Dissertations / Theses on the topic 'Blink detection'

Accidents caused by drivers’ drowsiness behind the steering wheel have a high fatality rate because of the discernible decline in the driver’s abilities of perception, recognition, and vehicle control abilities while sleepy. Preventing such accidents caused by drowsiness is highly desirable but requires techniques for continuously detecting, estimating, and predicting the level of alertness of drivers and delivering effective feedback to maintain maximum performance. The main objective of this research study is to develop a reliable metric and system for the detection of driver impairment due to drowsiness. More specifically, the goal of the research is to develop the best possible metric for detection of drowsiness, based on measures that can be detected during driving. This thesis describes the new studies that have been performed to develop, validate, and refine such a metric. A computer vision system is used to monitor the driver’s physiological eye blink behaviour. The novel application of green LED illumination overcame one of the major difficulties of the eye sclera segmentation problem due to illumination changes. Experimentation in a driving simulator revealed various visual cues, typically characterizing the level of alertness of the driver, and these cues were combined to infer the drowsiness level of the driver. Analysis of the data revealed that eye blink duration and eye blink frequency were important parameters in detecting drowsiness. From these measured parameters, a continuous measure of drowsiness, the New Drowsiness Scale (NDS), is derived. The NDS ranges from one to ten, where a decrease in NDS corresponds to an increase in drowsiness. Based upon previous research into the effects of drowsiness on driving performance, measures relating to the lateral placement of the vehicle within the lane are of particular interest in this study. Standard deviations of average deviations were measured continuously throughout the study. The NDS scale, based upon the gradient of the linear regression of standard deviation of average blink frequency and duration, is demonstrated as a reliable method for identifying the development of drowsiness in drivers. Deterioration of driver performance (reflected by increasingly severe lane deviation) is correlated with a decreasing NDS score. The final experimental results show the validity of the proposed model for driver drowsiness detection.

The thesis deals with eye blink detection, which is part of complex topic of face detection and recognition. The work intents on digital image processing. There is analyse of the topic and description of image databases for testing. Two main chapters describe design of eye blink detection with digital image processing with IR technology and without IR technology.

This master thesis deals with the issues of the eye blink recognition from video. The main task is to analyse algorithms dealing with a detection of persons and make a program that could recognize the eye blink. Analysis of these algorithms and their problems are in the first part of this thesis. In the second part design and properties of my program are described. The realization of the program is based on the method of move detection using the accumulated difference frame, which helps to identify the eye areas. The eye blink detection algorithm tests a match between a tresholded pattern of the eye area taken from the actual frame and the frame before. The resolution whether the eye blink happened or not, is based on the level of the match. The algorithm is designed for watching a sitting man, which is slightly moving. The background can be a little dynamic as well. An average quality video with a moderator and dynamic backround was used as a tested subject.

