A Neural Network (NN) based realization to obtain energy concentration along Instantaneous Freque... more A Neural Network (NN) based realization to obtain energy concentration along Instantaneous Frequencies (IFs) is proposed by compensating linear and non linear distortions present in Time Frequency Representations (TFRs). Blurry spectrograms and highly concentrated Wigner Distributions (WDs) of various signals constitute the training set. The input data is grouped according to some underlying feature present in TFR image to have better generalization ability of the trained NN. Blurry TFRs of multi component signals are then given as test cases to the trained NN. Effectiveness of the approach is established by comparing the information content in each input & resultant TFR.
In this paper we present a comparison of Neural Network Training Algorithms for obtaining a Time ... more In this paper we present a comparison of Neural Network Training Algorithms for obtaining a Time Frequency Distribution (TFD) of a signal whose frequency components vary with time. The method employs various algorithms used in NNs which are trained by using the spectrograms of several training signals as input and TFDs that are highly concentrated along the instantaneous frequencies (IFs) of the individual components present in the signal as targets. The trained neural networks are then presented with the spectrogram of unknown signals. We compute the entropy as a measure of the result obtained and carry out error and time analysis to compare the performance of algorithms used.
In this work, equalization for nonlinear dispersive channels is considered. Nonlinear communicati... more In this work, equalization for nonlinear dispersive channels is considered. Nonlinear communication channels can lead to significant degradations when nonlinearities are not taken into account at either the receiver or the transmitter. In many cases, the nonlinearity of the channel precludes the use of spectrally efficient signaling schemes to achieve high data-rates and the bandwidth efficiency. Satellite channel is a typical case of nonlinear channel that needs to be used efficiently. We develop two novel equalization strategies for a general class of nonlinear channels. Both strategies are based on iterating between decoding and equalization, termed as iterative equalization. The first strategy is a factor graph based equalizer that converts the nonlinear channel equalization problem into forward-backward algorithm on hidden Markov model (HMM). The equalizer is implemented via the sum-product algorithm on the factor graph representation of the channel and receiver blocks. The sec...
Traditional patient monitoring may not detect cerebral tissue hypoxia, and typical interventions ... more Traditional patient monitoring may not detect cerebral tissue hypoxia, and typical interventions may not improve tissue oxygenation. Therefore, monitoring cerebral tissue oxygen status with regional oximetry is being increasingly used by anesthesiologists and perfusionists during surgery. In this study, we evaluated absolute and trend accuracy of a new regional oximetry technology in healthy volunteers. A near-infrared spectroscopy sensor connected to a regional oximetry system (O3™, Masimo, Irvine, CA) was placed on the subject's forehead, to provide continuous measurement of regional oxygen saturation (rSO2). Reference blood samples were taken from the radial artery and internal jugular bulb vein, at baseline and after a series of increasingly hypoxic states induced by altering the inspired oxygen concentration while maintaining normocapnic arterial carbon dioxide pressure (PaCO2). Absolute and trend accuracy of the regional oximetry system was determined by comparing rSO2 against reference cerebral oxygen saturation (SavO2), that is calculated by combining arterial and venous saturations of oxygen in the blood samples. Twenty-seven subjects were enrolled. Bias (test method mean error), standard deviation of error, standard error of the mean, and root mean square accuracy (ARMS) of rSO2 compared to SavO2 were 0.4%, 4.0%, 0.3%, and 4.0%, respectively. The limits of agreement were 8.4% (95% confidence interval, 7.6%-9.3%) to -7.6% (95% confidence interval, -8.4% to -6.7%). Trend accuracy analysis yielded a relative mean error of 0%, with a standard deviation of 2.1%, a standard error of 0.1%, and an ARMS of 2.1%. Multiple regression analysis showed that age and skin color did not affect the bias (all P > 0.1). Masimo O3 regional oximetry provided absolute root-mean-squared error of 4% and relative root-mean-squared error of 2.1% in healthy volunteers undergoing controlled hypoxia.
