03 - IEEE Format - Distortion Reduction in FIR Filters by Approximation Through Window Factor On
Uploaded by
DEEN OVER DUNYA03 - IEEE Format - Distortion Reduction in FIR Filters by Approximation Through Window Factor On
Uploaded by
DEEN OVER DUNYAΜΎΝΒΗΝΎΛΗΊΝΒΘΗΊΕΓΘΞΛΗΊΕΘΏΛΘΊΌΊΜΝΎΗΐΒΗΎΎΛΒΗΐΜΎΝΒΓΎϜͨлͥΊϘϚϏωϒϋмͥтϖͨ
ЫлйкпΜΎΝϘχϠϏϒϏχϔΜϕωϏϋϚϟϕόΝϋϒϋϜϏϙϏϕϔΎϔύϏϔϋϋϘϏϔύͿΒΜΜΗΙϘϏϔϚͧлннптлнпͿΒΜΜΗΘϔϒϏϔϋͧлннптнмл
Distortion Reduction in FIR Filters by
Approximation through Window Factor on
Function Coefficients
Sergio Bimbi Junior
Agenor de Toledo Fleury
Ronaldo Ruas
Vitor Chaves de Oliveira
ΝώϏϙϕϖϋϔχωωϋϙϙχϘϚϏωϒϋϏϙϊϏϙϚϘϏψϛϚϋϊϛϔϊϋϘχΌϘϋχϚϏϜϋΌϕϓϓϕϔϙΊϚϚϘϏψϛϚϏϕϔΌΌϒϏωϋϔϙϋͨώϚϚϖͧͿͿϝϝϝͨϙϋϚͨϕϘύͨψϘͿϏϐψϋͿ
13
ϊϕϏͧкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨмΠϋψΕϏϔϑͧώϚϚϖͧͿͿϊϞͨϊϕϏͨϕϘύͿкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨм
ΜΎΝΒΗΝΎΛΗΊΝΒΘΗΊΕΓΘΞΛΗΊΕΘΏΛΘΊΌΊΜΝΎΗΐΒΗΎΎΛΒΗΐΜΎΝΒΓΎϜͨлͥΊϘϚϏωϒϋмͥтϖͨ
ЫлйкпΜΎΝϘχϠϏϒϏχϔΜϕωϏϋϚϟϕόΝϋϒϋϜϏϙϏϕϔΎϔύϏϔϋϋϘϏϔύͿΒΜΜΗΙϘϏϔϚͧлннптлнпͿΒΜΜΗΘϔϒϏϔϋͧлннптнмл
Distortion Reduction in FIR Filters by
Approximation through Window Factor on
Function Coefficients
Sergio Bimbi Junior, Agenor de Toledo Fleury, Ronaldo Ruas, Vitor Chaves de Oliveira
São Paulo State Technological Research Institute – IPT
Program - Master in Industrial Processes
Abstract—Filters are time-invariant linear systems which are
able to modify the characteristics of the signals connected to
their input, so that only a specific portion of the frequency Figure 1. Digital filter architecture.
components in a signal can reach the output of the filter. In
dynamic systems, digital filters are applied in order to improve If the signals to be processed are digital, the diagram can
system measurements with regards to performance and be shown in summary form, as seen in Figure 2 (21).
stability. The present article demonstrates a modification in a
low pass filters having Hamming window within the sample
space π. In this development, the sample space π is subdivided,
wherein equation plots are added within a polynomial of order
n. This technique provides the removal of unwanted frequency
components in small angular frequency windows, providing the Figure 2. Digital filter architecture in summary form.
signal with acceleration towards the target when as compared to
a low pass filter having Hamming window. In dynamic In order to implement a digital and time-invariant (LTI)
measurement systems, this feature is relevant, considering that filter, a DSP is required, in which computational algorithms
the system shall have grater approximation to its target values, are arranged. These can be represented in the form of block
thus implementing an average which indicates the value being
diagrams using basic structures such as unit delays, gains,
acquired in a more accurate and repetitive manner.
adders, feedbacks and delay structures in the block diagram,
Key Words- Digital Filter, Digital Signal Processing, Low Pass which is similar to the order of difference equations (transfer
Filter, High Pass Filter, Filter Pass Band, Band Reject Filter. function) of the filter known as Canonical structure.
II. DIGITAL FILTERS VERSUS ANALOG FILTERS
I. INTRODUCTION The comparison relationship between an analog and a
Filters are time-invariant linear systems which are able to digital filter is directly linked to the complexity, precision and
modify the characteristics of signals connected to their input, design adaptability (22). An analog filter is economically
so that only a specific portion of the frequency components more recommended than a digital filter, but once the system
in a signal can reach the output of the filter. Considering is implemented, due to its discrete components, the system
analog signals x(t) and y(t) and a filter having impulse becomes too complex for adaptations and improvements.