BACKGROUND.One group of theories aiming at providing a framework explaining the etiology, maintenance and phenomenology of anxiety disorders is classified as cognitive models of anxiety. These approaches assume that distortions in specific levels of information processing are relevant for the onset and maintenance of the disorder. A detailed knowledge about the nature of these distortions would have important implications for the therapy of anxiety, as the implementation of confrontative or cognitive elements precisely fitting the distortions might enhance efficacy. Still, these models and related empirical evidence provide conflicting assumptions about the nature of disorder-linked processing distortions. Many cognitive models of anxiety (e.g., Fox, Russo, & Dutton, 2002; Mathews & Mackintosh, 1998; Williams, Watts, MacLeod, & Mathews, 1997) postulate that anxiety-linked biases of attention imply hypervigilance to threat and distractibility from other stimuli in the presence of feared materials. This is convincingly confirmed by various experimentalclinical studies assessing attention for threat in anxious participants compared to non-anxious controls (for a review, seeMathews &MacLeod, 2005). In contrast, assumptions concerning anxiety-linked biased memory for threat are less convincing; based on the shared tendency for avoidance of deeper elaboration in anxiety disorders, some models predict memory biases only for implicit memory tasks (Williams et al., 1997) or even disclaim the relevance of memory in anxiety at all (e.g., Mogg, Bradley, Miles, & Dixon, 2004). Other theories restrict the possibility of measuring disorder-specific memory biases to tasks that require merely perceptual encoding of the materials instead of verbal-conceptual memory (e.g., Fox et al., 2002; Mathews &Mackintosh, 1998). On the one hand, none of these models has integrated all the inconsistencies in empirical data on the topic. On the other hand, the numerous empirical studies on memory in anxiety that have been conducted with varying materials, anxiety disorders, encoding and retrieval conditions do not allow final conclusions about the prerequisites for finding memory biases (for a review, see MacLeod & Mathews, 2004). A more detailed investigation of the complete spectrum of memory for threat utilizing carefully controlled variations of depth of encoding and materials is needed. In view of these inconsistencies, it is all the more surprising that one important part of this spectrum has so far remained completely uninvestigated: visual working memory (VWM). No study has ever differentially addressed VWM for threat in anxious vs. nonanxious participants and none of the cognitive models of anxiety provides any predictions concerning this stage of information processing. Research on cognitive biases in anxiety has thus far only addressed the two extremes of the processing continuum: attention and longer-term memory. In between, a gap remains, the bridging of which might bring us closer to defining the prerequisites of memory biases in anxiety. As empirical research has provided substantial and coherent knowledge concerning attention in anxiety, and as attention and VWM are so closely linked (see, for instance, Cowan, 1995), the thorough investigation of VWM may provide important clues for models of anxiety. Is anxiety related to VWM biases favoring the processing of threatening information, or does the avoidance presumed by cognitive models of anxiety already begin at this stage? RESEARCH AIMS. To investigate the relevance of biased VWM in anxiety, the present research focused in eight experiments on the following main research questions: (1) Is threat preferably stored in VWM in anxious individuals? (2) Does threat preference occur at the cost of the storage of other items, or is extra storage capacity provided? (3) Would the appearance of threat interrupt ongoing encoding of non-threatening items? (4) Does prioritized encoding of threat in anxiety occur strategically or automatically? (5) Are disorder-specific VWM biases also materials-specific? (6) Are VWM biases in anxiety modifiable through cognitive-behavioral therapy? METHODS. In Experiments 1-4, a spatial-sequential cueing paradigm was used. A subset of real-object display items was successively cued on each trial by a sudden change of the picture background for 150 ms each. After the cueing, one of the display pictures was hidden and probed for a memory test. On most trials, a cued item was tested, and memory accuracy was determined depending on the item’s position within the cue string and depending on its valence. In some cases, memory for an uncued item was tested. Experiment 1 and 2 were directed at discovering whether spider fearfuls and non-anxious controls would differ with respect to the accuracy in memorizing cued spiders and uncued spiders and, thus, reveal disorder-specific biases of VWM. In addition, the question whether the presence of a spider image is related to costs for the memorization of other images was tested. Experiment 3 addressed whether any disorder-specific VWM biases found earlier were specific to the feared spiders. Therefore, the critical stimuli here were a snake and a spider. Participants were spider fearfuls and non-anxious controls, both without snake anxiety. In Experiment 4, it was tested whether disorder-specific biases found in Experiment 1 and 2 were modifiable through cognitive-behavioral treatment. The critical stimulus was a spider image. Spider fearfuls were tested three times. Half of them received a cognitive-behavioral intervention after the first test, the other half only after the second test. In two additional experiments, VWM was assessed with a change-detection paradigm. The main aim was to clarify whether disorder-specific effects found in the previous experiments were associated with automatic or with strategic selective encoding of threatening materials, and whether any group differences in spider change detection were materials-specific to spiders, but not to snakes. In Experiment 5, several images were presented simultaneously in a study display for either 100 or 500 milliseconds. After a short interruption, a test display was presented including either the same items as the first one or one changed item. Participants’ accuracy in determining whether displays were the same or different was measured depending on the valence of the changed item, set size, and presentation time of the display. There were trials with and without spiders. If a change was made, it could involve either a non-spider or a spider item. Of specific interest was the condition in which a spider image was presented initially, but not in the test phase, as noticing this specific change would require storage of that image in VWM. Would group differences be particularly pronounced in the shorter encoding condition suggesting automatic encoding of threat, or would they occur in the longer encoding condition, suggesting strategic encoding of spiders? In Experiment 6, change detection accuracy for spiders vs. snakes was tested. The participants in both experiments were spider fearfuls vs. controls, but those of Experiment 6 were additionally required to lack snake anxiety. Moreover, a temporal VWM paradigm - an attentional blink task - was applied to assess whether a biased encoding of spider images in spider fearfuls would occur at the expense of non-threatening items undergoing concurrent processing, and whether this effect was specific to spiders, but not to snakes. Series of real-object pictures were presented at rates of 80 ms at the display center. The observer’s task was to identify and report the two target pictures indicated by a brighter background. In Experiment 7, the first target always depicted a neutral item. The valence of the second target was varied - either negative depicting a spider, positive, or neutral. Participants varied with respect to their spider anxiety. In Experiment 8, spider fearfuls and non-anxious controls, both without snake anxiety, were tested. The experiment was nearly the same as the previous one, but two negative target types were tested: disorder-relevant spiders and negative but not feared snakes. Of specific interest was whether the appearance of a threatening target would reduce the report probability of the earlier attended target, indicating the interruption of its VWM encoding in favor of the threat item. RESULTS. (1) Both anxious and non-anxious controls, showed VWM advantages for negative materials such as spider or snake images. (2) In addition, there were disorderspecific VWM biases: some effects were larger in spider fearfuls than in non-anxious controls and some effects occurred exclusively in spider fearfuls. (3) Group differences and, thus, disorder-specificity were particularly pronounced under competitive circumstances, that is, under the condition of numerous stimuli competing for processing resources: when only little orientation time was allowed, when only little time was provided for selecting and encoding items from a crowd, and when VWMfor the critical item required reflexive instead of voluntary attention. (4) Pronounced memory for task-relevant, voluntarily attended spiders was related to difficulties in disengaging attention from these items in the fearful group, reflected in reduced memory accuracy for the item following it. (5) Disorder-specific VWM biases seem to be based on attentional biases to threatening materials resulting in a very quick, automatic memory consolidation. However, this preferential encoding was not at the cost of neutral materials currently undergoing encoding processes. (6) All disorder-specific VWM biases occured only with fear-related materials, not with other negative materials. (7) Automatic and highly disorder-specific fear-related VWM biases – but not strategic VWM biases occuring in both groups - were modifiable through cognitive-behavioral intervention. CONCLUSIONS. This work provides additional information about informationprocessing distortions related to specific anxiety. With the experimental investigation of biased VWM, this work has been performed to fill a gap within research on cognitive biases in anxiety. Moreover, this dissertation contributes to cognitive theories of anxiety by proposing several recommendations for refinements of current theoretical approaches. Most important, it was suggested to extend existing models by a more detailed consideration of attention and memory. In view of numerous previous empirical studies on the topic and the conclusions of this dissertation, a differentiation of the attentional engagement and disengagement component appears inevitable. Even more important, in view of the data presented here predictions concerning VWM for threatening materials need to be taken into account. In addition, suggestions are provided for the differential consideration of biases occuring from prepotent threat value of negative stimuli vs. individual threat value. A proposal for a cognitive model of anxiety extended by all these aspects is provided to serve as an invitation of further research in the investigation of the nature of memory biases in anxiety disorders. REFERENCES: Cowan, N. (1995). Attention and Memory. An integrated framework.New York: Oxford University Press. Fox, E., Russo, R., & Dutton, K. (2002). Attentional bias for threat: Evidence for delayed disengagement from emotional faces. Cognition and Emotion, 16, 355-379. MacLeod, C., & Mathews, A. (2004). Selective memory effects in anxiety disorders: An overview of research findings and their implications. In D. Reisberg & P. Hertel (eds.), Memory and Emotion. Oxford: Oxford University Press. Mathews, A., & Mackintosh, B. (1998). A cognitive model of selective processing in anxiety. Cognitive Therapy and Research, 22 (6), 539-560. Mathews, A., & MacLeod, C. (2005). Cognitive vulnerability to emotional disorders. Annual Review of Clinical Psychology, 1, 167-195.Mathews, Mogg, May, & Eysenck (1989). Mogg, K., Bradley, B.P., Miles, F., & Dixon, R. (2004). Time course of attentional bias for threat scenes: Testing the vigilance avoidance hypothesis. Cognition and Emotion, 18(5), 689-700. Williams, J.M.G., Watts, F.N., MacLeod, C., & Mathews, A. (1997). Cognitive psychology and emotional disorders. Chichester: John Wiley.

This thesis compared the performance of a blind relay communication system with that of a non-blind relay communication system. For a blind system, a differential space-time block code (DSTBC) or a blind Viterbi detection were used, and for a non-blind system, a space-time block code (STBC) was used. Also, this thesis took another blind detection mechanism, the Zhang’s code, and compared its performance with that of the blind Viterbi detection under a Jakes’ fading environment. It was found that the case of a relay closer to the destination shows better performance than the other case of a relay closer to the destination. It was also observed that the blind Viterbi decoder shows better performance than the Zhang’s code under the Jakes’ fading environment.
Thesis (M.S.)--Wichita State University, College of Engineering, Dept. of Electrical Engineering and Computer Science.

Denna rapport är resultatet av ett examensarbete, på 20 högskolepoäng, som har utförts på Scania CV AB, Tekniskt centrum, Södertälje. Examensarbetet behandlar Side Blind Spot Detection och har resulterat i ett prototypsystem som detekterar objekt i döda vinklarna på sidorna av en lastbil. Systemet är ett aktivt säkerhetssystem som syftar till att förhindra olyckor och ge ökad trafiksäkerhet på vägarna.

