Learning Center
Plans & pricing Sign in
Sign Out

Resource Allocation in Wireless Networks



   Project progress reports due today.

   Homework 2 ready later today –
    due 6/2 (next Friday)

   Graded HW 1 and solutions ready

   Third paper summary on ad-hoc
    networks due next Wednesday.
         Ad-Hoc Wireless

   Main Characteristics
       Each node generates independent data
       Any node can communicate with any other.
       No centralized controller (self-configuring)
       Data transmitted in (short) packets
       Links typically symmetric.
       Nodes may be mobile and/or power constrained.
       Typically a large number of nodes
   Battlefield communications
   Wireless LANs
   Emergency infrastructures
   Short-term networks (e.g. convention)
   Sensor networks
       Medical applications (on-body)
       Buildings
       Wide area

   Cellular phone evolution
   Communication infrastructure for
    automated vehicles
       Automobiles
       Airplanes

Widely different channel characteristics,
distances, mobility, and rate requirements.
       Design Issues

   Link Layer design
   Channel sharing (MAC/reuse)
   Reliability/QOS
   Routing
   Network topology
   Network management/control

      Must exploit synergies
      between design layers
     Link Layer Issues
   Modulation and Coding
       Robustness
       Rate requirements
       Performance
       Adaptive techniques
            Rate, power, BER, code, framing.

   Bandwidth requirements
       Control and communication requirements

   Power control
       Typically distributed

   Antenna design
       Smart antennas
       Multipath mitigation
       Multiuser detection

   Connectivity
       Binary or adaptive.
      Channel Access

   Frequency-Division
   Time-Division
   DS Spread Spectrum
   FH Spread Spectrum
   Frequency reuse
     Bandwidth efficient
     Distributed allocation
     Dynamic channel allocation
      hard for packet data
    Frequency Division

   Fixed allocation inefficient
     Hard  to implement when node
      locations dynamically change

   Distributed dynamic channel
    allocation hard to do

   FD typically only used to
    create hierarchical networks

   Fixed allocation inefficient and
    impractical (as in FD)

   Aloha
       Inefficient
       No capture

   Carrier sensing
       Hidden nodes degrade performance
       Busy tone may interfere with
        transmission to other nodes.

                      Busy Tone
      Spread Spectrum
      Code Assignment
   Common spreading code for all nodes
      Collisions occur whenever receiver
       can “hear” two or more transmissions.
      Near-far effect improves capture.
      Broadcasting easy

   Receiver-oriented
      Each receiver assigned a spreading
       All transmissions to that receiver use
        the sequence.
       Collisions occur if 2 signals destined
        for same receiver arrive at same time.
            Can randomize transmission time.

       Little time needed to synchronize.
       Transmitters must know code of
        destination receiver
            Complicates route discovery.
            Multiple transmissions for broadcasting.
   Transmitter-oriented
       Each transmitter uses a unique
        spreading sequence
       No collisions
       Receiver must determine sequence of
        incoming packet
            Complicates route discovery.
            Good broadcasting properties
       Poor acquisition performance

   Preamble vs. Data assignment
       Preamble may use common code that
        contains information about data code
       Data may use specific code
       Advantages of common and specific
            Easy acquisition of preamble
            Few collisions on short preamble
            New transmissions don’t interfere with
             the data block
        Data link control
   Packet acknowledgements needed
      May be lost on reverse link
      Should negative ACKs be used.

   Combined ARQ and coding
      Retransmissions cause delay
      Coding may reduce data rate
      Balance may be adaptive

   Hop-by-hop acknowledgements
     Explicit acknowledgements
     Echo acknowledgements
            Transmitter listens for forwarded packet
            Not possible with directive antennas.
            Large delays in FIFO queues.
            More likely to experience collisions than a
             short acknowledgement.
       Hop-by-hop or end-to-end or both.

   Determining connectivity
     SNR measurements
     Bit/Packet error rate

   Connectivity control
     Linkcan adapt to maintain
      connectivity (adapt rate, power,…)
          Interaction with routing protocol.
     Power increase may affect other
      nodes (Bambos technique).

   How many connected nodes
    constitute a network
     Or,   take what you can get.
          Routing (1987)

   Flooding
     Broadcast packet to all neighbors
     Inefficient
     Robust for fast changing topologies.
     Little explicit overhead

   Point-to-point routing
     Routes follow a sequence of links
     Connection-oriented
            Explicit end-to-end connection
            Less overhead/less randomness
            Hard to maintain under rapid dynamics.
       Connectionless
            Packets forwarded towards destination
            Local adaptation
    Route dessemination

   Route computed at centralized node
       Most efficient route computation.
       Can’t adapt to fast topology changes.

   Distributed route computation
       Each node transmits connectivity
        information to other nodes.
       Nodes determine end-to-end route
        based on this local information.
       Adapts locally but not globally.

   Nodes exchange local routing tables
       Node determines next hop based on
        some metric.
       Deals well with connectivity dynamics.
       Routing loops common.
         Routing (1999*)

   Table-driven
        Destination-sequenced distance-vector
        Clusterhead gateway switch routing
        Wireless routing protocol

   On-Demand Routing
      On-demand distance vector routing
      Dynamic source routing
      Temporally ordered routing
      Associativity-based routing
      Signal stability routing

*”A review of current routing protocols for ad hoc mobile
wireless networks,” Royer and Toh, IEEE Personal
Communications Magzine, April 1999.
    Packet Forwarding

   Overhead information
     Routing information
     Packet identifiers
     Priority/delay information
     Tradeoffs in overhead size

   Synergies of routing and
    packet forwarding with link
       Other Network

   Network Capacity
   Admission Control
   Interface with wired networks
   Security
   Upgrades
     Software   changes
     Software   radios
    Network Capacity
   Capacity limits of ad-hoc 3D
     Data rates per node
     Number of nodes

   Assumptions
       N users uniformly distributed over the interior
        of a sphere.
       Each user communicates with another user
        randomly chosen among all users.
       Signal power decays based on free space path
       All users transmit at the same power.
       No channel separation or diversity.
       Interference acts as additive white Gaussian
        Capacity Bounds
   The total number of bits that may be
    transmitted by all users, per second,
              is approximately
               C K N       3

         Proportional to the cube root of N
   Lower Bound
       Based on deterministic routing scheme.

   Upper Bound
     Similar formula
     Uses convexity
    Lower Bound Proof
   Estimate the effects of interference in the
    limit of large N.
   Construct a series of cell tessellations with
    useful properties.
   Use the weak law of large numbers to
    prove the existence of one user in each
   Specify a routing and transmitting scheme
    using time sharing.
   Determine the capacity of this scheme,
    which lower bounds the capacity of the
    best scheme.
    What has changed
      since 1985?

   Signal processing is better,
    cheaper, and lower power.
   More powerful channel codes.
   Multiuser detection and smart
   Signal strength measuring
    techniques available in radios.

   How would we leverage these
    developments to make better
    ad-hoc networks?
    Sensor Networks

   Sensor Networks
       Data highly correlated in time and space.
       Low homogeneous rates.
       Links typically asymmetric.
       Data flows to centralized location.
       Energy is the driving constraint.
       1000-100,000 Nodes
       Have a common mission.
       Very different from typical ad-hoc networks

To top