More Info
									                                                                                          2008 ANNUAL REPORT        51

                        D. SIGNAL PROCESSING

                        Program Goals:
                        This program focuses on model-based signal processing. Research problems include physical
                        modelling, system identification, model validation, prediction, filtering, and signal recovery. In
                        the past, we have studied various signal processing problems in adaptive
                        control, Kalman filtering, communications channel equalization, and multi-user detection
                        for wireless communications. In 2008, we have focused our research in the following
                        areas: system identification, quantisation, dual-stage control, time-frequency analysis
                        and communication systems. Specific applications include commodity price modelling,
                        high-precision positioning and high-speed analog-top-digital converter design.

Minyue fu               D.1 SYSTEM IDENTIfICATION
Program Leader
                        Project Leader: J.-C. Agüero
                        Researchers:      G.J. Adams, J.-C. Agüero, G.C. Goodwin, C. Rojas (Student) and J.S. Welsh
                        External Academic Collaborators:
                        M. Alamir (Institut National Polytechnique de Grenoble, france),
                        P.M.J. Van den Hof (Delft University of Technology, The Netherlands)
                        J.I. Yuz (Universidad Técnica federico Santa María,Chile)

                        D.1.1 Experiment design for system identification
                        This research is partly funded under an ARC Discovery Grant.
                        We have a large research program aimed at gaining a better understanding of experiment
Juan-Carlos Agüero      design. There are several sub-projects within this research program.
Deputy Program Leader   One of the topics studied in 2008 is the equivalence between least costly and traditional
                        experiment design for control. We consider experiment design problems for both open and
                        closed loop systems. In open loop, equivalence is established for three specific cases,
                        relating to different parameterisations of the covariance expression (i.e. finite and high
                        order approximations) and model structure (i.e. dependent and independently parameterised
                        plant and noise models). In the closed loop setting, we consider only finite order covariance
                        expressions. H∞ performance specifications for control are used to determine the bounds
                        on the covariance expression for both the open and closed loop cases – see Rojas, Agüero,
                        Welsh and Goodwin (2008) in Journal Papers.
                        On the other hand, we also have developed fundamental integral limitations on the variance of
                        estimated parametric models, for both open and closed loop identification. As an application
                        of these results we have shown that, for multisine inputs, a well known asymptotic (in model
                        order) variance expression provides upper bounds on the actual variance of the estimated
                        models for finite model orders. The fundamental limitations established here give rise to a
                        ‘water-bed’ effect – see Rojas, Agüero, Welsh (2009) in Journal Papers.

                        D.1.2 On parameter estimation of the Schwartz-Smith short-term/long-term model
                        Researchers:      M. fu and Xin Tai (Visiting Student)
                        The short-term/long-term model proposed by Schwartz and Smith in 2000 is widely used
                        in modelling commodity prices. A key and nontrivial problem in this modelling technique is
                        how to estimate the model parameters. This work considers the parameter estimation problem
                        based on the maximum likelihood criterion and proposes a method to simplify the task.
                        Two components are contained in the proposed method: one to do with re-parameterization
                        and one to do with separating the parameter set so that one part can be solved directly
                        using least-squares and another part using nonlinear optimization. The effectiveness of
                        the proposed method has been demonstrated via numerical tests.
52     ARC Centre of Excellence for Complex Dynamic Systems and Control

