next hop

- Bandwidth- Limited as 1/Tschedule even if rest is unused


Scheduled Listen (Cont)

-- Time synchronization required– Frequency of time-synch protocol communication increases

with required accuracy and with variations in clocks.– If you lose synch, it is hard to regain it because you don’t

know when the node is listening. You must transmit continuously or wait for it to go into a watchdog full listen.

– Reducing listening cost (Tlisten) requires tightening synch accuracy.

– Having paid the cost, timestamping data is easy, but generally overkill.

– Nodes cannot “go silent” as they will lose synch.

-- Discovery/Join cost– generally requires long period of full listen or long period of

announce transmission.– The less the nodes are listening, the more costly the

discovery.

• Fragile (with scale)• Rigid (with scale and with variation)


Sampled Listen

• Relatively frequent, Very short “Listen Samples”

• Transmit (infrequent) increased to cover Tperiod

• Pave Psleep+ PlistenTsample / Tperiod + PxmitTperiod / Txmit-interval

• No time synchronization required.

• Rapid join and at low power.

• Scheduled sleep, wake times, time synch, etc. built on top for further reductions in power

P

P

Time

Time

Tperiod

Versatile Low Power Media Access for Wireless Sensor Networks, Polastre and Culler, Sensys04

http://www.cs.berkeley.edu/~culler/papers/sensys04-bmac.pdf


Sampling Mechanisms

Goal: Shift continuous listen burden to infrequent transmit

• Preamble Sampling (DARPA 1984) and Low-power Listen (UCB, 99)

– TX: extend preamble to listen period.

– RX: samples, hears preamble, Receives till START as usual.

– Requires low level link access (e.g., RFM1000)

• Repeated Transmit– Transmit packet multiple times to fill listen period.

– RX: CCA to sample, ON to receive packet

• Wake-up Packets– Transmit (small) wake up packets to cover sample, then packet

Preamble Header Data PayloadStart Trailer

Typical Frame Structure


Low Power Listening

• Extremely short CCA (500 us)

• Use short ‘chirp’ messages as preamble– Addressing minimizes overhearing cost

– Data time receive cost independent of sample period

• Communicate with anyone anytime– Always-on illusion manageability, IP friendly


Optimal Listen Sampling

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0 20 40 60 80 100 120 140 160 180

Listen Interval (packet times)

Com

m. C

ost

TX cost

RX cost

Listen cost

Total

L opt


Sampled Listen Trade-offs

+ Responsive– Unscheduled communication at Tperiod per hop, not Tschedule

+ Adaptive– Joins, leaves, changes in connectivity don’t require rescheduling– Rapid discovery

+ No synchronization cost– Can run time synch on top to further reduce costs– Don’t depend on it for correctness

+ Resilient– When forced to listen, do so at low power

+ Flexible– Tperiod configurable to workload, even on per-node basis– Permits a wide array of further optimizations to eliminate reduce wakeup costs and

listening samples

• Bandwidth+ Node can burst at full link bandwidth, not 1/Tschedule

- unoptimized bandwidth limited by 1/ Tperiod

- Transmit cost increases Tperiod

– So does receive cost if you do it naively


Link Layer – Optimizations

• Scheduling– Remove preamble cost for a subset of neighbors

– Reduce overhearing

• Packet pending– Efficiently transmit packet trains

• Use 802.15.4 data frame as acknowledgement– Addressing, secure, synchronization information

Basic LPL Phase-Synchronized LPL Packet Train


Link Layer – Abstraction

• Media Management Control– Local Media: Sample Period + Phase

– Remote Media: Sample Period + Phase + Pending

• Neighbor Connectivity Statistics– Signal Quality (RSSI)

– Success Rates (windowed average)

Lin

k

Media Management ControlLink StatsRemote Media Local Media

Data

Neighbor Table

Addr Period Phase Pending RSSI Success

Sample Period

Sample Phase

Ack

Ph

y


Communication Power Model

• Sleep Current = Is uA

• RX packet (n) = R0 + n* Rbyte uAs

• RX wake and receive = Rw + RXpacket(n)

• RX overhear and reject: Rrjt

• TX packet (n,c) = T0 + n* Tbyte + W* Iw uAs– Radio Wakeup, Clear Channel Assessment, Transmit

