The present invention relates to the field of digital signal processing. Specifically, it relates to finite impulse response (FIR) digital filters, and a novel method for setting their coefficients.
Digital filters play a central role in digital signal processing (DSP) which is one of the foundations of modern electronics technology. The use of linear phase digital filters is significant in applications requiring preserving the integrity of the time domain signal profile by minimizing distortions. Such applications might include digital communications, digital audio-visual signal processing, loT devices, VUI systems, and digital flight control systems. It is known in the art that certain FIR digital filter types exhibit an exact linear phase response and are therefore well suited for such applications.
Several circuit configurations are known in the art for realizing an FIR digital filter. FIG.1 represents the circuit configuration of a transversal type cascaded delay line structure for realizing a linear phase FIR filter. In circuit (1), the filter order is M and the number of taps is M+1, and there are M delay units (2-1 to 2-M) connected in cascade constituting a shift register. The filter input x[i] enters the first delay unit through terminal T1. Each delay unit z−1=e−jω, delays the input signal by one sampling period. The output of each delay unit is multiplied by a perspective multiplier h[n], where there are M+1 multiplier, (3-0 to 3-M). All multiplier outputs are added together in an adder circuit (4), where the output of the adder is the filter output y[i] at terminal T2.
The circuit described in
y[i]=Σ
n=0
M
h[n]x[i−n], (1)
where the current output y[i] is the summation of M+1 present and past inputs. The coefficient sequence h[n] is identified as the finite impulse response of this filter. Letting {tilde over (X)}(z) and {tilde over (Y)}(z) be the complex valued z-transforms of the input and output signals, respectively, equation (1) becomes
with the complex transfer function {tilde over (H)}(z) of the filter being defined as
Herein, the real-valued impulse response h[n] appears as the Inverse z-transform of {tilde over (H)}(z). It is customary to define the magnitude |{tilde over (H)}(z)|=|{tilde over (H)}(ejω) as the positive-valued amplitude response, and the argument arg({tilde over (H)}(z))=arg{tilde over (H)}(ejω) as the phase response. The real-valued zero-phase amplitude response, denoted as H(ω), can be defined for symmetric h[n] as
An FIR filter has a linear phase response if its impulse response h[n] is either even-symmetric or odd-symmetric in structure. Furthermore, the impulse response h[n] is real-valued only if |{tilde over (H)}(ejω)| is even-symmetric in frequency.
There are a few ways known in the art to design a linear phase band selective FIR filter, among which the windowing method has had historical significance. The method relies on first finding the ideal impulse response of an idealized model of the magnitude response of the filter. As an example, the magnitude response of an ideal low-pass filter can be defined as
where ±ωc are symmetric cutoff frequencies. This magnitude response is therefore even-symmetric giving a real-valued ideal impulse response hideal[n]. Applying the Inverse Discrete Time Fourier Transform (IDTFT) gives
This ideal impulse response has infinite length and is acausal being centered around n=0. The next step is truncating it to finite length M+1 and modulating it by applying a window function w[n], where there is a plethora of such functions used in the art. The impulse response becomes hw[n]=hideal[n]×w[n]. Next, for the case M is an even number, the impulse response is shifted by half of the filter order M/2, thereby making the impulse response causal and centered around M/2. The resulting causal windowed impulse response becomes
The transfer function of this practical windowed filter (ejω) is calculated as the Discrete Time Fourier Transform (DTFT) of hw[n]
(ejω)=Σn=0Mhw[n]e−jnω.
Because this is only a finite term transform, and because of the Gibbs phenomenon, |(ejω)| can only be an approximation of |
(ejω)|. In general, the larger the filter length M+1 (the number of taps), the more accurate |
(ejω)| will resemble the ideal filter (except for the Gibbs phenomenon which is an intractable problem for the discontinuous model used). Practical considerations require using the smallest number of taps to satisfy a set of filter specifications.