Examensarbetet har varit tvådelat för att uppnå detta mål. Denna rapport behandlar främst val av sensorteknik för att upptäcka objekt i de döda vinklarna på sidan av en lastbil. Den behandlar också hårdvara till användargränssnittet samt installation av användargränssnittet och sensorerna i en lastbil. Den andra delen av examensarbetet har bestått av utformning av användargränssnittet och programmering av systemet. Detta kan läsas i rapporten "Side Blind Spot Detection - System och användargränssnitt" författad av Jenny Hedenberg och Hanna Torell, Chalmers Tekniska högskola, 2005.

I rapporten har sex olika sensortekniker studerats och utvärderats. De sex sensorteknikerna är ultraljud, passiv IR, lidar (aktiv IR), kamera, IR kamera och radar. Resultatet av utvärderingen visade att radar är den mest lämpade sensortekniken för den här typen av applikationer och det är därför radar används som sensor i prototypsystemet.

Systemet har tre olika lägen beroende på hur mycket information föraren önskar få när ett objekt befinner sig i döda vinkeln vilket styrs av en systemknapp. Föraren får informationen från användargränssnittet som består av två LED- displayer som är placerade i dörrkarmarna på vardera sida. Förutom att visa varningarna visuellt i LED-displayen används också ljud vid varning. Resultatet blev som förväntat och visar de funktioner som användargränssnittet har på ett bra sätt. Vad gäller resultatet av hela prototypsystemet så visar det att radar är ett bra val för denna applikation för att den klarar av de krav som ställs. Tester av systemet visar dock att den införskaffade radarn har begränsningar som försvårar filtreringen. Detta leder till att systemet inte är helt tillförlitligt eftersom systemet ibland missar objekt och ibland felvarnar för objekt som inte finns eller för objekt som inte är relevanta.

Multicarrier transmission technology has shown tremendous potential in realizing high data rates for next generation broadband wireless communication systems. Multicarrier modulation schemes are robust to frequency selective fading which are inherent in broadband wireless channels. This thesis considers blind multiuser detection for the recently proposed time frequency spread multicarrier CDMA (TF-MC-CDMA) system in which the system uses both time and frequency domain spreading to achieve diversity in both time and frequency domains. The challenge for TF-MC-CDMA multiuser detection is to mitigate multiple access interference (MAI) in both the time and the frequency domain. The objective of this thesis is to develop and analyze the performance of blind multiuser detection algorithms for TF-MC-CDMA for three types of channel models: AWGN with MAI, slowly fading downlink Rayleigh multipath channels, and the slowly fading downlink multipath channels with Doppler shift induced intercarrier interference (ICI). A new suboptimal decoupling technique for blind multiuser detection for TF-MCCDMA in AWGN with MAI is proposed. It is found that the original TF-MC-CDMA blind multiuser detection problem can be suboptimally decoupled into two blind multiuser DS-CDMA problems. These two problems can be solved separately using blind DS-CDMA multiuser techniques. Our effort focused on using blind linear multiuser detectors in which we investigated into four types of blind detection methods: blind direct matrix inversion (DMI) method, blind CMOE RLS method, blind subspace multiuser detection method using SVD and PASTd adaptive subspace tracking. The suboptimal decoupling technique for blind multiuser detection for TF-MC-CDMA is extended to slowly fading downlink Rayleigh multipath channels known as type I detectors. Computer simulation results show that type I detectors do not work well in slowly fading multipath channels even though such a scheme provides very good performance in AWGN with MAI. In slowly fading channels, the orthogonality of the time domain signature sequences is preserved. We propose a type II detector which uses a cascade implementation with the time domain detection output acts as the input of the frequency domain detection. Computer simulations show that type II detectors provide much better performance than type I detectors. Blind multiuser detection for TF-MC-CDMA is further extended for slowly fading Rayleigh multipath channels with Doppler shift induced ICI. In mobile channels, Doppler shifts combined with multipath effects create random subcarrier frequency shifts which in turn cause subcarrier frequency mismatch. Such mismatch leads to the loss of orthogonality among subcarriers thus creating ICI. Our analysis shows that Doppler shifts induced ICI has the effect of destroying the common-channel property in downlink channels. The downlink TF-MC-CDMA signal becomes a quasi-uplink signal because of user dependent subchannel gains. The type II detector for slowly fading Rayleigh multipath channels is further extended to apply in such channels with Doppler shift induced ICI. It is shown through analysis and computer simulations that the proposed type II detectors implemented using CMOE RLS algorithm, blind subspace SVD algorithm, and the blind subspace PASTd adaptive subspace tracking algorithm, without modifications, provide robust performance in multipath channels with Doppler shift induced ICI.
Applied Science, Faculty of
Electrical and Computer Engineering, Department of
Graduate

Thermal neutron detectors, which are based on semiconductor material such as silicon coated with neutron reactive material like 10B and 6Li have been discussed for many decades. The performance of the thermal neutron detector system, GAMBE, which is based on two silicon sensors in a sandwich configuration is investigated. The results show that a single sandwich design with 6LiF film of (1.5 ± 0.6) mg/cm2 thick can achieve a total ("tn) and a coincidence ("cn) detection efficiency of 4% and 1% respectively. While, 6Li foil of (40 ± 10) μm thick is able to attain a ("cn) of (1.5 ± 0.9)% and a ("tn) of (9.2 ± 1.4)%. The coincidence that defines a true neutron hit is the simultaneous signal recorded by the two sensors facing the conversion layer. These coincidences provide a very good method for rejecting spurious hits coming from gamma-rays, which are usually present in the neutron field under measurement. This methodology results in a high gamma-ray rejection factor of 108. However, the price to pay is a reduction of the detection efficiency of the single sandwich detector. The thermal neutron detection efficiency of the detector is enhanced by using a stacked detector configuration and highdensity polyethylene (HDPE) sheets, as neutron moderators and reflectors. The GAMBE detector is positioned inside a box of HDPE with a lead window in the direction of the neutron flux for neutron moderation and a reduction of the effect of gamma-rays on the detector. The experimental layout was modeled in MCNP4C to investigate the contribution of HDPE to the thermal neutron flux (n/s/cm2). In this research, a stack of 4 silicon semiconductor sensors with two 6LiF films of an average thickness of (2.8 ± 0.6) mg/cm2 in a configuration of two sandwiches is shown to achieve a total and a coincidence detection efficiency of (27 ± 3)% and (4 ± 1)% respectively. This represents a significant improvement compared to a single detector. The effect of these stacked detectors for the development of a handheld thermal neutron detector, using 4 coated Si detectors is shown to have a 22% efficiency. Finally, this information is used to inform the optimised design of the handheld detector. The results based on GEANT4 and MCNP4C simulations indicate that the total detection efficiency of this portable detector with a stack of 7 sandwich detectors will increase up to 52% by using an optimal thickness of a 6LiF film of 17 μm (3.95 mg/cm2). This handheld detector has a highest total detection efficiency of 69% when using a 6Li foil of 36 μm thick.

