Search

Speech represents one of the most important characteristics of the human speicies, which sets humans apart from other living beings. Thus, speech represents highly important signal in digital communications. Since the transmission of the speech signal in digital communications began, the technology dedicated to the digitalization of speech took a significant place in digital signal processing. Still lasting growth in the usage of multimedia systems enables us to send and distribute digital speech and audio signals easier than ever. New implementation possibilities demand progress in coding techniques, which inspires researchers to develop new speech and audio signal coding algorithms. This dissertation will present different speech and audio signal coding algorithms, based on waveform signal coding. Also, the advantages and disabilities of implementing different coding techniques and quantizer types will be evaluated. The output signal quality will be evaluated by subjective and objective quality measures. Firstly, the implementation of subband coding techniques in speech and audio signal coding will be presented. Subband coding performs decomposing of the imput signal into subbands, which can be coded independently, by witch we can reduce the required bit rate. Four different subband coding algorithms will be presented, determined by different complexity levels and quantization techniques. This dissertation will present modified BTC(Block Truncation Coding) algorithm and its implementation in speech and audio signal coding. BTC algorithm was originally developed for applications in black and white image coding, while the modified algorithm exploits some of the common characteristics of black and white images and speech and audio signals. Also, a simple differential and backward adaptive coding algorithm will be presented. This algorithm will show the advantages of implementing aforementioned techniques, by witch we can obtakin high output signal quality, while preserving low computational complexity. ...

This PhD discusses the research related to the approximation and implementation of the twochannel hybrid filter banks. Special attention is paid to the analogue part, i.e. analysis part of the hybrid filter banks. Two approximations of the filter bank pair for analysis have been proposed. The first approximation of the transfer function of the low-pass filter is based on the simple adaptation of the orthogonal Jacobi polynomials in order to obtain the Pseudo-Jacobian polynomials. In relation to other known approximations, the Pseudo-Jacobian polynomial one has two prime parameters, which can adjust the characteristics of the filter in wide ranges. This approximation can be successfully applied for the realization of a complementary bank of filters. It is known that recursive double-complementary digital filter banks can be implemented with all-pass filters, and research has shown that double-complementary filter banks can also be realized in the analogue domain. The realization of the proposed filter bank has been done in two steps. In the first step, with the complementary decomposition, the prototype transfer function is obtained by two all-pass filters, while in the second step, by their addition or subtraction, transfer functions of lowpass and highpass filters are obtained. The advantage of such a system is that the same hardware can be used for realization of both low frequency and high frequency transfer functions. Monte Carlo simulation of the realization of a double complementary analog filter pair based on a parallel connection of two analogue all-pass filters showed that all-pass realization is characterized by a small sensitivity of the attenuation characteristics to the component tolerances in the filter pass-band, while the sensitivity in the stop-band is substantially higher compared to the case of a standard cascade realization of the low-pass filter and the high-pass filter. By a suitable selection of the analysis filter bank and the synthesis filter bank, a condition for suppressing the effects arising from the overlapping of the spectrum in banks for analysis and synthesis can be fulfilled. The all-pass complementarity of an analogue filter bank points to the fact that amplitude distortion, which is introduced by the analog bank of the analysis filters, can be completely suppressed, so that the non-linearity of the group delay characteristics is the predominant distortion. In order to achieve a near perfect reconstruction of the signal, a new realization of the group delay corrector was proposed, which makes it possible for the group delay to be constant in a flat sense, i.e. with a number of flatness at the origin. An analysis of the sensitivity has shown that the sensitivity of the correction of the group waveform in the filter pass-band that is proportional to the square of the Q -factor of the pole. In other words, the group delay corrector is very sensitive to the component tolerances.