response function h(t), as shown in Equation 1: Another interesting feature of analog filters, given that the
system is implemented by external components, is the change
ݕሺݐሻ ൌ ݄ሺݐሻ ݔ כሺݐሻ(1) in the characteristics of these devices with respect to
environmental factors such as temperature and humidity, thus
compromising their best theoretical performance. In turn,
In the frequency domain, the equation can be solved as seen digital filters, with respect to Analog to Digital conversion
in Equation 2. (AD) and Digital to Analog (DA) conversion and the
processing itself, have a poorer response time as compared to
ܻሺ݆߱ሻ ൌ ܪሺ݆߱ሻ ܺ כሺ݆߱ሻ(2) the analog filter. Digital filters also have errors inherent to the
quantization process (performed in AD conversion) and
Assuming an implemented digital filter is a Digital rounding errors due to the use of digital words having finite
Signal Processor (DSP) and the goal is to process an analog length. In recursive filters, this phenomenon can lead to
signal x(t), a digital filter system can be characterized as greater instability (23). On the other hand, the ability to
shown in Figure 1 (21). implement this filter within a processed system leads to
greater flexibility in the updating and maintenance of the
ΝώϏϙϕϖϋϔχωωϋϙϙχϘϚϏωϒϋϏϙϊϏϙϚϘϏψϛϚϋϊϛϔϊϋϘχΌϘϋχϚϏϜϋΌϕϓϓϕϔϙΊϚϚϘϏψϛϚϏϕϔΌΌϒϏωϋϔϙϋͨώϚϚϖͧͿͿϝϝϝͨϙϋϚͨϕϘύͨψϘͿϏϐψϋͿ
ϊϕϏͧкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨмΠϋψΕϏϔϑͧώϚϚϖͧͿͿϊϞͨϊϕϏͨϕϘύͿкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨм 14
ΜΎΝΒΗΝΎΛΗΊΝΒΘΗΊΕΓΘΞΛΗΊΕΘΏΛΘΊΌΊΜΝΎΗΐΒΗΎΎΛΒΗΐΜΎΝΒΓΎϜͨлͥΊϘϚϏωϒϋмͥтϖͨ
ЫлйкпΜΎΝϘχϠϏϒϏχϔΜϕωϏϋϚϟϕόΝϋϒϋϜϏϙϏϕϔΎϔύϏϔϋϋϘϏϔύͿΒΜΜΗΙϘϏϔϚͧлннптлнпͿΒΜΜΗΘϔϒϏϔϋͧлннптнмл
project. The architecture of a digital filter can be seen in IV. FIR FILTER
Figure 1. Finite Impulse Response (FIR) filters are those which show
a finitely long-lasting impulse response (25). This filter is
III. IIR FILTER characterized by a discrete transfer function which can be
Infinite Impulse Response (IIR) filters are those which seen in Equation 6.
have an infinitely long-lasting response to an impulse and
ಾ
ೖ ௭ ሺಾషೖሻ
ሺ௭ሻ
having recursive nature, thus, it can be concluded that this ൌ ೖసబ
(6)
ሺ௭ሻ ಾ
filter is characterized by relying on both the current input and
the previous input. Figure 3 illustrates the structure for This discrete function can be rewritten as a polynomial
developing an IIR filter, function x(k) is the input signal, function of negative powers of z. FIR filters have well-
values a1 to aQ and b0 to bP are coefficients representing the defined characteristics, which are:
type of filtering that is being performed (high-pass, low-pass,
band-pass) and function y(k) is the output signal as a result of 1) Finite memory, which stipulates that any transient has
the filtering of signal x(k) (24). limited duration.
2) Always developed as stable Bounded Input, Bounded
Output (BIBO).
3) It is possible to develop a desired magnitude response
having linear phase response.
In Figure 4, the structure of an FIR filter can be seen.
Figure 4.Structure for developing a digital FIR filter.
V. FILTER OR FIR IIR
The decision to evaluate which filter is the best for the
Figure 3. Structure for developing a digital IIR filter. application, i.e., whether to use an FIR filter or an IIR filter,
is directly connected to project-specific features. FIR filters
In mathematical aspects, the output of a digital IIR filter is have linear phase response, which implies that no phase
represented recursively as seen in Equation 3, where ak and distortion is produced in the filtered signal. This feature is
bk are filter coefficients. important in applications such as audio and image processing,
ஶ biomedicine and data transmission (26).
FIR filters are developed in non-recursive mode, thus they
ݕሾ݊ሿ ൌ ሾሿ כሾ െ ሿ ൌ
are always stable. This characteristic cannot be guaranteed for
ୀ
IIR filters. The effects of finite precision and quantization
errors are less severe for FIR filters (27).
ܾ כሾ݊ െ ݇ሿ ൌ IIR filters generally require less coefficients than FIR in
ୀ order to meet the same design specification. A lower order
filter has a shorter runtime. Analog filters may be converted
σெ
ୀଵ ܽ כ ݕሾ݊ െ ݇ሿ ൌ(3) to IIR digital filters fairly easily. In general, one can use an
IIR filter when the largest relevance is a very selective
By applying the z-transform we obtain the transfer response in the frequency domain, or when the
functions shown in Equation 4. implementation an analog filter is required. FIR filters may
be used when the number of coefficients is not very high (FIR
ሺ௭ሻ
ܪሺݖሻ ൌ (4) structure stability is guaranteed), and especially when the
ሺ௭ሻ
desired phase distortion is small (28).
