On Coverage and Routing in CDMA Ad-hoc Networks

R. Kempter, P. Amini, C. Schlegel and B. Farhang-Boroujeny

Brigham Young UniversityNovember 29, 2005

Outline

1) Random Channel Access in CDMA (Ad-hoc) networks→ ALOHA→ Spread ALOHA→ Random Packet-CMDA (RP-CDMA)

2) CDMA detection→ novel detector: Partitioned Spreading

3) Different detectors in the Ah-hoc scenario→ randomly generated networks

4) Impact of suboptimal detection on ad-hoc connectivity→ FLUX Mobile Robot Testbed

5) Conclusion

Random Channel Access: a brief overview

The maximum throughput of ALOHA isS=Gexp(-2G) → 0.18

(only 18% of a fully coordinated system)

For packet services, random access provides fundamental advantages

Advantages:• Random access is self regulating• Random access is load adaptive• Random access simplifies infrastructure

Disadvantages:• Load regulation is required• Channel sensing is required• Return channel is required

Random Channel Access: Spread ALOHA

Code-Division Multiple-Access(CDMA)CDMA uses spread-spectrum (SS)signals with unique signature se-quences for different users

CDMA:• Allows the co-existence of

signals on the same channel• CDMA is insensitive to

narrowband interference• CDMA is sensitive to unequal

received powers?

An SS signal is one whoseFourier bandwidth is muchlarger than its information bandwith

After despreading, one essentially obtains Classical ALOHA

S=Gexp(-2G) → 0.18

Packets create multiuserinterference!

Spread ALOHA is the combination ofsingle-code CDMA with ALOHA

But

Random Channel Access: novel idea → RP-CDMARandom Packet CDMA Packet StructureConsists of a distinct header and a (flexible) data portion

System Characteristics:• The Header Channel is essentially a low-traffic Spread ALOHA channel• The Data Channel needs to be able to handle a possibly large number of

concurrent transmissions

Random Channel Access: RP-CDMA

System Effect:The particular novel packet format effectively separates communications into a Header and a Data Channel

The ratio of header length Lh to data portion length Ld is a crucial system parameter.

Lh=50bits and internet traffic:50bytes, p=0.5 / 500bytes p=0.4 / 1500bytes p=0.1→ Average Ld/Lh=60

Example

RP-CDMA performance limits: packet collisions

Time

Collision at the base station? - a packet level point of view

Payload portionwith randomspreading

Header portionwith commonspreading

Packet from node x

Packet from node y

Header collisions do not automatically occur as soon as two or more headers overlap at the packet level.

As long as the headers do not overlap at the chip level, they are still separated by the processing gain.

Collision? -on the chip level

RP-CDMA performance limits: collisions, Nh=Np=10

• RP-CDMA promises great improvements in system performance• S-ALOHA is collision limited - MUD cannot improve performance

thus MF detection maximizes throughput

Collision perspective

RP-CDMA performance limits: interference, P/σ2=15dB

• RP-CDMA requires low SNR header detection• RP-CDMA requires Multiuser (Joint) detection

Interference perspective

Outline

1) Random Channel Access in CDMA (Ad-hoc) networks→ Spread Aloha→ Random Packet-CMDA (RP-CDMA)

2) CDMA detection→ novel detector: Partitioned Spreading

3) Different detectors in the Ah-hoc scenario→ randomly generated networks

4) Impact of suboptimal detection on ad-hoc connectivity→ FLUX Mobile Robot Testbed

5) Conclusion

Partitioned Spreading Detection

• Spreading waveform is partitioned into M sections and interleaved for transmission

• Partitions are individually decoded and iteratively combined:Turbo principle

Partitioned Spreading DetectionCritical System load: max. K/N such that a given QoS parameter can be fulfilled: Interference-free SNR

• Partitioned Spreading resolves (all) multiuser interference• Partitioned Spreading is near-far resistant• Partitioned Spreading is a non-linear detector front-end,

not an FEC decoder

RP-CDMA: proposed receiver structure

Outline

1) Random Channel Access in CDMA (Ad-hoc) networks→ Spread Aloha→ Random Packet-CMDA (RP-CDMA)

2) CDMA detection→ novel detector: Partitioned Spreading

3) Different detectors in the Ah-hoc scenario→ randomly generated networks

4) Impact of suboptimal detection on ad-hoc connectivity→ FLUX Mobile Robot Testbed

5) Conclusion

Comparison Between Different Receivers: ad-hoc scenario• nodes are placed randomly

on unit-square• focus on detectors: only payload• exponential pathloss, γ=2

• PRX =PTX

1+ d( )γ

• PTX /σ2=11dB

• choose detection threshold of 3dB→ fully meshed network

• N=6

As detector performance decreases with load, nodes at network perimeterloose connectivity → cell shrinking

• ratio of max. distance intact route/max. distance route as f(load)• number of network partitions as f(load)

Figures of merit

Comparison Between Different Receivers

Matched Filter Receiver:each signal is filtered with a filter matched to its own waveform.the signal-to-noise ratio depends on the number and powers of other users

Decorrelating Receiver:inverts the channel to completely eliminate interference. This resultsin a loss of energy of each user, dependant on the user population.interference no longer depends on the power of other users

Minimum Mean Square Error Receiver:establishes a filter with coefficients in the mmse-sense betweenthe packet’s sequence and the multi-access interference

