Pub Date : 1994-04-19DOI: 10.1109/ICASSP.1994.390028
S. Krishnamachari, W. J. Williams
A signal-adaptive kernel designed in the generalized marginals(GM) domain is introduced. This new kernel exploits the mechanism by which the cross-terms are created in the GM domain. It is shown that the cross-terms are created by a simple squaring process and the region of support for the cross terms is a subset of the region of support of the auto-terms. The generalized marginals of the Wigner distribution (WD) are always positive and real. The generalized marginals of all distributions which have a radially Gaussian kernel in the ambiguity domain are positive. This positivity is exploited for applying information measures in the construction of the adaptive kernel. The cross-term suppression is done in the GM domain and the time-frequency distribution is constructed using the filtered back-projection method. Moyal's formula is utilized to calculate the GM as the projections of the signal on linear chirps.<>
{"title":"Adaptive kernel design in the generalized marginals domain for time-frequency analysis","authors":"S. Krishnamachari, W. J. Williams","doi":"10.1109/ICASSP.1994.390028","DOIUrl":"https://doi.org/10.1109/ICASSP.1994.390028","url":null,"abstract":"A signal-adaptive kernel designed in the generalized marginals(GM) domain is introduced. This new kernel exploits the mechanism by which the cross-terms are created in the GM domain. It is shown that the cross-terms are created by a simple squaring process and the region of support for the cross terms is a subset of the region of support of the auto-terms. The generalized marginals of the Wigner distribution (WD) are always positive and real. The generalized marginals of all distributions which have a radially Gaussian kernel in the ambiguity domain are positive. This positivity is exploited for applying information measures in the construction of the adaptive kernel. The cross-term suppression is done in the GM domain and the time-frequency distribution is constructed using the filtered back-projection method. Moyal's formula is utilized to calculate the GM as the projections of the signal on linear chirps.<<ETX>>","PeriodicalId":290798,"journal":{"name":"Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124823945","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-04-19DOI: 10.1109/ICASSP.1994.389309
T. Kosaka, S. Sagayama
The paper proposes a tree-structured speaker clustering algorithm and discusses its application to fast speaker adaptation. By tracing the clustering tree from top to bottom, adaptation is performed step-by-step from global to local individuality of speech. This adaptation method employs successive branch selection in the speaker clustering tree rather than parameter training and hence achieves fast adaptation using only a small amount of training data. This speaker adaptation method was applied to a hidden Markov network (HMnet) and evaluated in Japanese phoneme and phrase recognition experiments, in which it significantly outperformed speaker-independent recognition methods. In the phrase recognition experiments, the method reduced the error rate by 26.6% using three phrase utterances (approximately 2.7 seconds).<>
{"title":"Tree-structured speaker clustering for fast speaker adaptation","authors":"T. Kosaka, S. Sagayama","doi":"10.1109/ICASSP.1994.389309","DOIUrl":"https://doi.org/10.1109/ICASSP.1994.389309","url":null,"abstract":"The paper proposes a tree-structured speaker clustering algorithm and discusses its application to fast speaker adaptation. By tracing the clustering tree from top to bottom, adaptation is performed step-by-step from global to local individuality of speech. This adaptation method employs successive branch selection in the speaker clustering tree rather than parameter training and hence achieves fast adaptation using only a small amount of training data. This speaker adaptation method was applied to a hidden Markov network (HMnet) and evaluated in Japanese phoneme and phrase recognition experiments, in which it significantly outperformed speaker-independent recognition methods. In the phrase recognition experiments, the method reduced the error rate by 26.6% using three phrase utterances (approximately 2.7 seconds).<<ETX>>","PeriodicalId":290798,"journal":{"name":"Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124884374","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-04-19DOI: 10.1109/ICASSP.1994.389977
J. Tsimbinos, K. Lever
Nonlinear systems usually cause spectral spreading resulting in an output signal bandwidth that is greater than the input signal bandwidth. When identifying and compensating such systems by digital processing methods, it has been common practice to see the sampling frequency at the Nyquist rate of the output signal. The aim of this paper is to show that sampling at the Nyquist rate of the output signal is usually not necessary, and that a nonlinear system can be identified and compensated at the Nyquist rate of the input signal. We do this by invoking Zhu's (see IEEE Trans. on Circuits and Systems-II: Analog and Digital Signal Processing., vol.39, no.8, p.587-588, 1992) generalised sampling theorem, and by giving three examples of nonlinear system identification and compensation. The first two examples involve known nonlinearities, the first memoryless, the second with memory. The third example deals with real data from an unknown nonlinearity in a radio frequency amplifier. For each example, identification and compensation are carried out for two input signal bandwidths, one causing the distortion terms of interest to be aliased, while for the other, they are not. The results show successful identification and compensation in both cases.<>