Oxidative stress during fetal development, delivery, or early postnatal life is a major cause of ... more Oxidative stress during fetal development, delivery, or early postnatal life is a major cause of neuropathology, as both hypoxic and hyperoxic insults can significantly damage the developing brain. Despite the obvious need for reliable cerebral oxygenation monitoring, no technology currently exists to monitor cerebral oxygen metabolism continuously and noninvasively in infants at high risk for developing brain injury. Consequently, a rational approach to titrating oxygen supply to cerebral oxygen demand -and thus avoiding hyperoxic or hypoxic insults -is currently lacking. We present a promising method to close this crucial technology gap in the important case of neonates on conventional ventilators. By using cerebral near-infrared spectroscopy (NIRS) and signals from conventional ventilators, along with arterial oxygen saturation, we derive continuous (breath-by-breath) estimates of cerebral venous oxygen saturation, cerebral oxygen extraction fraction, cerebral blood flow, and cerebral metabolic rate of oxygen. The resultant estimates compare very favorably to previously reported data obtained by non-continuous and invasive means from preterm infants in neonatal critical care.
Proceedings. IEEE International Multi Topic Conference, 2001. IEEE INMIC 2001. Technology for the 21st Century., 2001
ABSTRACT Quadrature Amplitude Modulation (QAM) is widely used in many digital communication syste... more ABSTRACT Quadrature Amplitude Modulation (QAM) is widely used in many digital communication systems such as high speed modems. Increasing the data rate by using larger constellations has brought a revolution in digital communication systems. Consequently, there is a need ...
8th International Multitopic Conference, 2004. Proceedings of INMIC 2004., 2004
The use of dual-tone multi frequency (DTMF) signaling is very common in telecommunication systems... more The use of dual-tone multi frequency (DTMF) signaling is very common in telecommunication systems such as telephone exchanges, private branch exchanges (PBXs), and call centers etc. One of the approaches of DTMF detection is based on the cycle estimation of the frequencies present in the DTMF signal. Any interference that lies in the frequency region of the DTMF signal disturbs
Proceedings of the IEEE Symposium on Emerging Technologies, 2005., 2005
In this paper we present a method of obtaining a Time Frequency Distribution (TFD) of a signal wh... more In this paper we present a method of obtaining a Time Frequency Distribution (TFD) of a signal whose frequency components vary with time. The method employs Neural Networks (NN) which are trained by using the spectrograms of several training signals as input and TFDs that are highly concentrated along the instantaneous frequencies of the individual components present in the signal as targets. The trained neural network is then presented with the spectrogram of unknown signals and highly concentrated TFDs are obtained.
Monitoring cerebrovascular state, including intracranial pressure (ICP) and the ability to regula... more Monitoring cerebrovascular state, including intracranial pressure (ICP) and the ability to regulate cerebral blood flow, is important for patient care in stroke, traumatic brain injury and other such conditions. However, current methodologies for direct measurement of ICP are highly invasive, and expose patients to the risk of infection. In addition, vascular properties such as resistance and compliance cannot be directly assessed. In this work, we employ a mathematical model-based approach to track variations in ICP and cerebrovascular properties from signals that can be acquired entirely non-invasively. The performance on simulation data indicates that the estimates track the desired quantities closely, thus suggesting that tests using clinical data are warranted.
2007 IEEE International Conference on Signal Processing and Communications, 2007
A Neural Network (NN) based approach to obtain energy concentration along Instantaneous Frequenci... more A Neural Network (NN) based approach to obtain energy concentration along Instantaneous Frequencies (IFs) of the individual components present in the signal, is proposed. Blurry spectrograms and highly concentrated Wigner Distributions (WDs) of various signals constitute the training set. The input data is grouped according to the underlying feature present in Time Frequency Representation (TFR) image to have better generalization ability of the trained NN. Blurry TFRs of multi component signals are then given as test cases to the trained NN. Effectiveness of the approach is established by comparing the information content in each input and resultant TFR.