In practice, channel coefficient information may not be available. Hence, a blind detection may be necessary. In addition, the direct link from a source to a destination may not be available. Therefore, information may have to flow through a relay, which is called a two-hop relay network. This thesis analyses the performance of the two-hop relay network by applying a blind detection assuming a decode-and-forward relay.
Thesis (M.S.)--Wichita State University, College of Engineering, Dept. of Electrical Engineering and Computer Science

This thesis considers a two-way relay system whereby two nodes exchange information via a relay node. Amplify-and-forward (AF) is applied as a relaying strategy.The channel information is not available at both receivers. The generalized likelihood ratio test (GLRT) and least squares error (LSE) detectors are derived in order to achieve blind unique recovery of the transmitted symbol. It is also proven that full diversity can be achieved. A constellation selection algorithm is designed to optimize the symbol error rate (SER) performance. This thesis also analyzes both theoretical procedures and simulation results of an existing differential detection scheme for two-way relay channels. Monte Carlo simulations demonstrate the performance of the proposed algorithm.
Thesis (M.S.)--Wichita State University, College of Engineering, Dept. of Electrical Engineering and Computer Science.

CONSELHO NACIONAL DE DESENVOLVIMENTO CIENTÍFICO E TECNOLÓGICO
O esquema OFDM (Orthogonal Frequency Division Multiplexing) é um esquema de transmissão de sinais por multiplexação em freqüência e sua adoção vem mostrando-se bastante promissora no que diz respeito à transmissão de sinais em canais seletivos em freqüência. Sua relativa robustez frente a canais com múltiplos percursos é conseguida efetuando-se a transmissão paralela de dados em subportadoras ortogonais estreitas. Os receptores OFDM, de uma forma geral, necessitam de estimativas do canal de comunicação para realizar a detecção coerente dos sinais recebidos. Estimativas do canal podem ser obtidas de forma assistida por meio da utilização de pilotos que consomem preciosa banda, ou de forma cega, valendo-se apenas do conhecimento de características estatísticas dos sinais transmitidos. Esta tese segue a linha de estimação cega de canal. São propostos dois estimadores cegos de canal, para sistemas OFDM com intervalo de guarda ZP ou CP. O primeiro baseia-se no casamento de correlação, a saber: a estimativa do canal corresponde ao argumento que minimiza a norma quadrática de Frobenius da diferença entre a matriz correlação teórica parametrizada dos sinais recebidos e uma estimativa desta matriz obtida por meio das observações no receptor. O segundo estimador é desenvolvido a partir de uma modificação do método de identificação de canal por meio da estimação do subespaço de ruído, lançando mão de uma estimativa para o operador projeção no subespaço de ruído. Técnicas para resolver a ambigüidade complexa presente nas estimativas do canal que resultam dos métodos propostos são abordadas. Os estimadores propostos são estendidos para sistemas OFDM ditos com intervalo de guarda insuficiente. Resultados de simulações ilustram o desempenho, tanto em termos de erro quadrático médio dos estimadores, quanto em taxa de erro de bit dos sistemas.
Orthogonal Frequency Division Multiplexing is a multi-carrier transmission technique suitable for wireless communications through frequency selective channels, thus making it an appealing scheme for next-generation applications demanding high data rates. Its robustness with respect to multipath channels is obtained by modulating a set of closely-spaced orthogonal sub-carriers. In order to coherently detect the received signals, Channel State Information must be available to OFDM receivers. Supervised channel estimation is achieved by multiplexing known pilot symbols and data symbols, thus reducing effective system throughput. On the other hand, unsupervised or blind estimation techniques rely solely on the knowledge of statistical characteristics of the transmitted signal in order to identify the channel. This thesis proposes two blind channel estimators, for OFDM systems with CP or ZP guard interval. The first estimator is based on correlation matching: a channel estimate is obtained as the argument which minimizes the Frobenius quadratic norm of the difference between the parameterized theoretical correlation matrix of the transmitted signals and an estimate of that matrix obtained by means of observations at the receiver. The second estimator is obtained by modifying the noise subspace based channel estimator and using an estimate for the noise subspace projector which relies on powers of the inverse correlation matrix. Techniques to eliminate the inherent complex ambiguity derived from the proposed blind channel estimators are addressed. The porposed estimators’ formulation is also extended to the case of OFDM systems with insufficient guard intervals. Simulation results depict Mean Square Error of the proposed estimators, as well as Bit Error Rate performance of systems using those estimators in various scenarios.

Dans cette thèse nous étudions la détection aveugle de sources pour des mélanges linéaires à un seul capteur et dans des situations dominées par l'interférence. Nos hypothèses de travail portent uniquement sur l'indépendance des sources et le caractère fini et uniforme de leur distribution. Centrant notre étude autour d'un algorithme itératif de la littérature, plusieurs méthodes d'initialisation sont proposées afin d'en améliorer la robustesse quant aux cas les plus défavorables. En particulier, la relation entre les minima locaux vers lesquels les algorithmes considérés peuvent converger et la géométrie des constellations sous-jacentes aux mélanges est étudiée. Plusieurs méthodes visant à intégrer cette information au processus de détection global sont analysées. Les résultats obtenus par simulation sur des configurations de mélange variées attestent de l'efficacité des stratégies proposées via à vis des algorithmes de détection aveugle traditionnels. Une extension de notre méthode à des canaux sélectifs en fréquence et pour des transmissions multi-porteuses est également effectuée, conduisant à de nouvelles propositions d'initialisation. Notre étude se conclut par des considérations plus générales quant à l'aptitude des algorithmes de classification à pouvoir discerner des ensembles de points exhibant une forte proximité
In this thesis we address the blind channel estimation and source detection of linear mixtures with a single sensor in scenarios strongly dominated by interference. In this framework our only assumptions relate to the sources mutual independence, as well as the discrete and uniform nature of their probability distribution. Based on existing iterative algorithms from the literature, we propose several initialization strategies so as to enhance both their overall performance and robustness to highly unfavorable mixture configurations. We provide a detailed analysis of the relation between the spurious fixed points these algorithms are known to possibly converge to and the underlying mixtures geometry. Possible strategies to account for this additional information in the overall detection process are discussed as well. Simulation results attest to a significant improvement of the achieved error rates compared to all tested traditional detection schemes. An extension of the method to the estimation of frequency-selective channels in multiuser and orthogonal multicarrier transmissions is then performed, along with several initialization propositions. We conclude our study by more general considerations on clustering algorithms and their ability to discriminate between partially entangled data classes

For a blind swimmer to be able to exercise in a swimming pool today, human assistance is required to notifying the swimmer in due time when and if she is too close to the edge. The purpose of this thesis was to see whether or not it is possible to replace and even improve the human intervention using a warning system built around a tablet mounted next to the pool edge. A secondary goal was to evaluate how suitable a tablet pc is for robotic applications. The system proposed utilizes a tablet’s built-in frontal facing camera, OpenCV as vision library, FM modules for the wireless warning system and is intended for the Android environment. Videos of the real scenario have been analyzed on a computer using OpenCV and a detection algorithm searching for the swimmer’s red swimming cap has been developed. Next, the algorithm was implemented on an Android tablet. The result obtained shows that it is perfectly possible to use a cheap tablet to accurately detect and notify the blind swimmer in due time when she is too close to the edge. If calibrated thoroughly, the likelihood of a missed detection is actually lower with this system as compared to human intervention, as humans can only warn the swimmer when she is above the water.

