MIMO Layered Multiuser Detection in Ad Hoc Networks

1
MIMO Layered Multiuser Detection in Ad Hoc
Networks

• Dr. Michele Zorzi, zorzi_at_cts.ucsd.edu
• University of California at S.Diego
• M. Zorzis work was supported under the MURI
• Work in collaboration with P. Casari and M.
  Levorato, Ph.D. students at the University of
  Padova, Italy

2
Preamble Spatial Multiplexingin point-to-point
links

• First introduced by Foschini 1, to enhance the
  raw available PHY bit rate in a point-to-point
  link by spatial superposition of multiple bit
  streams
• A TX terminal with A antennas sends a different
  stream over each antenna, and the RX terminal
  uses its own antennas to spatially separate and
  decode the incoming streams. Bit rate is A times
  higher
• Alternative approach multiple users, with
  multiused detection and interference cancellation
  at the RX 2

1 G. J. Foschini, Layered spacetime
architecture for wireless communication in a
fading environment when using multi-element
antennas, Bell Labs Tech. J., vol. 1, pp. 4159,
1996. 2 S. Sfar, R. D. Murch, and K. B.
Letaief, Layered spacetime multiuser detection
over wireless uplink systems, IEEE Trans.
Wireless Commun., vol. 2, no. 4, pp. 653668,
July 2003.
3
LAST-MUD Algorithm - 1
A antennas
K users

• Let each of the K users transmit with one antenna
• Each user sends its own transmission to the RX
• The RX sees a superposition of signals at each RX
  antenna and wants to separate them. Received
  signal
• Za is the channel vector, b the symbols, na the
  noise

4
LAST-MUD Algorithm - 2

• The decision statistics at the receiver is
• I is the space-filtered interference
• is the space cross
  correlation matrix
• n is the filtered Gaussian noise vector
• Note interfering signals in ra are jointly
  detected, those in I are unkown interference

5
LAST-MUD Algorithm - 3

• Subsequent decoding steps
• Calculate Rs pseudoinverse, i.e.,
• Re-order received symbols according to
  post-detection SNR
• Extract the index of the symbol with highest
  SNR
• Weigh the sufficient statistics vector components
  withthe -th column of to yield a scalar
  value
• Feed the scalar value into the decision block and
  estimate the corresponding symbol,
• Update vector M by erasure of symbol s
  contribution
• Update R by striking out the -th row and
  column
• Step to next symbol detection ( )

6
Ad hoc network scenario
A antennas
Used TX antennas
A antennas
A antennas
Used TX antennas
A antennas
Terminal 2
Terminal K

Terminal 1

• Each wireless terminal has A antennas available
• It uses a subset of these antennas to transmit
  (e.g.,2 transmits to 1 with 4 antennas, while K
  only uses 2)
• A RX terminal always uses all available antennas
  for detection purposes

7
LAST-MUD in ad hoc networks

• TX terminals may use more than one antenna
• Depends on data to send and channel/interf.
  conditions
• each antenna output is seen as a different user
  by PHY
• Goal rule radio access (schedule and rate) in
  order to exploit layered multiuser detection
  benefits
• Approach
• Make simulations of PHY behavior in an ad hoc
  network scenario to understand pros and cons of
  the technique
• Based on these simulations, provide design
  guidelines for a MAC layer protocol that exploits
  LAST-MUD
• Before TX, decide how much data to send and to
  whom
• During RX, decide which signals go into the
  interference cancellation term and which are
  unknown interference

8
Example of PHY results high load

• Here a terminal with 8 RX antennas listens to 4
  users transmitting 4 streams each for a total of
  16 streams
• At 50 m, the prob of correct decoding of all is
  very high
• High SNR results in good spatial separation of
  the wireless channels
• At larger distances, performance degrades because
  of the reduced SNR and resulting imperfect
  interference cancellation

Pno. of bit errors gt abscissa
9
PHY results more on high load

• The ability to receive many streams at short
  distance is very important
• In an ad hoc network, nodes will then be able to
• Decode many spatially superimposed data packets,
  if they come from near neighbors
• Decode many signaling messages (RTSs, CTSs, ACKs)
  as they are shorter and easier to decode
• Gain information about (local) network load and
  make scheduling, access, and rate decisions

Pno. of bit errors gt abscissa
See Paolo Casari, Marco Levorato, Michele Zorzi,
Some issues concerning MAC Design in Ad Hoc
Networks with MIMO communications, WPMC 2005,
Sep. 2005.
10
PHY results some conclusions

• In an ad hoc network scenario
• Signaling packets (transmitted with a single
  antenna) can be received without errors at
  significant distances
• Data packets (that require spatial multiplexing)
  may be decoded in high numbers only if coming
  from short distances
• Interference from unestimated streams is an
  issue, so that nodes should try to gain knowledge
  of traffic conditions in their neighborhood and
  decide what to decode and cancel and what to
  ignore
• Effect/benefits of channel coding still under
  study
• Some first results in the WPMC paper, more
  accurate characterization under way

