Paulo Diniz | Universidade Federal do Rio de Janeiro (UFRJ) (original) (raw)
Papers by Paulo Diniz
Online Learning and Adaptive Filters
Learn to solve the unprecedented challenges facing Online Learning and Adaptive Signal Processing... more Learn to solve the unprecedented challenges facing Online Learning and Adaptive Signal Processing in this concise, intuitive text. The ever-increasing amount of data generated every day requires new strategies to tackle issues such as: combining data from a large number of sensors; improving spectral usage, utilizing multiple-antennas with adaptive capabilities; or learning from signals placed on graphs, generating unstructured data. Solutions to all of these and more are described in a condensed and unified way, enabling you to expose valuable information from data and signals in a fast and economical way. The up-to-date techniques explained here can be implemented in simple electronic hardware, or as part of multi-purpose systems. Also featuring alternative explanations for online learning, including newly developed methods and data selection, and several easily implemented algorithms, this one-of-a-kind book is an ideal resource for graduate students, researchers, and professiona...
arXiv (Cornell University), Dec 8, 2020
A unified matrix formulation is presented for the analysis of intersymbol and intercarrier interf... more A unified matrix formulation is presented for the analysis of intersymbol and intercarrier interference in orthogonal frequency-division multiplexing (OFDM) systems. The proposed formulation relies on six parameters and allows studying various schemes, including those with windowing in the transmitter and/or in the receiver (called windowed OFDM systems), which may add cyclic suffix and/or cyclic prefix (CP), besides the conventional CP-OFDM. The proposed framework encompasses seven different OFDM systems. It considers the overlap-and-add procedure performed in the transmitter of windowed OFDM systems, being jointly formulated with the channel convolution. The intersymbol and intercarrier interference, caused when the order of the channel impulse response is higher than the number of CP samples, is characterized. A new equivalent channel matrix that is useful for calculating both the received signal and the interference power is defined and characterized. Unlike previous works, this new channel matrix has no restrictions on the length of the channel impulse response, which means that the study is not constrained to the particular case of two or three data blocks interfering in the received signal. Theoretical expressions for the powers of three different kinds of interference are derived. These expressions allow calculating the signal-to-interference-plus-noise ratio, useful for computing the data rate of each OFDM system. The proposed formulation is applied to realistic examples, showing its effectiveness through comparisons based on numerical performance assessments of the considered OFDM systems.
Foundations and Trends® in Signal Processing
2021 29th European Signal Processing Conference (EUSIPCO), 2021
As the volume of data keeps growing, the use of deep neural networks has been widespread in a var... more As the volume of data keeps growing, the use of deep neural networks has been widespread in a variety of applications, including image classification. This big available data has led to an increasing interest in designing more efficient systems. Most applications use all training data without taking into account their relevance. A mini-batch gradient descent algorithm is preferable in practice, but the chosen batch size is typically based on empirical tests, and it depends on the dataset characteristics. This work proposes a data-selection strategy applied to classification problems leading to computational savings and, in most cases classification error reduction. A few examples corroborate the effectiveness of the proposed approach.
This first volume, edited and authored by world leading experts, gives a review of the principles... more This first volume, edited and authored by world leading experts, gives a review of the principles, methods and techniques of important and emerging research topics and technologies in machine learning and advanced signal processing theory. With this reference source you will: * Quickly grasp a new area of research * Understand the underlying principles of a topic and its application* Ascertain how a topic relates to other areas and learn of the research issues yet to be resolved * Quick tutorial reviews of important and emerging topics of research in machine learning* Presents core principles in signal processing theory and shows their applications* Reference content on core principles, technologies, algorithms and applications * Comprehensive references to journal articles and other literature on which to build further, more specific and detailed knowledge * Edited by leading people in the field who, through their reputation, have been able to commission experts to write on a parti...