• Listen = Lw / W uA


Bottom Up Power Modeling

Receive 17 - Header

0.000

5.000

10.000

15.000

20.000

25.000

3.00E-02 3.20E-02 3.40E-02 3.60E-02 3.80E-02 4.00E-02 4.20E-02 4.40E-02

mA

Receive 127

0.000

5.000

10.000

15.000

20.000

25.000

2.00E-02 2.20E-02 2.40E-02 2.60E-02 2.80E-02 3.00E-02 3.20E-02 3.40E-02m

A

Receive packet with 15.4 header (17B) - one EUID64, one Short16

TelosB configuration: TI MSP430-F1611 + TI CC2420

Receive max 15.4 packet (127B)

Linear model:

Sta

rt O

scill

ator

War

m U

p -

AG

C

RX

byt

es

SP

I tra

nsfe

r

TX

ack


Energy Cost of the Primitives

Channel SampleChannel Sample OverhearOverhear

Transmit(26 bytes)

Transmit(26 bytes)

Transmit(127 bytes)

Transmit(127 bytes)

Receive(26 bytes)

Receive(26 bytes)

Receive(127 bytes)

Receive(127 bytes)

Preamble + Transmit(64 ms)





Link-Level Power Model


Empirical Power Model


MAC => Media Management

• Sampled Listening– Darpa Packet Radio, LPL, BMAC, XMAC, …

– Maintain Always-On Illusion, robust

• Scheduled Communication– S-MAC, T-MAC, PEDAMACS, TSMP, FPS, …– Low power when in sync, but fragile

• Hybrid Optimization– WiseMAC, SCP-MAC, …– Streaming – one wake up, many packets– Optimistic scheduling

• Learn when likely to listen• Use ack feedback to detect and repair

=> 802.15.4e ???


Open Issues for Research and Industry

• Sampled vs Synchronized– Is time synchronization required for the network to work at all?– Or is it an optimization

• Multi-channel usage– Frequency agility– Increased diversity– Impact on prediction, link assessment

• Distributed vs. Centralized– How much is determined locally vs network wide

• Fundamental lower bounds– energy is required to maintain structure– Where is the optimal point

• Visibility– How does network level inform link of relevance– How does link level inform network of status


Robustness and Reliability


Connectivity

• Much of the CS work on network protocols– Routing, cluster head formation, topology formation, …

assumes a unit disk model– If Distance < R, Connectivity = 1, otherwise 0

• EM models based on fading, signal-to-noise ratio (SNR), modulation, and coding.

– PRR (packet receive rate) for SNR (), frame size (f)…

– Nakagami and Rayleigh Fading

For Mica2 with FSK CC1000


Real World Example – open surface

• 2003 study of 100-200 first generation motes placed in regular grid in open tennis court.

• RFM 916 MHz ASK RF transceivers with simple whip antenna.

• Variation in Packet Receive Rate (PRR) from each transmitter.

Taming the Challenges of Reliable Multihop Routing in Sensor Networks, Alec Woo and David Culler, ACM SenSys Nov. 2003.

D. Ganesan, B. Krishnamachari, A. Woo, D. Culler, D. Estrin, and S. Wicker, "An Empirical Study of Epidemic Algorithms in Large Scale Multihop ," Intel Research, IRB-TR-02-003, Mar. 14, 2002


Analytical Models with Mulipath

Path loss with distance d

Bit error rate

19.2 kbps

30 kHz

PRR

Marco Zuniga, Bhaskar Krishnamachari, "Analyzing the Transitional Region in Low Power Wireless Links", IEEE SECON 2004.

http://ceng.usc.edu/~bkrishna/research/papers/channelmodellingSECON04.pdf


PRR vs Distance in practice


Example: TI CC2420

• IEEE 802.15.4 compliant

• 2400 – 2483.5 MHz RF tranceiver

• O-QPSK Direct Sequence Spread Spectrum (DSSS)– 250 kpbs data rate, 2 Mchips/s

• 0 dBm (1 mW) transmit power

• -95 dBm receive sensitivity

• 30/45 dB adjacent channel rejection

• 53/54 dB alternate channel rejection

• PIFA PCB antenna

http://focus.ti.com/lit/ds/symlink/cc2420.pdf

http://focus.ti.com/lit/ug/swru043/swru043.pdf


Channel Modulation

• ASK - Amplitude Shift Keying– Rene, Mica1 – RFM1000

• FSK - Frequency Shift Keying– Mica2 – CC1000

• O-PSK - Orthogonal Quadrature phase-shift keying

– Telos, TelosB, MicaZ - 802.15.4

FSK

http://upload.wikimedia.org/wikipedia/commons/3/39/Fsk.svg


O-QPSK, RSSI, CCI/LQI

• Chip Correlation Indicator


Real World Example

• Heavy metal environment

• Operating machinery

• Changing environment

• Electronic Equipment

• …


RSSI - Stationary


RSSI - Driving


Industrial Setting (Sexton)

