Directional Antennas for AdHoc Networks Andy Collins UW

Directional Antennas for Ad-Hoc Networks Andy Collins UW CSE 802. 11 Seminar 23 July 2003

Problem and opportunity • Ad-hoc networks are limited in capacity – More power = fewer hops but more interference • Technology will (soon) allow steerable antennas – Using phased arrays • High gain (10 X to 1000 X) • Electronically steerable (no physical rotation) • Less interference to nodes in other directions • Can we break the capacity bounds? – More gain ? = fewer hops and less interference

Major challenges • Neighbor discovery – Goal is to use hops that require directional antennas at one or both ends • How does sender know where to aim • How does receiver know to aim (and where)? • Hidden transmitters – Cannot hear nearby nodes pointing away • More interesting than geography because it changes fast • Must rethink MAC protocol

Antenna basics • Antennas are governed by two properties: – Efficiency: fraction of input power not lost as heat – Gain: concentration of energy in one direction • We focus only on gain – Can only reshape energy • The sum of power over the sphere must stay constant • Implies a direct tradeoff between gain and beamwidth • Some energy always ends up in “sidelobes” off the main beam

Antenna terminology • Gain is typically measured in decibels (d. B) – True d. B is a unitless ratio describing gain or loss through a component • Also used for amplifiers and attenuators, and path loss – Logarithmic: d. B = 10 log 10 (power out/power in) • 3 d. B is approx. twice as much power, 10 d. B exactly 10 X – Antenna gain is figured relative to isotropic (d. Bi) • An ideal omnidirectional radiator • This is the gain in the center of the beam

A gain antenna

Phased arrays • A phased array is a group of individual antennas grouped and fed together – Usually each element is omnidirectional – Pattern is determined by geometry and phase delay – Magic is in varying phase delay to each element to steer the beam without moving the elements – Long history of military use for radar – Can also use as omni receiver, and sense arrival direction • Very different from other sorts of directional antennas you’ve probably seen – Yagis, Log-periodic arrays, parabolic dishes

Modeling gain antennas • Both papers use a simplified model – Beam is uniform gain within its beamwidth – Single, uniform sidelobe • Can calculate gains and beamwidth – Max width a function of gain – Sidelobe gain a function of gain and selected beamwidth

Meanwhile, back in computer science… • Step 1: directional antennas for interference only – Use omni antennas for routing and MAC • Step 2: directional transmit / omni receive – Use geometry information to know where to send – Will use the term “DO” for “directional omni” • Step 3: double directional – Create a protocol to arrange for the receiver to aim back at the sender – Termed “DD” for “directional”

Interference reduction • What happens if we use normal omni MAC and directional data transmission? – Implies no improvement in hop count • Except for some ideas on power or processing tradeoffs – Can dramatically reduce the likelihood of collision • Especially if we also reduce power • Data transmission is most of the time spent • Must relax MAC rules to get this benefit • Ramanathan shows useful benefit

Ramanathan results Setup: 40 nodes, steered beams, “aggressive” CA, power control

Basic DMAC algorithm • A MAC algorithm for “DO” operation – Must know where receiver is to point beam – Can only get “half” the gain benefit • Although they also propose using receiver gain for data • Basically like 802. 11, but per-direction – Sender aims at receiver and sends RTS • As always, after doing aimed carrier sense – Non-receivers update DNAV for arrival direction – Receiver sends aimed CTS

DNAV table operation

Basic DMAC problems • Hidden terminals – “Asymmetry in gain”: can only hear RTS/CTS within “DO” range, but can interfere out to “DD” – Can’t hear nearby RTS/CTS when focused away – Can’t hear nearby transmissions in same direction • Deafness – Can’t tell when receiver is focused away and doesn’t hear RTS

Multi-hop RTS MAC algorithm • Can we use “DO” routing to set up “DD” hops? – Assume network is dense enough, and we have map

MMAC algorithm continued • Basically the same steps, but each repeated for “DO” and “DD” – Sender first aims a “DD” RTS towards receiver • Probably won’t be heard at receiver, unless it happens to be aimed properly for some reason • Point is to alert others along the path (who update DNAV for both arrival direction and opposite (where the receiver is) – Sender sends a “DO” RTS along the “DO” route to receiver • This is forwarded, but must be dropped rather than delayed in any queues – Receiver aims a “DD” CTS towards sender • This will be heard, because sender is expecting it • Also alerts intermediates near the receiver

MMAC problems • Hidden terminals – MMAC inherits all the DMAC hidden terminals • Directly, since it also does “DO” communication • Plus some “DD” analogs – “DD” signal may not reach all in-between nodes • It is not the case that all nodes between a pair of “DD” nodes can hear even one endpoint using “DO” reception – If gain > 6 d. Bi, then adding receiver directivity increases range by more than a factor of two – Nodes directly between “DD” pairs may not hear either end • But any node between a pair can interfere if it beamforms – And carrier sense will only work if it points towards the sender

Applicability of beamforming • What kinds of networks are a good match? – Nodes must know location and orientation • Although some algorithms may work purely by sensing incoming direction, and building only a logical map – Antenna arrays are larger than simple antennas • Typical setup is a circle of elements and 1/2 l spacing – Or about 16 cm diameter for 2. 4 GHz – But much smaller for X band above • Higher gain does require larger arrays – As well as more stable platforms • Fixed systems make a lot of sense – Vehicular too, if we can fold in mobility