COORDENAÇÃO DE APERFEIÇOAMENTO DO PESSOAL DE ENSINO SUPERIOR
Esta tese apresenta novas estruturas e algoritmos adaptativos para detecção às cegas de sinais DS-CDMA. São investigados receptores cegos com restrições lineares baseados nas funções custo de mínima variância (CMV) e módulo constante (CCM). Algoritmos adaptativos do tipo Affine-Projection para estimação dos parâmetros do receptor são desenvolvidos e seu desempenho em estado estacionário é analisado. Também são apresentados algoritmos adaptativos para estimação às cegas do canal de comunicações. Em seguida, novas estruturas de canceladores de interferência são propostas. Primeiramente um cancelador de interferência paralelo (PIC) linear baseado na função custo CCM é proposto. Em seguida é desenvolvido um novo esquema não-supervisionado de cancelamento sucessivo de interferência (SIC), baseado no conceito de arbitragem paralela. Por fim, é apresentado um esquema híbrido (HIC) que combina a estrutura SIC com uma estrutura multi- estágio, resultando em melhores estimativas para detecção e desempenho uniforme para os usuários do sistema.
This thesis presents new structures and adaptive algorithms for blind detection of DS-CDMA signals. Linearly constrained minimum variance (CMV) and constant modulus (CCM) receivers are investigated. Blind adaptive Affine- Projection like algorithms for receiver parameter estimation are derived and its steady-state performance is analyzed. Blind adaptive channel estimation algorithms are also presented. This work also proposes new interference cancellation structures. Firstly, a blind linear parallel interference canceller (PIC) based on the CCM cost function is proposed. Secondly, a new non-supervised serial interference canceller (SIC) based on the parallel arbitration concept is developed. Finally, an hybrid interference cancellation scheme (HIC) which combines SIC and multiple PIC stages is presented.

This study hypothesizes a greater ability by blind subjects in detecting deceptive communication from an audio channel only. Accuracy and confidence levels for the blind were compared with normally sighted undergraduate students' results. All subjects were requested to indicate their perception on several audible cues, including: speech errors, pauses, vocal segregates, response duration, vocal certainty, vocal tension, vocal pleasantness, speaking volume and rate. Subjects also indicated whether they thought the messages on the stimulus tapes were deceptive or truthful. Stimulus tapes containing deceptive statements were created by inducing a cheating incident. Undergraduate students in a lower-level communication course participated in a dot estimation task where they either performed on their own abilities or cheated with a confederate. Interviews immediately following the procedure resulted in deceptive responses from all subjects induced into cheating. A discussion of cheating as a methodology is presented in the final chapter. Results from this study indicate that blind participants tended to be more accurate at detecting deceptive communication than sighted subjects. Although vocal cues were rated similarly by both groups, the greater detection accuracy by the blind suggests sensory compensation may occur as a result of blindness. The final chapter suggests that with better measurement of audible cues used by the blind, future research may discover much about the importance of these deceptive cues.

Wide bandgap semiconductors have been broadly investigated for their potential to detect and emit high energy ultraviolet (UV) photons. Advancements in deep UV optoelectronic materials would enable the efficient and affordable realization of many medical, industrial and consumer UV optical devices. The traditional growth method, vacuum deposition, is an extremely complicated and expensive process. Sol-gel processing dramatically simplifies facility requirements and can be scaled to industrial size. The work presented here involves a novel study of the ternary wide bandgap material Ni1-xMgxO. Films were developed by sol-gel spin coating for investigation of material and electrical properties. This method produced films 200-600 nm thick with surface roughness below 4 nm RMS. Sintered films indicated an improvement from 60% to 90% transmission near the band edge. Additionally, compositional analysis was performed by X-ray Photoelectron Spectroscopy and film defects were characterized by photoluminescence using a continuous wave He-Cd UV laser, revealing the expected oxygen defect at 413nm. This film growth technique has produced thin polycrystalline films with low surface roughness and a high degree of crystalline orientation; crucial characteristics for semiconductor devices. These films have demonstrated the ability to be tuned over the full compositional range from the bandgap of NiO (3.6 eV) to that of MgO (7.8 eV). Optoelectronic devices produced by standard photolithographic techniques are discussed as well as the electrical transport properties of their metal contacts. Based on initial results, these films have demonstrated strong potential as solar blind detectors of UV radiation.
B.S.E.E.
Bachelors
Engineering and Computer Science
Electrical Engineering

'It is not a lack of spectrum. It is an issue of efficient use of the available spectrum"--conclusions of the FCC Spectrum Policy Task Force. There is growing interest towards providing broadband communication with high bit rates and throughput, especially in the ISM band, as it was an ignition of innovation triggered by the FCC to provide, to some extent, a regulation-free band that anyone can use. But with such freedom comes the risk of interference and more responsibility to avoid causing it. Therefore, the need for accurate interference detection and identification, along with good blind detection capabilities are inevitable. Since cognitive radio is being adopted widely as more researchers consider it the ultimate solution for efficient spectrum sharing [1], it is reasonable to study the cognitive radio in the ISM band [2]. Many indications show that the ISM band will have less regulation in the future, and some even predict that the ISM may be completely regulation free [3]. In the dawn of cognitive radio, more knowledge about possible interfering signals should play a major role in determining optimal transmitter configurations. Since signal identification and interference will be the core concerns [4], [5], we will describe a novel approach for a cognitive radio spectrum sensing engine, which will be essential to design more efficient ISM band transceivers. In this thesis we propose a novel spectrum awareness engine to be integrated in the cognitive radios. Furthermore, the proposed engine is specialized for the ISM band, assuming that it can be one of the most challenging bands due to its free-to-use approach. It is shown that characterization of the interfering signals will help with overcoming their effects. This knowledge is invaluable to help choose the best configuration for the transceivers and will help to support the efforts of the coexistence attempts between wireless devices in such bands.