By breaking down Y(z) and X(z), the transfer function
equation in an IIR filter is obtained as seen in Equation 5. VI. DESIGNING DIGITAL FILTERS
ಿ
ೖ ௭ షೖ
Digital Filter designs consist in determining a transfer
ܪሺݖሻ ൌ ೖసబ
ಾ (5) function that fits the frequency response specifications
ଵାೖసభ ೖ ௭ షೖ
ΝώϏϙϕϖϋϔχωωϋϙϙχϘϚϏωϒϋϏϙϊϏϙϚϘϏψϛϚϋϊϛϔϊϋϘχΌϘϋχϚϏϜϋΌϕϓϓϕϔϙΊϚϚϘϏψϛϚϏϕϔΌΌϒϏωϋϔϙϋͨώϚϚϖͧͿͿϝϝϝͨϙϋϚͨϕϘύͨψϘͿϏϐψϋͿ
15 ϊϕϏͧкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨмΠϋψΕϏϔϑͧώϚϚϖͧͿͿϊϞͨϊϕϏͨϕϘύͿкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨм
ΜΎΝΒΗΝΎΛΗΊΝΒΘΗΊΕΓΘΞΛΗΊΕΘΏΛΘΊΌΊΜΝΎΗΐΒΗΎΎΛΒΗΐΜΎΝΒΓΎϜͨлͥΊϘϚϏωϒϋмͥтϖͨ
ЫлйкпΜΎΝϘχϠϏϒϏχϔΜϕωϏϋϚϟϕόΝϋϒϋϜϏϙϏϕϔΎϔύϏϔϋϋϘϏϔύͿΒΜΜΗΙϘϏϔϚͧлннптлнпͿΒΜΜΗΘϔϒϏϔϋͧлннптнмл
required for a particular project or application. Within filter
applications, elements with well-defined characteristics exist: However, its behavior is best characterized as a function of
its frequency response as shown in Equation 8.
ஶ
1) Low-Pass Filter (LPF).
ఠ ሿ
2) High-Pass Filter (HPF). ܪሾ݁ ൌ ሾሿ݁ ିఠ ሺͺሻ
3) Band-Pass Filter (BPF). ୀିஶ
4) Band-Stop Filter (BSF).
Whereas h[n] can be shown by Equation 9.
Impulse responses to the ideal transfer functions are not గ
digitally achievable given that they have infinite length and ݄ሾ݊ሿ ൌ න ሺ݁ ఠ ሻ݁ ఠ ݀߱ሺͻሻ
are not causal. For FIR filters, one approach used is the ିగ
truncation of the impulse response of ideal filters (29).
For IIR filters, it is possible to map transfer functions With these determinations, ܪሺ݁ ఠ ሻ and h[n] pairs are
originally designed to analog filters to the z domain. obtained for ideal filters shown in Table 1.
VII. APPROXIMATION FOR FIR FILTERS
For implementing a filter, it is necessary to develop from
its transfer function, as shown in Equation 7.
ܪሾݖሿ ൌ ሾሿି ݖ ሺሻ
ୀିஶ
ΝώϏϙϕϖϋϔχωωϋϙϙχϘϚϏωϒϋϏϙϊϏϙϚϘϏψϛϚϋϊϛϔϊϋϘχΌϘϋχϚϏϜϋΌϕϓϓϕϔϙΊϚϚϘϏψϛϚϏϕϔΌΌϒϏωϋϔϙϋͨώϚϚϖͧͿͿϝϝϝͨϙϋϚͨϕϘύͨψϘͿϏϐψϋͿ
ϊϕϏͧкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨмΠϋψΕϏϔϑͧώϚϚϖͧͿͿϊϞͨϊϕϏͨϕϘύͿкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨм 16
ΜΎΝΒΗΝΎΛΗΊΝΒΘΗΊΕΓΘΞΛΗΊΕΘΏΛΘΊΌΊΜΝΎΗΐΒΗΎΎΛΒΗΐΜΎΝΒΓΎϜͨлͥΊϘϚϏωϒϋмͥтϖͨ
ЫлйкпΜΎΝϘχϠϏϒϏχϔΜϕωϏϋϚϟϕόΝϋϒϋϜϏϙϏϕϔΎϔύϏϔϋϋϘϏϔύͿΒΜΜΗΙϘϏϔϚͧлннптлнпͿΒΜΜΗΘϔϒϏϔϋͧлннптнмл
Filter type Response Impulse Figure 6. HPF behavior in the frequency domain.
magnitude response h[n]
ȁࡴሺࢋ࣓ ሻȁ
ఠ
1 for Ͳ ȁ߱ȁ ߱ for n = 0
గ
LPF ଵ
ሺ߱ ݊ሻfor n 0്
0 for߱ ൏ ȁ߱ȁ ߨ గ
ఠ
0 forͲ ȁ߱ȁ ߱ for n = 0
గ
HPF െ
ଵ
ሺ߱ ݊ሻfor n്
1 for߱ ൏ ȁ߱ȁ ߨ గ
0
ఠమ ିఠభ
0 forͲ ȁ߱ȁ ߱ଵ for n = 0
గ
1 for߱ଵ ȁ߱ȁ ଵ
BPF ߱ଶ
െ ሾ ሺ߱ଶ ݊ሻ െ
గ
ሾ ሺ߱ଵ ݊ሻሿ for n ്
0 for ߱ଶ ൏ ȁ߱ȁ ߨ 0
1 forͲ ȁ߱ȁ ߱ଵ ͳെቀ
ఠమ ିఠభ
ቁ for n =
గ
0
0 for ߱ଵ ȁ߱ȁ
BSF ߱ଶ ଵ
గ
ሾ ሺ߱ଵ ݊ሻ െ Figure 7. BPF behavior in the frequency domain.
1 for߱ଶ ൏ ȁ߱ȁ ߨ ሾ ሺ߱ଶ ݊ሻሿ for n ്
0
Table 1. Characteristics of ideal filters.
With Table 1 specifying the response behavior in
magnitude and the impulse response, one is able to check the
behavior of LPF, HPF, BPF and BSF filters respectively in
Figures 5, 6, 7 and 8.
Figure 8. BSF behavior in the frequency domain.
VIII. APPROXIMATION FOR FIR FILTERS BY USING
WINDOW FUNCTIONS
One way to have a better performance and overcome
limitations inherent to the realization of the impulse response
Figure 5. LPF behavior in the frequency domain.
function h[n] of ideal filters is to define an auxiliary sequence
h'(n) having a finite length of order M as described in
Equation 10.