D.1.3 Identifiability of errors in
      variables dynamic systems
Researchers:      J.-C. Agüero and
                  G.C. Goodwin
                                                      D.1.5 Virtual closed loop identification:
                                                            A generalized tool for
                                                            identification in closed loop
                                                      Researchers:        J.-C. Agüero, G.C. Goodwin
                                                                                                       D.1.7 Redundancy vs multiple starting
                                                                                                             points in nonlinear systems
                                                                                                             related inverse problems
                                                                                                       Researchers:     M. Alamir (france), G.C.
                                                                          and P.M.J. Van den Hof                        Goodwin and J.S. Welsh
There has been substantial research                                       (The Netherlands)
carried out on the errors in variables                                                                 Nonlinear inverse problems arising
(EIV) identifiability problem for dynamic             We have proposed a virtual closed loop           in dynamic system state/parameter
systems. These results are spread across              model parameterization to perform system         estimation are generally nonconvex
a significant volume of literature. We                identification. This parameterization            and possess multiple local minima that
have developed a single theorem which                 is designed to achieve specific goals.           may threaten the convergence of global
compactly summarizes many of the known                We show that the method includes,                optimisation routines. This problem is
results. The theorem also covers several              as special cases, known methods for              generally addressed by a multiple starting
cases which we believe to be novel. We                closed loop identification and also offers       point based algorithm. An alternative
have analysed single input single output              additional flexibility. We have analysed         approach is to use the recently proposed
systems using second order properties.                the ramifications of the new tailor-made         concept of safe redundancy in order
We have also extended the results to a                parameterization for systems operating           to derive an algorithm that crosses
class of multivariable systems.                       in closed loop. The approach exploits a          singularities by exploiting the particular
                                                      property of Box-Jenkins models in order          nature of dynamic inverse problems. In fact,
We have also developed novel results                  to minimize the bias arising from feedback       it is possible to combine the two algorithms
related to the identifiability of EIV dynamic         and noise model mismatch – see Agüero,           such that the properties of each are utilised
systems based on exploiting properties                Goodwin and Van den Hof (2008) in                in a hybrid algorithm. The underlying idea
of non-stationary data. We have analysed              Conference Papers.                               in safe redundancy is not to choose a
single-input single-output systems using                                                               good cost function but to have different
second order properties. Our results show             D.1.6 Relative error issues in sampled           cost functions, that share the same global
that, it is possible to establish identifiability           data models                                minimum but have no reason to share
of EIV systems under mild conditions when                                                              all the local minimum, and make them
                                                      Researchers:        J.-C. Agüero, G.C. Goodwin
the data is non-stationary – see Agüero                                                                cooperate to reach the desired solution –
                                                                          and J.I. Yuz (Chile)
and Goodwin (2008) in Journal Papers and                                                               see Alamir, Welsh and Goodwin in
Conference Papers.                                    Most real world systems operate in               Journal Publications.
                                                      continuous time. However, to store,
D.1.4 Robust identification of process                analyse or transmit data from such               D.1.8 On useful redundancy in robust
      models from plant data                          systems the signals must first be sampled.             optimal experiment design for
Researchers:      G.J. Adams, J.-C. Agüero,           Consequently there has been on-going                   nonlinear system identification
                  G.C. Goodwin, C.R. Rojas,           interest in sampled data models for              Researchers:     M. Alamir (france),
                  J.S. Welsh and                      continuous time systems. The emphasis                             G.C. Goodwin and
                  J.I. Yuz (Chile)                    in the literature to-date has been on three                       J.S. Welsh
                                                      main issues namely the impact of folding,
A precursor to any advanced control                   sampled zero dynamics and the associated         This research is partly funded under an
solution is the step of obtaining an accurate         model error quantification. Existing error       ARC Discovery Grant.
model of the process. Suitable models                 analyses have almost exclusively focused
can be obtained from phenomenological                                                                  In particular, a recently introduced
                                                      on unnormalised performance. However, in
reasoning, analysis of plant data or a                                                                 redundancy property associated to inverse
                                                      many applications relative errors are more
combination of both. We have a large                                                                   problems, has been for dynamic systems
                                                      important. for example, high performance
research program aimed at the problem                                                                  is heavily exploited to guarantee global
                                                      controllers tend to invert the system
of estimating (or calibrating) models from                                                             convergence. The project considers general
                                                      dynamics and consequently relative errors
plant data. A key goal is to achieve robust                                                            discrete-time nonlinear systems in which
                                                      underpin closed loop performance issues
identification. By robust we mean that small                                                           measurements are affected by bounded
                                                      including robustness and stability. This has
errors in the hypotheses should lead to                                                                noise – see Alamir, Welsh and Goodwin in
                                                      motivated us to examine the relative errors
small errors in the estimated models. We                                                               Conference Publications.
                                                      associated with several common sampled
argue that, in some circumstances, it is              data model types. This analysis reveals that
essential that special precautions, including         the inclusion of appropriate zero dynamics
discarding some part of the data, be taken            is essential to ensure that the relative error
to ensure that robustness is preserved. We            converges to zero as the sampling period
have illustrated these ideas using practical          is reduced – see Goodwin, Yuz and Agüero
case studies – see Yuz and Goodwin                    (2008) in Conference Papers.
(2008) in Chapters in Books, Goodwin,
Agüero, Welsh, Yuz, Adams and Rojas
(2008) in Journal Papers.
                                                                                                                 2008 ANNUAL REPORT        53