The advent of spectrum sharing has increased the need for robust interference rejection methods. The Citizens Broadband Radio Service (CBRS) band is soon to be occupied by LTE waveforms and License Assisted Access (LAA) will have LTE signals coexisting with other signals in the 5 GHz band. In anticipation of this need, we present a method for interference rejection of cyclostationary signals, which can also help avoid interference through better detection of low power co-channel signals. The method proposed in this thesis consists of a frequency-shift (FRESH) filter which acts as a whitening filter, canceling the interference by exploiting its cyclostationarity. It learns the cyclostationary characteristics of the interferer blindly, through a property restoration algorithm which aims to drive the spectrum to white noise. The property restoration algorithm, inspired by the constant modulus algorithm (CMA), is applied to each frequency bin to determine the optimal coefficients for the proposed CMA FRESH whitening filter (CFW). The performance of the CFW in interference rejection is compared to a time-invariant version, and proposed use cases are analyzed. The use cases consist of the rejection of a high powered, wider bandwidth interferer which is masking the signal-of-interest (SOI). The interferer is rejected blindly, with no knowledge of its characteristics. We analyzed signal detection performance in the case that the SOI is another user with much lower power, for multiple types of SOIs ranging from BPSK to OFDM. We also deal with the case that the SOI is to be received and demodulated; we recover it and compare resulting bit error rates to state of the art FRESH filters. The results show significantly better signal detection and recovery.
Master of Science
Wireless communication is complicated by the fact that multiple radios may be attempting to transmit at the same frequency, time and location concurrently. This scenario may be a due to malicious intent by certain radios (jamming), or mere confusion due to a lack of knowledge that another radio is transmitting in the same channel. The latter scenario is more common due to congested wireless spectrum, as the number of devices increases exponentially. In either case, interference results. We present a novel interference rejection method in this work, one that is blind to the properties of the interferer and adapts to cancel it. It follows the philosophy of property restoration as extolled by the constant modulus algorithm (CMA) and is a frequency shift (FRESH) filter, hence the name. The process of restoring the wireless spectrum to white noise is what makes it a whitening filter, and is also how it adapts to cancel interference. Such a filter has myriad possible uses, and we examine the use case of rejecting interference to detect or recover the signal-of-interest (SOI) that we are attempting to receive. We present performance results in both cases and compare with conventional time-invariant filters and state of the art FRESH filters.

There has been a recent explosive growth in mobile data consumption. This, in turn, imposes many challenges for mobile services providers and regulators in many aspects. One of these primary challenges is maintaining the radio spectrum to handle the current and upcoming expansion in mobile data traffic. In this regard, a radio spectrum regulatory framework based on secondary spectrum access is proposed as one of the solutions for the next generation wireless networks. In secondary spectrum access framework, secondary (unlicensed) systems coexist with primary (licensed) systems and access the spectrum on an opportunistic base. In this thesis, aspects related to finding the free of use spectrum portions - called spectrum opportunities - are treated. One way to find these opportunities is spectrum sensing which is considered as an enabler of opportunistic spectrum access. In particular, this thesis investigates some topics in blind spectrum sensing where no priori knowledge about the possible co-existing systems is available. As a standalone contribution in blind spectrum sensing arena, a new blind sensing technique is developed in this thesis. The technique is based on discriminant analysis statistical framework and called spectrum discriminator (SD). A comparative study between the SD and some existing blind sensing techniques was carried out and showed a reliable performance of the SD. The thesis also contributes by exploring sensing parameters optimization for two existing techniques, namely, energy detector (ED) and maximum-minimum eigenvalue detector (MME). For ED, the sensing time and periodic sensing interval are optimized to achieve as high detection accuracy as possible. Moreover, a study of sensing parameters optimization in a real-life coexisting scenario, that is, LTE cognitive femto-cells, is carried out with an objective of maximizing cognitive femto-cells throughput. In association with this work, an empirical statistical model for LTE channel occupancy is accomplished. The empirical model fits the channels' active and idle periods distributions to a linear combination of multiple exponential distributions. For the MME, a novel solution for the filtering problem is introduced. This solution is based on frequency domain rectangular filtering. Furthermore, an optimization of the observation bandwidth for MME with respect to the signal bandwidth is analytically performed and verified by simulations. After optimizing the parameters for both ED and MME, a two-stage fully-blind self-adapted sensing algorithm composed of ED and MME is introduced. The combined detector is found to outperform both detectors individually in terms of detection accuracy with an average complexity lies in between the complexities of the two detectors. The combined detector is tested with measured TV and wireless microphone signals. The performance evaluation in the different parts of the thesis is done through measurements and/or simulations. Active measurements were performed for sensing performance evaluation. Passive measurements on the other hand were used for LTE downlink channels occupancy modeling and to capture TV and wireless microphone signals.

QC 20150209

碩士
國防大學
網路安全碩士班
107
Eye blink detection can be applied to driver fatigue detection, disease diagnosis, face anti-spoofing and etc.. With the advancement of image analysis technology, the facial landmark technique which support real-time analysis have been proposed in recent years. The EAR (Eye Aspect Ratio) method is a representative eye blink detection method based on facial landmark by observing the change of the eye aspect ratio while blinking. In this paper, we proposed the EBF (Eye Blink Factor) method based on facial landmark technique to capture the feature points coordinates around the eyes. The multi-layer perceptron (MLP) composited with various hidden layers, neurons and activation functions is used as the classifier for training and testing phases in the experiments. Experimental results shows the proposed EBF method with appropriate MLP structure can effectively detect eye blinks.

碩士
國立交通大學
資訊科學系所
93
The information about eye blinks shows one’s fatigue. Detecting blink is useful to monitoring system or warning system. Automatically detecting fatigue of drivers may save them from accidents. We propose a fast algorithm for eye location and eye blink detection. This algorithm is adapted to the complex background and the situation which people wear glasses. In the first place, we classify pixels as skin color and non-skin color to build a skin mask of original image. Second, we patch the skin mask with morphological operations and locate the face bounding block. Applying horizontal projection to the bounding block and filtering with face feature conditions, we locate eye region. After gathering statistics of pixels in eye region, we can find out which frames contain eye blinks.

碩士
國立臺中科技大學
多媒體設計系碩士班
100
Our study is focus on eye detection, identification, tracking and locating. All of study topics are popular and widely varied from human computer interaction in nowadays. Human eye research’s devices are already commercially available, however, the cost of these devices and systems are usually expensive. Finding a low-cost architecture to replace existence high-cost device is an achievable way. In hardware aspect, we could use low-cost web camera with visible light to accomplish. Therefore, in software aspect, the system has to possess efficiency and precision. We proposed an eye detection process and improved approach of eye identification. In addition, we also improve the a method of eyeball centre location. First, We used the action of eyes blinking; then we analyzed relative eye behaviors and features information. The second is reformed the computation of eye positioning algorithm. To seek out the eyeball’s best centre location integrates with data values and positions. The evaluation referred to two papers that belonged to human eye technology and published in 2010 and 2011. One is eye state identification and the other is eyeball centre location. We evaluated and compared with them. Then, the result revealed that our approach is more accurate in any kind of appraisal datum. Moreover, our result data are both higher than two papers that we mentioned.