In this PhD thesis two-channel QMF IIR filter banks have been discussed. As a part of this research, direct synthesis of a new IIR filter class, as a lowpass protoype function which satisfies condition for complementary decomposition, has been proposed. Since attenuation characteristic of the new filter class lies in the area between the Butterworth’s and the Elliptic’s attenuation characteristics of the same order, new class has been referred to as Transitional Butter-Elliptic filters. Transfer function of the the lowpass prototype function was separated into two allpass filters by complementary decomposition. Then, lowpass and highpass filters, as power symmetric filter pair, are obtained using those allpass filters. The main advantage of complementary IIR filters based on parallel connection of two allpass filters, is its smaller sensitivity to digital word length change compared to standard filters and low-cost implementation since the same hardware is used for the lowpass and the highpass transfer function. Within the analysis of the new filter class, transfer functions of the odd order with one, two, three and four extremal values of attenuation characteristics in the passband and in the stopband, has been discussed. All types of Transitional filters have been compared with the Butterworth’s and the Elliptic’s filters of the same order. It is shown that in comparison with Butterworth filter, the Transitional filter has better selectivity but lower selectivity than Elliptic filter. However, the proposed new filter class has pure imaginary poles and accuracy in complementary decomposition is at very high level, which the Elliptic filter cannot achieve. Also, the practical realization of the new filter class is easier than for the Elliptic filter since the new class has conjugate-complex zeros on the unit circle and one multiple zero at z = −1. Further, poles and zeros of the Elliptic filter are accumulated near the passband edge and the distance between them decreases as the filter order increases. This requires a very long digital word length for high-order filter realization. The new class filter has approximately equidistant arrangement of poles so it can be realized with shorter digital word length than it is required for the Elliptic filter realization. For Transitional filters class with one extremum of multiplicity L in the passband and in the stopband, closed-form expression for the transfer function coefficients has been obtained. Application of these kind of filters is significant since for sample rate conversion the lowestcomplexity filter is required. With appropriate selection of filters in analysis and synthesis section, the conditions for ii removing the aliasing effects are fulfilled. Magnitude distortions in IIR filter bank can also be completely eliminated and therefore output signal of the QMF bank has only phase distortion. In order to obtain nearly perfect reconstruction of the output signal, the allpass equalizer has been proposed for group delay compensation. Also, effect of the equalizer order on the accuracy of signal reconstruction has been analyzed.

This article presents a method for increasing the performance of the progressive coding algorithms for the subbands of images, by representing the coefficients with a code that reduces the truncation error.

Submitted by Joana Lopes Dias (jdias@ubi.pt) on 2015-07-15T11:00:39Z No. of bitstreams: 1 dissertacao-20101025-final.pdf: 3757760 bytes, checksum: 83b5bf0cc044d59135c415fa0c2d75ce (MD5) Made available in DSpace on 2015-07-15T11:00:39Z (GMT). No. of bitstreams: 1 dissertacao-20101025-final.pdf: 3757760 bytes, checksum: 83b5bf0cc044d59135c415fa0c2d75ce (MD5) Previous issue date: 2010

The authors suggest a parallel processing structure using the proposed international standard for visual telephony (CCITT P*64 kbs standard) as processing elements, to compress digital high definition television (HDTV) pictures. The basic idea is to partition an HDTV picture, in space or in frequency, into smaller sub-pictures and then compress each sub-picture using a CCITT P*64 kbs coder. This seems to be a cost-effective solution to the HDTV hardware. Since each sub-picture is processed by an independent coder, without coordination these coded sub-pictures may have unequal picture quality. To maintain a uniform quality HDTV picture, the following two issues are studied: sub-channel control strategy (bits allocated to each sub-picture); and quantization and buffer control strategy for individual sub-picture coders. Algorithms to resolve these problems and their computer simulations are presented. >