Cambridge University Press eBooks, Jun 15, 2012
Anais do XXII Simpósio Brasileiro de Telecomunicações
Resumo-Este trabalho investiga o conteúdo tempo-freqüência de frames. Iniciamos mostrando que a s... more Resumo-Este trabalho investiga o conteúdo tempo-freqüência de frames. Iniciamos mostrando que a soma dos conteúdos tempofreqüenciais de todos os elementos de um conjunto de funções ser positiva é uma condição suficiente para que este conjunto gere um frame em L 2 (R). A seguir deriva-se que para frames de Weyl-Heisenberg {E mb Tnag(t)} n,m∈Z gerados a partir de uma função par g(t) os máximos e mínimos de seu conteúdo tempo-freqüência encontram-se em (na, mb) e (na + a/2, mb + b/2), respectivamente; e que para g(t) ímpar teremos os máximos localizados em (na, mb + b/2) e os mínimos em (na + a/2, mb). Estes resultados fornecem uma forma efetiva de gerar frames mais apertados por entrelaçamento, ao custo dobrar a cardinalidade dos frames. Os frames construídos pela abordagem apresentada são avaliados para utilização como dicionários em decomposições vorazes de sinais.
6th International Congress on Sound and Vibration, ICSV6, 1999
2018 26th European Signal Processing Conference (EUSIPCO), 2018
The conjugate gradient (CG) adaptive filtering algorithm is an alternative to the more widely use... more The conjugate gradient (CG) adaptive filtering algorithm is an alternative to the more widely used Recursive Least Squares (RLS) and Least Mean Square (LMS) algorithms, where the former requires more computations, and the latter leads to slower convergence. In recent years, some adaptive filtering algorithms have been equipped with data selection mechanism to classify if the data currently available consists of an outlier or if it brings about enough innovation. In both cases the data could be discarded avoiding extra computation and performance degradation. This paper proposes a data selection strategy to the CG algorithm and verifies its effectiveness in simulations utilizing synthetic and real data.
2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2017
In this paper, we propose some sparsity aware algorithms, namely the Recursive least-Squares for ... more In this paper, we propose some sparsity aware algorithms, namely the Recursive least-Squares for sparse systems (S-RLS) and l0-norm Recursive least-Squares (l0-RLS), in order to exploit the sparsity of an unknown system. The first algorithm, applies a discard function on the weight vector to disregard the coefficients close to zero during the update process. The second algorithm, employs the sparsity-promoting scheme via some non-convex approximations to the l0-norm. In addition, we consider the respective versions of these algorithms in data-selective versions in order to reduce the update rate. Simulation results show similar performance when comparing the proposed algorithms with standard Recursive Least-Squares (RLS) algorithm while the proposed algorithms require lower computational complexity.
2021 IEEE Statistical Signal Processing Workshop (SSP), 2021
Orthogonal frequency-division multiplexing (OFDM) systems have championed the elimination of inte... more Orthogonal frequency-division multiplexing (OFDM) systems have championed the elimination of inter-symbol interference (ISI) and inter-block interference (IBI) originated from multi-path fading. By introducing some redundant symbols at the transmitter such as zero padding (ZP), spectral efficiency is reduced. The amount of redundancy is related to the channel-model order, an information carrying some uncertainty in practical situations, particularly when one is willing to increase data transmission. The recent trend of utilizing neural networks to address some communication issues sparkled the idea of exploiting machine-learning (ML) to improve the performance of ZP-OFDM transceivers whenever the channel order is not known. This work presents a novel application of ML to address ZP-OFDM physical layer issues. The simulation results show that the ML ZP-OFDM brings about performance improvements, such as reduced bit-error-rate (BER), when the amount of redundancy is insufficient and some form of nonlinearity is present at the transmitter end.
2017 IEEE Global Conference on Signal and Information Processing (GlobalSIP), 2017
Anais do XXVI Simpósio Brasileiro de Telecomunicações, 2008
Resumo-Este trabalho analisa a influência da técnica SAIC (Single-Antenna Interference Cancellati... more Resumo-Este trabalho analisa a influência da técnica SAIC (Single-Antenna Interference Cancellation) de cancelamento de interferência cocanal sobre os parâmetros dos algoritmos de controle dinâmico de potência e de adaptação do enlace de rádio, em uma conexão downlink do GPRS. Com o desenvolvimento de novas técnicas, torna-se comum a utilização de algoritmos com parâmetros que foram otimizados sob condições diferentes das efetivamente utilizadas. Devidoà alta complexidade de um sistema de comunicação móvel, isto pode interferir negativa ou positivamente nas métricas de desempenho para outras partes constituintes do sistema. Com o intuito de verificar a influência do SAIC sobre outros algoritmos, implementamos um simulador da camada de enlace do sistema GSM/GPRS. Os resultados obtidos indicam que podemos ajustar os parâmetros do algoritmo de controle de potência para obter uma economia de potência, quando utilizamos o SAIC. Verificamos também que o throughput resultante pode ser maior no SAIC se ajustarmos devidamente o algoritmo de adaptação do enlace de rádio. Essas conclusões são válidas somente em ambientes cuja interferência predominante seja a CCI.