Project Leader: Minyue fu
Researchers:     M. fu, A. Salton (Student) and J. Zheng
                                                                                                   for a hard disk drive
                                                                                             D.2.3 Nonlinear tracking control

                                                                                                                   M. fu, C. Du (Singapore),
                                                                                                                   Y. Wang (Singapore)
                                                                                                                   and J. Zheng
External Academic Collaborators:
Y. Wang (Nanyang Technological University, Singapore)                                        This work studies a nonlinear tracking
C. Du (Agency for Science, Technology and Research (A*STAR)                                  control method for a hard disk drive dual-
Data Storage Institute (DSI), Singapore)                                                     stage actuator (DSA) system that consists
Y. Guo (Nanyang Technological University, Singapore)                                         of a voice coil motor (VCM) actuator
L. Xie (Nanyang Technological University, Singapore)                                         and a piezoelectric (PZT) microactuator.
                                                                                             Conventional track seeking controllers for
D.2.1 Development of a linear motion dual-stage actuator                                     DSA systems were generally designed to
                                                                                             enable the VCM actuator to approach the
Researchers:     J. Zheng and M. fu                                                          target track without overshoot. However,
A dual-stage actuator (DSA) servo system is characterized by a structural design with        we observe that this strategy is unable to
two actuators connected in series along a common axis. The primary actuator (coarse          achieve the minimal settling time when
actuator) is of long travel range but with poor accuracy and slow response time. The         the target tracks are beyond the PZT
secondary actuator (fine actuator) is typically of higher precision and faster response      actuator stroke limit. To further reduce the
but with a limited travel range. By combining the DSA system with properly designed          settling time, we design the VCM actuator
servo controllers, the defects of one actuator are compensated by the merits of the other,   controller to yield a closed-loop system
resulting in an improved performance. We have designed a linear motion DSA where the         with a small damping ratio for a fast rise
primary stage is driven by a linear motor (LM) and the secondary stage is driven by a        time and certain allowable overshoot.
piezoelectric actuator (PZT). Experimental results demonstrate a significant improvement     Then, a composite nonlinear control law is
both in settling time reduction and disturbance rejection when comparing the proposed        designed for the PZT actuator to reduce
design with a traditional single-stage actuator system.                                      the overshoot caused by the VCM actuator
                                                                                             as the system output approaches the target
                                                                                             track. Experimental results show that the
D.2.2 Development of a rotational motion dual-stage actuator                                 proposed dual-stage servo outperforms the
Researchers:     J. Zheng, A. Salton and M. fu                                               conventional dual-stage servo in short-span
                                                                                             seeking and, additionally, achieves better
While the use of dual-stage actuators (DSAs) is popular in hard disk drives (HDDs), some
                                                                                             track following accuracy than the VCM only
other applications that make use of this structural design include XY positioning tables,
                                                                                             single-stage servo.
macro/micro robot manipulators, machine tools, wafers alignment in microlithography,
among others. Motivated by the fact that several other applications may benefit from long
range over a compact space (such as HDDs), we have designed and manufactured the
rotational motion dual-stage actuator as depicted in figure 34. This system is based in
the same principle as the linear motion DSA, where the primary stage, driven by a voice
coil motor (VCM), provides the system with a large range, and the secondary stage, driven
by a piezoelectric actuator (PZT), achieves a higher precision and faster response. The
mechanical design of the rotational DSA, however, is different and more sophisticated than
the one of the linear motion DSA. In order to achieve a rotational motion the secondary
stage must bend the structure of the actuator, which requires the system to have complex
mechanical structure. To overcome this difficulty we have made use of a finite element
method software tool to aid our mechanical design. Simulation and experimental results
demonstrate the effectiveness of the proposed DSA system.

                                                                                             figure 35: Illustration of the proposed
                                                                                             control strategy.

