MUD: Multi-user detection

Jetmir Palushi
Stevens Institute of Technology
EE613 DSP for Communications
 Overview
• MUD in CDMA systems
• MUD Features
• Modulation and Implementation Concept
• Multiple Access Interference (MAI)
• MUD algorithms
– Linear
– Non-Linear
– Optimal MLSE
• MUD Detectors
– Decorrelating
– Minimum Mean Squared Error (MMSE)
– Blind Adaptive MMSE
– Multistage
– Decision Feedback
– Successive Interference Cancellation
• Detector Performance
• Limitations of MUD
• Conclusion
 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 Verd
´u in the early 1980’s.
 MUD in CDMA Systems
•Multi-User Detection considers all users as signals for each other
leading to joint detection
- Reduced interference leads to capacity increase
- It alleviates the near and far problems
•System capacity is limited by interference threshold that a detector
uses to make its decision
•Multiple Access Interference increases with the number of users
•MUD suppresses the MUI increasing the system capacity
•Can be implemented at either base station, mobile or both
- Size and weight requirements are not stringent for base
station
- Therefore it is currently being implemented for mobile to
base station.
 MUD Features
• Quasi-synchronous transmissions is easier than other methods

• It has the capability to reject the interference created by the narrow band

• Capable to achieve diversity in frequency

• It tremendously reduces the complexity and it increases the spectral efficiency

• Robustness to multipath fading

• The use of modern DSP makes MC-CDMA implementation feasible and

attractive

• MC-CDMA translates the time operations to the frequency domain

• Effect of ISI and delay spread is mitigated

Modulation and Implementation

Transmitter Receiver
 Modulation and Implementation
• Using Binary Phase Shift Keying (BPSK) technique we have:
– The Signal for the kth user given by the formula:

uk t    xk i  ck i  sk t  iT   k 
i 0
– Where
• The i-th input symbol of the from the k-th user is x(i)
• c(i) is the gain of channel
• s(t) is the waveform that contains the PN sequence
 k is the transmission delay; for synchronous CDMA, k=0 for all
users
– And the received signal is given by the formula:
K
y t    uk t   z t 
– Where k 1

• K is the number of users on the network at a given time t

• z(t) is the complex White Gaussian Noise
 Modulation and Implementation
• The Sampled output of the matched filter of kth user is given by the formula:
T
yk   y t sk t dt
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
– The 1st term means the desired information to be recovered
– The 2nd term is the Multi-Access Interference (MAI)
– And the 3rd term is the complex White Gaussian Noise
• If we assume a two-user channel we have
T
r   s1 t s2 t dt
0
 Modulation and Implementation
• From the general equation given on the slide above we solve for user1 and
user2 received signals which are y1 and y2 respectively:
y1  c1 x1  rc 2 x2  z1 y2  c2 x2  rc1 x1  z 2

•  
This results on the detected symbol of k-th user: xˆ k  sgn yk
• Suppose the signal of user 1 is stronger than the signal of user 2 then we
have the near/far problem, the MAI term rc1x1 present in the signal of user 2 is
very large
 Multiple Access Interference (MAI)
• Imperfect cross-correlation characteristics of the spreading
codes
• Multi-path fading contributes to MAI
• Causes severe degradation in the performance of the system
• Capacity is interference limited
• MAI is a function of:

– Number of Users
– Cross-Correlation between users
– Amplitude of Interfering Signals

• MAI is due to the non-orthogonality between users

 MUD Algorithms
•Linear detectors apply linear
transformations to matched
filter outputs to minimize MAI.
Simple to implement but can
get complex.

