Ho Chi Minh City University of Technology
Faculty of Electrical and Electronics Engineering
Department of Telecommunications
----------------------------------------------------------
Multiuser Detection
for CDMA Systems
Lectured by Ha Hoang Kha, Ph.D.
Ho Chi Minh City University of Technology
Email:
[email protected] Content
Multiuser detection for CDMA
H. H. Kha 2
References
Ta Tri Nghia, Wireless Communications, lecture notes.
T.S. Rappaport, Wireless Communications, Prentice Hall
PTR, 1996.
A. Goldsmith, Wireless Communications, 2004.
H. H. Kha 3
DS/CDMA Systems
A conventional DS/CDMA system treats each user
separately as a signal, with other users considered as
noise or MAI – multiple access interference
Capacity is interference-limited
Near/far effect: users near the BS are received at
higher powers than those far away
• those far away suffer a degradation in performance
• Need tight power control
H. H. Kha 4
Multiuser Detection
Multiuser detection considers all users as signals for
each other -> joint detection
• Reduced interference leads to capacity increase
• Alleviates the near/far problem
MUD can be implemented in the BS or mobile, or both
In a cellular system, base station (BS) has knowledge
of all the chip sequences
Size and weight requirement for BS is not stringent
Therefore MUD is currently being envisioned for the
uplink (mobile to BS)
H. H. Kha 5
Concept of MUD
Simplified system model (BPSK)
• Baseband signal for the kth user is:
∞
uk (t ) = ∑ xk (i ) ⋅ ck (i ) ⋅ sk (t − iT − τ k )
i =0
- xk(i) is the ith input symbol of the kth user
- ck(i) is the real, positive channel gain
- sk(t) is the signature waveform containing the PN sequence
- τk is the transmission delay; for synchronous CDMA, τk=0 for all users
- T: symbol time
• Received signal at baseband
K
y (t ) = ∑ uk (t ) + z (t )
k =1
- K number of users
- z(t) is the complex AWGN
H. H. Kha 6
Concept of MUD
Sampled output of the matched filter for the kth user:
T
yk = ∫ y (t )sk (t )dt Consider for a particular bit
0
K T T
= ck xk + ∑ x j c j ∫ sk (t )s j (t )dt + ∫ sk (t )z (t )dt
j≠k 0 0
• 1st term - desired information
• 2nd term - MAI
• 3rd term - noise
Assume two-user case (K=2), and
T
r = ∫ s1 (t )s2 (t )dt
0
H. H. Kha 7
Concept of MUD
Outputs of the matched filters are:
y1 = c1 x1 + rc2 x2 + z1 y2 = c2 x2 + rc1 x1 + z 2
Detected symbol for user k: xˆ k = sgn ( yk )
If user 1 is much stronger than user 2 (the near/far problem), the MAI
term rc1x1 present in the signal of user 2 is very large
Successive Interference Cancellation
• decision is made for the stronger user 1: xˆ1 = sgn ( y1 )
• subtract the estimate of MAI from the signal of the weaker user:
xˆ2 = sgn ( y2 − rc1 xˆ1 )
= sgn (c2 x2 + rc1 (x1 − xˆ1 ) + z 2 )
• all MAI can be subtracted from user 2 signal provided estimate is correct
MAI is reduced and near/far problem is alleviated
H. H. Kha 8
MUD in CDMA systems
The primary idea of Multi User Detection (MUD)
techniques is to cancel the interference caused by
other users.
This is done by exploiting the available side
information of the interfering users, rather than
ignoring the presence of other users like in Single
User Detection (SUD) techniques.
The idea of MUD was proposed by Sergio Verdu in
the early 1980’s.
H. H. Kha 9
MUD Algorithms
Linear detectors apply
linear transformations to
matched filter outputs to
minimize MAI. Simple to
implement.
