Wireless Communications Current Issues, Future Solutions

1
Wireless Communications Current Issues, Future
Solutions

• Dr. Teng Joon (T. J.) Lim
• Assistant Professor
• Dept. of Elect. Comp. Eng.
• Univ. of Toronto

Presented at Nortel Networks, April 29, 2005
2
Seminar Outline

• Background and Research Focus
• Current Issues in Wireless Communications
• Recent Results
• QoS-Constrained Precoding in Downlink Channels
• Phase Noise Estimation in OFDM
• Some Results in Cooperative Diversity Networks
• Current Interests
• Applying approximate statistical inference
  methods in multi-user communications including
  CDMA and OFDMA.
• Practical issues in precoding e.g. channel
  estimation and feedback

3
Personal History

• Bachelors degree from Natl Univ. Singapore
  1992 PhD from Univ. of Cambridge 1996.
• Member Tech. Staff at Centre for Wireless Comms
  (CWC), Singapore, 1995 2000.
• Assistant Professor, Univ. of Toronto, 12/2000
  present.

4
Research Background

• PhD Adaptive IIR (infinite impulse response)
  filtering for system identification applications
  e.g. acoustic echo cancellation.
• Post-PhD Wireless Communications
• 1995present Multiuser detection (Adaptive
  filtering, Kalman filters, interference
  cancellation, turbo detection, ad hoc networks)
• 2000present Spatial domain processing
  (Space-time multi-access, precoding)
• 1997present OFDM and OFDMA (PAR reduction,
  parameter estimation)
• 1997present Performance analysis in fading
  channels.

5
Research Focus

• Transceiver design for multi-user wireless
  systems e.g. cellular, WLAN, BWA.
• Transmitter precoding
• Detection methods (interference cancellation,
  adaptive filters, etc.)
• Parameter estimation (e.g. frequency and phase
  offset)
• Cooperative diversity
• Performance analysis of systems in fading
  channels.

6
Future Wireless Networks

• Multimedia portable terminals
• Users serviced by one base station may need
  different data rates
• Processing power limited so as much complexity as
  possible should be located at access points/base
  stations.
• Very high data rates
• Severe intersymbol interference
• Highly accurate synchronization necessary.

7
Future Wireless Networks

• Wireless and Mobile Are Not Synonyms
• Many fixed wireless applications previously
  unimagined e.g. backhaul networks, WLANs.
• Transmitter able to match itself to the channel?
• Nodes specially configured as relays only?
• Bandwidth Efficiency More Critical Than Ever
• More high-rate users in given spectrum
• Design system that expects and deals with
  non-orthogonal multi-access?
• Or use cognitive radio concept to share time and
  frequency efficiently?

8
Precoding Problem
Ch. 1
Rx 1
Txer
Ch. 2
Rx 2

• Channels MIMO due to multiple antennas for
  instance. Represented as matrix Hk.
• Transmit to all users at the same time e.g. SDMA,
  CDMA.
• ? Multi-user interference present.

9
Precoding Problem

• Scenario 1 (QoS-Constrained) Fixed wireless
  network, transmitter knows Hk exactly all the
  time, design transmitter to minimize power while
  meeting users individual data rate requirements.

MU interference exists, but we dont care as long
as each user gets its required data rate.
Channel 1
Channel 2
Channel 3
BS
10
Precoding Scenarios
Scenario 2 (Zero-Forcing) Again transmitter has
full knowledge of channels, and forces multi-user
interference to zero at each receiver.
No co-channel interference at any
receiver. Possible only if no. of Tx antennas gt
total no. of Rx antennas Not practical. Each
users Tx power can be adjusted to give required
QoS, but this scheme will not minimize total
transmitted BS power Not academic.
Can be seen as a more traditional PHY approach
fixed transmit power, try to reduce BER by ZF
or MMSE.
11
Precoding Scenarios

• Scenario 3 (Single-User Channels) No multi-user
  interference (e.g. TDMA), transmitter knows first
  and second order statistics of channels, tries to
  minimize bit error rate.

Main Idea If spatial channels are correlated,
space-time codes dont work so well. BUT if
correlations are known, then make use of that
information to improve performance as far as
possible.
12
QoS-Constrained Precoding

• Examine Scenario 1 more closely
• K users Nt antennas at downlink transmitter.
• 1 antenna at each receiver (practical).
• User k requires data rate Rk.
• QAM modulation assumed for all users.
• Find method to minimize total transmitted power
  while satisfying all rate requirements.