ܯ
ሼܪሾ݊ሿܹሾ݊ሿǡ ȁ݊ȁ
݄Ʋሺ݊ሻ ൌ ൞ ʹ ሺͳͲሻ
ܯ
Ͳǡ ȁ݊ȁ
ʹ
The obtained response, despite having finite length, is not
causal. However, it can be transformed into causal by
multiplying by ି ݖெȀଶ , without distorting the magnitude
response and without destroying the linear phase property.
Sequence w[n] is known as the "window function" (30).
ΝώϏϙϕϖϋϔχωωϋϙϙχϘϚϏωϒϋϏϙϊϏϙϚϘϏψϛϚϋϊϛϔϊϋϘχΌϘϋχϚϏϜϋΌϕϓϓϕϔϙΊϚϚϘϏψϛϚϏϕϔΌΌϒϏωϋϔϙϋͨώϚϚϖͧͿͿϝϝϝͨϙϋϚͨϕϘύͨψϘͿϏϐψϋͿ
17 ϊϕϏͧкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨмΠϋψΕϏϔϑͧώϚϚϖͧͿͿϊϞͨϊϕϏͨϕϘύͿкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨм
ΜΎΝΒΗΝΎΛΗΊΝΒΘΗΊΕΓΘΞΛΗΊΕΘΏΛΘΊΌΊΜΝΎΗΐΒΗΎΎΛΒΗΐΜΎΝΒΓΎϜͨлͥΊϘϚϏωϒϋмͥтϖͨ
ЫлйкпΜΎΝϘχϠϏϒϏχϔΜϕωϏϋϚϟϕόΝϋϒϋϜϏϙϏϕϔΎϔύϏϔϋϋϘϏϔύͿΒΜΜΗΙϘϏϔϚͧлннптлнпͿΒΜΜΗΘϔϒϏϔϋͧлннптнмл
IX. FIR FILTERS USING WINDOW
The central idea of digital filter design by using windows X. DISTORTION REDUCTION WITHIN WINDOW IN
is to select a suitable filter having an ideal frequency (which AN FIR FILTER WITH POLYNOMIAL APPLIED IN
is always non-causal and has infinite impulse response) and RADIANS
then truncate its impulse response in a window in order to get The distortion reduction within the window performed in a
a causal FIR, linear phase filter. Considering an ideal low filter is an important factor with regards to rejection of
pass filter Hd(݁ ௪ ) having magnitude 1 and linear phase with unwanted points in present lobes, which are not acceptable
a pass band and zero response at the cutoff band, as seen in beyond the main lobe; this is because they represent
Equation 11 (31): frequencies that are not required for that filter. As the number
of M coefficients increases, the width of each sidelobe
ͳǤ ݁ ିఈ௪ ǡ ȁܹȁ ܹ decreases, but the area over them remains constant. This
݀ܪሺ݁ ௪ ሻ ൌ ൜ ሺͳͳሻ causes ripples to experience peaks near band edges. This fact
Ͳǡ ܹ ൏ ȁܹȁ ߨ
is referred to as "Gibbs Phenomenon."
The impulse response of this filter is infinite, given by In order to perform the proposed distortion reduction
Equation 12 and 13. within one window out of the many applicable, adaptation of
ͳ గ the window equation is necessary; in this article, the window
݄݀ሾ݊ሿ ൌ ࣣ ିଵ ሾܪௗ ሺ݁ ௪ ሻሿ ൌ න ܪሺ݁ ఠ ሻ݁ ఠ ݀߱ሺͳʹሻ used is the Hamming window shown in Equation 19, where
ʹߨ ିగ ௗ n is equal to the current filter coefficient and M is the total
number of coefficients.
ͳ ჯ ݊݁ݏሾჯ ሺ݊ െ ߙሻሿ
݄݀ሾ݊ሿ ൌ න ͳǤ ݁ ఠ ݁ ఠ ݀߱ ൌ ሺͳ͵ሻ
ʹߨ ିჯ ߨሺ݊ െ ߙሻ ଶగ
ݓሾ݊ሿ ൌ ͲǡͷͶ െ ͲͶ ሺ ) (19)
ெ
In order to obtain a causal linear phase FIR filter h[n] of
length M, as shown in Equation 14. The 2ߨ factor represents the entire spectrum of radian axes.
The technique enables the inclusion of new points to the
݄ ሾ݊ሿǡ Ͳ݊ ܯെͳ windowing; this occurs by dividing the 2ߨ radian spectrum
݄ሺ݊ሻ ൌ ൜ ௗ ሺͳͶሻ into smaller points in order to obtain a more efficient filter.
Ͳǡ ݈݁݁ݏ
This takes place with the implementation of an n-th
Where ߙ is represented by Equation 15. polynomial applied to 2ߨ, in addition to the 0.46 factor in
order to maintain linearity. In Figure 9 and 10, respectively,
ሺ ܯെ ͳሻ the full radian axis and the division of the total radian
ߙ ൌ ሺͳͷሻ spectrum can be viewed (33).
ʹ
This operation is called windowing. In general, h[n] may
be characterized by the product of ܪௗ ሾ݊ሿ in a window w[n] as
shown in Equation 16.
݄ሾ݊ሿ ൌ ܪௗ ሾ݊ሿǤ ݓሾ݊ሿሺͳሻ
Where w[n] is a symmetric function with respect to
ߙwithin the range Ͳ ݊ ܯെ ͳ, with zero being absent
from this range. It is useful to mention that, depending on how
w[n] is obtained, different filter techniques can be obtained as
shown in Equation 17, defining a rectangular window (32).