In dieser Arbeit wird mit einem informationstheoretischen Ansatz der Zusammenhang zwischen grundlegenden Eigenschaften sensorischer Informationsverarbeitung und der Statistik natürlicher Eingangssignale untersucht. Im ersten Teil der Dissertation werden theoretische Überlegungen zur mehrkanaligen Bildverarbeitung durch ortsfrequenzselektive Neurone im visuellen Kortex angestellt. Dazu wurden für das charakteristische Leistungsdichtespektrum natürlicher Bilder unterschiedliche Teilbandzerlegungen im Rahmen der Shannonschen Rate-Distortion-Theorie untersucht. Die Ergebnisse zeigen, dass eine logarithmische, selbstähnliche Zerlegung mit deutlich geringerer Komplexität zur gewünschten Signalqualität führen kann, als eine lineare Partitionierungsstrategie. Eine weitere Analyse hat erbracht, dass es eine leicht modifizierte, optimale Zerlegung gibt, die zu einem noch besseren Trade-off zwischen Aufwand und Leistung führt. Diese Kanalaufteilung ist in erster Näherung selbstähnlich, zeigt jedoch eine systematische Abnahme der logarithmischen Bandbreiten mit zunehmender Mittenfrequenz. Exakt dieser Effekt wurde auch in neuro- und biowissenschaftlichen Experimenten beobachtet. Damit kann die spezifische Struktur der Ortsfrequenzzerlegung im Sehsystem als das Ergebnis einer effizienten Anpassung an die in natürlichen Bildern vorhandene Statistik zweiter Ordnung gedeutet werden. Das Prinzip der Angepasstheit sensorischer Systeme an die statistischen Redundanzen der natürlichen Umgebung gilt auch für multisensorische Integrationsmechanismen, wie in mehreren psychophysischen Experimenten mit akustisch-visuellem Stimulusmaterial im zweiten Teil der Arbeit demonstriert wird. Dabei werden Inkrementalschwellen gemessen, die abhängig von der ökologischen Validität konsistent eine deutliche Sensitivitätssteigerung unter redundanten, gleichlautenden Stimulusbedingungen aufweisen und damit auf einen hochselektiven, neuronalen Summationsmechanismus hindeuten. Ein mehrstufiges On/Off-Modell ist in der Lage, die experimentellen Daten zu erklären. Es konnte damit gezeigt werden, dass die modalitätsspezifischen Signale in Raum und Zeit kohärente Stimuluskonfigurationen darstellen müssen, um effektiv integriert zu werden. Diese Versuchsergebnisse liefern ein klares Gegenargument zu der klassischen Idee von grundlegend getrennten, modalitätsspezifischen sensorischen Kanälen, die nur in Form von Bias-Effekten oder auf höherer Entscheidungsebene miteinander interagieren. Using an information-theoretic approach, this thesis analyzes the relationship between basic properties of sensory information processing and the statistics of natural input signals. The first part of the thesis deals with the multi-channel decomposition by spatial frequency selective neurons in the visual cortex. To understand the specific biological partitioning scheme, I analyzed the rate-distortion behaviour of several subband decompositions with respect to the characteristic power spectral density of natural images. The results show that a logarithmic self-similar decomposition can yield the desired degree of distortion with far less complexity than a linear decomposition. A closer analysis revealed that there exists a slightly modified optimum decomposition which can yield even better results. This decomposition is roughly self-similar, but with a weak systematic decrease of log-bandwidth with increasing center frequency. This effect has also been observed for visual neurons. My results thus corroborate the hypothesis that biological sensory systems are efficiently adapted to the statistical redundancies of the natural environment. If this adaptation to the statistical covariances of the sensory input is a universal mechanism that determines not only the information processing within but also between different modalities, substantial crossmodal sensory interacttions should exist. To test this prediction, I conducted several psychophysical discrimination experiments in the second part of my thesis. It was measured whether joint auditory-visual sensitivity functions show intersensory interaction effects matched to the multimodal statistical covariance structure. The experimental results provide significant evidence for a neural integration of multisensory information at an early level. Moreover, this type of multisensory integration is highly selective for ecologically valid stimulus combinations and depends on mutual properties of crucially on the auditory and visual signals used in the sense of being co-incident and co-localized. The results thus provide evidence against an independent, modality-specific processing in early sensory stages offering a new perspective on the nature and role of multisensory interactions.