•Non-Linear detectors are more

complex calculation wise than
linear detectors due to
nonlinearity, however they
perform better under severe
conditions
 Optimal MLSE detector
• The Maximum-likelihood sequence estimation (MLSE)
is too complex to be implemented
• For synchronous CDMA, it searches over 2K possible
combinations of the bits in vector x. Where x is given
by:

 
xˆ  arg  max 2 y T Wx  bT WRWb 
 x1, 1K 
• R and W matrices are described under Decorrelating
detector section and y is the received signal
• For asynchronous CDMA it uses Viterbi algorithm with
2K-1 states which is very complicated to be
implemented in practice as well
 Linear Algorithms
• Linear mapping algorithms are applied to the
outputs of the matched filters
• Less complexity than optimal ML receiver
• Practical Linear Algorithms:
- Decorrelating Detector
- Minimum-mean squared error (MMSE)
- Blind (adaptive non-adaptive) techniques
 Decorrelating Detector
Matrix Representation:

y1
y  RWx  z
Matrix
y2 Filter
The matrix R is of
yk R-1 the form:

– where y=[y1,y2,…,yK]T, R and W are KxK

matrices
– Components of R are given by cross-
correlations between signature waveforms
sk(t)
– W is diagonal with component Wk,k given by
the channel gain ck of the kth user
– z is a colored Gaussian noise vector
 Decorrelating Detector
• We can solve for x by inverting matrix R
~
y  R 1 y  W x  R 1 z  xˆk  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
 Minimum Mean Squared Error
(MMSE) Detector
•Transmitted Signal can be Modeled as:

- The MMSE detector

takes the background
noise in to account
and utilizes the
knowledge of the
received signal powers
•Then the receive Signal can be Modeled as:
-It minimizes the mean
squared error between
the actual data and the
soft outputs of the
conventional detectors
 Minimum Mean Squared Error
(MMSE) Detector
•The MMSE detector output y_k is therefore:

Advantage: Better error

probability performance, and
no noise enhancement

Disadvantage: Requires
estimation of received
amplitudes, and matrix
inversion
Blind Adaptive MMSE Detector
• Blind adaptive detector characteristics:
- The detector doesn’t require the training sequence in order to calculate
the channel impulse response
- Requires the knowledge of the signature waveforms and timing
information of the desired user
- The limitation is that it works only for short codes
• The major disadvantage of the adaptive MMSE detector over the
“blind” adaptive MMSE is that it requires the training sequences this
results on a waste of the bandwidth which is populated with signals
that do not carry any communication data. Therefore for the “Blind”
adaptive we have a clear benefit when it is compared to other
detectors since it does not require any training sequence that’s why
is called “blind”.
• Adaptive MMSE detectors also are advantageous over other non-
adaptive detectors because they can adapt to unknown and time-
varying channel conditions
 Non-Linear Algorithms
• Non-Linear Algorithms:
Estimate the interference caused by each
user on the others, re-spread and cancel from
the received signal. This is done through multitude of
stages.
• Practical Non-Linear Detectors:
- Multistage Detector
- Decision Feedback Detector
- Subtractive Interference cancellation
Successive Interference Cancellation (SIC)
Parallel Interference Cancellation (PIC)
Selective Parallel Interference Cancellation
 Multistage Detector

The concept of the multistage detector is to make a decision in

every stage as the name indicates. As shown on the diagram
above the received signal is y and the detector produces the
decisions x(1) x(2) up to x(n). Where:
 
x1 2   sgny1  rc 2 x2 1
 
x2 2   sgny2  rc1 x1 1
Decision Feedback Detector

•As shown on the diagram there are 2 matrix transformation:

forward filter and feedback filter
•Pretty much same performance as the Decorrelator detector
 Successive Interference Cancellers
• The SIC detectors start to subtract off
the strongest remaining signals in a
successive fashion from the rest of the
signals (See diagram to the right)
– By canceling the strongest signal from
the rest we gain most of the benefit
and it is the most reliable cancellation