2006 IEEE International Multitopic Conference, 2006
Network Architecture, for a Time Frequency application. Varying the number of neurons and hidden ... more Network Architecture, for a Time Frequency application. Varying the number of neurons and hidden layers has been found to greatly affect the performance of Neural Network (NN), trained via various blurry spectrograms as input over highly concentrated Time Frequency Distributions (TFDs) as targets, of the same signals. Number of neurons and hidden layers are varied during training and the impact is observed over test spectrograms of unknown multi component signals. Entropy and Mean Square Error (MSE) is the decision criteria for the most optimum solution.
The Proceedings of the Multiconference on "Computational Engineering in Systems Applications", 2006
A realization of obtaining energy concentration along instantaneous frequency (IF) for blurry Tim... more A realization of obtaining energy concentration along instantaneous frequency (IF) for blurry Time Frequency Images (TFIs) is proposed using Neural Networks (NNs). NN captures the fundamental principle at work once trained with various spectrogram TFIs by defining various edges present in input data. Unknown spectrogram TFIs of multi component signals, are then given as test images to the trained NN. Measuring the information content in each input & resultant TFI thus manifest the effectiveness of NNs in TFI analysis.
2007 International Conference on Information and Emerging Technologies, 2007
In this paper we present advantage of training MNCD for obtaining Time localized Frequencies (als... more In this paper we present advantage of training MNCD for obtaining Time localized Frequencies (also called IF), which is one useful concept for describing the changing spectral structure of a time-varying signal, arising so often in Time Frequency Distribution (TFD) theory. It has been found that training does not give the same results every time; this is because the weights are initialized to random values and high validation error may end up training early. Moreover once a network is trained with selected input, its performance improves significantly as opposed to the one that does not receive selected input data for training. The performance of MNCD can be compared by computing the Entropy, Mean Square Error (MSE) and time consumed for convergence.
Time-frequency distributions (TFDs) that are highly concentrated in the time-frequency plane are ... more Time-frequency distributions (TFDs) that are highly concentrated in the time-frequency plane are computed using a Bayesian regularised neural network model. The degree of regularisation is automatically controlled in the Bayesian inference framework and produces networks with better generalised performance and lower susceptibility to over-fitting. Spectrograms and Wigner transforms of various known signals form the training set. Simulation results show that regularisation, with input training under Mackay's evidence framework, produces results that are highly concentrated along the instantaneous frequencies of the individual components present in the test TFDs. Various parameters are compared to establish the effectiveness of the approach.
IEEE Journal on Selected Areas in Communications, 2000
Satellite channels are generally nonlinear and dispersive in nature, due to amplifiers being driv... more Satellite channels are generally nonlinear and dispersive in nature, due to amplifiers being driven close to saturation. These effects can cause significant degradations when they are not taken into account at either the receiver (equalization) or at the transmitter (pre-distortion). State-of-the-art equalizers rely on the forward-backward algorithm and yield excellent performance. However, they have unreasonable complexity and storage requirements, especially for highly dispersive channels and/or large constellations. In this paper, we derive an equalization strategy for nonlinear channels based on Monte Carlo methods. We present a detailed performance, complexity and storage analysis. A significant performance gain compared to the linear equalizer is reported, and the proposed technique results in a significant reduction in both complexity and storage, compared to the forward-backward equalizer.
EURASIP Journal on Advances in Signal Processing, 2009
We present a review of the diversity of concepts and motivations for improving the concentration ... more We present a review of the diversity of concepts and motivations for improving the concentration and resolution of timefrequency distributions (TFDs) along the individual components of the multi-component signals. The central idea has been to obtain a distribution that represents the signal's energy concentration simultaneously in time and frequency without blur and crosscomponents so that closely spaced components can be easily distinguished. The objective is the precise description of spectral content of a signal with respect to time, so that first, necessary mathematical and physical principles may be developed, and second, accurate understanding of a time-varying spectrum may become possible. The fundamentals in this area of research have been found developing steadily, with significant advances in the recent past.