IEEE Transactions on Wireless Communications, 2019
As wireless services proliferate, the demand for available spectrum also grows. As a result, spec... more As wireless services proliferate, the demand for available spectrum also grows. As a result, spectral efficiency is still an issue being addressed by many researchers aiming at improving the quality of service to a growing number of users. Massive multiple-input multiple-output (MIMO) has been presented as an attractive technology for the next wireless systems since it can alleviate the expected spectral shortage. Nevertheless, such a technique requires a dedicated chain of radio frequency (RF) components for each antenna element which result in high costs at base station (BS) side. To reduce the number of RF chains, we propose several transmit antenna selection schemes aiming at minimizing the mean square reception error and also reducing the transmission power which is one of the main contributions of our work. The proposed strategies are inspired by the matching pursuit technique and its quantized version, named matching pursuit with generalized bit planes. The presented results show that reliable reception can be accomplished with low computationally intensive algorithms for antenna selection.
IEEE Transactions on Circuits and Systems II: Express Briefs, 2017
This letter introduces new data selective adaptive filtering algorithms for trinion and quaternio... more This letter introduces new data selective adaptive filtering algorithms for trinion and quaternion spaces T and H. The work advances the set-membership trinion and quaternion-valued normalized least mean square (SMTNLMS and SMQNLMS) and the set-membership trinion and quaternion-valued affine projection (SMTAP and SMQAP) algorithms. We derive set-membership trinion algorithms and then, as special cases, obtain trinion algorithms not employing the set-membership strategy. Prediction simulations based on recorded wind data are provided, showing the improved performance of the proposed algorithms in terms of reduced computational complexity. Then the quaternion based SMQAP and SMQNLMS algorithms are derived and their improved performances are verified in an adaptive beamforming problem.
Cambridge University Press eBooks, Jun 15, 2012
Cambridge University Press eBooks, Jun 15, 2012
Online Learning and Adaptive Filters
Learn to solve the unprecedented challenges facing Online Learning and Adaptive Signal Processing... more Learn to solve the unprecedented challenges facing Online Learning and Adaptive Signal Processing in this concise, intuitive text. The ever-increasing amount of data generated every day requires new strategies to tackle issues such as: combining data from a large number of sensors; improving spectral usage, utilizing multiple-antennas with adaptive capabilities; or learning from signals placed on graphs, generating unstructured data. Solutions to all of these and more are described in a condensed and unified way, enabling you to expose valuable information from data and signals in a fast and economical way. The up-to-date techniques explained here can be implemented in simple electronic hardware, or as part of multi-purpose systems. Also featuring alternative explanations for online learning, including newly developed methods and data selection, and several easily implemented algorithms, this one-of-a-kind book is an ideal resource for graduate students, researchers, and professiona...
arXiv (Cornell University), Dec 8, 2020
A unified matrix formulation is presented for the analysis of intersymbol and intercarrier interf... more A unified matrix formulation is presented for the analysis of intersymbol and intercarrier interference in orthogonal frequency-division multiplexing (OFDM) systems. The proposed formulation relies on six parameters and allows studying various schemes, including those with windowing in the transmitter and/or in the receiver (called windowed OFDM systems), which may add cyclic suffix and/or cyclic prefix (CP), besides the conventional CP-OFDM. The proposed framework encompasses seven different OFDM systems. It considers the overlap-and-add procedure performed in the transmitter of windowed OFDM systems, being jointly formulated with the channel convolution. The intersymbol and intercarrier interference, caused when the order of the channel impulse response is higher than the number of CP samples, is characterized. A new equivalent channel matrix that is useful for calculating both the received signal and the interference power is defined and characterized. Unlike previous works, this new channel matrix has no restrictions on the length of the channel impulse response, which means that the study is not constrained to the particular case of two or three data blocks interfering in the received signal. Theoretical expressions for the powers of three different kinds of interference are derived. These expressions allow calculating the signal-to-interference-plus-noise ratio, useful for computing the data rate of each OFDM system. The proposed formulation is applied to realistic examples, showing its effectiveness through comparisons based on numerical performance assessments of the considered OFDM systems.