                                                 figure 34: The rotational motion
                                                 dual stage actuator.
54     ARC Centre of Excellence for Complex Dynamic Systems and Control

D.2.4 Development of an extended
      reset controller and its
      experimental demonstration
Researchers:      M. fu, Y. Guo (Singapore),
                                                      Simulations and experimental results are
                                                      used to demonstrate the effectiveness of
                                                      the proposed control method for tracking
                                                      position reference commands and its
                                                                                                       a global minimiser of the 2-norm of the
                                                                                                       reconstruction error. The first optimality
                                                                                                       condition is related to the design of the
                                                                                                       feedback quantiser, and can always be
                  Y. Wang (Singapore),                robustness to system uncertainties.              achieved. The second condition depends
                  L. Xie (Singapore) and                                                               only on the reconstruction vectors, and is
                  J. Zheng                                                                             given explicitly in terms of the Gram matrix
                                                      D.3 QUANTISATION                                 of the reconstruction frame. As a by-product,
Reset control aims at enhanced                        Project Leader: D.E. Quevedo                     we also show that the first condition alone
performance that cannot be obtained by                                                                 characterizes scalar feedback quantisers
linear controllers. The conventional reset            Researchers:        G.C. Goodwin, S. Derpich     that yield the smallest MSE, when one
control is simple for implementation by                                   (Student), D. Marelli,       models quantisation noise as uncorrelated,
resetting some of its controller states to                                J. Østergaard,               identically distributed random variables.
zero when its input meets a threshold.                                    D.E. Quevedo and
However, it is found that in some cases the                               E.I. Silva (Student)         We have also characterized the rate-
enhanced performance of conventional                                                                   distortion function for zero-mean stationary
                                                      External Academic Collaborators:                 Gaussian sources under the MSE fidelity
reset control is still limited such as with
                                                      H. Bölcskei (ETH Zurich, Switzerland)            criterion and subject to the additional
only partial reduction of the overshoot in a
step reference response. Thus, the stability                                                           constraint that the distortion is uncorrelated
analysis and design of the reset control              D.3.1 Feedback quantisers                        to the input. The solution is given by two
system are extended, where the reset time                                                              equations coupled through a single scalar
                                                      Researchers:        S. Derpich,
instances are prespecified and the controller                                                          parameter. This has a structure similar
                                                                          J. Østergaard,
states are reset to certain non-zero values,                                                           to the well known water-filling solution
                                                                          G.C. Goodwin,
which are calculated online in terms of the                                                            obtained without the uncorrelated distortion
                                                                          D.E. Quevedo, E.I. Silva
system states for optimal performance.                                                                 restriction. Our results fully characterize the
Experimental results on a piezoelectric               We have developed novel results on               unique statistics of the optimal distortion.
positioning stage demonstrate that the                perfect reconstruction feedback quantisers       We also show that, for all positive distortions,
extended reset control can further reduce             (PRfQs), i.e., noise-shaping, predictive and     the minimum achievable rate subject to the
the overshoot and thus achieve shorter                sigma-delta A/D converters whose signal          uncorrelation constraint is strictly larger
settling time than the conventional reset             transfer function is unity. Our analysis of      than that given by the un-constrained rate-
control. Robustness tests against various             this class of converters is based upon an        distortion function. This gap increases with
step levels, disturbance and sensor noise             additive white noise model of quantisation       the distortion and tends to infinity an zero,
have also been developed.                             errors. Our key result is a formula that         respectively, as the distortion tends to zero
                                                      relates the minimum achievable MSE of such       and infinity – see Derpich, Silva, Quevedo
                                                      converters to the signal-to-noise ratio (SNR)    and Goodwin (2008) in Journal Papers,
D.2.5 A reset state estimator using an                of the scalar quantiser embedded in the          Derpich, Quevedo and Goodwin (2008) in
      accelerometer for enhanced                      feedback loop. This result allows us to obtain   Conference Papers, Silva, Quevedo and
      motion control with sensor                      analytical expressions that characterize         Goodwin (2008) in Conference Papers.
      quantisation                                    the corresponding optimal filters. We also
Researchers:      M. fu and J. Zheng                  show that, for a fixed SNR of the scalar         D.3.2 Quantisation of filter bank
                                                      quantiser, the end-to-end MSE of an optimal            frame expansions through
Sensor quantisation is a key factor that
                                                      PRfQ which uses the optimal filters (which             moving horizon optimisation
deteriorates the tracking performance of
                                                      for this case turn out to be IIR) decreases
positioning systems with low-resolution                                                                Researchers:      H. Bölcskei (Switzerland),
                                                      exponentially with increasing oversampling
optical encoders. This work presents a                                                                                   G.C. Goodwin and
                                                      ratio. Key departures from earlier work
method to improve the performance of                                                                                     D.E. Quevedo
                                                      include the fact that fed back quantisation
such systems by merging an accelerometer
                                                      noise is explicitly taken into account and       We have developed a novel approach to
of low cost. firstly, to reject the external
                                                      that the order of the converter filters is not   quantisation in oversampled filter banks.
disturbance, friction force and system
                                                      a-priori restricted.                             The new technique is based on moving
perturbations, we design a disturbance
observer (DOB) based on acceleration                  We have also developed results on scalar         horizon optimisation, does not rely on an
signals. Secondly, a reset kinematic                  feedback quantisation (SQf) with uniform         additive white noise quantisation model and
state estimator (RKSE) is designed                    quantisers. We focus on general SfQ              allows one to explicitly enforce stability of
using acceleration signals to make the                configurations where reconstruction is           the associated nonlinear feedback loop.
state estimate immune to both system                  via a linear combination of frame vectors.       Moreover, the quantisation structure that
perturbations and input disturbances. Thirdly,        Using a deterministic approach, we derive        we have proposed includes sigma-delta and
a state feedback controller is designed               two necessary and sufficient conditions          linear predictive subband quantisers as a
based on the internal model principle (IMP)           for SfQ to be optimal, i.e., to produce, for     special case and, in general, outperforms
for accurate sinusoidal reference tracking.           every input, a quantised sequence that is        them – see Quevedo, Bolcskei and Goodwin
                                                                                                       (2008) in Journal Papers.
                                                                                                                            2008 ANNUAL REPORT     55