ͳǡ Ͳ݊ܯെͳ
ݓሺ݊ሻ ൌ ቄ ሺͳሻ
Ͳǡ ݈݁݁ݏ
Thus, it is necessary to define a finite h[n] sequence for the
digital filter from ܪௗ ሾ݊ሿǤ The Fourier transform is a discrete
sequence in time, whereas H(݁ ௪ ሻ is continuous. The causal
FIR filter H(݁ ௪ ሻ response is obtained in the frequency
domain by the convolution of H(݁ ௪ ሻ and the window
response W(݁ ௪ ሻ, as shown in Equation 18. Figure 9. Full 2ߨ radian spectrum.
ܪሾ݁ ఠ ሿ ൌ ܪ൫݁ ௪ ൯ ܹ כሺ݁ ௪ ሻ
ͳ గ
ൌ න ሺ݁ ఠ ሻܪௗ ሺ݁ ሺ௪ିఒ ሻ݀ߣሺͳͺሻ
ʹߨ ିగ
ΝώϏϙϕϖϋϔχωωϋϙϙχϘϚϏωϒϋϏϙϊϏϙϚϘϏψϛϚϋϊϛϔϊϋϘχΌϘϋχϚϏϜϋΌϕϓϓϕϔϙΊϚϚϘϏψϛϚϏϕϔΌΌϒϏωϋϔϙϋͨώϚϚϖͧͿͿϝϝϝͨϙϋϚͨϕϘύͨψϘͿϏϐψϋͿ
ϊϕϏͧкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨмΠϋψΕϏϔϑͧώϚϚϖͧͿͿϊϞͨϊϕϏͨϕϘύͿкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨм 18
ΜΎΝΒΗΝΎΛΗΊΝΒΘΗΊΕΓΘΞΛΗΊΕΘΏΛΘΊΌΊΜΝΎΗΐΒΗΎΎΛΒΗΐΜΎΝΒΓΎϜͨлͥΊϘϚϏωϒϋмͥтϖͨ
ЫлйкпΜΎΝϘχϠϏϒϏχϔΜϕωϏϋϚϟϕόΝϋϒϋϜϏϙϏϕϔΎϔύϏϔϋϋϘϏϔύͿΒΜΜΗΙϘϏϔϚͧлннптлнпͿΒΜΜΗΘϔϒϏϔϋͧлннптнмл
Figure 12. Conduction window in the time domain when
changing w[n] (Equation 20).
By applying the discrete Fourier transform, one is able to
verify the behavior of both implementations in the frequency
Figure 10. Full 2ߨ radian spectrum divided into smaller
domain. In Figure 13, the filter behavior in the frequency
points.
domain can be seen; it is possible to check the pass region and
cut-off frequency as a function of system specifications. One
The central focus of this process is to minimize the effect
can also verify peaks at band regions, or "Gibbs phenomenon"
of lower lobes not included in the cutoff frequency
(34).
determined within the system, which consists of the sum of
window w[n] in a polynomial having factor nx-1 for nx ≤ 1.
The described implementation can be seen in Equation 20.
ଶೣషభ గ
ݓሾ݊ሿ ൌ ሾͲǡͷͶ െ ͲǡͶ௫ିଵ ሺ )]
ெ
ଶೣషమ గ
+ሾͲǡͷͶ െ ͲǡͶ௫ିଶ ሺ )]
ெ
ଶೣషయ గ
+ ሾͲǡͷͶ െ ͲǡͶ௫ିଷ ሺ )]
ெ
௫ି௫ ଶೣషೣ గ
+ሾͲǡͷͶ െ ͲǡͶ ሺ )] (20)
ெ
Equation 20 describes the sum of a polynomial applied to
w[n] applied to a Hamming window as an example.
In Figures 11 and 12, respectively, the driving windows
can be seen in the time domain, applied to Equation 19 and
20, respectively.
Figure 13. Filter having Hamming window in the
frequency domain, with w[n] unchanged.
In Figure 14, the filter behavior in the frequency domain
can be seen, with a pass frequency region better defined at its
cutoff frequency, as well as the reduction in "Gibbs
phenomenon" effects, where it is present in the lower than
zero region, so that, by applying the modulus, this
phenomenon will be eliminated.
Figure 11. Conduction window in the time domain
without changing w[n] (Equation 19).
ΝώϏϙϕϖϋϔχωωϋϙϙχϘϚϏωϒϋϏϙϊϏϙϚϘϏψϛϚϋϊϛϔϊϋϘχΌϘϋχϚϏϜϋΌϕϓϓϕϔϙΊϚϚϘϏψϛϚϏϕϔΌΌϒϏωϋϔϙϋͨώϚϚϖͧͿͿϝϝϝͨϙϋϚͨϕϘύͨψϘͿϏϐψϋͿ
19 ϊϕϏͧкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨмΠϋψΕϏϔϑͧώϚϚϖͧͿͿϊϞͨϊϕϏͨϕϘύͿкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨм
ΜΎΝΒΗΝΎΛΗΊΝΒΘΗΊΕΓΘΞΛΗΊΕΘΏΛΘΊΌΊΜΝΎΗΐΒΗΎΎΛΒΗΐΜΎΝΒΓΎϜͨлͥΊϘϚϏωϒϋмͥтϖͨ
ЫлйкпΜΎΝϘχϠϏϒϏχϔΜϕωϏϋϚϟϕόΝϋϒϋϜϏϙϏϕϔΎϔύϏϔϋϋϘϏϔύͿΒΜΜΗΙϘϏϔϚͧлннптлнпͿΒΜΜΗΘϔϒϏϔϋͧлннптнмл
Figure 14. Filter having Hamming window in the frequency Figure 16. Filter having modified Hamming window in
domain with changes in w[n]. the time domain.