1

4

3

2

6

5

IEEE 802.5.4 Channels


Channel Fading

• Multipath effects– Varies by position

– Varies by frequency

– Varies over time

• Overcome with diversity– Time diversity

» Retransmission

– Spatial Diversity

» Multiple antennas

– Path diversity

» Alternative receivers

– Frequency diversity

» Spreading, Multiple channels

Radio Channel Quality in Industrustrial Wireless

Environments, Dan Sexton, et. al SICON'05


WIFI Relationship


Variations over time

•Understanding the causes of packet delivery success and failure in dense wireless sensor networks, Kannan Srinivasan, Prabal Dutta, Arsalan Tavakoli, and Philip Levis


Received Signal Strength ?


Noise

• Periodic WiFi beacon


802.11 / 802.15.4 Interference


Classic Media Access Control• CSMA wireless MAC

– MACA, MACAW, IEEE 802.11

– Listen (CCA - clear channel assessment)

– If channel busy, back off (exponentially) and retry

– Send RTS – request to send

– Wait for destination to respond with CTS (clear to send)

– On CTS, send data packet

– and on overhearing CTS to another backoff

• TDMA– Divide time into periodic slots

– Assign slots for individual nodes to transmit

A B C

Hidden Terminal

A B C

Exposed Terminal

D


Lesson 1: Listen

• MACs deal with mitigating high contention– In low duty cycle networks, contention is low regardless– Can occur due to highly correlated behavior

» Example: all node sample periodically» Separate sample from report, shift phase

• WSN packets are not the only thing on the air • CCA essential for determining noise also

– Don’t transmit over a noisy (or busy) channel– Required even with TDMA techniques

• Low-Power Wireless Packets are small– IEEE 802.15.4 limit 127 bytes– PRR drops rapidly with frame size!– Handshake is extremely energy expensive

• In a mesh, hidden terminals and exposed terminals are EVERYWHERE


Lesson 2: Link Level Retransmission

• There will be packet loss, even on good links

• PRR(h hops) = PRRh

• Link-level acknowledgements are essential.– Provides ability to estimate the quality of the link!

0

5

10

15

20

25

-15 -10 -5 0 5 10 15

ms

mA

CCA Receive Ack


Lesson 3 – Asymmetric Links

• Asymmetric Links are common– Non-isotropic antenna, propagation, multipath

– Variations in transmit power or receive sensitivity among nodes

– Variations in noise level at receiver and transmitter

• Cannot assume the reverse link is good– Verify it!

– Continuously


Lesson 4 – Variation in Time

• Link quality varies in time due to many factors

• Changing physical environment.

• Changing RF noise

• Changing traffic from other nodes– Essentially additional noise, but right in the channel

• Cannot expect to determine connectivity in advance and just use it.

– Zigbee route determination?

– TSMP connectivity “survey”?


Lesson 5: Reception => Link

• Reception of a packet from a node does not imply the link is good.

• Will infrequently receive good packets even on a bad link.

• Many nodes far away => will frequently receive a packet from one of them.

• Neighbor table cannot contain an entry for everyone that a node has heard from!

• Must track only a small important subset?


Lesson 6: No Reception => No Link

• Capture Effect

• A node will receive packets with low signal strength (at least if the noise is low)

• If a strong packet appears at the receiver while it is in the middle of processing a weak packet

– Both will be lost

• And collision => loss– If a weak packet arrives while receiving a strong packet, the

strong packet will be received if the weak does not exceed the SNR threshold.

Exploiting the Capture Effect for Collision Detection and RecoveryWhitehouse, K.; Woo, A.; Jiang, F.; Polastre, J.; Culler, D.Embedded Networked Sensors, 2005. EmNetS-II. The Second IEEE Workshop onVolume , Issue , 30-31 May 2005 Page(s): 45 - 52


Lesson 7: Determining Link Quality

• RSSI, LQI, and Packet Sequence numbers tell receiver about the inbound link.