                                                      D.4 TIME-fREQUENCY ANALYSIS
                                                      Project Leader: M. fu
                                                      Researchers:          M. fu, K. Mahata and D.E. Marelli
                                                      External Academic Collaborators:
                                                      M. Aramaki (CNRS-Institut de Neurosciences Cognitives
                                                      de la Méditerranée, Marseille, france)
                                                      P. Balazs (Austrian Academy of Sciences, Austria)
                                                      R. Kronland-Martinet (CNRS-Laboratoire de Mécanique et d’Acoustique, Marseille, france.
                                                      P. Majdak (Austrian Academy of Sciences, Austria)
                                                      I. Raeburn (University of Wollongong, Australia)
                                                      C. Verron (Orange Labs, Lannion, france.

                                                      D.4.1 Linear system modelling in the subband domain
                                                      Researchers:          P. Balazs (Austria), M. fu, P. Majdak (Austria) and D.E. Marelli
                                                      The subband technique permits a numerically efficient implementation of a linear system.
                                                      This is done by putting in cascade an analysis filterbank, followed by a transfer matrix
                                                      (the subband model) and a synthesis filterbank. To exploit the potential of this technique
                                                      we designed the subband model, the analysis and the synthesis filterbanks, to minimize
                                                      a statistical least-mean squares criterion; and we included in the design procedure, a
                                                      sparsity criterion on the subband model. By doing so, we minimize the number of non-zero
                                                      coefficients of the subband model, and consequently its implementation complexity. In
                                                      figure 36 we show the performance of the proposed subband design when implementing
                                                      head-related transfer functions (HRTf). The proposed approach yields significant
                                                      computational savings in comparison with pole-zero models, or finite impulse response
                                                      models implemented using fast convolution algorithms.

                                                       MoDEl                APPRoxiMATion ERRoR               CoMPlExiTy [mult./sample]
                                                       fIR                  40.63e-3                          135
                                                       PZ                   15.29e-3                          81
                                                       SB                   16.55e-3                          37



                                                       Amplitde [dB]

                                                                                                        True HRTF
                                                                   −20                                  FIR model
                                                                                                        Pole−zero model
  figure 36: Head-related transfer function (HRTf)
     implementation using a finite impulse response                −30                                  Subband model
(fIR) model, a pole-zero model (PZ) and a subband
       (SB) model. Left: frequency response. Right:

                Approximation error and complexity.