A consistent theory of optimum subband coding of zero mean wide-sense cyclostationary signals with N-periodic statistics is presented in this paper. Blocked polyphase representation of the analysis and synthesis filter banks is introduced as an effective way of multirate subband coder description. Optimum energy compaction using Nyquist-M process is presented as a solution for maximizing the coding gain of the coder. In two definitions and four theorems the author proves that Nyquist-M filters fulfill necessary and sufficient conditions imposed on subband signals. Results from Matlab simulations are presented to support theoretical conclusions.

A novel architecture for image compression is proposed, which is based on a suitable combination of subband filtering and linear neural networks. This combination permits efficient coding, together with the advantages of the neural-network-based approach. The architecture is described, and results of simulations are presented. The architecture is shown to perform well, notwithstanding the reduced complexity of the approach. The structure is highly parallel, so that high computation rates are possible; this property can be useful if sequences of images are to be compressed. >

oz YÜKSEK LİSANS TEZİ DECOMPOSITION BY TREE STRUCTURE TO OBTAIN THE BEST SUBBAND DIVISIONS BY USING UNIFORM FILTER BANKS TUGAY AKBAŞ ÇUKUROVA ÜNİVERSİTESİ FEN BİLİMLERİ ENSTİTÜSÜ ELEKTRİK-ELEKTRONİK ANABİLİM DALI Danışman : Yrd. Doç. Dr. Sami Anca Yıl: 2001, Sayfa: 70 Jüri : Yrd. Doç. Dr. Sami Anca Prof. Dr. Süleyman Güngör Yrd. Doç. Dr. Murat Aksoy Doğrusal dönüşümler, görüntü işlemede kullanılan, görüntü analizi, görüntü kodlaması gibi pek çok uygulama için baz teşkil etmektedir. Altband dönüşümler de bu uygulamalar için kullanışlı özellikleri bulunan doğrusal dönüşümlerin bir alt sınıfıdır. Kodlama performansını altband süzgeçlerini işaretin doğasına uygun olarak seçmekle arttırmak mümkün olduğundan altband kodlama, görüntü işleme uygulamaları için verimli bir yoldur. Bu çalışmada önerilen diğer bir metod da önce işareti istatistiksel bir model olan birinci dereceden autoregressive (AR(1)) modeline uyarlamak ve en iyi kodlama kazancını veren band bölüşümlerini hesaplamaktır. Bu oranları elde etmek için de ağaç yapısı kullanılmıştır. Bu yöntem durağan görüntülere de uygulanmıştır. Anahtar Kelimeler: Süzgeç Çatısı, Kayıplı Veri Sıkıştırma, Altband Kodlama ve Kazanç n ABSTRACT MSc THESIS DECOMPOSITION BY TREE STRUCTURE TO OBTAIN THE BEST SÜBBAND DIVISIONS BY USING UNIFORM FILTER BANKS TUGAY AKBAŞ DEPARTMENT OF ELECTRICAL AND ELECTRONICS ENGINEERING INSTITUTE OF NATURAL AND APPLIED SCIENCES UNIVERSITY OF ÇUKUROVA Supervisor : Assoc. Prof. Dr. Sami Anca Year: 2001, Pages:70 Jury : Assoc. Prof. Dr. Sami Anca Prof. Dr. Süleyman GÜNGÖR Assoc. Prof. Dr. Murat AKSOY Linear transforms are the basis for many applications used in image processing, image analysis, and image coding. Subband transforms are a subclass of linear transforms, which offer useful properties for these applications. Since, it is possible to enhance its quality by changing subband filters adaptive to signal characteristics, subband coding is an efficient way for image processing applications. It is proposed as an alternative method that first to sample signal by the first order autoregressive (AR (1)) model and the finding the best band splits which maximize the coding gain. Tree structure is used to obtain the best band fractions. Then, the structure is applied to still images. Key Words: Filter Bank, Lossy Data Compression, Subband Coding and Gain I 70