13
QoS-Constrained Precoding

• Shannon Capacity max. rate (bps/Hz) for which
  decoding error probability can be made
  arbitrarily small.
• In multi-user channels, for a given power
  constraint, we define a capacity region if rates
  (R1,,RK) yield zero error probability, then the
  vector (R1,,RK) is in the capacity region.

14
QoS-Constrained Precoding

• Uplink two-user rate region with individual power
  constraints is a pentagon

R2
All rate vectors inside the red region can be
achieved, with user 1 transmitting at power below
P1, user 2 using less than P2.
A
Pt. A Decode user 1, then 2 Pt. B Decode user
2, then 1
B
R1
15
QoS-Constrained Precoding

• Downlink two-user capacity region with total
  power constraint is a union of many pentagons.

R2
R1
16
QoS-Constrained Precoding

• Key duality result
• Every point on the boundary of the downlink
  capacity region corresponds to a vertex of an
  uplink capacity region.
• By reversing the decoding order at that vertex,
  we obtain the downlink encoding order at that
  point.
• Knowing the input covariance matrices for the
  virtual uplink at a boundary point is equivalent
  to knowing the input covariance matrices for the
  downlink.

17
QoS-Constrained Precoding

• Observe If optimal encoding used for all users,
  then rate requirements can be met with minimal
  power if and only if the desired rate vector is
  located on the boundary of the capacity region.
• Reason Using more power moves boundary outwards
  and desired rate will be met for sure using less
  power moves boundary inwards and desired rate
  cannot be met.

18
QoS-Constrained Precoding

• Sub-problem 1 For any given rate vector, find
  the transmission scheme that will place it on the
  boundary of a rate region.
• Combination of successive dirty-paper coding and
  uplink-downlink capacity region duality.
• Quite easy for arbitrarily fixed encoding order
  and single-antenna downlink receivers.
• Very hard (till our recent work) to find optimal
  encoding order.

19
QoS-Constrained Precoding
20
QoS-Constrained Precoding
Gist of Solution

• Arbitrarily choose P and (a1,a2). Maximize a1R1
  a2R2 s.t. total power lt P.
• If decoding order chosen according to relative
  magnitudes of a1 and a2, then problem is convex.
• Compare solution of max. problem to desired
  rates. If R1 gt R1(target) then lower a1 ditto
  for other users.
• Repeatedly update as until all rates above their
  targets, or all rates below their targets.
• Then update P increase P if all rates below
  targets, decrease P if all rates above targets.
  Go back to step 1.
• Iterate until sufficiently close to target rates.

21
QoS-Constrained Precoding

• Sub-problem 2 Assuming QAM modulation, with
  given allowable BER and required rates, design
  base station transmitter for minimum power.
• Novel multi-user gap to capacity formulation
  allows translation of desired rate and BER to a
  virtual data rate.
• Design system (according to information theoretic
  principles) using virtual rates.

22
Precoder Block Diagram
23
Tomlinson-Harashima Precoding
Encoder k
To Encoder k-1.
Symbol stream for user k
Tx Beamformer
mod M
Cancel User (k1)
Design these!

• Tx beamformer determines power used in
  transmission.
• Int. cancellation block requires channel
  knowledge.

From Encoder k1.
24
Challenges in QoS-Constrained Precoding

• How do we obtain good channel information at the
  transmitter?
• How do we make the precoder less sensitive to
  errors in channel estimation?
• Given a fixed bandwidth on the feedback channel
  b/w rxers and txer, what is the best
  information to feed back, and how do we use that?
• If we only have statistical information about
  channels, is that useful at all?
• Detailed complexity analysis needed to determine
  feasibility.

25
Other Approaches to QoS-Constrained Precoding

• Can also formulate problem in terms of signal to
  interference ratio (SIR) requirements.
• Each user has its own SIR needs
• Goal is to minimize total BS transmitted power to
  all users.
• Multi-antenna receivers pose a difficult
  technical challenge in this case (but not in
  capacity formulation, because of geometry of
  capacity region).
• We have devised an approximate method to handle
  the multi-antenna case with SIR requirements.