                                                                      0           531          1255     2711                 6414          26646
                                                                                               Frequency [Hz]
56    ARC Centre of Excellence for Complex Dynamic Systems and Control

D.4.2 High-speed analog-to-digital
      converter design
Researchers:     M. fu, K. Mahata and
                 D.E. Marelli
                                                     D.4.3 Efficient sound synthesis for
                                                           real-time applications
                                                     Researchers:        M. Aramaki (france),
                                                                         R. Kronland-Martinet
                                                                                                      D.4.4 Proper group actions in abstract
                                                                                                            harmonic analysis
                                                                                                      Researcher:      D.E. Marelli, I. Raeburn
                                                                         (france), D.E. Marelli and
High speed analog-to-digital converters                                                               Abstract harmonic analysis generalizes
                                                                         C. Verron (france)
(ADCs) can be realized using the so-called                                                            the study of continuous-, discrete-time or
multi-channel architecture, which combines           The inverse fast fourier transform (IffT)        multidimensional functions, to functions
the outputs of a number of slow speed                synthesis is a flexible and efficient method     with domain on a topological group (space),
ADCs. A research challenge is how to                 for synthesizing digital sound. It consists      which take values (have range) on a
combine the output from different channels           in approximating the spectral shape of the       C-star algebra rather than on the real or
to reconstruct the desired samples. This             desired sound in a sequence of overlapping       complex numbers. A research line within
design is typically done by assuming that            blocks of samples. The application of the        this context studies the properties of these
the input signal is bandlimited, leading             IffT method is limited by its inherent time/     functions when a group action (a set of
to a computaionally expensive design.                frequency resolution tradeoff, since its         automorphisms with a group structure) is
We proposed an alternative approach by               block size needs to be longer than the           applied on their range. A famous theorem
dropping the bandlimited assumption and              noise auto-correlation function, which is        of Green asserts that in the case of
doing the design in a statistically optimal          often not compatible with the generation         commutative C-star algebras, if a proper
sense. We have applied this approach                 of short transient signals. We have shown        and free action acts on the maximal ideal
in both, time-interleaved ADCs, and for              that the IffT method can be re-casted            space, then the crossed product between
hybrid-filterbank ADCs. The proposed                 as the implementation of a time-varying          the acting group and the C-star algebra
designs lead to superior performances in             filter in the subband domain. Using this, we     is Morita equivalent to the functions on
terms of complexity when compared with               were able to overcome the aforementioned         the orbit space. Rieffel used this result to
other available approaches.                          limitation by designing an appropriate           abstract the concepts of properness and
                                                     subband model for the time-varying filter.       freeness (which he called saturation) to
                                                                                                      group actions acting on general (non-
                                                                                                      commutative) C-star algebras. When these
                                                                                                      concepts are applied to a particular case
                                                                                                      of non-commutative C-star algebras called
                                                                                                      graph algebras, they exhibit a striking
                                                                                                      parallel with the commutative case. We
                                                                                                      studied the properties of proper actions
                                                                                                      which are not saturated. In the commutative
                                                                                                      case, we showed that not the whole
                                                                                                      crossed product, but a proper ideal of it
                                                                                                      satisfies the above Morita equivalence,
                                                                                                      if and only if the action is not free. Also,
                                                                                                      we identified this ideal by describing the
                                                                                                      primitive ideals which contain it. In the non-
                                                                                                      commutative case, we showed that while
                                                                                                      the ideal can certainly be proper, there
                                                                                                      are situations where the action is not free
                                                                                                      but the ideal is the whole crossed product.
                                                                                                      Moreover, even in the later case, Morita
                                                                                                      equivalence may not hold. Our results
                                                                                                      suggest that the aforementioned parallel
                                                                                                      completely breaks down in the absence
                                                                                                      of saturation.
                                                                                                             2008 ANNUAL REPORT        57