XI. EQUATION APPLICATION IN THE TIME DOMAIN XII. CONCLUSION
The development of a low pass filter with Hamming The inclusion of a modification in Equation 19, turning it
window computationally consists of the implementation of a into Equation 20, provides benefits in the implementation of
"First In - First Out", wherein its coefficients are accumulated digital filters having windowing, which one is able to verify
in the time domain and multiplied by the calculated filter in the plots shown in Figures 13 and 14 with respect to the
coefficient. The example shown in Figure 15 consists of a low frequency spectrum, in addition to a reduction of the "Gibbs
pass filter having Hamming window, with a cutoff frequency phenomenon". The minimization of spurious frequencies
of 1Hz and 100Hz sampling rate. To that filter, 136 with the polynomial calculation within the filter window
coefficients are applied to a signal that emulates a harmonic minimizes the passing of spurious frequencies and enables
developed by using an approximation sequence of the number the reduction of the number of coefficients applied; this
ߨǤ aspect computationally reduces processing time in what we
envision as dynamic systems that require limited time, such
as dynamic measurement systems. From a time domain point
of view, one can verify the increase in response time and
linearity with regards to the faster convergence of the
obtained reading towards the actual value to be measured,
thus removing spurious frequencies from these samples.
Figure 15. Filter having Hamming window in the time
domain.
Figure 16 shows the same computational implementation
mentioned in Figure 14 with the modification of the
coefficients calculated in accordance with Equation 20
(modified Hamming - Bimbi). This will demonstrate the
increased linearity and speed of convergence to the target
point.
ΝώϏϙϕϖϋϔχωωϋϙϙχϘϚϏωϒϋϏϙϊϏϙϚϘϏψϛϚϋϊϛϔϊϋϘχΌϘϋχϚϏϜϋΌϕϓϓϕϔϙΊϚϚϘϏψϛϚϏϕϔΌΌϒϏωϋϔϙϋͨώϚϚϖͧͿͿϝϝϝͨϙϋϚͨϕϘύͨψϘͿϏϐψϋͿ
ϊϕϏͧкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨмΠϋψΕϏϔϑͧώϚϚϖͧͿͿϊϞͨϊϕϏͨϕϘύͿкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨм 20
ΜΎΝΒΗΝΎΛΗΊΝΒΘΗΊΕΓΘΞΛΗΊΕΘΏΛΘΊΌΊΜΝΎΗΐΒΗΎΎΛΒΗΐΜΎΝΒΓΎϜͨлͥΊϘϚϏωϒϋмͥтϖͨ
ЫлйкпΜΎΝϘχϠϏϒϏχϔΜϕωϏϋϚϟϕόΝϋϒϋϜϏϙϏϕϔΎϔύϏϔϋϋϘϏϔύͿΒΜΜΗΙϘϏϔϚͧлннптлнпͿΒΜΜΗΘϔϒϏϔϋͧлннптнмл
REFERENCES [23] Cabezas, J. C. E. & Diniz, P. S. R. (1990). FIR filters using
interpolated prefilters and equalizers. IEEE Transactions on
[1] Adams, J.W. (1991a). FIR digital filters with least-squares Circuits and Systems, 37, 17–23.
stopbands subject to peak-gain constraints. IEEE Transactions [24] Chang, T.-L. (1981). Suppression of limit cycles in digital filters
on Circuits and Systems, 39, 376–88. designed with one magnitude-truncation quantizer. IEEE
[2] Adams, J. W. (1991b). A new optimal window. IEEE Transactions on Circuits and Systems, CAS-28, 107–11.
Transactions on Signal Processing,39, 1753–69. [25] Charalambous, C. & Antoniou, A. (1980). Equalisation of
[3] Adams, J.W. & Willson, Jr., A. N. (1983).Anew approach to FIR recursive digital filters. IEE Proceedings – Part G, 127, 219–25.
digital filters with fewer multipliers and reduced sensitivity. 865 References
IEEE Transactions on Circuits and Systems, CAS-30, 277–83. [26] Chen, W.-H., Smith, C. H. & Fralick, S. C. (1977). A fast
[4] Adams, J.W. &Willson, Jr., A. N. (1984). Some efficient digital computational algorithm for the discrete cosine transform. IEEE
prefilter structures. IEEE Transactions on Circuits and Systems, Transactions on Communications, COM-25, 1004–9.
CAS-31, 260–5. [27] Cheney, E.W. (1966). Introduction to Approximation Theory.
[5] Ahmed, N., Natarajan,T.&Rao, K. R. (1974). Discrete cosine NewYork,NY: McGraw-Hill.
transform. IEEE Transactionson Computers, C-23, 90–3. [28] Churchill, R.V. (1975). Complex Variables and Applications.
[6] Akansu, A. N. & Medley, M. J. (1999). Wavelets, Subband and NewYork,NY: McGraw-Hill.
Block Transforms in Communications and Multimedia. Boston, [29] Claasen, T. A. C. M., Mecklenbrauker,W. F. G. & Peek, J. B. H.
MA: Kluwer Academic Publishers. (1975). On the stability of the forced response of digital filters
[7] Antoniou,A. (1982). Accelerated procedure for the design of with overflow nonlinearities. IEEE Transactions on Circuits and
equiripple nonrecursive digital filters. IEE Proceedings – Part Systems, CAS-22, 692–6.