碩士
國立彰化師範大學
車輛科技研究所
105
Vehicle safety equipment has become one of the important components of modern vehicle design. Meanwhile, active safety equipment is the mainstream in recent years. In this study, an active safety system that is combined with a brain wave and eye blinking detection technology is proposed. The architecture of this system is divided into two parts: the first part is the brain fatigue warning technology, and the second part is the eye blinking detection technology. In the system implementation, the brainwave electroencephalogram (EEG) fatigue warning device is developed on the ThinkGear system, where the signal can be divided into attention (concentration) and meditation (relaxation). According to the brainwave eSense indication defines attention and meditation to do their judgment, the situation is transmitted using HC-05 Bluetooth to the Arduino Control Panel. For the blinking detection technology a webcam is proposed to capture images. The system is developed by MATLAB programing and the Open Source Computer Vision Library (Open CV). The image from the webcam is processed using Haar facial features and Adaptive Boosting (AdaBoost) machine learning methods to filter the correct facial features. After the images are imported into the histogram of oriented gradients (HOG), it can identify if the eyes are open or closed. Integrating the two parts of the system, the alerting system will sound an alert to remind the driver if their eyes are closed for more than two seconds or if their brainwave concentration is lower than the default value. Experimental results indicate that both the brainwave and eye blink detection systems are useful in driving safety. This active system development costs are very low and effectively reduce the incidence of fatigue driving.

碩士
國立臺灣科技大學
自動化及控制研究所
93
In many applications of human-computer interfaces, the eye location and tracking techniques play a quite important role; for example, gaze-based communication and control, iris-based biometric recognition, and so on. In addition to this, the incorporation of blinking detection has made such human-computer interfaces more diversified like the eye-control human computer interface, computer-user fatigue detection, and driver fatigue detection in the intelligent traffic system. In this thesis, we have developed a non-contact eye’s blinking detection system that uses a video camera on the top of a computer monitor to capture the image of the user as well as the environment in real time. This system consists of three main processing phases: face detection, eye location and tracking, and blinking detection. Regarding the face detection, a light compensation technique is first employed to diminish the influence of light sources on color images. After color model converting, the hue component is adopted to extract skin colors and build up a skin binary image. Mathematical morphological operations and connected component labeling methods are further exploited to find candidate face regions which are verified by the priori facial feature information. In the eye location and tracking phase, a support vector machine is applied to searching eye regions which provide geometric information for eye tracking by a particle filter in image sequences. As to the blinking detection, the color histograms of eye regions are used to distinguish opening eyes from closing ones. The experimental results reveal that our developed system can trace eyes and detect the blinking effectively.

碩士
中華大學
生物資訊學系(所)
97
In this study, we developed a wisdom system that can actively remind the computer user to let his/her eyes take a break. This system uses a PC camera to capture the image of the user and the background. It applies recognition technology to validate the user. At the same time, the system uses a built-in timer to record how long the user has worked on the computer. When the user is using the computer, the system not only detects the visual distance but also count the eye’s blinks by the facial feature. If the user doesn’t keep a proper visual distance from the monitor of the computer, his/her eyes’ blinks too often or too less, his/her has used the computer for a long time, illumination is too bright or too dark according to the background lighting behind the user, or the user uses computer not within a proper time, the system will pop up a warning message via words /voice to inform the user. Therefore, this system can initiatively be used to protect computer user’s eyes from harm. This system consists of three parts: the face detection, the face recognition, the visual distance detection and the blink detection technology. In the face detection part, a PC camera is used to capture the color image, and then the environment noise is removed by doing connected component labeling and morphology process to find human face. In the face recognition part, we use Principal Component Analysis technology to extract the feature of human face, and then use Euclidean distance technology to validate the user. In the visual distance detection part, Ellipse mask method is used for obtaining face information and detecting user visual distance from the monitor of computer. In the blink detection part, set the eyes region to count the blinks frequency to estimate the user weary degree.

碩士
國立雲林科技大學
電機工程系
102
Conventional face recognition authentication system is unable to discriminate real-world human faces from faces in the photo or in the video, and is easily faked by faces in the photo or in the video. So live face recognition authentication system with accurate live detection is getting fundamental and crucial. In order to accomplish real-time live face recognition authentication system, live detection based on blink detection and some extra simple live motion analysis is a cost-effective and high-practicability auxiliary method, especially the blink detection is the most intuitive and noninvasive. 　　This thesis proposes Parallel-AdaBoost blink detection to perform AdaBoost eyes detection stage and AdaBoost open-eyes detection stage simultaneously through multithreaded parallelism. That is, AdaBoost open-eyes detection works on the region of interest of eyes candidate detected by AdaBoost eyes detection simultaneously and iteratively. If the detection result of AdaBoost open-eyes detection is true, the eyes are open. If otherwise, the eyes are half-closed or completely-closed. Live blink motion is determined by two alternatively-happened detection results of AdaBoost open-eyes detection. 　　Experimental results show the detection accuracy and execution speed of proposed Parallel-AdaBoost blink detection are better than conventional ones. The proposed Parallel-AdaBoost blink detection can prevent face recognition authentication system from being broken by faces in the photo, and some extra simple live motion request and analysis can be integrated to further prevent face recognition authentication system from being cheated by faces in the video. 　　On the other hand, on Android embedded platform, this thesis implements and integrates the proposed Parallel-AdaBoost blink detection to accomplish “Portable 4-Factor Authentication Device Accelerated By Multi-Thread Pipeline Parallelism” for applications of mobile payment, mobile authentication, and mobile locker.