A Neural Network (NN) based realization to obtain energy concentration along Instantaneous Freque... more A Neural Network (NN) based realization to obtain energy concentration along Instantaneous Frequencies (IFs) is proposed by compensating linear and non linear distortions present in Time Frequency Representations (TFRs). Blurry spectrograms and highly concentrated Wigner Distributions (WDs) of various signals constitute the training set. The input data is grouped according to some underlying feature present in TFR image to have better generalization ability of the trained NN. Blurry TFRs of multi component signals are then given as test cases to the trained NN. Effectiveness of the approach is established by comparing the information content in each input & resultant TFR.
In this paper we present a comparison of Neural Network Training Algorithms for obtaining a Time ... more In this paper we present a comparison of Neural Network Training Algorithms for obtaining a Time Frequency Distribution (TFD) of a signal whose frequency components vary with time. The method employs various algorithms used in NNs which are trained by using the spectrograms of several training signals as input and TFDs that are highly concentrated along the instantaneous frequencies (IFs) of the individual components present in the signal as targets. The trained neural networks are then presented with the spectrogram of unknown signals. We compute the entropy as a measure of the result obtained and carry out error and time analysis to compare the performance of algorithms used.
In this work, equalization for nonlinear dispersive channels is considered. Nonlinear communicati... more In this work, equalization for nonlinear dispersive channels is considered. Nonlinear communication channels can lead to significant degradations when nonlinearities are not taken into account at either the receiver or the transmitter. In many cases, the nonlinearity of the channel precludes the use of spectrally efficient signaling schemes to achieve high data-rates and the bandwidth efficiency. Satellite channel is a typical case of nonlinear channel that needs to be used efficiently. We develop two novel equalization strategies for a general class of nonlinear channels. Both strategies are based on iterating between decoding and equalization, termed as iterative equalization. The first strategy is a factor graph based equalizer that converts the nonlinear channel equalization problem into forward-backward algorithm on hidden Markov model (HMM). The equalizer is implemented via the sum-product algorithm on the factor graph representation of the channel and receiver blocks. The sec...
Traditional patient monitoring may not detect cerebral tissue hypoxia, and typical interventions ... more Traditional patient monitoring may not detect cerebral tissue hypoxia, and typical interventions may not improve tissue oxygenation. Therefore, monitoring cerebral tissue oxygen status with regional oximetry is being increasingly used by anesthesiologists and perfusionists during surgery. In this study, we evaluated absolute and trend accuracy of a new regional oximetry technology in healthy volunteers. A near-infrared spectroscopy sensor connected to a regional oximetry system (O3™, Masimo, Irvine, CA) was placed on the subject's forehead, to provide continuous measurement of regional oxygen saturation (rSO2). Reference blood samples were taken from the radial artery and internal jugular bulb vein, at baseline and after a series of increasingly hypoxic states induced by altering the inspired oxygen concentration while maintaining normocapnic arterial carbon dioxide pressure (PaCO2). Absolute and trend accuracy of the regional oximetry system was determined by comparing rSO2 against reference cerebral oxygen saturation (SavO2), that is calculated by combining arterial and venous saturations of oxygen in the blood samples. Twenty-seven subjects were enrolled. Bias (test method mean error), standard deviation of error, standard error of the mean, and root mean square accuracy (ARMS) of rSO2 compared to SavO2 were 0.4%, 4.0%, 0.3%, and 4.0%, respectively. The limits of agreement were 8.4% (95% confidence interval, 7.6%-9.3%) to -7.6% (95% confidence interval, -8.4% to -6.7%). Trend accuracy analysis yielded a relative mean error of 0%, with a standard deviation of 2.1%, a standard error of 0.1%, and an ARMS of 2.1%. Multiple regression analysis showed that age and skin color did not affect the bias (all P > 0.1). Masimo O3 regional oximetry provided absolute root-mean-squared error of 4% and relative root-mean-squared error of 2.1% in healthy volunteers undergoing controlled hypoxia.