G, 129, 1–10. [30] Cochran, W., Cooley, J., Favin, D. et al. (1967). What is the fast
[8] Antoniou, A. (1983). New improved method for design of Fourier transform? IEEE Transactions on Audio and
weighted-Chebyschev, nonrecursive, digital filters. IEEE Electroacoustics, AU-15, 45–55.
Transactions on Circuits and Systems, CAS-30, 740–50. [31] Cohen, A., Daubechies, I. & Feauveau, J. C. (1992).
[9] Antoniou, A. (1993). Digital Filters: Analysis, Design, and Biorthogonal bases of compactly supported wavelets.
Applications, 2nd edn. New York, NY: McGraw-Hill. Communications on Pure and Applied Mathematics, XLV, 485–
[10] Antoniou, A. (2006). Digital Signal Processing: Signals, 560.
Systems, and Filters. New York, NY: McGraw-Hill.Antoniou, [32] Constantinides, A. G. (1970). Spectral transformations for digital
A. & Lu, W.-S. (2007). Practical Optimization: Algorithms and filters. IEE Proceedings, 117, 1585–90.
Engineering Applications. New York, NY: Springer. [33] Cooley, J.W. &Tukey, J.W. (1965).An algorithm for the machine
[11] Antoniou, A. & Rezk, M. G. (1977). Digital filters synthesis computation of complex Fourier series. Mathematics of
using concept of generalizedimmitance converter. IEE Journal Computation, 19, 297–301.
of Electronics Circuits, 1, 207–16. Antoniou, A. & Rezk, M. G. [34] Cortelazzo, G.&Lightner, M. R. (1984). Simultaneous design of
(1980).Acomparison of cascade and wave fixed-point both magnitude and group delay of IIR and FIR filters based on
digitalfilter structures. IEEE Transactions on Circuits and multiple criterion optimization. IEEE Transactions on
Systems, CAS-27, 1184–94. Acoustics, Speech, and Signal Processing, ASSP-32, 9949–67.
[12] Apostol, T. M. (1967). Calculus, 2nd edn., volume I. Toronto:
Xerox College Publishing. Avenhaus, E. (1972). On the design
of digital filters with coefficients of limited wordlength. IEEE
Transactions on Audio and Electroacoustics, AU-20, 206–12.
Sergio Junior Bimbi (e-mail: [email protected])
Bauer, P. H. & Wang, J. (1993). Limit cycle bounds for floating- is pursuing a Master's degree in Electrical Engineering (IPT)
point implementations of second-order recursive digital filters. and holds a Bachelor’s degree in Telecommunications
IEEE Transactions on Circuits and Systems II: Analog and Engineering from the FIEO University Center. He is currently
Digital Signal Processing, 39, 493–501. Corrections on 41, 176,
February 1994.
an Electronics Engineer at Masipack and University
[13] Belevitch, V. (1968). Classical Network Theory. San Francisco, Professor at the São Paulo State Technological College
CA: Holden-Day. (FATEC). Sergio has experience in Electrical Engineering
[14] Benvenuto, N., Franks, L. E. & Hill, Jr., F. S. (1984). On the with emphasis in Embedded Electronics and
design of FIR filters with power-of-two coefficients. IEEE
Transactions on Communications, COM-32, 1299–307.
Telecommunications, developing hardware for numerical
[15] Bhaskaran, V. & Konstantinides, K. (1997). Image and Video controls, industrial automation and VHDL programming. He
Compression Standards: Algorithms and Architectures. Boston, has developed research at the UNIFIEO University Center
MA: Kluwer Academic Publishers. Bomar, B.W. (1989). On the with FPGAs that are currently used in many subsequent
design of second-order state-space digital filter sections. IEEE
Transactions on Circuits and Systems, 36, 542–52.
projects. He has expertise with developing measurement
[16] Bomar, B.W. & Joseph, R. D. (1987). Calculation of L∞ norms systems, automatic weighing scales, PLC programming,
in second-order state-space digital filter sections. IEEE control systems, weight checkers, digital filters for the field
Transactions on Circuits and Systems, CAS-34, 983–4. of measurements, etc. He develops research in the field of
[17] Bomar, B.W., Smith, L. M. & Joseph, R. D. (1997). Roundoff
noise analysis of state-space digital filters implemented on
SDR (QPSK) with ZYNQ family integrated circuits having
floating-point digital signal processors. IEEE Transactions on dynamic partial reconfiguration.
Circuits and Systems II: Analog and Digital Signal Processing,
44, 952–5. Vitor Chaves de Oliveira (e-mail: [email protected]) is a
[18] Boyd, S. & Vandenberghe, L. (2004). Convex Optimization.
Cambridge, UK: Cambridge University Press. Bracewell, R. N.
doctoral candidate in Electrical Engineering (Mackenzie),
(1984). The fast Hartley transform. Proceedings of the IEEE, 72, Master in Electrical Engineering (PUCC), attending
1010–18. specialization in Electrical Engineering and Power Systems
[19] Bracewell, R. N. (1994). Aspects of the Hartley transform. (UNISAL), Bachelor of Computer Science (UNISAL) and
Proceedings of the IEEE, 82, 381–6.
