THE WIRELESS REVOLUTION:

1
Federal Communications Commission May 29, 2001
THE WIRELESS REVOLUTION A Signal Processing
Perspective Vince Poor (poor_at_princeton.edu)
May 29, 2001 - The Wireless Revolution
2
OUTLINE

• The Role of Signal Processing in Wireless
• Some Recent Signal Processing Advances
• Space-time Multiuser Detection
• Turbo Multiuser Detection
• Quantum Multiuser Detection
• Conclusion

May 29, 2001 - The Wireless Revolution
3
THE ROLE OF SIGNAL PROCESSING IN WIRELESS
May 29, 2001 - The Wireless Revolution
4
Motivating Factors

• Telecommunications is a 1012/yr. business
• c. 2005 wireless gt wireline
• gt 109 subscribers worldwide
• Explosive growth in wireless services
• Use of a public resource (the radio spectrum)
• Convergence with the Internet

The Role of Signal Processing in Wireless
5
Wireless Applications

• Mobile telephony/data/multimedia (3G)
• Nomadic computing
• Wireless LANs
• Bluetooth (piconets)
• Wireless local loop
• Wireless Internet/m-commerce

The Role of Signal Processing in Wireless
6
Wireless is Rapidly Overtaking Wireline
Source The Economist Sept. 18-24, 1999
The Role of Signal Processing in Wireless
7
Traffic Increasingly Consists of Data
Source http//www.qualcomm.com
The Role of Signal Processing in Wireless
8
Demand Growing Exponentially
Source CTIA
- As of 05/01/01, there were 114,546,113, in
U.S., according to www.wow-com.com - Every 2.25
secs., a new subscriber signs up for cellular in
U.S.
The Role of Signal Processing in Wireless
9
Theres Plenty of Room to Grow - I
Mobile Phones Subscribers per 100 inhabitants,
1998
The Role of Signal Processing in Wireless
10
Theres Plenty of Room to Grow - II
Mobile Phones Market Penetration, 2000
Courtesy of Tom Sugrue (FCC)
The Role of Signal Processing in Wireless
11
Wireless Challenges

• High data rate (multimedia traffic)
• Networking (seamless connectivity)
• Resource allocation (quality of service - QoS)
• Manifold physical impairments
• Mobility (rapidly changing physical channel)
• Portability (battery life)
• Privacy/security (encryption)

The Role of Signal Processing in Wireless
12
Wireless Channels

• Fading Wireless channels behave like linear
  systems whose gain depends on time, frequency and
  space.
• Limited Bandwidth (data rate, dispersion)
• Dynamism (random access, mobility)
• Limited Power (on at least one end)
• Interference (multiple-access, co-channel)

The Role of Signal Processing in Wireless
13
Not Growing Exponentially

• Spectrum - 3G auctions!
• Battery power
• Terminal size

The Role of Signal Processing in Wireless
14
Moores and Evereadys Laws
Courtesy of Ravi Subramanian (MorphICs)
The Role of Signal Processing in Wireless
15
Signal Processing to the Rescue

• Source Compression
• Transmitter Diversity (Fading Countermeasures)
  
• Spread-spectrum CDMA, OFDM (frequency
  selectivity)
• Temporal error-control coding (time selectivity)
• Space-time coding (angle selectivity)
• Advanced Receiver Techniques
• Interference suppression (multiuser detection -
  MUD)
• Multipath combining space-time processing
• Equalization
• Channel estimation
• Improved Micro-lithography (phase-shifting masks)

The Role of Signal Processing in Wireless
16
Advances in ASIC Technology
Microns
5/30/00 25 nm gate announced with optical
lithography using phase-shifting masks (T.
Kailath, et al.).
.8
.5
.35
Courtesy of Andy Viterbi
.25
.18
Time
1991
Future
1998
1997
1995
The Role of Signal Processing in Wireless
17
Signal Processing for Wireless (v 1.0)
Fleming Valve 1910
Helical Transformer 1919
Marconi Crystal Receiver 1919
DeForest Tubular Audion 1916
The Role of Signal Processing in Wireless
18
SOME RECENT SIGNAL PROCESSING ADVANCES

• Introduction
• Space-time Multiuser Detection (3G)
• Turbo Multiuser Detection (2.5G)
• Quantum Multiuser Detection (?G)