Successive Canceller employing strong code:successively canceling the interference of already decoded packets from remaining packets, optimal iff exponential power distribution

Detector Performance: matched filter

Detector Performance: decorrelator

Detector Performance: successive canceler

Detector Performance: MMSE

Detector Performance: partitioned spreading

Outline

1) Random Channel Access in CDMA (Ad-hoc) networks→ Spread Aloha→ Random Packet-CMDA (RP-CDMA)

2) CDMA detection→ novel detector: Partitioned Spreading

3) Different detectors in the Ah-hoc scenario→ randomly generated networks

4) Impact of suboptimal detection on ad-hoc connectivity → FLUX Mobile Robot Testbed

5) Conclusion

Recap: ad-hoc networks, facts

RP-CDMA with Partitioned Spreadingaddresses and possibly solves important problems in the ad-hoc scenario

• The RP-CDMA packet format is “self-contained”, a necessity for (mobile) ad-hoc networks where no resource allocation (code exchange) is possible/desired

• Compared to Spread ALOHA, RP-CDMA offers greatly reduced collision probability and multiuser detection can be used

• Joint detection can eliminate the hidden terminal problem

• Partitioned Spreading is near-far resistant: reliable detectionin varying environments (without power control)

Example: ad-hoc network based on FLUX Mobile robotsSnapshot of the FLUX Mobile Robot Testbed equipped with 900MHz Mica2 Motesas well as IEEE 802.11b cards. The testbed is open for experimenters [FLUX]

[FLUX] David Johnson, Tim Stack, Russ Fish, Dan Flickinger, Rob Ricci, Jay Lepreau, “TrueMobile: A Mobile Robotic Wireless and Sensor Network Testbed,” University of Utah Flux Group Technical Note 2005-02, April 2005. Revised version to appear in INFOCOMM 2006, www.flux.utah.edu

Solve for RP-CDMAwith Partitioned Spreading. Compare to different MU-detectors

Idea

FLUX Mobile Ad-Hoc Robots: physical layer

• Assume transmission power of 15dBm, comparable to IEEE802.11 cards

• Header length Lh=50bits, constant Ld/Lh=25

• Nh=Np=14

• Pathloss for Mica2 motes [HaePuc05]: Prx=min(1,10d-3)

• Assume detection threshold/noise of 8dB

→

[HaePuc05] Martin Haenggi, Daniele Puccinelli, “Routing in Ad Hoc Networks: A Case for Long Hops,” IEEE Communications Magazine, October 2005

Max. transmission radius of R=3m, leading to a fully connected multihop network

FLUX Mobile Ad-Hoc Robots: network layer

• Source-based shortest hop routing, random source and destination

• In ad-hoc networks, the probability of packet transmission:

Ptrans. packet = Pgenerate packet + Pforward packet

Pforward is a function of node connectivity, thus network specific

FLUX Mobile Ad-Hoc Robots: Ptrans and packet collisions

• due to packet forwarding, Ad-hoc networks are quickly highly loaded!

• in average only 0.4 packets/node arelost due to collisions:→ Compare to hidden node problem in IEEE 802.11!

Observation

FLUX Mobile Ad-Hoc Robots: intact links

FLUX Mobile Ad-Hoc Robots: number of network partitions

• Only Partitioned Spreading maintains original network connectivity. All other detectors lead to severe connectivity loss (isolated subnets).

• In the case of the mobile robot ad-hoc network this can render the network useless

Observation

FLUX Mobile Ad-Hoc Robots: conclusions

In contrast to IEEE 802.11 networks, the sender has no means of knowing if its transmission might overload the sender

Feedback from the receiver required

Addition of “busy” indicator on network or link layer.As sender receives indicator, all transmissions have to be stalled

• Narrowband busy tone: not too much interference for data (CDMA)• In the case of AODV: add busy bit to hello and any other broadcast

messages such as RREQ

The weakest link determines the performance of a subnet

Less capable nodes should be restricted to thenetwork perimeter

Conclusion 1/2

• Spread ALOHA is collision limited, thus multiuser detectioncannot improve system performance

• Random Packet CDMA (RP-CDMA), allows for asynchronous, connectionless data transfer in ad-hoc and base station centric CDMA networks

• RP-CDMA is not collision limited, thus multiuser detection can be used to improve system performance

• Partitioned Spreading CDMA, where spreading sequences are partitioned and interleaved before transmission was presented.

At the receiver, the partitions are individually filtered and iteratively decoded → Turbo principle.

• CDMA with Partitioned Spreading detection is near-far resistant

Conclusion 2/2

• PS compared to the Matched Filter, the Decorrelator, the MMSEand a Successive Canceler in a randomly generated ad-hoc network and in a scenario based on the FLUX mobile robot testbed

• Only Partitioned Spreading detection can maintain original network coverage and network fragmentation can be avoided

• All other (multi-) user detectors are comparably suboptimal!

• In contrast to IEEE 802.11, in CDMA networks senders do not knowwhen overloading the receiver.→ “busy” indicator needs to be added to CDMA (ad-hoc)

protocols - either on the routing (AODV) or link layer

• Less capable nodes (MF, Deco, MMSE, Succ. Canceler) should be restricted to the perimeter of the ad-hoc network

The End

Spread ALOHA