Oxidative stress during fetal development, delivery, or early postnatal life is a major cause of ... more Oxidative stress during fetal development, delivery, or early postnatal life is a major cause of neuropathology, as both hypoxic and hyperoxic insults can significantly damage the developing brain. Despite the obvious need for reliable cerebral oxygenation monitoring, no technology currently exists to monitor cerebral oxygen metabolism continuously and noninvasively in infants at high risk for developing brain injury. Consequently, a rational approach to titrating oxygen supply to cerebral oxygen demand -and thus avoiding hyperoxic or hypoxic insults -is currently lacking. We present a promising method to close this crucial technology gap in the important case of neonates on conventional ventilators. By using cerebral near-infrared spectroscopy (NIRS) and signals from conventional ventilators, along with arterial oxygen saturation, we derive continuous (breath-by-breath) estimates of cerebral venous oxygen saturation, cerebral oxygen extraction fraction, cerebral blood flow, and cerebral metabolic rate of oxygen. The resultant estimates compare very favorably to previously reported data obtained by non-continuous and invasive means from preterm infants in neonatal critical care.
Proceedings. IEEE International Multi Topic Conference, 2001. IEEE INMIC 2001. Technology for the 21st Century., 2001
ABSTRACT Quadrature Amplitude Modulation (QAM) is widely used in many digital communication syste... more ABSTRACT Quadrature Amplitude Modulation (QAM) is widely used in many digital communication systems such as high speed modems. Increasing the data rate by using larger constellations has brought a revolution in digital communication systems. Consequently, there is a need ...
8th International Multitopic Conference, 2004. Proceedings of INMIC 2004., 2004
The use of dual-tone multi frequency (DTMF) signaling is very common in telecommunication systems... more The use of dual-tone multi frequency (DTMF) signaling is very common in telecommunication systems such as telephone exchanges, private branch exchanges (PBXs), and call centers etc. One of the approaches of DTMF detection is based on the cycle estimation of the frequencies present in the DTMF signal. Any interference that lies in the frequency region of the DTMF signal disturbs
Proceedings of the IEEE Symposium on Emerging Technologies, 2005., 2005
In this paper we present a method of obtaining a Time Frequency Distribution (TFD) of a signal wh... more In this paper we present a method of obtaining a Time Frequency Distribution (TFD) of a signal whose frequency components vary with time. The method employs Neural Networks (NN) which are trained by using the spectrograms of several training signals as input and TFDs that are highly concentrated along the instantaneous frequencies of the individual components present in the signal as targets. The trained neural network is then presented with the spectrogram of unknown signals and highly concentrated TFDs are obtained.
Monitoring cerebrovascular state, including intracranial pressure (ICP) and the ability to regula... more Monitoring cerebrovascular state, including intracranial pressure (ICP) and the ability to regulate cerebral blood flow, is important for patient care in stroke, traumatic brain injury and other such conditions. However, current methodologies for direct measurement of ICP are highly invasive, and expose patients to the risk of infection. In addition, vascular properties such as resistance and compliance cannot be directly assessed. In this work, we employ a mathematical model-based approach to track variations in ICP and cerebrovascular properties from signals that can be acquired entirely non-invasively. The performance on simulation data indicates that the estimates track the desired quantities closely, thus suggesting that tests using clinical data are warranted.
2007 IEEE International Conference on Signal Processing and Communications, 2007
A Neural Network (NN) based approach to obtain energy concentration along Instantaneous Frequenci... more A Neural Network (NN) based approach to obtain energy concentration along Instantaneous Frequencies (IFs) of the individual components present in the signal, is proposed. Blurry spectrograms and highly concentrated Wigner Distributions (WDs) of various signals constitute the training set. The input data is grouped according to the underlying feature present in Time Frequency Representation (TFR) image to have better generalization ability of the trained NN. Blurry TFRs of multi component signals are then given as test cases to the trained NN. Effectiveness of the approach is established by comparing the information content in each input and resultant TFR.