26
Detection and Estimation

• Multi-user detection jointly detect all
  interfering signals e.g. CDMA uplink.
• Multi-stage interference cancellation parallel,
  successive, hybrid. Linear variants very well
  understood non-linear ones still being studied
  in academic circles.
• Adaptive detection use adaptive filters to
  suppress interference, if short spreading codes
  used.
• Kalman multi-user detection can be used for
  joint detection and channel estimation. Provides
  a way to implement asynchronous linear MMSE
  detector.
• Collaborated with Oki of Japan in late 90s.

27
Detection and Estimation

• MUD methods can also be applied to spatial
  multiplexing (BLAST), and many other scenarios
  with co-channel interference that is partially
  known to the receiver.
• With coding and interleaving, turbo MUD can be
  developed single-user performance possible at
  reasonable SNR.
• Current work uses statistical mechanics ideas
  (variational free energy) to develop better
  decoders.

28
Detection and Estimation

• OFDM and OFDMA Industry interest in OFDM for
  WLAN, BWA, DVB and so on.
• But there are still major issues with phase
  noise, frequency error, timing synchronization
  which all significantly degrade performance.
• Especially problematic in OFDMA
• Each user gets a subset of carriers
• Each user introduces its own freq. offset, time
  of arrival, etc.
• How do we estimate and then account for these
  non-idealities in a detector?

29
OFDM Phase Noise Estimation

• OFDM can be severely affected by phase noise.
• Problem arises from phase jitter in local
  oscillators at transmitter and receiver.
• Not easy to handle because phase noise
  time-varying phase offsets. Constant phase offset
  can be handled relatively easily.
• If d.c. component of phase offset removed,
  residual phase noise can still be damaging.

30
OFDM Phase Noise Estimation
Even residual phase errors of a couple of degrees
can cause significant problems, in particular an
error floor.
31
OFDM Phase Noise Estimation

• 64 carriers
• VCO has rms phase error of 3 degrees.
• 64-QAM on each carrier
• Rayleigh fading channel with 3 taps
• Phase noise generated according to 802.11g specs.

32
OFDM Phase Noise Estimation

• Principle is to estimate vector of phase errors q
  (in time) using observations r.
• Best (MAP) method Derive f(qr) and maximize
  that w.r.t. q, if pilot symbols transmitted. If
  transmitted symbols need to be estimated too,
  then maximize joint distribution f(q,xr), where
  x is the txed signal.
• However MAP technique not feasible because of
  complicated form of f(q,xr) e.g. function has
  more than one local max.
• Variational approach Let Q(q,x) be some
  tractable distribution (Gaussian), and then lets
  try to minimize some measure of distance
  between Q(q,x) and f(q,xr).

33
OFDM Phase Noise Estimation

• Distance between two probability distributions
  can be measured using the Kullback-Leibler
  divergence D(Qf) EQlog(Q/f).
• D(Qf) has the same expression as variational
  free energy in statistical mechanics.
• So minimizing D(Qf) w.r.t. the parameters of
  the Q(.) distribution is known as the variational
  approach to probabilistic inference.
• Note that Q(.) is only an approximation of f(.)
  so in general, variational estimate will not be
  the MAP estimate.

34
OFDM Phase Noise Estimation

• BUT
• Usually the estimates are close to optimal
• The min. of D(Qf) can be found through setting
  its derivatives w.r.t. the free parameters of
  Q(.) to zero.
• E.g. if Q(.) is Gaussian, then its mean value
  maximizing value. Minimizing D(Qf) w.r.t. the
  mean and covariance of Q(.) gives the variational
  estimate of q as the mean of Q(.).
• Not magic D(.) is usually a multi-modal function
  of parameters of Q(.). So good initialization
  needed in the form of training.

35
Variational OFDM Phase Noise Estimator
Best postulated distribution
True Distribution
Parameter value
36
OFDM Phase Noise Estimation

• Algorithm derived in this way has complexity
  O(N3) per OFDM frame.
• Requires some training (not blind).
• Simplified algorithm partitions OFDM frame into
  N/K groups of K (time) samples each.
• By estimating phase errors in each group
  separately, complexity is brought down to O(NK2).
• Performance degrades, but big saving in
  complexity if N is large.

37
Current OFDM Work

• Extend variational framework to estimation of
  frequency offsets and fading channels.
• Consider the multi-user OFDM (or OFDMA) problem
• Straightforward method use multi-user
  detection, but complexity is high.
• Different method use approximate inference
  (including graphical models like Bayes nets,
  factor graphs) to figure out a practical approach
  to joint estimation and detection.