Foundations and Trends® in Signal Processing
2021 29th European Signal Processing Conference (EUSIPCO), 2021
As the volume of data keeps growing, the use of deep neural networks has been widespread in a var... more As the volume of data keeps growing, the use of deep neural networks has been widespread in a variety of applications, including image classification. This big available data has led to an increasing interest in designing more efficient systems. Most applications use all training data without taking into account their relevance. A mini-batch gradient descent algorithm is preferable in practice, but the chosen batch size is typically based on empirical tests, and it depends on the dataset characteristics. This work proposes a data-selection strategy applied to classification problems leading to computational savings and, in most cases classification error reduction. A few examples corroborate the effectiveness of the proposed approach.
This first volume, edited and authored by world leading experts, gives a review of the principles... more This first volume, edited and authored by world leading experts, gives a review of the principles, methods and techniques of important and emerging research topics and technologies in machine learning and advanced signal processing theory. With this reference source you will: * Quickly grasp a new area of research * Understand the underlying principles of a topic and its application* Ascertain how a topic relates to other areas and learn of the research issues yet to be resolved * Quick tutorial reviews of important and emerging topics of research in machine learning* Presents core principles in signal processing theory and shows their applications* Reference content on core principles, technologies, algorithms and applications * Comprehensive references to journal articles and other literature on which to build further, more specific and detailed knowledge * Edited by leading people in the field who, through their reputation, have been able to commission experts to write on a parti...
Cambridge University Press eBooks, Jun 15, 2012
Anais do XXII Simpósio Brasileiro de Telecomunicações
Resumo-Este trabalho investiga o conteúdo tempo-freqüência de frames. Iniciamos mostrando que a s... more Resumo-Este trabalho investiga o conteúdo tempo-freqüência de frames. Iniciamos mostrando que a soma dos conteúdos tempofreqüenciais de todos os elementos de um conjunto de funções ser positiva é uma condição suficiente para que este conjunto gere um frame em L 2 (R). A seguir deriva-se que para frames de Weyl-Heisenberg {E mb Tnag(t)} n,m∈Z gerados a partir de uma função par g(t) os máximos e mínimos de seu conteúdo tempo-freqüência encontram-se em (na, mb) e (na + a/2, mb + b/2), respectivamente; e que para g(t) ímpar teremos os máximos localizados em (na, mb + b/2) e os mínimos em (na + a/2, mb). Estes resultados fornecem uma forma efetiva de gerar frames mais apertados por entrelaçamento, ao custo dobrar a cardinalidade dos frames. Os frames construídos pela abordagem apresentada são avaliados para utilização como dicionários em decomposições vorazes de sinais.
6th International Congress on Sound and Vibration, ICSV6, 1999
2018 26th European Signal Processing Conference (EUSIPCO), 2018
The conjugate gradient (CG) adaptive filtering algorithm is an alternative to the more widely use... more The conjugate gradient (CG) adaptive filtering algorithm is an alternative to the more widely used Recursive Least Squares (RLS) and Least Mean Square (LMS) algorithms, where the former requires more computations, and the latter leads to slower convergence. In recent years, some adaptive filtering algorithms have been equipped with data selection mechanism to classify if the data currently available consists of an outlier or if it brings about enough innovation. In both cases the data could be discarded avoiding extra computation and performance degradation. This paper proposes a data selection strategy to the CG algorithm and verifies its effectiveness in simulations utilizing synthetic and real data.
2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2017
In this paper, we propose some sparsity aware algorithms, namely the Recursive least-Squares for ... more In this paper, we propose some sparsity aware algorithms, namely the Recursive least-Squares for sparse systems (S-RLS) and l0-norm Recursive least-Squares (l0-RLS), in order to exploit the sparsity of an unknown system. The first algorithm, applies a discard function on the weight vector to disregard the coefficients close to zero during the update process. The second algorithm, employs the sparsity-promoting scheme via some non-convex approximations to the l0-norm. In addition, we consider the respective versions of these algorithms in data-selective versions in order to reduce the update rate. Simulation results show similar performance when comparing the proposed algorithms with standard Recursive Least-Squares (RLS) algorithm while the proposed algorithms require lower computational complexity.
