Chip level equalisation for W- CDMA Stephen McLaughlin Dave Cruickshank, Sacha Spangenberg
Chip level equalisation for W-CDMA
Stephen McLaughlin
Dave Cruickshank, Sacha Spangenberg
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Structure of PresentationBackground
Adaptive filters for MUD
Receiver Architectures for MUD
Performance comparison
Conclusions
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Background Work focused on next generation terminals for
mobile communication based on CDMA.
Integration of new services like Web browsing Video conferencing & Real Audio GPS & Traffic Guidance
and increasing number of subscribers result in increasing demands on system resources
Interference (ISI/MAI) will reduce Quality of Services unless countermeasures are taken.
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
CDMA Systems: CDMA multiplexed signals share the same
frequency band, but are separated by their distinctive spreading codes
In the ideal case, the spreading codes can be orthogonal to each other and the transmitted signal can be received and de-multiplexed using a simple receiver !
The communication channel however usually destroys the orthogonality of the codes resulting in inter-user interference, an effect more known as Multiple Access Interference, MAI. This degrades the systems performance.
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Received CDMA Signal: The impulse response of the communication
channel also introduces an effect known as Intersymbol Interference, ISI, which results in the undesirable overlapping between at least two independent signals.
Finally, the presence of additive random noise, ARN, at the receiver is unavoidable.
Therefore, the wanted user’s received signal, rw, consists of a number of unwanted terms:
rw = sw + MAI + ISI + ARN
There are several types of CDMA receiver structures with varying performance and complexity (very simple (MF) - very complicated (ML)).
Unwanted termsDesired term
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
)1(d
)(kd
)(Kd
)1(c
)(kc
)( Kc
+
C
s )(kh +)(kH
)(kA
)(knChannel
)(kx
d
Noise
Background
d(k) : Data vector user k d: Combined data vector c(k) : CDMA code user k C: Code delay matrix h(k) : Channel impulse response (CIR) user k H : CIR delay matrix A: System matrix
x(k)=A(k)d+n(k)
Downlink CDMADownlink CDMA
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
MUD basicsWhy multiuser detection (MUD)?
MUD addresses the interference problem by cancelling or suppressing interfering users and multipath effects from the desired users signal.
In the Base Station (uplink) … knowledge of all users codes, data and CIRs can be used
to enhance the signal of a specific user.
In the Mobile Station (downlink) … limited knowledge of users codes and estimates CIR,
hence sub-optimum approaches need to be considered.
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
MUD basics
Required knowledge: CDMA codes of all K users and Channel impulse responses of all K users.
Depending on the data detection scheme: Covariance matrix of transmitted data Covariance matrix of noise vector
MUD Receiver
User 1
Prior user knowledge,Feedback
Multi-pathchannel
Noise CD
MA
cod
es
Ch
ann
elre
spo
nse
No
ise
Co
varia
nce
Mat
rixD
ata
Co
varia
nce
Mat
rix
receivedsignal
User k
User K
User 1
User k
User K
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
General Classification of MUD’s:
Cancellation
Parallel Successive
Hybrid
DECORPICMMSEPIC DECORSICPSMLMUD
Equalizer
MMSE DECOR
MUD
ML MF
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Adaptive Filters for MUDWhy do we need adaptive filters
in MUD receivers?
For the estimation of
time-varying channel impulse response
time-varying user profile
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Adaptive Filters for MUDWhat adaptive filter types do we use?
Linear or non-linear? LINEAR due to complexity savings
What linear filters are suitable? Transversal filter
LMS based RLS based Stochastic Newton Class, i.e. SFAEST?
Lattice structure Systolic Arrays
QR-decomposition based RLS
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Adaptive Filters for MUDTrade-off between
Performance
Computational complexity
Stability
We focus on transversal filters and investigate in their ability to
support reliable and accurate equalisation in a mobile communication scenario.
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Adaptive Filters for MUDMost popular algorithms are used first
Least Mean Square
Recursive Mean Square
Traditionally the combined channel impulse response (CCIR) is estimated
one filter that estimates system matrix A
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
)1(d
)(kd
)(Kd
)1(c
)(kc
)( Kc
+
C
s )(kh +)(kH
)(kA
)(knChannel
)(kx
d
Noise
Background
d(k) : Data vector user k d: Combined data vector c(k) : CDMA code user k C: Code delay matrix h(k) : Channel impulse response (CIR) user k H : CIR delay matrix A: System matrix
x(k)=A(k)d+n(k)
Downlink CDMADownlink CDMA
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Receiver Architecturesfor MUD (1)
The Conventional Architecture (CA)
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Performance Evaluation (CA) Apart from the time-varying channel the dynamic
user profile adds to the complexity of the equalisation task.
Users switching on and off can cause error bursts in the desired users signal as the equaliser needs time to adapt the filter coefficients. We refer to this as the birth/death problem.
Convergence rates of adaptive filter algorithms therefore are crucial to maintain a suitable bit error ratio.
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Performance Evaluation (CA)
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Performance Evaluation (CA) Simulations show
LMS is too slow in convergence and cannot cope with interference bursts due to birth/death scenario
RLS can handle this but is rather complex
A better solution is required which reduces the complexity of the adaptive filter without performance loss by
a) using an adaptive algorithm with lower computational complexity
b) modifying the task of the adaptive filter to reduce complexity
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Receiver Architecturesfor MUD (2)
The New Architecture (NA)
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Architecture comparisonConventional Architecture
Training of combined channel impulse response i.e. convolution of code and
channel impulse response
Filter length M+P-1
Training at symbol level hence slow convergence
New Architecture
Training of channel impulse response by means of filter
Pre-calculated multiuser detector performs despreading of desired user
Filter length P
Training at chip level hence fast convergence
M = CDMA codelength, P = # of channel taps
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Simulation parametersChannel parameters
AWGN 6-tap static FIR Coefficients
0.6608, 0.5287, -0.3965, 0.2643, -0.1983, 0.1322
COST207 6-tap Typical Urban 2 Mchip/s datarate Desired vehicle speed
0-540 km/h
Simulated Doppler Frequencies
50 Hz 27 km/h
100 Hz 54 km/h
Adaptive Filter parameters Memory Length
64 symbols CA64 symbols NA (=1024 chips)
Other essential parameters 16-chip Random Codes Dynamic User Profile Cycle
32 symbols Error threshold
10000 Errors or BER=1e-6 with > 10 Errors
Signal -to-Noise Ratio 6 dB for Ensemble BER 0-10 dB for Average BER
NA uses combined signal training
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Performance comparison
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Performance comparisonBit error ratio for RLS in new architecture
Different training lengths, PG=16, 8 user, 6 tap channel
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Performance comparisonBit error ratio for NLMS in new architecture
Different training lengths, PG=16, 8 user, 6 tap channel
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Ensemble BER performance6-tap TU COST207
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Ensemble BER performance6-tap TU COST207
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Average BER performance6-tap TU COST207
10th October 2000 Signals and Systems Group, The University of Edinburgh
Chip Level Equalisation for W-CDMA
Stephen McLaughlin
Conclusions The new architecture shows improved
performance in time-varying scenario. Average BER is improved Ensemble BER is improved
=> Error burst behavior can be tackled
Any adaptive LS algorithm can be used with the new architecture - RLS shows best results amongst tested algorithms.
NA requires only short filter length for CIR estimation particularly useful when long spread codes are used