Michael Muma

A new method for heart rate monitoring during physical exercise using photoplethysmographic signals

Accurate and reliable estimation of the heart rate using wearable devices, especially during physical exercise, must deal with noisy signals that contain motion artifacts. We present an approach that is based on photoplethysmographic (PPG) signals which are measured with two wrist-type pulse oximeters. The heart rate is related to intensity changes of the reflected light. Our proposed method suppresses the motion artifacts by adaptively estimating the transfer functions of each of the three-axis acceleration signals that produce the artifacts in the PPG signals. We combined the output of the six adaptive filters into a single enhanced time-frequency domain signal based on which we track the heart rate with a high accuracy. Our approach is real-time capable, computationally efficient and real data results for a benchmark data set illustrate the superior performance compared to a recently proposed approach.

The Role of Cardiopulmonary Signals in the Dynamics of the Eye’s Wavefront Aberrations

The role of cardiopulmonary signals in the dynamics of wavefront aberrations in the eye has been examined. Synchronous measurement of the eyes wavefront aberrations, cardiac function, blood pulse, and respiration signals were taken for a group of young, healthy subjects. Two focusing stimuli, three breathing patterns, as well as natural and cycloplegic eye conditions were examined. A set of tools, including time-frequency coherence and its metrics, has been proposed to acquire a detailed picture of the interactions of the cardiopulmonary system with the eyes wavefront aberrations. The results showed that the coherence of the blood pulse and its harmonics with the eyes aberrations was, on average, weak (0.4 ± 0.15), while the coherence of the respiration signal with eyes aberrations was, on average, moderate (0.53 ± 0.14). It was also revealed that there were significant intervals during which high coherence occurred. On average, the coherence was high (>0.75) during 16% of the recorded time, for the blood pulse, and 34% of the time for the respiration signal. A statistically significant decrease in average coherence was noted for the eyes aberrations with respiration in the case of fast controlled breathing (0.5 Hz). The coherence between the blood pulse and the defocus was significantly larger for the far target than for the near target condition. After cycloplegia, the coherence of defocus with the blood pulse significantly decreased, while this was not the case for the other aberrations. There was also a noticeable, but not statistically significant, increase in the coherence of the comatic term and respiration in that case. By using nonstationary measures of signal coherence, a more detailed picture of interactions between the cardiopulmonary signals and eyes wavefront aberrations has emerged.

Distributed robust labeling of audio sources in heterogeneous wireless sensor networks

A novel algorithm for distributed labeling of speech sources is proposed. We consider a wireless sensor network comprising devices that are equipped with multiple microphones, which can “hear” a number of speech signals. The labeling task is performed in a decentralized fashion with a new two-step approach. The first step corresponds to the distributed extraction of proper source-specific features from the mixed signals. In the second step, these features are exploited via a distributed unsupervised learning technique. We present approaches that can be used in hierarchically organized or in non-hierarchically organized network configurations. Numerical examples using real data display the performance of the proposed technique.

Robust source number enumeration for r-dimensional arrays in case of brief sensor failures

There has been much activity on model selection for multi-dimensional data in recent years under the assumption of a Gaussian noise distribution. However, methods which are optimal for Gaussian noise are very sensitive against brief sensor failures. We suggest two robust model order selection schemes for multi-dimensional data based on the MM-estimator of the covariance of the r-mode unfoldings of the complex valued data tensor. Simulation results are given for 2-D and 3-D uniform rectangular arrays based source enumeration, both for Gaussian noise and a brief sensor failure.

Robust diffusion-based unsupervised object labelling in distributed camera networks

Recently, a new ICT paradigm emerged, which considers Multiple Devices that cooperate in Multiple Tasks (MDMT). Under this paradigm, cooperation among the nodes can be beneficial when subsets of the nodes share common interests or observations. For cooperation to be successful, it is thus necessary to account for a decentralized labelling scheme that allows to uniquely identify every object of interest. Such labelling not only ensures proper data exchange among the nodes but also allows the formation of interest-specific clusters and hence, might also be beneficial from a communications cost perspective. The research question addressed in this paper is to develop robust distributed labelling strategies in the context of camera networks where no central unit is available for fusing all the information. Simulation results demonstrate that a high labelling accuracy can be achieved in the considered setup (planar scene) with a correct classification performance close to the centralized solution. The proposed methodology is a promising strategy for distributed clustering in camera networks that can be extended to more complex scenarios.