38
Co-operative Diversity

• Main idea N relay nodes help one source node to
  transmit gt act together like an N-antenna
  transmitter, use transmit diversity (space-time
  coding) techniques.

Source
Destination
Relays
39
Co-operative Diversity

• Laneman et al. extended ideas to multiple sources
  (orthogonal channels) in clusters, and analyzed
• Decode and Forward relays decode, then
  re-encode and transmit.
• Amplify and Forward relays correct for phase
  shift on source-relay channel, then transmit
  signal w/o decoding.
• Selection Diversity relay forwards only if
  source-relay channel SNR is above threshold.
• Space-Time Coded Cooperation each relay acts
  like one antenna in a multi-antenna transmitter.
• Incremental Cooperation Destination tells nodes
  whether packet received correctly if not, relays
  forward packet.

40
Cooperative Diversity
Cluster
2
D
1
3
4
6
5

• Two phases
• Node k transmits on channel k in phase I (---
  lines).
• Other nodes in cluster re-transmit all other
  nodes information in phase II (solid lines).
• Repetition-based cooperation simply
  re-transmit or decode, re-encode and
  re-transmit.
• Space-time coded cooperation treat cluster as
  antenna array.

41
Co-operative Diversity

• Recently we found the outage probability
  expression valid for all SNRs in a DF system

42
Co-operative Diversity

• Whats interesting is that at SNR 15 dB for
  instance, best result is when two nodes (out of
  8) are co-operating.
• Because when SNR is low, decoding error
  probability is high and so signal transmitted by
  relays have a good chance of being unhelpful.
• Asymptotic analysis would not reveal this.
• Implies that choice of nodes in a cluster depends
  on SNR.

43
Co-operative Diversity

• Currently working on power allocation among
  relays.
• If destination knows channels from relays, but
  not source-relay channels, how would it calculate
  Tx powers for each relay?
• If destination knows complete channel from source
  to destination?
• Also, biggest question is how to transfer
  information from one node to another for sharing.

44
Summary

• There are many exciting new ideas in wireless
  today that can potentially lead to quantum leaps
  in performance (mobility, rate, power, etc.)
• Our research group is right at the forefront of
  many of these developments, contributing
  especially to knowledge in cross-layer design and
  receiver design.
• In this talk, we briefly described a few of the
  most recent results in precoding, detection,
  phase estimation and cooperative diversity.
• More technical details can be found in the
  references, and through discussions with the
  speaker.

45
References

• Precoding
• Fung, Yu, Lim, Precoding for the Multi-Antenna
  Downlink Multi-user Gap Approximation and
  Optimal User Ordering, submitted to IEEE Trans.
  Comms., Apr 2005.
• Doostnejad, Lim, Sousa, Precoding and
  Beamforming Design for MIMO Broadcast Channels
  with Multiple Antennas at Each Receiver,
  submitted to IEEE Trans. Comms., Feb 2005.
• Variational Approach to Detection/Estimation
• Lin, Zhao, Lim, OFDM phase noise cancellation
  via approximate probabilistic inference,
  presented at IEEE Wireless Comms Networking
  Conf. (WCNC), New Orleans, LA, Mar 2005.
• Lin, Lim, A variational free energy minimization
  interpretation of multiuser detection in CDMA,
  submitted to IEEE Globecom 2005.

46
References

• Multiuser Detection
• Wu, Juntti, Lim, Detectors and asymptotic
  analysis for bandwidth-efficient space-time
  multiple-access systems, submitted to IEEE
  Trans. Comms., Jan 2004 revised Jan 2005.
• Lau, Lim, A low-complexity enhancement to
  sub-optimal CDMA receivers, IEEE Trans. Wireless
  Comms., Nov 2004.
• Albeanu, Lim, Optimization of linear iterative
  interference cancellation receivers for CDMA
  communications, IEEE Trans. Comms., Mar 2004.
• Wu, Lim, Turbo multiuser detection for
  differentially modulated CDMA, IEEE Trans.
  Wireless Comms., Mar 2004.

47
References

• Cooperative Diversity
• Zhao, Adve, Lim, Outage probability at arbitrary
  SNR with cooperative diversity, to appear in
  IEEE Comm. Letters.
• Zhao, Lim, Adve, Relay power allocation in a
  cooperative diversity network, in preparation.