Non-Linear detectors are
more complex
calculation wise than
linear detectors due to
nonlinearity, however
they perform better
under severe conditions
H. H. Kha 10
Decorrelator
Matrix representation
y = RW x + z
• where y=[y1,y2,…,yK]T, R and W are KxK matrices
• Components of R are cross-correlations between
codes
• W is diagonal with Wk,k given by the channel gain ck
• z is a colored Gaussian noise vector
H. H. Kha 11
Optimal MLSE Detector
Maximum-likelihood sequence estimation (MLSE) is
the optimal detector (Verdú, 1984)
For synchronous CDMA, search over 2K possible
combinations of the bits in vector x
=xˆ arg max 2 y Wx − x WRWx
T T
x∈{−1,+1}
K
For asynchronous CDMA, use Viterbi algorithm with 2K-
1 states
Both too complex for practical implementation
H. H. Kha 12
Decorrelating Detector
Matrix Representation:
=y RWx + z
y1
y2 Matrix Filter
R-1
yk The matrix R is of
the form:
d = R −1y 1
ρ 1,2
ρ
1,3
R = ρ ρ
– where y=[y1,y2,…,yK]T, R and W are KxK
1
matrices 2,1 2,3
ρ 3,1 ρ
– Components of R are given by cross-
1
correlations between signature waveforms
sk(t)
3,2
– W is diagonal with component Wk,k given by
the channel gain ck of the kth user
– z is a colored Gaussian noise vector
H. H. Kha 13
Decorrelating Detector
We can solve for x by inverting matrix R
~
y = R −1 y = W x + R −1 z ⇒ xˆ k = sgn ( ~
yk )
The matrix representation method is analogous to zero-
forcing (ZF) equalizers for ISI channels
Advantages:
• Does not require knowledge of users’ powers
Disadvantages:
• Noise enhancement
H. H. Kha 14
Multistage Detectors
Decisions produced by 1st stage are x1 (1), x2 (1)
2nd stage:
x1 (2 ) = sgn[ y1 − rc2 x2 (1)]
x2 (2 ) = sgn[ y2 − rc1 x1 (1)]
and so on…
H. H. Kha 15
Subtractive Interference Cancellers
Successively subtracting off the strongest remaining
signal
• Cancelling the strongest signal has the most benefit
• Cancelling the strongest signal is the most reliable
cancellation
An alternative called the Parallel Interference
Cancellers simultaneously subtract off all of the users’
signals from all of the others
• works better than SIC when all of the users are received
with equal strength (e.g. under power control)
H. H. Kha 16
Successive Interference Cancellers
H. H. Kha 17
Successive Interference Cancellers - SIC
H. H. Kha 18
Parallel Interference Cancellers - PIC
H. H. Kha 19
Parallel Interference Cancellers - PIC
H. H. Kha 20
Successive Interference Cancellers
SIC VS. PIC
The main disadvantages are:
1) If the strongest estimate is not highly 1) More vulnerable to near-far
reliable it results on performance issues
degradation 2) Complicated circuitry
2) As the power profile changes the
signals must be reordered
3) Every stage introduces a delay
The main advantages are:
1) The weakest user will see a
tremendous signal gain from the
MAI reduction since all of the 1) Because of the parallel
interfering channel will add up as nature no delays/stage
signals to the weakest user. required!
Hence every user is on a win-win 2) Simpler than other linear
situation. detectors
2) For severe conditions if we remove
the strongest user the rest of
weaker users will benefit hence
the signal can be recovered
3) Can recover from near-far effects
H. H. Kha 21
Performance of MUD
H. H. Kha 22
Limitations of MUD
Issues in practical implementation
• Processing complexity
• Processing delay
• Sensitivity and robustness
Limitations of MUD
• Potential capacity improvements in cellular systems are
not enormous but certainly nontrivial (2.8x upper bound)
• Capacity improvements only on the uplink would only be
partly used anyway in determining overall system
capacity
• Cost of doing MUD must be as low as possible so that
there is a performance/cost tradeoff advantage
H. H. Kha 23