[20] Burrus, C. S. & Parks, T. W. (1970). Time domain design of
Computer Technician (SENAI). University Professor
recursive digital filters. IEEE Transactions on Audio and working in the fields of Electrical Engineering and Computer
Electroacoustics, AU-18, 137–41. Science; Magazine editor for the Brazilian Society of
[21] Butterweck, H. J. (1975). Suppression of parasitic oscillations in Television Engineering (SET); IT/ERP/Telecom
second-order digital filters by means of a controlled-rounding
arithmetic. Archiv Elektrotechnik und Übertragungstechnik, 29,
Infrastructure Consultant at Coach IT Consulting. He is an
371–4. author of books in the cloud computing field and has
[22] Butterweck, H. J., van Meer, A. C. P. & Verkroost, G. (1984). published dozens of articles in scientific journals and at
New second-order digital filter sections without limit cycles. national and international conferences. His research interests
IEEE Transactions on Circuits and Systems, CAS-31, 141–6.
are focused on Telecommunications: IP networks, Radio
ΝώϏϙϕϖϋϔχωωϋϙϙχϘϚϏωϒϋϏϙϊϏϙϚϘϏψϛϚϋϊϛϔϊϋϘχΌϘϋχϚϏϜϋΌϕϓϓϕϔϙΊϚϚϘϏψϛϚϏϕϔΌΌϒϏωϋϔϙϋͨώϚϚϖͧͿͿϝϝϝͨϙϋϚͨϕϘύͨψϘͿϏϐψϋͿ
21 ϊϕϏͧкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨмΠϋψΕϏϔϑͧώϚϚϖͧͿͿϊϞͨϊϕϏͨϕϘύͿкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨм
ΜΎΝΒΗΝΎΛΗΊΝΒΘΗΊΕΓΘΞΛΗΊΕΘΏΛΘΊΌΊΜΝΎΗΐΒΗΎΎΛΒΗΐΜΎΝΒΓΎϜͨлͥΊϘϚϏωϒϋмͥтϖͨ
ЫлйкпΜΎΝϘχϠϏϒϏχϔΜϕωϏϋϚϟϕόΝϋϒϋϜϏϙϏϕϔΎϔύϏϔϋϋϘϏϔύͿΒΜΜΗΙϘϏϔϚͧлннптлнпͿΒΜΜΗΘϔϒϏϔϋͧлннптнмл
Broadcasting (Digital TV), Electromagnetic Theory, Control
Algorithms and Mathematical Modeling of Systems.
Agenor de Toledo Fleury (e-mail: [email protected] )
Bachelor of Mechanical Engineering from ITA / Aeronautics
Technological Institute (1973), Master of Mechanical
Engineering from the University of São Paulo (1978) and
PhD in Mechanical Engineering from the University of São
Paulo (1985). He is currently a professor at FEI University
Center, where he is Coordinator for the Mechanical
Engineering Graduate Program and a PhD professor at the
Polytechnic School of the University of São Paulo. He has
experience in various projects related to Mechanical
Engineering, with emphasis on Dynamic and Control
Systems. His most recent projects address the modeling and
control of nonlinear systems, optimal control and estimation
of dynamic system states in Biomechanics, Robotics,
Automotive Engineering and Embedded Systems
applications. He received the SAE Brazil 2010 Engineering
Education Award.
Ronaldo Ruas (e-mail: [email protected])
Bachelor’s degree in Material Processes and Electronic
Components from São Paulo State Julio de Mesquita Filho
University (1998). Master's degree in Electrical Engineering
from the University of São Paulo (2001). PhD in Electrical
Engineering from the University of São Paulo (2006).
Currently collaborates as a guest researcher at the integrated
systems laboratory (LSI), in the plasmas and new materials
field, at the Polytechnic School of the University of São
Paulo. He works at Protec Brazil as a technical manager in
the development of plasma-assisted thin film processes
(PVD, PECVD). He teaches courses in the fields of
electronics, Microprocessors and Applied Electricity at the
Osasco Technological College, where he is a coordinator for
the Bachelor of Industrial Automation Technology course.
θХаСаФХЯНЮаХПШСΓ
ηХЩОХοЮΔΑψСЮУХЫΑλШСбЮеΑζУСЪЫЮРСωЫШСРЫΑχбНЯΑχЫЪНШРЫНЪРРС
τШХвСХЮНΑϋХаЫЮθФНвСЯΒӔӒӓӘΔιХЯаЫЮаХЫЪχСРбПаХЫЪХЪλξχλХШаСЮЯ
ОеζЬЬЮЫдХЩНаХЫЪаФЮЫбУФόХЪРЫгλНПаЫЮЫЪλбЪПаХЫЪ
θЫСТТХПХСЪаЯΔψκωξσωκχσζωξτσζροτϊχσζρτλ
ηχτζιθζψωκσμξσκκχξσμΔξψψσυЮХЪаΓӔӖӖӘάӛӔӖӘξψψσ
τЪШХЪСΓӔӖӖӘάӛӖӕӔΔРЫХΓӓӒΔӓӚӗӚӒΫЯСаХЦОСΔӔӒӓӘΔӕΔόСОρХЪЧΓФааЬΓΫΫ
РдΔРЫХΔЫЮУΫӓӒΔӓӚӗӚӒΫЯСаХЦОСΔӔӒӓӘΔӕ
ΝώϏϙϕϖϋϔχωωϋϙϙχϘϚϏωϒϋϏϙϊϏϙϚϘϏψϛϚϋϊϛϔϊϋϘχΌϘϋχϚϏϜϋΌϕϓϓϕϔϙΊϚϚϘϏψϛϚϏϕϔΌΌϒϏωϋϔϙϋͨώϚϚϖͧͿͿϝϝϝͨϙϋϚͨϕϘύͨψϘͿϏϐψϋͿ
ϊϕϏͧкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨмΠϋψΕϏϔϑͧώϚϚϖͧͿͿϊϞͨϊϕϏͨϕϘύͿкйͨксосйͿϙϋϚϏϐψϋͨлйкпͨм 22
