M Tech Dissertations
Permanent URI for this collectionhttp://ir.daiict.ac.in/handle/123456789/3
Browse
7 results
Search Results
Item Open Access Vocal tract length normalization for automatic speech recognition(Dhirubhai Ambani Institute of Information and Communication Technology, 2014) Sharma, Shubham; Patil, Hemant A.Various factors affect the performance of Automatic Speech Recognition (ASR) systems. In this thesis, speaker differences due to variations in vocal tract length (VTL) are taken into account. Vocal Tract Length Normalization (VTLN) has become an integral part of ASR systems these days. Different methods have been studied to compensate for these differences in the spectral-domain. In this thesis, various state-of-the-art methods have been implemented and discussed in detail. For example, method of Lee and Rose uses a maximum likelihood-based approach. It implements a grid search over a range of values of warping factors to obtain optimal warping factors for different speakers. On the other hand, method by Umesh et al. uses scale transform to obtain VTL normalized features. Frequency warping is the basis of such normalizing techniques. Mel scale warping is the most acceptable for compensating the speaker differences as it is inspired from the hearing process of human ear. Use of Bark scale–based warping is proposed in this thesis. Bark scale is based on the perception of loudness by human ear in contrast with the Mel scale which is based on pitch perception. Bark scale-based warping provides improved recognition accuracy in case of mismatched conditions (i.e., training on male (or female) speakers and testing on female (or male) speakers). Performances of different methods have been tested for different ASR tasks in English, Gujarati and Marathi languages. TIMIT database is used for English language and details of database collection for Gujarati and Marathi languages have been discussed. The performance provided by using VTLN has shown improvement over state-of-the-art MFCC features alone for almost all applications considered in this thesis. One of the major tasks done in this thesis is the development of Phonetic Engines (PE) using VTLN in three different modes of speech, viz., read, spontaneous and lecture mode in Gujarati and Marathi languages. Lee-Rose method is used for the design of PEs. Improved accuracy is achieved using VTLN-based method as compared to MFCCs. In addition, template matching experiment is performed using various VTL-normalized features under study and MFCCs for application of spoken keyword spotting. Better precision and lower equal error rates (EER) are obtained using VTL-normalized Scale Transform Cepstral Coefficients (STCC). This suggests that VTLN-based features can be useful for bigger applications such as audio search and spoken term detection (STD).Item Open Access FIR filter for high speed 60 GHz wireless communications.(Dhirubhai Ambani Institute of Information and Communication Technology, 2013) Dholariya, Tarun; Sen, SubhajitIn this thesis report the design and implementation of the FIR filter at 5 GS/s presented. IEEE 802.15.3c standard declared unlicensed band from 57 to 66 GHz. This 60 GHz wireless communication transmitter uses interpolator filter part of this works at 5 GS/s. In this report 6 order 5 GS/s FIR filter implementation is described. In 180nm technology speed requirement cannot be achieved. Using the 45 nm technology this FIR filter is implemented and tested. The FIR filter structure is designed in bit pipelined structure. The key part are CPL adder and TSPC Flip-Flop. These adder and FF enabled very high frequency operation at 7.35 GHz. This led to implementation of 5 GS/s FIR filter using power of 2 coefficients.Item Open Access Application of compressive sensing to tow-way relay channel estimation(Dhirubhai Ambani Institute of Information and Communication Technology, 2012) Nair, Rahit R.; Chakka, VijaykumarAn Amplify and Forward Two-Way Relay Network is one where two nodes transmit data to each other via an intermediate relay. The relay amplifies the superimposed data from both the nodes before sending it to both the nodes. A method for the estimation of channel is proposed for Amplify and Forward Two-Way Relay Network (AF-TWRN). The proposed method utilizes the fact that the channel in the case of AF-TWRN shows sparse characteristic. The sparse multipath channel is estimated in the case of AF-TWRN using compressive sensing (CS) reconstruction algorithm, namely Iterative Hard Thresholding (IHT). MSE based performance of these methods in estimating the composite AF-TWRN channel was calculated and compared to that using Compressive Sampling Matching Pursuit (CoSaMP) and Orthogonal Matching Pursuit (OMP). IHT and CoSaMP are seen to perform slightly better than OMP with lesser computational complexity than OMP. It was also shown that all three CS based estimation methods perform better than the traditional Least Squares (LS) method in the estimation of Sparse AF-TWRN channel. A low complexity detection strategy was proposedItem Open Access Built-in self test architecture for mixed signal systems(Dhirubhai Ambani Institute of Information and Communication Technology, 2009) Jain, Mahavir Rajmal; Mandal, Sushanta; Nagchoudhuri, DipankarBuilt-in self test architecture or BIST as we call them, are the necessity of time since the shrinking sizes of component on-chip with advance in IC technology are making it BIST artistries are being rapidly developed and used for digital circuitry due to well defied fault models and advanced designs tools and techniques available. But with more analog circuitry being built on same platform with digital circuitry, the necessity for BIST architecture of mixed signal system on chip is increasing. The proposed BIST scheme is developed to test the data converters, both DAC and ADC on chip as well as other analog IP modules depending on specification of design without/least affecting the architecture of actual design and without making use of any complex DSP circuitry. The concept of internet node access based testing of digital blocks is also used for dynamic parameter of DAC like offset voltage and gain error, monotonicity and linearity non-specific BIST scheme. Letter the digital control logic protion of SAR ADC is tested with scan-chain insertion and thus overall functionality of SRA ADC is verified. A simple comparison based method is also proposed for other type ADCs. For other anlog IPs, we propose IP-based testing where digital to analog converted test signals can be applied depending on specifications of IP design from vendor without affecting architecture of the design. The output from IPs are taken at different nodes and applied directly to ADC on chip. The digitized output response is then compared with expected response to test the functionality of the DUT and find out its deviation from desired value to achieve pre-defined level of accuracy. The design failing any of the sequence of afore-mentioned test is discarded