Σκοπός της παρούσας Διπλωματικής Εργασίας είναι η ανάπτυξη και εφαρμογή τεχνικών ψηφιακής επεξεργασίας εικόνων για τον εντοπισμό του οφθαλμού και τον υπολογισμό συγκεκριμένων παραμέτρων που συνδέονται με την κατάσταση του χρήστη. Συγκεκριμένα, χρησιμοποιήθηκε ένα ολοκληρωμένο Σύστημα Εντοπισμού Οφθαλμού που περιλαμβάνει τα υποσυστήματα της CMOS κάμερα, της μεταφοράς δεδομένων – εικόνων, της ψηφιοποίησης των δεδομένων, και τέλος το υποσύστημα της επεξεργασίας εικόνων οφθαλμού και του υπολογισμού παραμέτρων. Στα πλαίσια του τελευταίου αυτού υποσυστήματος αναπτύχθηκαν δύο μεθοδολογίες που βασίστηκαν στην εφαρμογή αλγορίθμων ψηφιακής επεξεργασίας εικόνων. Η πρώτη μεθοδολογία βασίστηκε στον υπολογισμό της μέσης φωτεινότητας για την άνω και την κάτω περιοχή του οφθαλμού. Η χρονική μεταβολή των δύο τιμών της φωτεινότητας χρησιμοποιήθηκε για την εξαγωγή πληροφοριών για την κατάσταση του οφθαλμού (ανοιχτός ή κλειστός). Η δεύτερη μεθοδολογία στηρίχτηκε σε ένα συνδυασμό τεχνικών ψηφιακής επεξεργασίας εικόνων. Η επεξεργασία κάθε εικόνας της ακολουθίας video περιλαμβάνει τέσσερα βασικά βήματα: (α) ευθυγράμμιση της εικόνας σε σχέση με ένα κοινό σύστημα αναφοράς, (β) εφαρμογή δύο φίλτρων για την ανίχνευση των κορυφών και των κοιλάδων της εικόνας, (γ) σύντηξη των δύο φιλτραρισμένων εικόνων που προκύπτουν και (δ) μετατροπή της εικόνας σύντηξης σε δυαδική με εφαρμογή κατάλληλου κατωφλίου. Η καταμέτρηση των λευκών εικονοστοιχείων της δυαδικής εικόνας στην περιοχή του οφθαλμού καθορίζει την κατάσταση του οφθαλμού (ανοικτός ή κλειστός). Τέλος, και μέσω του λογισμικού, υπολογίζονται οι σχετικές παράμετροι της κατάστασης του οφθαλμού όπως ο αριθμός ανοιγο-κλεισίματος οφθαλμού, η διάρκεια κάθε ανοιγο-κλεισίματος οφθαλμού και οι χρονικές αποστάσεις μεταξύ των προσδιορισμένων ανοιγο-κλεισιμάτων σε μια αλληλουχία συλλεγμένων εικόνων.
The scope of the thesis was the development and application of digital image processing techniques in order to detect human eye in video sequences and determine parameters related to the user’s state. Specifically, an integrated Eye-Tracking System was used in order to obtain the necessary image frames for further processing. The System consists of four modules, the CMOS camera module, the transfer module, the digitization module and the software module. The software module was based on the application of image processing techniques to detect the eye and calculate specific parameters. Two image processing techniques were developed and tested throughout this thesis. The first method was based on the calculations of the mean brightness of the upper and lower eye region for each frame of the video sequence. The temporal variation of this mean value provided useful information for the eye state (open/closed). The second method was based on a combination of various image processing techniques. The processing of each video frame comprises of four basic steps: a) registration of the image in relation to the first frame of the video sequence, b) filtering in order to detect the peaks and valleys of the image being processed, c) fusion of the filtered images, and d) binarization of the fused image by thresholding. The calculation of the number of white pixels in the eye region of the binary image indicates the state of the eye (open/closed) and allows the determination of the blink parameters related to the user’s state (vigilance/somnolence). The parameters being measured throughout this thesis were the number of eye blinks, the blink duration and the blink interval.

碩士
國立中正大學
電機工程研究所
88
In this thesis, two different blind detectors, the complementary subspace-based blind LDD and compressed minimum variance (CMV) blind MMSE detector, are proposed. It is shown that both of these schemes can be adopted with the prior knowledge of only the desired user’s signature waveform and its timing. The former is based on complementary subspace concept. Our approach here uses a geometric operator that generates the orthogonal complement of the interfering subspace to cancel the interference. By estimating the orthogonal complement of the interfering subspace, the LDD can be obtained blindly. The latter is based on a compressed notion and the minimum mean-square-error (MMSE) concept to find the close form of the blind detector. Compared with the previous subspace-based blind multiuser detection scheme, it is seen that both of the proposed methods have lower computational complexity. Furthermore, they also have better resistance against the orthogonality error of the estimated eigenspace.

碩士
國立成功大學
電腦與通信工程研究所
100
For wireless communications, the orthogonal-frequency division multiplexing (OFDM) system is frequently applied due to its robustness to the multipath environment and its low-complexity frequency-domain equalization. In order to increase the data throughput, we can use the blind method instead of the pilot-assisted method for the signal detection in OFDM systems. In this thesis, we propose a blind detection algorithm for the OFDM system with normalized symbol amplitudes. We use channel impulse response (CIR) to model the channel response (CR) variation across sub-channels. Then we apply the branch-and-bound principle to solve the non-linear integer programming problem. The proposed blind detection algorithm can effectively reduce the number of visiting nodes. We also use a polynomial to model the time variation of CRs in every sub-channel, such that we can use the predicted CRs to obtain a better initial data sequence in the blind detection algorithm. The better data sequence can obviously reduce the number of visiting nodes in the tree search. Simulation results validate that the proposed blind detection algorithm can reduce the complexity while maintain the same error rates.

Signal processing techniques such as multi user detection (MUD) have the capability of greatly enhancing the performance and capacity of future generation wireless communications systems. Blind adaptive MUD's have many favourable qualities and their application to OS-COMA systems has attracted a lot of attention. The constant modulus algorithm is widely deployed in blind channel equalizations applications. The central premise of this thesis is that the constant modulus cost function is very suitable for the purposes of blind adaptive MUD for future generation wireless communications systems. To prove this point, the adaptive performance of blind (and non-blind) adaptive MUD's is derived analytically for all the schemes that can be made to fit the same generic structure as the constant modulus scheme. For the first time, both the relative and absolute performance levels of the different adaptive algorithms are computed, which gives insights into the performance levels of the different blind adaptive MUD schemes, and demonstrates the merit of the constant modulus based schemes. The adaptive performance of the blind adaptive MUD's is quantified using the excess mean square error (EMSE) as a metric, and is derived for the steady-state, tracking, and transient stages of the adaptive algorithms. If constant modulus based MUD's are suitable for future generation wireless communications systems, then they should also be capable of suppressing multi-rate DS-COMA interference and also demonstrate the ability to suppress narrow band interference (NBI) that arises in overlay systems. Multi-rate DS-COMA provides the capability of transmitting at various bit rates and quality of service levels over the same air interface. Limited spectrum availability may lead to the implementation of overlay systems whereby wide-band COMA signal are collocated with existing narrow band services. Both overlay systems and multi-rate DS-COMA are important features of future generation wireless communications systems. The interference patterns generated by both multi-rate OS-COMA and digital NBI are cyclostationary (or periodically time varying) and traditional MUD techniques do not take this into account and are thus suboptimal. Cyclic MUD's, although suboptimal, do however take the cyclostationarity of the interference into account, but to date there have been no cyclic MUD's based on the constant modulus cost function proposed. This thesis thus derives novel, blind adaptive, cyclic MUD's based on the constant modulus cost function, for direct implementation on the FREquency SHift (FRESH) filter architecture. The FRESH architecture provides a modular and thus flexible implementation (in terms of computational complexity) of a periodically time varying filter. The operation of the blind adaptive MUD on these reduced complexity architectures is also explored.· The robustness of the new cyclic MUD is proven via a rigorous mathematical proof. An alternate architecture to the FRESH filter is the filter bank. Using the previously derived analytical framework for the adaptive performance of MUD's, the relative performance of the adaptive algorithms on the FRESH and filter bank architectures is examined. Prior to this thesis, no conclusions could be drawn as to which architecture would yield superior performance. The performance analysis of the adaptive algorithms is also extended in this thesis in order to consider the effects of timing jitrer at the receiver, signature waveform mismatch, and other pertinent issues that arise in realistic implementation scenarios. Thus, through a careful analytical approach, which is verified by computer simulation results, the suitability of constant modulus based MUD's is established in this thesis.
Thesis (Ph.D.)-University of KwaZulu-Natal, Durban, 2004.