May 29, 2001 - The Wireless Revolution
19
INTRODUCTION
Some Recent Signal Processing Advances
20
First, A Few Words About MUD

• Multiuser detection (MUD) refers to data
  detection in a non-orthogonal multiplex its of
  interest, e.g., in
• CDMA channels
• TDMA channels with channel imperfections
• DSL with crosstalk
• MUD can potentially increase the capacity (e.g.,
  bits-per-chip) of interference-limited systems
  significantly
• MUD comes in various flavors
• Optimal (max-likelihood, MAP)
• Linear (decorrelator, MMSE)
• Nonlinear interference cancellation

Some Recent Signal Processing Advances
21
Some Recent Developments

• The basic idea of MUD is to exploit (rather than
  ignore) cross-correlations among signals to
  improve data detection. Recent developments in
  this area
• Space-Time MUD
• Joint exploitation of spatial and temporal
  structure.
• Turbo MUD
• Joint exploitation of temporal structure induced
  by channel coding, and the multi-access channel.
  
• Quantum MUD
• Joint exploitation of quantum measurements and
  the multi-access channel.

Some Recent Signal Processing Advances
22
SPACE-TIME MUD
Some Recent Signal Processing Advances
23
Multi-Access, Antenna, Path Channel
Space-Time MUD
24
Single-Antenna Reception
Non-orthogonal signaling, multipath, fading,
dispersion, dynamism, etc.
Space-Time MUD
25
Space-Time MA Signal Model

• Transmitted signal due to the k-th user

bk(i) data symbol sk(t) signaling waveform

• Vector channel (impulse response) of the k-th
  user

tkl path delay gkl path gain akl
array response

• Received signal

Space-Time MUD
26
A Sufficient Statistic Space-Time Matched
Filter Bank

• Log-likelihood function of the received signal
  r(t)

tkl path delay gkl path gain akl
array response

• H is a matrix of cross-correlations among the
  received signals
• Sufficient statistic yk(i) space-time
  matched filter output

Space-Time MUD
27
Space-Time Multiuser Receiver
Maximum Likelihood Sequence Detection OR Itera
tive Interference Cancellation
Space-Time MUD
28
Optimal Space-Time MUD

• Maximum likelihood sequence detection maximizes
  (over b)

D multipath delay spread

• Computational complexity O(2KD)

Space-Time MUD
29
Linear S-T Interference Cancellers
Decorrelator sgn(Re H-1y) MMSE sgn(Re
(Hs2I)-1y)
Problem
with
Solve

• Gauss-Seidel Iteration (Serial IC)

• Jacobi Iteration (Parallel IC)

• Computational complexity O(K D mmax)

Space-Time MUD
30
Simulation K 8 L 3 P 3
Direct-sequence spread-spectrum (16 chips/bit).
Space-Time MUD
31
Nonlinear S-T Interference Cancellers

• Decision Feedback

Cholesky Decomposition

• Successive Cancellation

• EM/SAGE-Based IC (Interfering symbols are
  hidden data)

• Turbo MUD - Coded channels (b has
  constraints).

Space-Time MUD
32
TURBO MUD
Some Recent Signal Processing Advances
33
MUD The Decoding of Error-Control Codes

• Recall the basic idea of MUD is to exploit
  cross-correlations among signals to improve data
  detection.
• Similarly, error-control coding exploits
  dependencies among channel symbols to improve
  data detection.
• Turbo MUD is a technique for jointly exploiting
  these two types of dependencies.

Turbo MUD
34
Coded Multiple-Access Channel
Basic Idea of Turbo MUD

• The convolutional code the multiaccess channel
  form a concatenated code.
• Like other concatenated codes, this code can be
  efficiently decoded via a turbo-style receiver.