{"title":"Sampling frequency requirements for identification and compensation of nonlinear systems","authors":"J. Tsimbinos, K. Lever","doi":"10.1109/ICASSP.1994.389977","DOIUrl":"https://doi.org/10.1109/ICASSP.1994.389977","url":null,"abstract":"Nonlinear systems usually cause spectral spreading resulting in an output signal bandwidth that is greater than the input signal bandwidth. When identifying and compensating such systems by digital processing methods, it has been common practice to see the sampling frequency at the Nyquist rate of the output signal. The aim of this paper is to show that sampling at the Nyquist rate of the output signal is usually not necessary, and that a nonlinear system can be identified and compensated at the Nyquist rate of the input signal. We do this by invoking Zhu's (see IEEE Trans. on Circuits and Systems-II: Analog and Digital Signal Processing., vol.39, no.8, p.587-588, 1992) generalised sampling theorem, and by giving three examples of nonlinear system identification and compensation. The first two examples involve known nonlinearities, the first memoryless, the second with memory. The third example deals with real data from an unknown nonlinearity in a radio frequency amplifier. For each example, identification and compensation are carried out for two input signal bandwidths, one causing the distortion terms of interest to be aliased, while for the other, they are not. The results show successful identification and compensation in both cases.<<ETX>>","PeriodicalId":290798,"journal":{"name":"Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125079599","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-04-19DOI: 10.1109/ICASSP.1994.389462
Jungwoo Lee, B. Dickinson
We present a new subband video coding algorithm with temporally adaptive motion interpolation. In the approach proposed, the reference frames for motion estimation are adaptively selected using temporal segmentation in the lowest spatial subband. Variable target bit allocation for each picture type in a group of pictures is used to allow variable number of reference frames with the constraint of constant output bit rate. Block-wise DPCM, PCM, and run-length coding combined with truncated Huffman coding are used to encode the quantized data in the subbands. Simulation results of the adaptive scheme compare favorably with those of a non-adaptive scheme.<>
{"title":"Subband video coding with temporally adaptive motion interpolation","authors":"Jungwoo Lee, B. Dickinson","doi":"10.1109/ICASSP.1994.389462","DOIUrl":"https://doi.org/10.1109/ICASSP.1994.389462","url":null,"abstract":"We present a new subband video coding algorithm with temporally adaptive motion interpolation. In the approach proposed, the reference frames for motion estimation are adaptively selected using temporal segmentation in the lowest spatial subband. Variable target bit allocation for each picture type in a group of pictures is used to allow variable number of reference frames with the constraint of constant output bit rate. Block-wise DPCM, PCM, and run-length coding combined with truncated Huffman coding are used to encode the quantized data in the subbands. Simulation results of the adaptive scheme compare favorably with those of a non-adaptive scheme.<<ETX>>","PeriodicalId":290798,"journal":{"name":"Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing","volume":"v 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129401286","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-04-19DOI: 10.1109/ICASSP.1994.389470
Y. Chiu, S. Yau
This paper considers the reconstruction of time-varying object by computer tomography. In practice, projections at different directions are measured in sequence. Thus, when the object is time-varying, the projections at different directions are obtained at different time and will not correspond to those of the same distribution. The reconstructed images will therefore have motion artifacts. A new image reconstruction method based on a priori knowledge of the projections is introduced to solve this problem. A novel iterative algorithm is developed and its validity is demonstrated by computer simulation results.<>
{"title":"Tomographic reconstruction of time-varying object from linear time-sequential sampled projections","authors":"Y. Chiu, S. Yau","doi":"10.1109/ICASSP.1994.389470","DOIUrl":"https://doi.org/10.1109/ICASSP.1994.389470","url":null,"abstract":"This paper considers the reconstruction of time-varying object by computer tomography. In practice, projections at different directions are measured in sequence. Thus, when the object is time-varying, the projections at different directions are obtained at different time and will not correspond to those of the same distribution. The reconstructed images will therefore have motion artifacts. A new image reconstruction method based on a priori knowledge of the projections is introduced to solve this problem. A novel iterative algorithm is developed and its validity is demonstrated by computer simulation results.<<ETX>>","PeriodicalId":290798,"journal":{"name":"Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing","volume":"53 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129410620","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-04-19DOI: 10.1109/ICASSP.1994.389986
