Estimação de fasores harmônicos baseada em banco de filtros polifásicos: implementação em FPGA com processador embarcado
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidade Federal de Juiz de Fora (UFJF)
DOI
Abstract
The growing number of nonlinear loads and power electronics based sources makes
the voltages and currents in the power grid move away from the ideal sinusoidal waveform,
so it becomes important to estimate not only the fundamental component but also the
harmonic components of the signal. This task becomes harder when the fundamental
frequency deviates from its nominal value, since the phase error grows with the harmonic
order. This work presents and implements a harmonic phasor estimation method for
signals with varying frequency, made up of three stages: frequency estimation by zero
crossing, smoothed by a Kalman filter; dynamic resampling of the signal by B-Spline
interpolation in a Farrow structure; and harmonic extraction by a Flat-Top filter bank
with polyphase decomposition, followed by an IDFT. Thanks to the resampling, the filter
bank always works with the nominal number of samples per cycle and its coefficients do
not need to be recomputed. The method was first developed as a reference model in
Python and then implemented on two soft-core processors of the SAPHO platform, one for
the interpolator and one for the filter bank, which communicate through FIFO memories
and are coordinated by a state machine. During the implementation, some stages had
to be adjusted because of the 32-bit arithmetic of the processor, such as the Kalman
filter and the period calculation in the zero crossing estimator. The implementation was
validated by functional simulation and compared with the reference model in tests based
on the IEC/IEEE 60255-118-1 standard, with off-nominal frequency, frequency ramp and
modulation. The frequency was estimated accurately in all tests. In the off-nominal case,
the reference model stayed below the 1% TVE limit over the whole range evaluated, and
the embedded implementation only exceeded this limit at the highest harmonic orders.
The results showed that this difference comes mainly from the time misalignment between
the signal and the frequency used in the phase correction, and not from the numerical
precision of the filter bank. In the ramp test, on the other hand, the reference model itself
exceeds the limit at the highest orders, which points to a limitation of the method under
continuous frequency variation. Overall, the results show that the method is feasible on
an embedded platform and indicate where it can still be improved.
Description
Citation
Endorsement
Review
Supplemented By
Referenced By
Creative Commons license
Except where otherwise noted, this item's license is described as Acesso Aberto