ÖZET Yüksek Lisans Tezi YA YILI İZGE TEKNİKLERİ KULLANAN KABLOSUZ YEREL ALAN AĞLARI Ahmet AKBULUT Ankara Üniversitesi Fen Bilimleri Enstitüsü Elektronik Mühendisliği Anabil im Dalı Danışman: Prof. Dr. Mümtaz YILMAZ Bu çalışmada, kablosuz yerel alan ağlan ve kullandıkları yayılı izge teknikleri incelenmiş, doğrudan dizili yay ıh izge tekniğinin alt bant kodlanmış video sinyallerinin iletiminde kanal bozulmalarına karşı etkileri bilgisayar benzetimleri kullanılarak ortaya konmuştur. Video sinyalinin sıkıştırılmasında alt bant kodlama tekniği kullanılarak alıcı için iki farklı çözünürlükte görüntü seçeneği sunan ölçeklenebilirlik özelliği sağlanmıştır. Bu görüntüler taşıdıkları öneme göre kanal bozulmalarına karşı farklı korunmuşlardır. Böylelikle kanal koşulları kötüyken bile alıcıda kabul edilebilir bir kalitede görüntü elde edilebilmektedir. 2000, 53 sayfa ANAHTAR KELİMELER: Yerel alan ağları, Yayılı izge. Alt bant video kodlama ABSTRACT M.S. Thesis WIRELESS LOCAL AREA NETWORKS USING SPREAD SPECTRUM TECHNIQUES Ahmet AKBULUT Ankara University Graduate School of Natural and Applied Sciences Department of Electronics Engineering Supervisor: Prof.Dr.Mûmtaz YILMAZ In this study, Wireless Local Area Networks (WLANs) and the spread spectrum techniques which are used in WLANs have been examined. The effects of direct sequence spread spectrum techniques for the use of subband coded video signal transmission over wireless channel have been computer simulated. Subband coding technique, which enables the scalability property, is used to compress video signals. Two different image resolutions for subband coding has been attempted. These images have different priority levels against channel distortion according to their information content Consequently resonable image quality has been obtained even in adverse channel conditions. 2000, 53 pages Key Words: Local area networks, Spread spectrum, Subband video coding vcvUsaaGOTbuenfflfr 53

oz DOKTORA TEZİ THE BEST SUBBANDS WHICH MAXIMIZE CODING GAIN WITH APPLICATION TO STILL IMAGES SAMI ARICA ÇUKUROVA ÜNİVERSİTESİ FEN BİLİMLERİ ENSTİTÜSÜ ELEKTRİK-ELEKTRONİK ANABİLİM DALI Danışman : Doç. Dr. Hakan Çağlar Yıl: 1999, Sayfa: 107 Jüri : Doç. Dr. Hakan Çağlar Prof. Dr. Süleyman Güngör Yrd. Doç. Dr. Ulus Çevik Altband kodlama en etkin kayıplı veri sıkıştırma yöntemlerinden biridir. Kodlama performansım altband süzgeçlerim işaretin doğasına uygun olarak seçmekle artırmak mümkündür. Bu çalışmada önerilen alternatif bir metod da önce işareti bir istatistiksel model olan birinci dereceden autoregressive (AR(1)) modeline uyarlamak ve kodlama kazancını en iyi yapacak band oranlarını hesaplamaktır. Bu yöntem iki boyutlu (2D) işaretlere de yani durağan görüntülere de uygulan abilir. Görüntünün sütunları ve satırları bağımsız olarak kabul edilebilir ve AR(1) ile modellenir. Daha sonra sütunları ve satırları ayrı olarak alt bileşenlerine ayırmak için en iyi band oranlarım sağlayacak bir ağaç yapısı oluşturulur. En iyi band oranlan kodlama kazancım en büyük yaptığından altband kodlamanın başarımını bu ayrıştırma ile artırılmış olacaktır. Anahtar Kelimeler: Sıkıştırma, Süzgeç Çatısı, Altband Kodlama, Dal gacık. II ABSTRACT Ph.D. THESIS THE BEST SUBBANDS WHICH MAXIMIZE CODING GAIN WITH APPLICATION TO STILL IMAGES SAMI ARICA DEPARTMENT OF ELECTRICAL AND ELECTRONICS ENGINEERING INSTITUTE OF NATURAL AND APPLIED SCIENCES UNIVERSITY OF ÇUKUROVA Supervisor : Assoc. Prof. Dr. Hakan Çağlar Year: 1999, Pages: 107 Jury : Assoc. Prof. Dr. Hakan Çağlar Prof. Dr. Süleyman Güngör Assis. Prof. Dr. Ulus Çevik Subband coding is an efficient tool for lossy data compression. It. is possible to enhance its performance by changing subband filters adaptive to signal charac teristics. An alternative method which is proposed in this study is first to model signal by the first order autoregressive (AR(1)) statistical model and finding the best band fractions which maximize coding gain. A tree structure is employed to maintain the best band fractions. This method can also be applied to two dimensional (2D) signals like still images. Columns and rows of the image are considered to form an ensemble and the ensemble is modelled by AR(1). Later a tree is adapted to fit the best band fractions and columns and rows are de composed separately. Because the best band fractions maximize the coding gain subband coding performance is improved by the decomposition. Key Words: Compression, Filter Bank, Subband Coding, Wavelet. 107

