Dominik Rzepka

Fixed-budget kernel least mean squares

In this paper a new version of kernel normalized least mean squares algorithm is proposed, allowing for maintaining a fixed memory budget. Constant growth of the support vectors dictionary, inherent to online kernel methods is constrained using pruning criterion. After admitting a new input vector to the dictionary, the least important entry is found and removed. This allows for competition among the input vectors and dynamic adaptation of the dictionary content. In the task of time-varying system identification this method has the clear advantage over a coherence criterion which does not modify vectors already present in the dictionary. Experimental results confirm that algorithm performs well in time-varying environment and it can maintain a performance comparable to the state-of-the-art algorithms, using smaller number of support vectors, with linear complexity.

Bandwidth Estimation From Multiple Level-Crossings of Stochastic Signals

Motivated by a broad range of potential applications in event-driven signal processing inference, we examine a class of spectral moments estimators derived from multiple level-crossings of a stationary Gaussian stochastic signal as well as non-Gaussian signals being monotone transformations of the Gaussian signal. The established spectral moments estimators are used to obtain estimates of the stochastic signal bandwidth. Both the mean-squared bandwidth and the absolute bandwidth are taken into account. Also the notion of the power bandwidth is introduced and its estimation is examined. We prove that the proposed estimates are consistent as the length of the observed time interval is increasing. The accuracy of the proposed bandwidth estimation methods from multiple level-crossings is assessed in simulation studies.

Optimized design of successive approximation time-to-digital converter with single set of delay lines

The paper addresses the problems of design of picosecond resolution time-to-digital converter based on successive approximation (SA-TDC). The principle of the conversion process in SA-TDC consists in successive delaying the events defining a start and a stop of the input time interval by the use of binary-weighted delays. The paper is focused on optimization of particular components of the SA-TDC architecture with a single set of delay lines in order to reduce differential (DNL) and integral (INL) nonlinearities. In particular, the paper contribution is an improvement of time resolution of the converter from 25 ps to 12.5 ps (i.e., by one extra bit) in 180 nm CMOS technology through enhancements of design of circuit components which results in a reduction of conversion errors.

Sample-and-Hold Asynchronous Sigma-Delta Time Encoding Machine

In this brief, a new modulation scheme called sample-and-hold asynchronous sigma-delta modulation (SH-ASDM), the extension of the classical asynchronous sigma-delta scheme (ASDM), is introduced. The SH-ASDM is aimed to provide an instantaneous time encoding instead of the mean value encoding realized in the ASDM. This eliminates the effect of the dynamic range reduction of output pulses lengths, which appears for high-frequency input signals in ASDM. SH-ASDM outperforms the ASDM by about 3 dB in the signal-to-noise ratio quantization characteristics, in respect to the worst case input signal. The new analytic model of ASDM is also proposed, which decomposes this type of time-encoding into a linear filtering, sampling, and nonlinear sample transformation.

Compressive sampling of stochastic multiband signal using time encoding machine

In this paper we present an improvement of architecture of the Asynchronous Sigma-Delta time encoding machine, which makes it suitable for sampling the wideband sparse analog signals. Since the values of the samples are encoded into sampling instants, this is a signal-dependent sampling scheme. By superseding the commonly used multitone model of the input signal with multiband stochastic model, the randomness of the sampling instants is obtained. This allows for the reconstruction of signal using compressive sampling methods without additional randomization hardware. The simulation results validate the presented approach.

Estimation of varying bandwidth from multiple level crossings of stochastic signals

This papers examines the problem of a local bandwidth recovery for non-stationary stochastic signals when the only available information is given in terms of level crossings. The use of multiple level crossings is a fundamental paradigm for the recently investigated event-based sampling approach. In fact, level crossings allows us to exploit local signal properties and to avoid unnecessarily fast sampling when the local signal intensity is low. The paper proposes the least-square based method for the local intensity estimation from level crossings for the class of signals being the time-warped version of the stationary and bandlimited Gaussian processes. This result is then related to the concept of the local mean bandwidth and finally to the local power bandwidth. The smooth convolution estimate of the local intensity is proposed and its positivity corrected version is introduced. The latter is achieved by the truncation argument and next by the method of alternating projections onto convex sets.

Operator-theoretic approach to minimal-norm bandlimited interpolation of nonuniform samples

Methods for the minimal-norm bandlimited interpolation of nonuniform samples are known to have serious numerical issues even though this interpolation is mathematically well defined. Moreover, they have been formulated only for finite numbers of samples. By a rigorous revisit of this problem from an operator-theoretic viewpoint in Hilbert spaces, we devise an algorithm for this interpolation that applies to an infinite number of samples with no condition, with guaranteed stability at least for finite numbers of samples, and with a recursive part that can be rigorously implemented in discrete time.

Clockless signal-dependent compressive sensing of multitone signals using time encoding machine

In this paper we use modified sample-and-hold asynchronous sigma-delta modulator (ASDM) time encoding machine (TEM) for compressive sensing (CS) of multitone signals. Nouniform sampling, suitable for CS, is inherent property of TEM, however it is not random as it is common in CS schemes, but signal-dependent. The irregularity of TEM sampling instants is analyzed using theory of dynamical systems and recognized as periodic or chaotic, depending on the input signal. We show experimentally that such behavior deteriorates CS effectiveness, but it can be avoided using randomization by adding simple bipolar signal with sufficiently high amplitude to the input. Finally, a signal-dependent, asynchronous randomization circuit for generation of pseudo-random sequences by the use of binary encoded output of TEM is introduced.

Breath sensor based on conductive foam: Idea, construction, and characterisation

This paper deals with breath sensor constructed based on conductive, polyurethane foam, which is usually used for storage of electronics devices. A piece of the foam with especially prepared terminals is placed under tight T-shirt at the chest of a monitored person. The paper presents the idea of the sensor, investigation on its construction and problems being faced with. Parameters assessment processes and results are described too, so finally well characterized electronics component, ready to use, is presented. Some examples of use of the sensor are presented at the end of the paper.

Signal recovery from level-crossing samples using projections onto convex sets

In this paper, a new method of recovery of a bandlimited signal from level-crossing samples is presented. In contrast to known methods, not only the level-crossing samples (explicit information) but also the fact that the signal does not cross the levels in the time intervals between the samples (implicit information) are used for the recovery. The reconstruction algorithm is iterative and adopts the framework of Projections onto Convex Sets (POCS). The projections are realized by alternating lowpass filtering and hardlimiting operations. Numerical experiments demonstrate the efficiency of the proposed recovery method.