• Need to reply to the sender for it to know if that is a good link.

• The 2-way exchange provides way to filter out asymmetric links.

• Link ACKs inform sender about outbound link.

• The more channels you track the less you know about each

• Need a simple metric for network layer to use


Summary

• Many of the best protocols are opportunistic – Use whatever connectivity occurs

• Topology determination and route selection is a constant and gentle process

– Passive monitoring wherever possible

– Use every piece of information available to track quality

• Concentrate link estimation on the few important candidates.

• Additional network density helps reliability– If the media management is done right


Power-aware Routing

• Cost-based Routing– Minimize number of hops

– Minimize loss rate along the path

– Perform local retransmissions, minimize number along path

• Energy balance– Cost-function prefers nodes with larger energy resources

• Utilize redundancy– Nodes near the sink route more traffic, hence use more energy

– Give them bigger batteries or provide more of them and spread the load

– Randomize routes

• Utilize heterogeneity– Route through nodes with abundant power sources


Example Mesh Routing Implementations

• TinyOS– MintRoute (2003) – cost based collection routing over Low Power Listening on CC1000 – MultiHop LQI (2005) – LQI cost-based collection on 15.4– Drain / Drip – (2006)

• ZIGBEE– Star, Tree, and Mesh defined– Tree generally rejects– Only mesh profiles appear to be utilized– AODV (build tree from any node with full bcast)– No power management defined

• Ember– Proprietary extensions / simplifications of zigbee

• Millennial– Powered backbone

• Crossbow xmesh / TrueMesh – extends TinyOS mint-route with local slots– Preamble Chirps on slot edge, listens only on edges

• Dust SmartMesh– Survey all pairs connectivity– Compute off-line schedule of Source->Dest opportunities– Maintain global time slots with single pair per slot– Periodic collection to single root at 1 packet per minute

• Arch Rock– 15.4 packet-based Sampled Listening– IP routing over 6LoWPAN, triply-redundant mesh


In-network Processing

• Best way to reduce communication cost is to not communicate

• Compute at the sensor– Only communicate important events

• Compute over localized regions of the network– Distributed detection

– Validation


Previous Work on Duty-Cycling – Dual-channel MAC: PAMAS [SR98]

• Synchronization of PAMAS: – Each node sends and receives RTS/CTS messages over

control channel, which is always turned on.

• Power Saving of PAMAS:– Data channel is turned on when activity is expected.

• Pros: Easy to implement.

• Cons: Requires dual-channel, control channel still consumes power

Node 1

listen

active

sleepData Channel

Control Channel(always on)

Node 2Data Channel


RTS CTS

Node 3Data Channel


active

sleep

RTS CTS

listen

sleep

sleep sleep


Previous Work on Duty-Cycling – Virtual Clustering (S-MAC, T-MAC)

• Power Saving of S-MAC:– Each node is turned on only for its time slot.

• Synchronization of S-MAC:– Each node sets up its own schedule by

– (1) Sending its SYNC packet when it hasn’t found neighbor.

– (2) Following schedule of a neighbor whose schedule is earlier.

– Channel contention is addressed by RTS/CTS.

• Pros: Algorithm can be applied to any platform.

• Cons: Overhead of RTS/CTS, Atmel specific implementation.

Synchronization of S-MAC

Node 1

active

sleep

Node 2

sleep

activeTd1

Td1

Follow the schedule of Node 1

active

Rebroadcast the schedule of Node 2 after a random delay

active

Td1

Td2 Td2

sleepsleep

SYNC packet


Previous Work on Duty-Cycling – Low-power listening (B-MAC)

• Power Saving for B-MAC:– Each node sleeps after listen with no channel activity.

• Synchronization for B-MAC:– Preamble from sender node is long enough to span Tperiod.

• Pros: No separate synchronization step is needed.

• Cons: Long preamble is not supported on Trio node.

time

125 ms 125 ms 125 ms 125 ms

ReceiverReceiver

Sender preamble

data

data

data

Source: XMesh Routing Layer, Martin Turon et al. TinyOS Tech Exchange 2005.

Tperiod

David E. Culler University of California, Berkeley

Documents

Transcript of David E. Culler University of California, Berkeley