ÖZET ÇOK UÇLU EKG VERİLERİNİN ÇOK ÖRNEKLEME FREKANSLI İŞARET İŞLEME VE DÖNÜŞÜM DÜZLEMİ KODLAMA TEKNİKLERİ KULLANILARAK SIKIŞTIRILMASI M. Cengiz Aydın Elektrik ve Elektronik Mühendisliği Bölümü Yüksek Lisans Tez Yöneticisi: Doç. Dr. A. Enis Çetin Temmuz 1991 Bu tezde sayısal elektrokardiogram (EKG) işaretlerinin sıkıştırılması için iki ayrı yöntem sunulmaktadır. İlk yöntem tek uçlu EKG verilerinin sıkıştırıl masını amaçlar. Bu amacı gerçekleştirmek için Alt-Bant-Kodlaması, Dönüşüm Düzleminde Kodlama ve Değişken Uzunluklu Kodlama gibi çeşitli yöntemler kullanılmıştır. EKG veri tabanlarının problemi olan veri hacmi azaltılması uy gulama olarak seçilmiş ve benzetim sırasında başarılı sonuçlar elde edilmiştir. İkinci yöntem çok uçlu EKG verilerinin sıkıştırılması için önerilmektedir. Sekiz EKG kanalına ait veriler lineer bir dönüşümden geçirilirler. Bu dönüşüm sonucu oluşan işaretler Çok Örnekleme Frekanslı İşaret İşleme ve Dönüşüm Düzleminde Kodlama teknikleri kullanılarak sıkıştırılırlar. Böylece, tek kanallı sıkıştırma yöntemlerine göre daha yüksek sıkıştırma oranlarına ulaşılmaktadır. Anahtar Sözcükler: EKG verilerinin sıkıştırılması, Alt-Bant-Kodlaması, Dönüşüm Düzleminde Kodlama. iv ABSTRACT MULTILEAD ECG DATA COMPRESSION BY MULTIRATE SIGNAL PROCESSING AND TRANSFORM DOMAIN CODING TECHNIQUES M. Cengiz Aydın M.S. in Electrical and Electronics Engineering Supervisor: Assoc. Prof. Dr. A. Enis Çetin July 1991 This thesis presents two data compression methods for digital electrocardio gram (ECG) signals. The first method is proposed for the compression of single lead ECG data. Various coding techniques such as Sub-Band,Transform and Variable Length Codings are used to realize that purpose. Data size reduction which is a problem in ECG databases is chosen as an application and successful results are achieved during simulations. The second algorithm is proposed for the multilead ECG data compression. A linear transform is applied to samples of the eight standard ECG lead signals. The resulting transform domain sig nals are compressed using various coding methods, including multi-rate signal processing and transform domain techniques. Higher compression ratios are achieved compared to the single channel compression techniques. Keywords: ECG Data Compression, Sub-Band Coding, Transform Coding. m 53