Turbo MUD
35
Rate-R-Coded Multiaccess Signal Model
Received Signal

• K active users.
• B channel symbols per frame
• dk set of RB data symbols transmitted by user
  k
• bk(dk) vector of channel symbols obtained by
  encoding dk
• pk recd waveform of user k 1/T per-user
  signaling rate.
• n(t) unit AWGN s noise intensity

Turbo MUD
36
Sufficient Statistic
As before, the vector y of matched-filter outputs
is sufficient for inferring b1(d1) b2(d2) ...
bK(dK) and d1 d2 ... dK.
Turbo MUD
37
Optimal MUD/Decoding
ML Detection (b)/Decoding (d)
MAP Detection/Decoding
Complexity per Symbol (Assume Binary Symbols)
O(2KD) - uncoded symbols, delay spread D MLSD
MAP MUD
O(2n) - convolutionally encoded symbols,
constraint length n orthogonal signaling
BCJR, Viterbi algo, etc.
Turbo MUD
38
Turbo MUD The Main Idea

• For constraint-length-n convolutionally coded
  transmission on an asynchronous K-user
  multiaccess channel, optimal detection/decoding
  has complexity O(2Kn) Giallorenzi Wilson.
• This complexity can be reduced to O(2K) O(2n)
  via the turbo principle Moher.
• I.e., iterate between MUD and channel decoding,
  exchanging soft (channel) symbol information at
  each iteration.

Turbo MUD
39
Multiaccess Channel Turbo Receiver
Channel Output
SISO MUD
De-Int.
Int.
Output Decision
SISO Decoders

• Soft-input/soft-output (SISO)
• Iterative
• Interleaving removes correlations

vs.
Turbo MUD
40
SISO MUD

• To get posterior probabilities from the multiuser
  detector, we should use MAP MUD.
• MAP MUD is prohibitively complex O(2K) K
  users
• This differs from standard turbo decoding, in
  which the constituent decoders are of similar
  complexity.
• Many lower complexity approaches Alexander et
  al. Honig et al., Lu Wang, Müller Huber,
  Naguib Sheshadri, Reed et al., Schlegel,
  Tarköy, Wang Chen, Wang Poor (COM99),
  others

Turbo MUD
41
Recall Low Complexity MUD
Recall the Model

• MUD fits this model to the observations.
• As noted before, in addition to ML/MAP, there are
  many low-complexity techniques for doing this
  e.g.,
• Linear MUD decorrelator, MMSE, bootstrap (v.
  efficient iterative implementation as linear
  interference cancellers (ICs))
• Nonlinear ICs successive cancellation,
  multistage, EM/SAGE
• Generally, these dont allow the computation of
  the posterior probabilities needed for turbo MUD.
  

Turbo MUD
42
Low Complexity SISO MUD

• Conventional MMSE MUD
• MMSE output desired symbol Gaussian error
  Poor Verdú, IT97
• From this, posterior probabilities can be
  estimated from the MMSE detector output.
• This yields an effective low-complexity SISO MUD.
• MMSE w/ Priors

Turbo MUD
43
Simulation Example K 4 r 0.7
Rate-1/2 convolutional code constraint length 5
128-long random interleavers
Turbo MUD
44
QUANTUM MUD
Some Recent Signal Processing Advances
45
Quantum MUD

• A basic element of MUD is the matched-filter-bank
  sufficient statistic.
• With quantum measurements, observation is
  restricted (uncertainty principles apply).
• In this case, the observation instrument must be
  designed jointly with the detector.
• Photon counting is usually not optimal.

Quantum MUD
46
Quantum MUD Design Problem
Quantum MUD
47
A Two-User Quantum Channel
Quantum MUD
48
Two-User Example Error Probabilities
Quantum MUD
49
Conclusion

• The transformation from wireless voice to
  wireless data is causing exponentially increasing
  demand for wireless capacity.
• Signal processing is the great enabler
• Source compression
• Fading countermeasures/transmitter diversity
• Interference suppression/space-time processing
• Micro-lithography
• Recent advances

May 29, 2001 - The Wireless Revolution
50
Conclusion - Contd

• MUD exploits signal cross-correlations to
  substantially improve data detection.
• Space-time MUD
• Combines exploitation of temporal spatial
  cross-correlations.
• Turbo MUD
• Combines exploitation of cross-correlations
  introduced by the channel with exploitation of
  dependence introduced by coding.
• Quantum MUD
• Combines exploitation of cross-correlations with
  the instrument design for the quantum channels.
• Some Open Issues
• Space-time MUD Hardware implementation
• Turbo MUD Adaptivity, convergence behavior
• Quantum MUD Relevance in applications

May 29, 2001 - The Wireless Revolution
51
THANK YOU!
May 29, 2001 - The Wireless Revolution