Project Leader: G.C. Goodwin
Researchers:    J.A. De Doná, M. fu,
                                              and to apply a finite alphabet constraint
                                              over a sliding optimisation window. The
                                              approach leads to embellishments of the
                                              usual (non-adaptive) decision feedback
                                                                                            two new optimization criteria unique to
                                                                                            non-binary codes. for each individual
                                                                                            short cycle, we assign non-binary values
                                                                                            to the associated non-zero elements in the
                G.C. Goodwin, J. Ning,        equalizer and its multi-step extensions.      parity-check matrix in order to optimise
                D.E. Quevedo and              It includes a provision for degrees of        the decoding performance. for overlapped
                M. Wang,                      belief in past estimates, which addresses     short cycles, the non-zero elements are
                                              the problem of error propagation – see        assigned by first improving the minimum
External Academic Collaborators:              Quevedo, Goodwin and De Dona (2008)           distance and then optimising the decoding
B.S. Krongold (The University of              in Journal Papers.                            performance. The additional optimization
Melbourne, Australia)                                                                       requires only moderate increase in the code
                                                                                            design complexity. Simulations results are
                                              D.5.3 EM-based receiver design for            used to demonstrate the effectiveness of
D.5.1 A complex-baseband active-set                 uplink MIMO-OFDMA systems               the proposed method.
      approach for tone reservation
                                              Researchers:    G.C. Goodwin,
      PAR reduction in OFDM systems
                                                              D.E. Quevedo and
Researchers:    G.C. Goodwin,                                 M.Wang                        D.5.5 Outer code design for serially
                B.S. Krongold                                                                     concatenated continuous phase
                (Australia), D.E.             We have proposed an iterative receiver              modulation with symbol-wise
                Quevedo and M. Wang           for uplink multiple-input multiple-output           interleading
                                              (MIMO) orthogonal frequency division          Researchers:     M. fu and J. Ning
We have developed an active-set approach      multiple access (OfDMA) systems based
for PAR reduction via time reservation in     on the expectation maximization (EM)          Serially concatenated continuous phase
complex-baseband orthogonal frequency         algorithm. Iterating between the E-step       modulation (SCCPM) schemes containing
division multiplexing (OfDM) systems.         and the M-step, the EM-based receiver         a binary outer code are good candidates
In the complex-baseband model, the            updates the channel estimates and, refines    in achieving both high spectral efficiency
optimisation problem of tone reservation      data detection by increasing the likelihood   and high power efficiency. Recently, it is
turns out to be a quadratically constrained   function. Practical implementation issues     shown that SCCPM systems with symbol-
quadratic program (QCQP), which is            have also been considered: space-time         wise interleaving can achieve substantial
computationally prohibitive for practical     block-coding (STBC) is incorporated to        improvement in BER performance over
implementations. To address this problem,     improve system performance against            their counterparts with bit-wise interleaving.
we consider the complex-valued samples of     fading; a reduced-complexity algorithm        The modulation index and the duration
an OfDM symbol as sectors in the complex      is proposed, which simplifies the             of the shape function in the modulation
plane and develop an iterative clipping       computation whilst not compromising           are important parameters for spectral
algorithm. In our approach, PAR reduction     performance. Simulation results show          efficiency. Conventionally, the outer code
is achieved by minimizing the radius of       that the performance of the proposed          with order-2 is assumed to be superior to
the circle that enclosed all samples, thus    receiver can approach coherent detection      those with higher orders despite of the
minimizing the maximum power peak.            with affordable computational cost – see      modulation index and the duration. This
Simulation results show that the proposed     Wang, Goodwin and Quevedo (2008) in           work considers symbol-wise interleaved
algorithm can achieve near-optimal            Conference Papers.                            SCCPM schemes and optimises the outer
performance with fast convergence and                                                       codes using extrinsic information transfer
limited complexity – see Wang, Quevedo,       D.5.4 A new method for lowering the           (EXIT) chart for different modulation
Goodwin and Krongold (2008)                         error floors of non-binary              indices and the durations. Also an
in Conference Papers.                               LDPC codes                              additional intra-symbol interleaving
                                              Researchers:    M. fu and J. Ning             is used to achieve the best decoding
D.5.2 A multi-step detector for linear                                                      performance. Simulation results are used
      ISI-channels incorporating de           Non-binary low-density parity-check           to show the superior performance of the
      grees of belief in past estimates       (LDPC) codes are good candidates for          proposed schemes.
Researchers:    J.A. De Doná,                 forward-error-control code with short block
                G.C.Goodwin and               lengths (normally less than 5,000 bits) and
                D.E. Quevedo                  can substantially outperform their binary
                                              counterparts. In this work, we propose a
We have formulated the channel                new method for improving the performance
equalization problem in the framework of      of non-binary LDPC codes of short block
constrained maximum likelihood estimation.    lengths at the high signal-to-noise ratio
This allows us to highlight key issues        (SNR) region. In addition to the standard
including the need to summarize past data     short cycle removal methods, we consider

To top