Robustness Analysis of Spatial Time-Frequency Distributions Based on the Influence Function

Standard spatial time-frequency distribution (STFD) estimators, derived based on the Gaussian noise assumption, are known to have poor performance in the case of impulsive noise. Recently, different STFD estimators have been proposed, which, based on simulations, are claimed to be robust. In this paper, we provide an influence function robustness analysis of STFD estimators. We derive the influence functions for the asymptotic and for the finite-sample case and study robustness of the standard, as well as for some recently proposed robust STFD estimators. The empirical influence function gives practitioners a simple way to pre-select STFD estimators for their scenario. Our analysis confirms that, unlike for the standard estimator, the proposed robust estimators yield a bounded influence function and are robust over a broad class of distributions. Future research on STFD estimation will allow for the design of robust and efficient estimators based on the influence function.

Robust and adaptive diffusion-based classification in distributed networks

Distributed adaptive signal processing and communication networking are rapidly advancing research areas which enable new and powerful signal processing tasks, e.g., distributed speech enhancement in adverse environments. An emerging new paradigm is that of multiple devices cooperating in multiple tasks (MDMT). This is different from the classical wireless sensor network (WSN) setup, in which multiple devices perform one single joint task. A crucial first step in order to achieve a benefit, e.g., a better node-specific audio signal enhancement, is the common unique labeling of all relevant sources that are observed by the network. This challenging research question can be addressed by designing adaptive data clustering and classification rules based on a set of noisy unlabeled sensor observations. In this paper, two robust and adaptive distributed hybrid classification algorithms are introduced. They consist of a local clustering phase that uses a small part of the data with a subsequent, fully distributed on-line classification phase. The classification is performed by means of distance-based similarity measures. In order to deal with the presence of outliers, the distances are estimated robustly. An extensive simulation-based performance analysis is provided for the proposed algorithms. The distributed hybrid classification approaches are compared to a benchmark algorithm where the error rates are evaluated in dependence of different WSN parameters. Communication cost and computation time are compared for all algorithms under test. Since both proposed approaches use robust estimators, they are, to a certain degree, insensitive to outliers. Furthermore, they are designed in a way that they are applicable to on-line classification problems.

An online approach for intracranial pressure forecasting based on signal decomposition and robust statistics

Intracranial pressure (ICP) is an important physiological signal for patients with traumatic brain injuries. Accurate ICP forecasting enables active and early interventions for more effective control of ICP levels. To achieve high accuracy, most existing methods require a high sampling rate (100 Hz), which is infeasible for online medical applications. Therefore, we propose an online ICP forecasting method requiring only low rate signal sampling (0.1 Hz). Our ARIMA based forecasting method applies empirical mode decomposition (EMD) to remove non-stationarities from the ICP signal, and robust estimation to mitigate the influence of motion induced artifacts. Experimental performance assessment with simulated and clinically collected data demonstrate that the proposed method is more accurate compared to previously proposed and standard methods.

Robust and computationally efficient diffusion-based classification in distributed networks

Todays wireless sensor networks provide the possibility to monitor physical environments via small low-cost wireless devices. Given the large amount of sensed data, efficient and robust classification becomes a critical task in many applications. Typically, the devices must operate under stringent power and communication constraints and the transmission of observations to a fusion center (FC) is, in many cases, infeasible or undesired. A challenging research question in such cases is the design of data clustering and classification rules when each sensor collects a set of unlabelled observations that are drawn from a known number of classes. We propose two robust distributed hybrid classification algorithms, i.e., the Diffusion K-Medians and the Communicationally Efficient Distributed K-Medians. An extensive performance analysis in comparison to a benchmark algorithm is provided that investigates the error rates in dependence of different parameters of a distributed sensor network, and also considers communication cost. Our proposed algorithms, which are insensitive to outliers and various parameters, are applicable to on-line classification problems and scale well w.r.t. the number of classes.

In-network adaptive cluster enumeration for distributed classification and labeling

A crucial first step for signal processing decentralized sensor networks with node-specific interests is to agree upon a common unique labeling of all observed sources in the network. The knowledge “who observes what” is required, e.g. in node-specific audio or video signal enhancement to form node clusters of common interest. Recently proposed in-network distributed adaptive classification and labeling algorithms assume knowledge on the number of objects (clusters), which is not necessarily available in real-world applications. Thus, we consider the problem of estimating the number of data-clusters in the distributed adaptive network set-up. We propose two distributed adaptive cluster enumeration methods. They combine the diffusion principle, where the nodes share information within their local neighborhood only (without fusion center), with the X-means and the PG-means cluster enumeration. Performance is evaluated via simulations and the applicability of the methods is illustrated using a distributed camera network where moving objects appear and disappear from the Line-of-Sight (LOS) and the number of clusters becomes time-varying.