2021 IEEE Statistical Signal Processing Workshop (SSP), 2021
Orthogonal frequency-division multiplexing (OFDM) systems have championed the elimination of inte... more Orthogonal frequency-division multiplexing (OFDM) systems have championed the elimination of inter-symbol interference (ISI) and inter-block interference (IBI) originated from multi-path fading. By introducing some redundant symbols at the transmitter such as zero padding (ZP), spectral efficiency is reduced. The amount of redundancy is related to the channel-model order, an information carrying some uncertainty in practical situations, particularly when one is willing to increase data transmission. The recent trend of utilizing neural networks to address some communication issues sparkled the idea of exploiting machine-learning (ML) to improve the performance of ZP-OFDM transceivers whenever the channel order is not known. This work presents a novel application of ML to address ZP-OFDM physical layer issues. The simulation results show that the ML ZP-OFDM brings about performance improvements, such as reduced bit-error-rate (BER), when the amount of redundancy is insufficient and some form of nonlinearity is present at the transmitter end.
2017 IEEE Global Conference on Signal and Information Processing (GlobalSIP), 2017
Anais do XXVI Simpósio Brasileiro de Telecomunicações, 2008
Resumo-Este trabalho analisa a influência da técnica SAIC (Single-Antenna Interference Cancellati... more Resumo-Este trabalho analisa a influência da técnica SAIC (Single-Antenna Interference Cancellation) de cancelamento de interferência cocanal sobre os parâmetros dos algoritmos de controle dinâmico de potência e de adaptação do enlace de rádio, em uma conexão downlink do GPRS. Com o desenvolvimento de novas técnicas, torna-se comum a utilização de algoritmos com parâmetros que foram otimizados sob condições diferentes das efetivamente utilizadas. Devidoà alta complexidade de um sistema de comunicação móvel, isto pode interferir negativa ou positivamente nas métricas de desempenho para outras partes constituintes do sistema. Com o intuito de verificar a influência do SAIC sobre outros algoritmos, implementamos um simulador da camada de enlace do sistema GSM/GPRS. Os resultados obtidos indicam que podemos ajustar os parâmetros do algoritmo de controle de potência para obter uma economia de potência, quando utilizamos o SAIC. Verificamos também que o throughput resultante pode ser maior no SAIC se ajustarmos devidamente o algoritmo de adaptação do enlace de rádio. Essas conclusões são válidas somente em ambientes cuja interferência predominante seja a CCI.
IEEE Transactions on Wireless Communications, 2019
As wireless services proliferate, the demand for available spectrum also grows. As a result, spec... more As wireless services proliferate, the demand for available spectrum also grows. As a result, spectral efficiency is still an issue being addressed by many researchers aiming at improving the quality of service to a growing number of users. Massive multiple-input multiple-output (MIMO) has been presented as an attractive technology for the next wireless systems since it can alleviate the expected spectral shortage. Nevertheless, such a technique requires a dedicated chain of radio frequency (RF) components for each antenna element which result in high costs at base station (BS) side. To reduce the number of RF chains, we propose several transmit antenna selection schemes aiming at minimizing the mean square reception error and also reducing the transmission power which is one of the main contributions of our work. The proposed strategies are inspired by the matching pursuit technique and its quantized version, named matching pursuit with generalized bit planes. The presented results show that reliable reception can be accomplished with low computationally intensive algorithms for antenna selection.
IEEE Transactions on Circuits and Systems II: Express Briefs, 2017
This letter introduces new data selective adaptive filtering algorithms for trinion and quaternio... more This letter introduces new data selective adaptive filtering algorithms for trinion and quaternion spaces T and H. The work advances the set-membership trinion and quaternion-valued normalized least mean square (SMTNLMS and SMQNLMS) and the set-membership trinion and quaternion-valued affine projection (SMTAP and SMQAP) algorithms. We derive set-membership trinion algorithms and then, as special cases, obtain trinion algorithms not employing the set-membership strategy. Prediction simulations based on recorded wind data are provided, showing the improved performance of the proposed algorithms in terms of reduced computational complexity. Then the quaternion based SMQAP and SMQNLMS algorithms are derived and their improved performances are verified in an adaptive beamforming problem.
Cambridge University Press eBooks, Jun 15, 2012
Cambridge University Press eBooks, Jun 15, 2012