Mu-Huo Cheng, V. Stonick
This paper presents a unified approach to analyze the convergence properties of Steiglitz-McBride(1966) method (SMM) in general environments. SMM is formulated as a successive substitution equation. Using results from fixed point theory enables a unified analysis of SMM in both white and colored noise, and sufficient and insufficient order cases. This analysis provides us with several new results. Specifically, for sufficient order filters in white noise environments, the convergence rate of SMM can be predicted by the signal-power to noise-power ratio (SNR) at plant output. For sufficient order filters in colored noise, SMM may diverge or converge depending on the initial estimate and SNR at plant output. If SMM converges, the convergence point is near the unbiased solution. SNR again determines the bias magnitude. For insufficient order filters, in addition to the possible multiple convergence points, we also demonstrate the existence of diverging fixed points of SMM. These diverging fixed points can be used to separate the convergence region, and identify the convergence points for each initial estimate.<>
{"title":"Convergence, convergence point and convergence rate for Steiglitz-McBride method; a unified approach","authors":"Mu-Huo Cheng, V. Stonick","doi":"10.1109/ICASSP.1994.389986","DOIUrl":"https://doi.org/10.1109/ICASSP.1994.389986","url":null,"abstract":"This paper presents a unified approach to analyze the convergence properties of Steiglitz-McBride(1966) method (SMM) in general environments. SMM is formulated as a successive substitution equation. Using results from fixed point theory enables a unified analysis of SMM in both white and colored noise, and sufficient and insufficient order cases. This analysis provides us with several new results. Specifically, for sufficient order filters in white noise environments, the convergence rate of SMM can be predicted by the signal-power to noise-power ratio (SNR) at plant output. For sufficient order filters in colored noise, SMM may diverge or converge depending on the initial estimate and SNR at plant output. If SMM converges, the convergence point is near the unbiased solution. SNR again determines the bias magnitude. For insufficient order filters, in addition to the possible multiple convergence points, we also demonstrate the existence of diverging fixed points of SMM. These diverging fixed points can be used to separate the convergence region, and identify the convergence points for each initial estimate.<<ETX>>","PeriodicalId":290798,"journal":{"name":"Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129464215","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-04-19DOI: 10.1109/ICASSP.1994.389999
Terence Wang, Chin-Liang Wang
We present a new 2-D optimum block stochastic gradient (TDOBSG) algorithm for 2-D adaptive finite impulse response (FIR) filtering. Unlike the 2-D optimum block adaptive (TDOBA) algorithm derived from a truncated Taylor's series expansion, which is in fact a suboptimum one, the TDOBSG algorithm exactly minimizes the squared norm of the a posteriori estimation error vector in a given block by optimally choosing the convergence factor of the adaptive filter. The optimum convergence factor can be computed from input signals at the same order of computational complexity as that of the TDOBA algorithm. Computer simulations based on the configuration of adaptive image noise cancellation show that the TDOBSG algorithm has better convergence speed and accuracy than those of the TDOBA algorithm.<>
{"title":"A new two-dimensional block adaptive FIR filtering algorithm","authors":"Terence Wang, Chin-Liang Wang","doi":"10.1109/ICASSP.1994.389999","DOIUrl":"https://doi.org/10.1109/ICASSP.1994.389999","url":null,"abstract":"We present a new 2-D optimum block stochastic gradient (TDOBSG) algorithm for 2-D adaptive finite impulse response (FIR) filtering. Unlike the 2-D optimum block adaptive (TDOBA) algorithm derived from a truncated Taylor's series expansion, which is in fact a suboptimum one, the TDOBSG algorithm exactly minimizes the squared norm of the a posteriori estimation error vector in a given block by optimally choosing the convergence factor of the adaptive filter. The optimum convergence factor can be computed from input signals at the same order of computational complexity as that of the TDOBA algorithm. Computer simulations based on the configuration of adaptive image noise cancellation show that the TDOBSG algorithm has better convergence speed and accuracy than those of the TDOBA algorithm.<<ETX>>","PeriodicalId":290798,"journal":{"name":"Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing","volume":"103 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128314228","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-04-19DOI: 10.1109/ICASSP.1994.389989
D. Lun
The traditional approach in realizing the prime factor discrete cosine transform (PFDCT) often suffers from two problems. First, although only the Ruritanian mapping is used for input indexing, it requires to perform a series of complicated tests and additions which even outweigh the computational effort of the PFDCT. Second, the additions mentioned above are not carried out in an in-place form. This implies that an auxiliary data array is required to buffer the temporary results generated during the additions. Otherwise, erroneous results will be obtained. We propose an efficient indexing scheme for the computation of the PFDCT. By suitably swapping the data, all the additions can be carried out in an in-place form. Furthermore the number of tests required to perform on the indices of the data is greatly reduced. They are achieved by considering the special properties of the Ruritanian mapping.<>
