L2 norm provides better results with moderate
and low noise levels although it requires high sampling
frequency, high digitizing resolution, and higher
computing time to achieve its full performance
L1 needs both low cost
hardware and lower computing time using conventional
architectures. Also, results show that 6-8 b of resolution
are enough for most applications
while L1 and L2 norms
L2 method requires a high resolution (10-12 b)
from the A/D converter to reach maximum accuracy (Fig.
5). In contrast, correlation
L1 norm show no increase
in performance when digitizing with 6 or more bits. This
fact allows the use of low resolution (cost) converters for
applications requiring a moderate level of accuracy
Fig. 4). However,
for the L1 and L2 algorithms, errors tend to be rather constant
if the relation between the sampling frequency and
the fundamental signal frequency f, lfbecomes higher than
twice the Nyquist criterion. Thus, taking at least five samples
per period makes the S/N ratio to be the limiting
factor, which allows us to reduce the A/D converter cost,
memory requirements, and computing time.
Fig. 3 shows the performance of the different
digital signal processing (DSP) methods as a function
of S/N ratio. As can be expected, errors increase with
decreasing S/N ratios, this effect being much more noticeable
when the envelope is extracted by peak tracking.
When high noise levels are present, correlation yields the
best results. It is interesting to note that the error obtained
with this method levels out at about 0.1 mm. This fact is
a consequence of the limiting factor of the sampling period.
With S/N ratios above 20 dB, errors which are about
two orders of magnitude lower than those found with conventional
threshold techniques can be expected
Correlation
L2 norm provides better results with moderateand low noise levels although it requires high samplingfrequency, high digitizing resolution, and highercomputing time to achieve its full performanceL1 needs both low costhardware and lower computing time using conventionalarchitectures. Also, results show that 6-8 b of resolutionare enough for most applicationswhile L1 and L2 normsL2 method requires a high resolution (10-12 b)from the A/D converter to reach maximum accuracy (Fig.5). In contrast, correlationL1 norm show no increasein performance when digitizing with 6 or more bits. Thisfact allows the use of low resolution (cost) converters forapplications requiring a moderate level of accuracyFig. 4). However,for the L1 and L2 algorithms, errors tend to be rather constantif the relation between the sampling frequency andthe fundamental signal frequency f, lfbecomes higher thantwice the Nyquist criterion. Thus, taking at least five samplesper period makes the S/N ratio to be the limitingfactor, which allows us to reduce the A/D converter cost,memory requirements, and computing time.Fig. 3 shows the performance of the differentdigital signal processing (DSP) methods as a functionof S/N ratio. As can be expected, errors increase withdecreasing S/N ratios, this effect being much more noticeablewhen the envelope is extracted by peak tracking.When high noise levels are present, correlation yields thebest results. It is interesting to note that the error obtainedwith this method levels out at about 0.1 mm. This fact isa consequence of the limiting factor of the sampling period.With S/N ratios above 20 dB, errors which are abouttwo orders of magnitude lower than those found with conventionalthreshold techniques can be expectedCorrelation
การแปล กรุณารอสักครู่..
