The paper presents a method of suppressing offset induced error in comparators used in stochastic converters. A stochastic converter has two analog voltage inputs, and the output is proportional to the integral of input product. This converter is used to build a precision watt-hour meter prototype at University of Novi Sad. A stochastic method uses a high oversampling frequency and fast, off-the-shelf commercial comparators. These comparators have relatively large offset, but we show that it does not affect the quality of measurement. Mathematical theory, simulation and testing results are presented. The results show offset error reduction up to three orders of magnitude.

This paper describes 4-bit Stochastic Flash ADC based instrument, with its low-resolution compensated by adding dither (uniform noise) to the input signal and calculation of the output average value over a finite time period, resulting in low part count, high sampling speed and low error. Precision of this instrument is user controlled - the longer the period of measurement, higher the measurement precision. This enables measurement of electrical quantities over long time intervals, such as electrical energy and RMS voltage in a power grid. For 60-second measurement period, precision is 58 ppm, and for 180 s it is 34 ppm.

This paper presents a proposal of a new way to generate analog noise, discrete in time, with uniform probability density function. Instead of the usual way of generating discrete analog uniform noise using a uniform pseudorandom number generator and a digital-to-analog converter, generating based on non-equidistant sampling of sawtooth voltage is suggested. The original hardware structure that implements this idea is suggested and its work is simulated. The simulations have confirmed the validity of the idea. The main gain is the unlimited resolution of the analog output of the uniform noise generator suggested. The resolution of standard uniform noise generators is limited by the resolution of the digital-to-analog converter used.

The paper presents the influence of signal stationarity on digital stochastic measurement method implementation. The implementation method is based on stochastic voltage generators, analog adders, low resolution A/D converter, and multipliers and accumulators implemented by Field-Programmable Gate Array (FPGA). The characteristic of first implementations of digital stochastic measurement was the measurement of stationary signal harmonics over the constant measurement period. Later, digital stochastic measurement was extended and used also when it was necessary to measure time-series of non-stationary signal over the variable measurement time. The result of measurement is the set of harmonics, which is, in the case of non-stationary signals, the input for calculating digital values of signal in time domain. A theoretical approach to determine measurement uncertainty is presented and the accuracy trends with varying signal-to-noise ratio (SNR) are analyzed. Noisy brain potentials (spontaneous and non-spontaneous) are selected as an example of real non-stationary signal and its digital stochastic measurement is tested by simulations and experiments. Tests were performed without noise and with adding noise with SNR values of 10dB, 0dB and -10dB. The results of simulations and experiments are compared versus theory calculations, and comparasion confirms the theory.