{"title":"On efficient software realization of the prime factor discrete cosine transform","authors":"D. Lun","doi":"10.1109/ICASSP.1994.389989","DOIUrl":"https://doi.org/10.1109/ICASSP.1994.389989","url":null,"abstract":"The traditional approach in realizing the prime factor discrete cosine transform (PFDCT) often suffers from two problems. First, although only the Ruritanian mapping is used for input indexing, it requires to perform a series of complicated tests and additions which even outweigh the computational effort of the PFDCT. Second, the additions mentioned above are not carried out in an in-place form. This implies that an auxiliary data array is required to buffer the temporary results generated during the additions. Otherwise, erroneous results will be obtained. We propose an efficient indexing scheme for the computation of the PFDCT. By suitably swapping the data, all the additions can be carried out in an in-place form. Furthermore the number of tests required to perform on the indices of the data is greatly reduced. They are achieved by considering the special properties of the Ruritanian mapping.<<ETX>>","PeriodicalId":290798,"journal":{"name":"Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128526730","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-04-19DOI: 10.1109/ICASSP.1994.389915
R. K. Jarrott, T. A. Soame
This paper presents an overview of a significant and computationally demanding application of signal processing, viz. over-the-horizon radar (OTHR). The Jindalee Operational Radar Network is an integrated network of OTHRs which depend heavily for their performance, in both air and surface mode, on signal processing. It outlines the characteristics of OTHR data that need to be recognised in the design of its processing algorithms, and then describes the chosen signal processing techniques and the implementation architecture. Comment is made, where underlying assumptions in the theory are not valid and a view is given on where additional development in OTHR signal processing is required.<>
{"title":"The processing of HF skywave radar signals","authors":"R. K. Jarrott, T. A. Soame","doi":"10.1109/ICASSP.1994.389915","DOIUrl":"https://doi.org/10.1109/ICASSP.1994.389915","url":null,"abstract":"This paper presents an overview of a significant and computationally demanding application of signal processing, viz. over-the-horizon radar (OTHR). The Jindalee Operational Radar Network is an integrated network of OTHRs which depend heavily for their performance, in both air and surface mode, on signal processing. It outlines the characteristics of OTHR data that need to be recognised in the design of its processing algorithms, and then describes the chosen signal processing techniques and the implementation architecture. Comment is made, where underlying assumptions in the theory are not valid and a view is given on where additional development in OTHR signal processing is required.<<ETX>>","PeriodicalId":290798,"journal":{"name":"Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing","volume":"48 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128261293","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1994-04-19DOI: 10.1109/ICASSP.1994.389846
R. S. Roberts, H. Loomis
Real-time cyclic spectral analysis is useful in many applications, but is difficult to achieve because of its computational complexity. This paper studies the distribution of complex multipliers in multiprocessor cyclic spectrum analyzers, with the objective of obtaining computational balance. Computationally balanced implementations efficiently use hardware so that computational bottlenecks are reduced and a smooth flow of data between computational sections of the analyzer is maintained. Tables are presented that give the number of complex multipliers required in each section of the analyzer to obtain computational balance.<>
{"title":"Computational balance in real-time cyclic spectral analysis","authors":"R. S. Roberts, H. Loomis","doi":"10.1109/ICASSP.1994.389846","DOIUrl":"https://doi.org/10.1109/ICASSP.1994.389846","url":null,"abstract":"Real-time cyclic spectral analysis is useful in many applications, but is difficult to achieve because of its computational complexity. This paper studies the distribution of complex multipliers in multiprocessor cyclic spectrum analyzers, with the objective of obtaining computational balance. Computationally balanced implementations efficiently use hardware so that computational bottlenecks are reduced and a smooth flow of data between computational sections of the analyzer is maintained. Tables are presented that give the number of complex multipliers required in each section of the analyzer to obtain computational balance.<<ETX>>","PeriodicalId":290798,"journal":{"name":"Proceedings of ICASSP '94. IEEE International Conference on Acoustics, Speech and Signal Processing","volume":"iv 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1994-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129354065","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：