ÖZET VEKTÖREL NİCEMLEMENİN SES İŞARETİNİN ALT-KUŞAK KODLANMASINA UYGULANMASI ÜZERİNE BİR ÇALIŞMA UĞUR, Mehmet Sedat Yüksek Lisans Tezi, Elektrik ve Elektronik Mühendisliği Bölümü Tez Yöneticisi: Doç. Dr. Melek YÜCEL Tez Eş-yöneticisi: Öğr. Gör. Tuğrul MEMİŞOĞLU Eylül 1991, 85 sayfa Bu çalışmada ses işaretinin alt-kuşak kodlanmasına vektörel nicemleme uygulanması incelenmiştir. Ses işaretinin bir dizi kuşak-geçiren süzgeçle süzülmesi sonucu elde edilen alt-kuşak işaretlerinin birbirinden bağımsız olarak nicemlenmesi yerine, nicemleme birimi olarak bu işaretlerden oluşan vektör alınmıştır. Bu amaçla, değişik erkek ve kadın konuşmacılardan alınan ses kayıtlarından derlenmiş uzun bir belletici dizi kullanılarak farklı kodlama oranları için kod kitapçıkları tasarımlanmıştır. Vektörel nicemleme için başarımı en-iyi olmamakla birlikte, daha az işlemle gerçekleştirilebilmesin-den ötürü ağaç üzerinde arama algoritması tercih edilmiştir. Vektörel nicemlemeye dayalı alt-kuşak kodlama dizgesi, sayıl nicemlemeye dayalı değişik alt-kuşak kodlama dizgeleriyle öznel ve nesnel nitelik açısından deneysel olarak karşılaştırılmıştır. Nesnel nitelik ölçüsü olarak işaret-nicemleme gürültüsü oranı, öznel nitelik ölçüsü olarak ise dinleyicilerin ortalama kanısı kabul edilmiştir. Bunların yanı sıra, ses işaretlerini alt-kuşak işaretlerine ayırmak için dördül ayna gölgeli süzgeç ve eşlenik dördül süzgeç kullanmanın yeniden oluşturulan sesin niteliği üzerindeki etkileri deneysel olarak sınanmıştır. Anahtar Sözcükler : Alt-Kuşak Kodlama, Vektörel Nicemleme, Dördül Ayna Gölgeli Süzgeçler, Eştlenik Dördül Süzgeçler, Ses Kodlama, Çok-Oranlı işaret İşleme. Bilim Dalı Sayısal Kodu: 609.02.08 iv ABSTRACT A STUDY ON THE APPLICATION OF VECTOR QUANTIZATION TO SUB-BAND CODING OF SPEECH SIGNALS UĞUR, Mehmet Sedat M.S. in Electrical and Electronics Engineering Supervisor: Assoc. Prof. Dr. Melek YÜCEL Co-Supervisor: Instr. Tuğrul MEMİŞOĞLU September 1991, 85 pages Application of vector quantization to sub-band coding of speech signals is studied in this thesis. The vector of sub-band signals obtained by band pass filtering the speech signal via a bank of filters and then decimating is taken as a unit for quantization, instead of quantizing the sub-band signals separately. For this purpose, codebooks of different rates are designed using a long training sequence assembled from the speech recordings of various male and female speakers. Tree-searched vector quantization is preferred due to its computational efficieny, although it yields slightly suboptimum results. The objective and subjective quality of the vector quantization based sub- band coding scheme and various scalar quantization based sub-band coding schemes are compared experimentally, where signal-to-quantization noise ratio and mean opinion score of human subjects are considered as objective and subjective quality measures, respectively. Also, the effect of employing a bank of quadrature mirror filters and conjugate mirror filters to split the speech signal into sub-bands on the quality of the reconstructed speech is tested experimentally. Key Words : Sub-Band Coding, Vector Quantization, Quadrature Mirror Filters, Conjugate Quadrature Filters, Speech Coding, Multirate Signal Processing. Science Code : 609.02.08 111 85