2006 IEEE International Multitopic Conference, 2006
Network Architecture, for a Time Frequency application. Varying the number of neurons and hidden ... more Network Architecture, for a Time Frequency application. Varying the number of neurons and hidden layers has been found to greatly affect the performance of Neural Network (NN), trained via various blurry spectrograms as input over highly concentrated Time Frequency Distributions (TFDs) as targets, of the same signals. Number of neurons and hidden layers are varied during training and the impact is observed over test spectrograms of unknown multi component signals. Entropy and Mean Square Error (MSE) is the decision criteria for the most optimum solution.
The Proceedings of the Multiconference on "Computational Engineering in Systems Applications", 2006
A realization of obtaining energy concentration along instantaneous frequency (IF) for blurry Tim... more A realization of obtaining energy concentration along instantaneous frequency (IF) for blurry Time Frequency Images (TFIs) is proposed using Neural Networks (NNs). NN captures the fundamental principle at work once trained with various spectrogram TFIs by defining various edges present in input data. Unknown spectrogram TFIs of multi component signals, are then given as test images to the trained NN. Measuring the information content in each input & resultant TFI thus manifest the effectiveness of NNs in TFI analysis.
2007 International Conference on Information and Emerging Technologies, 2007
In this paper we present advantage of training MNCD for obtaining Time localized Frequencies (als... more In this paper we present advantage of training MNCD for obtaining Time localized Frequencies (also called IF), which is one useful concept for describing the changing spectral structure of a time-varying signal, arising so often in Time Frequency Distribution (TFD) theory. It has been found that training does not give the same results every time; this is because the weights are initialized to random values and high validation error may end up training early. Moreover once a network is trained with selected input, its performance improves significantly as opposed to the one that does not receive selected input data for training. The performance of MNCD can be compared by computing the Entropy, Mean Square Error (MSE) and time consumed for convergence.
Time-frequency distributions (TFDs) that are highly concentrated in the time-frequency plane are ... more Time-frequency distributions (TFDs) that are highly concentrated in the time-frequency plane are computed using a Bayesian regularised neural network model. The degree of regularisation is automatically controlled in the Bayesian inference framework and produces networks with better generalised performance and lower susceptibility to over-fitting. Spectrograms and Wigner transforms of various known signals form the training set. Simulation results show that regularisation, with input training under Mackay's evidence framework, produces results that are highly concentrated along the instantaneous frequencies of the individual components present in the test TFDs. Various parameters are compared to establish the effectiveness of the approach.
IEEE Journal on Selected Areas in Communications, 2000
Satellite channels are generally nonlinear and dispersive in nature, due to amplifiers being driv... more Satellite channels are generally nonlinear and dispersive in nature, due to amplifiers being driven close to saturation. These effects can cause significant degradations when they are not taken into account at either the receiver (equalization) or at the transmitter (pre-distortion). State-of-the-art equalizers rely on the forward-backward algorithm and yield excellent performance. However, they have unreasonable complexity and storage requirements, especially for highly dispersive channels and/or large constellations. In this paper, we derive an equalization strategy for nonlinear channels based on Monte Carlo methods. We present a detailed performance, complexity and storage analysis. A significant performance gain compared to the linear equalizer is reported, and the proposed technique results in a significant reduction in both complexity and storage, compared to the forward-backward equalizer.
EURASIP Journal on Advances in Signal Processing, 2009
We present a review of the diversity of concepts and motivations for improving the concentration ... more We present a review of the diversity of concepts and motivations for improving the concentration and resolution of timefrequency distributions (TFDs) along the individual components of the multi-component signals. The central idea has been to obtain a distribution that represents the signal's energy concentration simultaneously in time and frequency without blur and crosscomponents so that closely spaced components can be easily distinguished. The objective is the precise description of spectral content of a signal with respect to time, so that first, necessary mathematical and physical principles may be developed, and second, accurate understanding of a time-varying spectrum may become possible. The fundamentals in this area of research have been found developing steadily, with significant advances in the recent past.
Uploads
Papers by Faisal Kashif