Rafael Canetti

Convergence analysis of the least-squares identification algorithm with a variable forgetting factor the time-varying linear systems

Abstract This paper studies the convergence of the least-squares identification algorithm with a variable forgetting factor. The evolution of the parameter estimator with respect to initial conditions, actual parameter changes and stochastic perturbations is analysed. Bounds on the deterministic and stochastic parts of their estimator error are found and their relation to a persistent excitation condition is discussed. This relation explains the possibility of having “bursting” phenomena during the identification process. Special attention is paid to the consistency of these results with the classical ones obtained for the ordinary least-squares algorithm when the forgetting factor tends to unity.

Correlations between frequency-domain HRV indices and lagged Poincaré plot width in healthy and diabetic subjects

The conventional Poincaré plot for heart rate variability (HRV) analysis is a scatterplot of successive (lag 1) pairs of RR intervals (intervals between heartbeats), and its width (SD1) is considered a measure of short-term variability. It has been shown that SD1 correlates better with HF than with LF (high- and low-frequency bands of the spectrum, respectively). Our aim was to assess how these correlations were affected when SD1 was obtained for longer lags. 10 min ECGs were used to construct Poincaré plots with lags of 1-10 heartbeats in two groups of subjects, one with normal HRV and the other with impaired HRV (control and diabetic groups respectively, N = 15 each). SD1 was quantified for these subjects and HRV spectral indices were estimated. The diabetic group had lower LF, HF and SD1 than the control group (p < 0.05). In both groups, SD1 tended to increase as the lag increased. In the control group, SD1 for lags 1 and 2 was highly correlated with HF (r(s) > 0.9), while SD1 for lags 4 correlated better with LF (r(s) 0.9) than with HF (0.65 <or= r(s) <or= 0.73). However, in the diabetic group, the correlation results did not change in that way for different lags (correlation results between HF and SD1: r(s) 0.95 for lags 1-10). In conclusion, the comparative strength of the correlations between lagged Poincaré widths and spectral indices might be useful to distinguish normal from pathological HRV.

NeuroFPGA-implementing artificial neural networks on programmable logic devices

An FPGA implementation of a multilayer perceptron neural network is presented. The system is parameterized both in network related aspects (e.g.: number of layers and number of neurons in each layer) and implementation parameters (e.g.: word width, pre-scaling factors and number of available multipliers). This allows to use the design for different network realizations, or to try different area-speed trade-offs simply by recompiling the design. Fixed point arithmetic with pre-scaling configurable in a per layer basis was used. The system was tested on an ARC-PCI board from altera/spl trade/ several examples from different application domains were implemented showing the flexibility and ease of use of the obtained circuit. Even with the rather old board used, an appreciable speed-up was obtained compared with a software-only implementation based on Matlab neural network toolbox.

Heart rate variability: short-term studies are as useful as holter to differentiate diabetic patients from healthy subjects.

Background: The definitive incorporation of heart rate variability (HRV) as a clinical tool depends on the development of more confident techniques of measurement. The length of the studies is a critical issue. Whereas Holter studies allow the monitorization at different hours and activities, short‐term recordings allow the control of environmental conditions. Recording length is also strongly related to the procedure of analysis; for instance, some time‐domain indexes are strongly affected by the duration of the study. Meanwhile, spectral analyses require stationary conditions, only achieved in short‐term studies. Our main goal was to determine if HRV indexes obtained from short‐term analyses were as useful as those from Holter monitoring for diagnosis of reduced HRV in diabetes.

Calibration of an inertial measurement unit

This paper presents a fast and low cost way to calibrate different inertial measurement sensors. In particular the calibration of an accelerometer and a gyroscope using nonlinear least squares is presented. A model of the sensors which includes the main errors that MEMS devices present is used. A calibration method is proposed for estimating the static parameters of the model and a temperature adjustment is proposed and implemented.

Implementation of Adaptive Logic Networks on an FPGA board

This work is part of a project that studies the implementation of neural network algorithms in reconfigurable hardware as a way to obtain a high performance neural processor. The results for Adaptive Logic Network (ALN) type binary networks with and without learning in hardware are presented. The designs were made on a hardware platform consisting of a PC compatible as the host computer and an ALTERA RIPP10 reconfigurable board with nine FLEX8K, FPGAs and 512 KB RAM. The different designs were run on the same hardware platform, taking advantage of its configurability. A software tool was developed to automatically convert the ALN network description resulting from the training process with the ATREE 2.7 for Windows software package into a hardware description file. This approach enables the easy generation of the hardware necessary to evaluate the very large combinatorial functions that results in an ALN. In an on-board learning version, an ALN basic node was designed optimizing it in the amount of cells per node used. Several nodes connected in a binary tree structure for each output bit, together with a control block, form the ALN network. The total amount of logic available on-board in the used platform limits the maximum size of the networks from a small to medium range. The performance was studied in pattern recognition applications. The results are compared with the software simulation of ALN networks.

Fisho: A cost-effective intelligent autonomous robot fish

We developed a cost-effective fish-like Autonomous Underwater Vehicle (AUV) with a sophisticated intelligence. The robot fish, named Fisho, is able to avoid obstacles in a swimming pool, reflecting the usual behavior of a biological fish. As a general rule, a robot captures external stimulus by means of electrical sensors; Fisho has infrared sensors to find the distance to close objects and pressure sensors to estimate its vertical physical position (depth). Its brain is a small-sized 200-MHz Single Board Computer (SBC), that collects information from the sensors and sends decisions to electro-mechanical servos (in order to act on its tail and flippers, thus control its trajectory). Fishos behavior is signed by its velocity and the vibration mode of its tail. Inspired in real fish, we designed its intelligence with a three-layer Markov chain by means of a full characterization of tasks, purposes and fish-moods. Its design and physical structure is both non-expensive and simple, trying to emulate a real fish, whereas addressing all typical challenges in an AUV. In this paper we discuss Fishos physical implementation: sensing platforms, electro-mechanical structure and intelligence.

Short-term studies of heart rate variability: comparison of two methods for recording

Heart rate variability (HRV) is often analysed using short-term studies. Our objective was to compare two of them in a group of diabetic patients (reduced HRV) and in a control group. From the same 10 min surface electrocardiogram (ECG) two recordings were obtained. In one of them the whole signal was acquired through an A/D converter (post-event method). In the other (real-time method), an interface between the electrocardiograph and a parallel port of a computer was used to perform real-time processing of the ECG signal. The R-R intervals were measured after a visual validation in the post-event method. In the real-time method, the stored R-R intervals were automatically filtered. For both methods HRV indexes were calculated using the same software. The values of mean R-R intervals for each subject were almost identical regardless of the method. Accordingly, we found a high correlation between HRV indexes obtained from both methods (all Spearman values > or = 0.9441 and P < 0.0001). In addition, we found similar P values in the comparisons between the diabetic and control groups. We conclude that both methods are suitable for HRV analysis. Therefore, the selection of method can be based on other considerations such as the capability to store the ECG of the post-event method or the speed of analysis and lower cost of the real-time one.