11
Simulation of an ad hoc networkmain assumptions

• MATLAB simulator for LAST-MUD in ad hoc nets
• Main assumptions
• Collision avoidance mechanism based on RTS and
  CTS transmission correct DATA reception is
  confirmed by ACK
• Completely connected, grid-topology network,
  nodes within carrier sensing range of each other
  (a sort of worst case).
• Transmissions are in frames all RTSs are
  transmitted simultaneously in a slot, and so are
  CTS, DATA and ACK
• Receivers track channels for the incoming signals
  (for MUD)
• Only a limited number (32 in our results need
  to decide which ones!), the others will be
  unknown interference

RTS
CTS
ACK
DATA

RTS
CTS
ACK
DATA
12
Simulation of an ad hoc network MAC layer
operations

• Each node keeps a queue of unsent packets
• RTSs are sent to request permission to TX packets
• Antennas allocated according to length of packet
  (1000 bits per antenna per transmission). RTS
  contains this info.
• Receivers can estimate the traffic in the
  upcoming frame, thanks to the ability to receive
  multiple (all?) RTSs
• CTSs are issued based on the received RTSs
• Based on traffic and interference/channel
  estimates, potential receivers decide how many
  transmissions (and which ones) to allow
• MUD Interference cancellation
• Each receiver allocates its degrees of freedom to
  receive intended data and track and cancel (some
  of) the interferers
• How this is done is important for good
  performance!

13
Simulation of an ad hoc network MUD policies

• NFT (do Not Follow Traffic)
• The node only tracks the streams directed to
  itself and neither decodes nor cancels other
  streams all transmissions meant for other nodes
  are part of the unknown interference term
• PFT (Partially Follow Traffic)
• Each node first uses its degrees of freedom to
  track the channels of all incoming streams, and
  then uses the remaining resources (if any) to
  detect and cancel interference coming from other
  streams
• FT (Follow Traffic)
• Each node tracks the channel of the strongest
  signals (with the obvious constraint that at
  least one must be meant for it), thereby
  providing maximum interference capabilities and
  greatly improving the reception performance
  (traded off for throughput)
• Note in PFT and FT, decisions on which signals
  to receive translate into how CTSs are issued
• Cross layer approach
• MAC makes access decisions based on PHY
  conditions (e.g. power)
• How PHY detects signals is directed by the MAC
  decisions

14
MAC results throughput

• PFT and FT perform better (as expected), as they
  estimate and cancel interference as well
• NFT very poor
• FT has the best performance, as it balances
• The need to decode useful pkts
• The need to protect them by canceling unwanted
  interference
• FTs max throughput is 8 pkts/slot
• As many as there are RX antennas, and the network
  is fully connected good result
• Need to explore this further for better
  understanding

15
More results

• Other results obtained
• Packet success ratio (percentage of the attempted
  transmissions that actually get through)
• Queue length (how packets get backlogged)
• Protocol efficiency (percentage of the throughput
  that actually corresponds to useful traffic
  accounts for protocol overhead)
• In all cases, the FT scheme is seen to be clearly
  superior
• Not shown here for lack of time, please see
• Paolo Casari, Marco Levorato, Michele Zorzi, On
  the implications of layered space-time multiuser
  detection on the design of MAC protocols for ad
  hoc networks, IEEE PIMRC 2005, Sep. 2005.

16
Conclusions

• Layered multiuser detection is a very promising
  technique to be deployed in ad hoc networks with
  multiple antennas
• The use of LAST-MUD paves the way for effective
  cross-layer design of MAC protocols, which in
  turn leads to better protocol performance
• Protocol enhancements may be obtained only if a
  correct knowledge of the physical layer behavior
  is properly taken into account

17
Future work

• Carry out a more detailed performance study
• Study the effect of coding and other PHY issues
  (waveform?)
• Work on PHY approx. to avoid bit-level
  simulations
• Include multihop operation in the evaluation
• Consider other (better) MAC policies
• We do not claim FT and PFT are optimal
• Study the broadcast problem with MIMO/directional
  antennas
• Study the effect of asynchronous packet
  transmission

18
Backup slides follow
19
MAC results success ratio

• Here, the average ratio of successfully decoded
  streams to sent streams is depicted
• As before, NFT has the worst perf., and FT with
  exp backoff has the best one
• This is because FT with exp backoff is a good
  coupling, that both achieves good decoding
  performance (FT) and still solves traffic
  congestion (exp backoff)

20
MAC results queue length

• Here the average node queue length is depicted
• As before, FT is best, in the sense that its
  ability to manage traffic and decoding procedures
  lets packets be handled effectively if traffic is
  not too high

21
MAC results protocol efficiency

• Here the average protocol efficiency (i.e., the
  average ratio of correctly decoded data bits to
  the total sent bits) is depicted.
• For the already cited motivations, FT shows the
  best performance
• Better decoding perf. and better traffic handling
  means less wasted data streams and higher
  efficiency