faulty. The circuitry is designed and evaluated at schematic level using TSMC complementary metal oxide-semiconductor (CMOS) 0.5um technology.Item Open Access Effect of channel asymmetry on reputation based cooperation mechanisms in mobile ad-hoc networks(Dhirubhai Ambani Institute of Information and Communication Technology, 2008) Vasavada, Tejas; Srivastava, SanjayEnforced cooperation among MANET nodes is an active research issue. In applications, where the users have different goals and there is no central authority to control them, users may become selfish. These nodes may not forward packets sent by others and thus affect the basic functionality of the network. One proposed class of protocols to handle such scenarios is based on reputation functions. In reputation based schemes, nodes maintain reputation values of other nodes. Reputation value of a node is high for nodes that forward the received packets and low otherwise. Nodes with very low reputation values are identified as selfish nodes and isolated from network operations. In such schemes, every node has to observe whether its neighbour is forwarding packets or not. A node after sending a packet to its neighbour to forward further, increases reputation of neighbour if it overhears the same packet from the neighbour. If it does not overhear the packet, reputation of neighbour is reduced. This is the basic method to observe whether neighbour node is cooperative or not. Here basic assumption is that the channel between two neighbour nodes is always symmetric. This assumption does not hold true due to two reasons: (1) Nodes are moving. Neighbour node may have forwarded the packet but by the time it forwards, either observing node or neighbour or both might move out of each others range. (2) Even if both are in each others range, due to multipath fading, observing node may not overhear the packets forwarded by neighbour node. Thus sometimes even honest nodes may be considered selfish and isolated due to this asymmetry. This reduces throughput of honest nodes. In this thesis work we have examined the probability of channel asymmetry as a function of ratio r(of inter node distance and transmission range), for given values of relative average velocity of nodes V and Ricean parameter K. Ricean parameter K represents type of the environment i.e. obstructed or unobstructed. We have proposed an enhancement of existing reputation protocol OCEAN. In the enhanced protocol, observing node probabilistically updates reputation of neighbour when it does not overhear. This probabilistic update takes into account probability of channel asymmetry. We have tried to minimize false positives, i.e. honest nodes being detected as selfish. We have shown through simulations that false negatives, i.e. selfish nodes being detected as honest, do not increase much. We have also shown that how false positives and false negatives change as degree of dishonesty of selfish nodes change. At last, we have shown that throughput levels of honest nodes in original OCEAN protocol and modified OCEAN protocol are almost same.Item Open Access Architecture design for preliminary ECG analysis system using new DFT based analysis technique(Dhirubhai Ambani Institute of Information and Communication Technology, 2007) Chaurey, Vasudha; Nagchoudhuri, DipankarThis thesis proposes a hardware architecture of an ASIC for a portable ECG analysis system. The device is meant to record and analyze ECG signals in real time so as to detect the presence of abnormalities. In order to achieve this, a totally new approach for the analysis of ECG signals using Discrete Fourier Transform (DFT) is developed as a part of the thesis. An important finding of the project, through experiments performed on real ECG data in MATLAB, is that the phases of first 8 DFT coefficients of beat-wise ECG signals give distinguishing patterns for normal and abnormal beats. With this idea of variable size DFT as the basis, a much simplified form of the technique with fixed 32 point DFT is derived without significantly disturbing the patterns so as to make it suitable for hardware implementation. The translation of the algorithm to hardware architecture has been done in a way so as to achieve optimization in terms of area as well as power by minimizing the number of computations. One of the important features of the proposed architecture is the synchronization of the complete system which processes an asynchronous signal, i.e. the ECG beats, in real time. The thesis gives the custom Register Transfer Level architecture for the processing block of the system, which is meant for selecting fixed number of samples from every beat and do a customized DFT phase computation for detecting abnormalities. The project proposes a very simple technique for the beat detection as a part of optimization. Since the incoming signal is much slower than the processing rate, the proposed architecture is designed with the flexibility of adding extra functionalities in the system other than ECG analysis which if possible can use the same processing hardware during the wait periods.Item Open Access Representation theory in signal processing: some connections through gap(Dhirubhai Ambani Institute of Information and Communication Technology, 2005) Dixit, Parth; Sinha, Virendra P.This thesis is concerned with the study of some of the recent developments in the area of signal processing that arise from the connections of discrete transforms and their inherent symmetries with the representation theory of groups. There are two main aspects of these developments, one, that of unification and generalization and, two, that of constructing fast algorithms for the implementation of matrix transforms. The focus of this thesis is on the second line of developments - that of constructing fast algorithms. The central theme of recent work in this direction is that of defining symmetries of a transform in terms of a pair of group representations, and then decomposing the representations along a chain of subgroups. Such a decomposition leads to a factorization of transform matrices into product of sparse matrices, which provide a means of fast transform computations. The decomposition process relies on results of group representation theory not commonly familiar to workers in the area of digital signal processing. Furthermore, the algebraic computations involved would be enormously difficult without the use of computer algebra tools and software packages of recent origin whose potential is only now beginning to be realized. On both counts, there is a need for providing a bridge between what is commonly understood of fast transforms and what the recent developments mean for them. With this need in mind, an attempt has been made in this thesis to (a) identify and discuss the relevant representation theory results, and (b) present results of hands-on experience, in the form of examples, with the software package called GAP (acronym for Groups, Algorithms and Programmes) in transform factorization.