ÖZET ALT BANT tSBÜ KODLAMASI TEKNİ?İ KULLANILARAK GÖRÜNTÜ BİLGİSİNİN SIKIŞTIRILMASI KURU, Murat Yüksek Lisans Tezi. Elektrik ve Elektronik Hüh. Ortadoğu Teknik üniversitesi. Ankara Tez Yöneticisi : Doç. Dr. Melek D. Yücel Eylül 1990, 103 sayfa Bu tezde alt bant kod lama tekniği ile türevsel darbe kodlamalı modülasyon (DPCM) teknikleri açıklanmış ve görüntünün kaynak kod lamasında kullanılmıştır. SBC tekniği, bir tüm bant görüntüden izgesel olarak ayrık alt bantlar elde etmektir. Elde edilen alt bant görüntüler tek tek ele alınıp, herbirine kendi varyanslarının diğerlerinin varyanslarına göre bir fonksiyonuna uyarak görüntü elemanı başına bit sayıları verilmektedir. Bundan sonra her alt bant kendisine verilen belirli bit sayısına uyularak DPCM yöntemi. ile kodlanmaktadır. Görüntü elemanı başına bit sayısının dahafazla azaltılabilmesi için kaydırma kod laması (shift coding) ve süreç uzunluğu kod laması (run length coding) adlı iki dağıntı kodlama (entropy coding) tekniği kullanılmaktadır. Dağıntı kodlaması» kod kelimelerinin istatiklerini kullanarak eşit olmayan uzunlukta yeni kod kelimeleri bularak görüntünün daha düşük bit sayısı ile gösterilmesidir. Sistem performansı, ortalama karesel hata ve boylandırılmış ortalama karesel hata ölçütleri ile ölçülmüştür. Ayrıca performans hakkında öznel çıkarımlarda da bulunulmuştur. Anahtar Sözcükleri alt bant kodlaması, türevsel darbe kod lama lı modülasyon, dağıntı kodlaması, görüntü bilgisi sıkıştırılması vi 103

In this paper, we present a characterization of MRA biorthogonal wavelet filters with full frequency supports. Based on this characterization, it is established that wavelet ramp filters are biorthogonal wavelets if the original wavelets are sufficiently regular. An efficient subband coding algorithm is developed for wavelet filtering in filtered backprojection, which is the most popular method in computed tomography (CT). Computer simulation suggests that this wavelet filtering process is a useful tool for improving image quality and reducing computational time in local CT reconstruction.

In recent years the ``Internet Multicast Backbone'', or MBone, has risen from a small, research curiosity to a large scale and widely used communications infrastructure. A driving force behind this growth was the development of multipoint audio, video, and shared whiteboard conferencing applications. Because these real-time media are transmitted at a uniform rate to all the receivers in the network, a source must either run at the bottleneck rate or overload portions of its multicast distribution tree. We overcome this limitation by moving the burden of rate-adaptation from the source to the receivers with a scheme we call Receiver-driven Layered Multicast, or RLM. In RLM, a source distributes a hierarchical signal by striping the different layers across multiple multicast groups and receivers adjust their reception rate by simply joining and leaving multicast groups. In this paper we describe a layered video compression algorithm which, when combined with RLM, provides a comprehensive solution for scalable multicast video transmission in heterogeneous networks. In addition to a layered representation, our coder has low-complexity (admitting an efficient software implementation) and high loss resilience (admitting robust operation in loosely controlled environments like the Internet). Even with these constraints, our hybrid DCT/wavelet-based coder exhibits good compression performance. It outperforms all publicly available Internet video codecs while maintaining comparable run-time performance. We have implemented our coder in a ``real'' application --- the UCB/LBL video conferencing tool ``vic.'' Unlike previous work on layered video compression and transmission, we have built a fully operational system that is currently being deployed on a very large scale over the MBone.

Video coding has been investigated for the novel application of video transmission over packed-switched networks. The underlying design goals are presented, together with a software implementation of a coding scheme, which divides the input signal into frequency bands in all three dimensions, seems promising in that it lends itself to parrallel implementation, is robust enough to handle errors due to lost packets, and yields high compression with sustained good quality. Moreover, it may be well integrated with the network to handle issues is given, together with the results obtained from simulated system.