• The other similar alternative is the PIC

method. This starts to simultaneously
subtract off all of the users’ signals
from all of the others unlike the serial
cancellation that starts with the
strongest signal user (See diagram
under Successive Interference Cancellers
– PIC)
– It works better than SIC when all of the
users are received with equal strength
since it is much easier to detect them
and hence decreases the probability of
error
Successive Interference Cancellers
- SIC
Successive Interference Cancellers
- PIC
Successive Interference Cancellers
- PIC
Successive Interference Cancellers
SIC VS. PIC
The main disadvantages are:
1) If the strongest estimate is not 1) More vulnerable to near-far
highly reliable it results on issues
performance 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 1) Because of the parallel
tremendous signal gain from the nature no delays/stage
MAI reduction since all of the required!
interfering channel will add up as 2) Simpler than other linear
signals to the weakest user. detectors
Hence every user is on a win-win
situation.
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
Detector Performance
Detector Performance
Detector Performance
 Limitations of MUD
Limitations with implementation
•Sensitivity and robustness
•Processing delay
•Processing complexity

Limitations of MUD
•System capacity improvements are not enormous and not trivial
•Cost must be kept low in order to increase performance/cost
tradeoff
•Capacity improvements only in the uplink would be partly used in
determining the overall system capacity
- Need to use MUD in both uplink and downlink
- Implementing MUD in mobiles is still a challenge
 Conclusions
• MUD has many advantages over other communications techniques
however they are limited by the complexity of their implementations and
a simple implementation is needed. As the DSP field progresses further
and more calculations can be performed with ease more of these
advantages will be implemented in future work.
• Current investigations involve implementation and robustness issues
• MUD research is still in a phase that would not justify to make it a
mandatory feature for 3G WCDMA standards
• Currently other techniques such as smart antenna seem to be more
promising
• Though MUD has not been a mandatory feature of the wireless
standards so far, the rapid advances in DSP architectures promise the
evolution of MUD as integrated feature of future wireless standards to
provide better capacity and data rates
• Feasible VLSI implementations for Mobiles
 References
• R. Lupas, S. Verdu, "Linear multi-user detectors for synchronous code-division multiple-access
channels," IEEE Trans. Information Theory, vol. 35, no. 1, pp. 123-136, Jan. 1989.
• R. Lupas, S. Verdu, "Near-far resistance of multi-user detectors in asynchronous channels," IEEE
Trans. Comm. Vol. 38, no. 4, pp. 496-508, April 1990.
• M. Honig, U. Madhow, S. Verdu, "Blind adaptive multi-user detection," IEEE Trans. Info. Theory, vol.
41, no. 4, pp. 944-960, July 1995.
• H. V. Poor, S. Verdu, "Probability of error in MMSE multi-user detection," IEEE Trans. Info. Theory,
vol. 43, no. 3, pp. 858-871, May 1997.
• M. Honig, M. K. Tsatsanis, "Adaptive techniques for multi-user CDMA receivers," IEEE Signal
Processing Magazine, pp. 49-61, May 2000. [Review]
• X. Wang, H. V. Poor, "Blind multi-user detection: a subspace approach," IEEE Transactions on
Information Theory, vol. 44, no. 2, pp. 677-690, March 1998.
• http://wsl.stanford.edu/~ee360/mud_peter.ppt#4
• X. Wang, H.V. Poor, "Space-time multi-user detection in multi-path CDMA channels," IEEE
Transactions on Signal Processing, vol. 47, no. 9, September 1999.
• H. V. Poor, S. Verdu, "Single-user detectors for multi-user channels," IEEE Trans. Comm. Vol. 36,
no. 1, pp. 50-60, Jan. 1988.
• Duel-Hallen, J. Holtzman, Z. Zvonar, "Multi-user detection for CDMA Systems," IEEE Personal
Communications, pp. 46-58, April 1995. [Review]
• Yi Wang, Zhimin Du, Lu Gao and Weiling Wu, “Performance Analysis of MMSE Multiuser Detection”,
Beijing University of Posts and Telecommunications
• Shimon Moshavi, “Multi-User Detection for DS-CDMA Communications”, Bellcore
• Deepak Das, “Blind Adaptive Multiuser Detection for Cellular Systems Using Stochastic
Approximation With Averaging”, IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL.
20, NO. 2, FEBRUARY 2002