An alternative method popular in the art for the independent control of passband and stopband specifications is the Parks-McClellan method based on the iterative Remez exchange algorithm. Herein, the maximum tolerance errors in the filter bands are minimized in the Chebyshev equiripple sense. The method is optimal for obtaining the minimum number of taps to realize any target set of specifications. However, the method has the disadvantage that the iterative algorithm employed is not always guaranteed to converge. Another disadvantage of the Parks-McClellan method is having the built-in artifact of passband equiripple behavior which introduces a level of distortion to the filtered signal. In addition, concern has been expressed that the passband ripples cause pre-echoes and post-echoes in the time domain impulse response with noticeable undesirable effects (Multirate Signal Processing for Communication Systems, Fredrick Harris, Prentice Hall PTR, 2004 and Dunn, Julian. “Anti-Alias and Anti-Image Filtering: The Benefits of 96 kHz Sampling Rate Formats for those who cannot hear above 20 kHz” AES 104th Convention, May 16-19, 1998. Amsterdam, The Netherlands. Preprint 4734).
For the reasons discussed in [0010], methods have been proposed to modify the Remez algorithm to produce flatter passbands (K. Steiglitz, “Optimal design of FIR digital filters with monotone passband response,” IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-27, pp. 643-649, December 1979, and P. P. Vaidyanathan Optimal Design of Linear-Phase FIR Digital Filters with Very Flat Passbands and Equiripple Stopbands, IEEE Transactions on Circuits and Systems, VOL. CAS-32, NO. 9, 1985 and FIR Filter and setting Method of Coefficients Thereof, Y. Mogi, K. Nishibori, Soni Corporation, USOO7028061B2).
The present invention provides a method for designing narrow transition bandwidth FIR filters, based on the windowing method, characterized by having passbands that are flatter than Parks-McClellan equiripple filters, being free from the equiripple artefact.
A major aspect of this invention is producing a finite impulse response h[n] representing a modification of a conventional windowing impulse response hw[n], wherein a compensated impulse response hc[n] is first obtained by summing an auxiliary impulse response ha[n] to the windowing impulse response hw[n] to obtain hc[n]=hw[n]30 ha[n], wherein hc[n] may be subject to further manipulation. The auxiliary impulse response ha[n] is purposefully selected as to induce sought improvements in the filter amplitude response.
According to another aspect of the present invention, the finite impulse response of the filter h[n] is obtained by applying a discrete time modulation m[n] to the compensated impulse response as h[n]=m[n]×hc[n], written alternatively as
h[n]=m[n]×(hw[n]+ha[n])=m[n]×hw[n]+m[n]×ha[n].
According an embodiment of this invention, the said discrete time modulation maybe an identity modulation, in which case m[n]=1, resulting in h[n]=hc[n].
Another aspect of this invention is using the windowing impulse response hw[n] as a starting point in the process for obtaining h[n]·hw[n] can be calculated in the prior art manner described by equations (2)-(4) for the case of a low-pass filter. However, hc[n] is not limited to only this type of filter, any ideal even-symmetric magnitude response (ejω) could be used to obtain a corresponding ideal impulse response hideal[n], followed by the application of a window function w[n].
Another aspect of this invention is that ha[n] and hw[n] can always be made to have the same length M+1. A unified length is needed to perform the sum hw[n]+ha[n]. Starting with impulse responses ha[ ] and hw[ ] that are not of the same length, zero-valued impulses can be added at the edges of the shorter of the two as needed to increase its length to be the same as the longer one (zero padding).
Another aspect of this invention is using an auxiliary impulse response ha[n] having an auxiliary transfer function (ejω)=Σn=0Ma[n]e−jnω with a phase response that is identical to the phase response of the transfer function of the windowed impulse response
(ejω)=Σn=0Mhw[n]e−jnω), thereby enabling a well-defined interference of the amplitude responses of the two transfer functions upon the sum hw[n]+ha[n].
Another aspect of the invention that follows from is that the compensated transfer function (ejω) given by the compensated impulse response hc[n], and being equal to
(ejω)=
(ejω)+
(ejω), has a zero-phase amplitude response that is the sum of the zero-phase amplitude responses, Hc(ω)=Hw(ω)+Ha(ω). This well-defined interference allows for straightforward setting the input parameters of Ha(ω) and Hw(ω) to achieve the desired response of Hc(ω).
According to an embodiment of the present invention the transfer function of the filter {tilde over (H)}(ejω) may be calculated as the convolution of the DTFT of the discrete time modulation function {tilde over (M)}(ejω)with the compensated transfer function (ejω), namely {tilde over (H)}(ejω)={tilde over (W)}(ejω)*
(ejω). In this case, the impulse response of the filter h[n] is obtained as the IDTFT of {tilde over (H)}(ejω).
Another aspect of this invention is using an auxiliary impulse response ha[n] producing an auxiliary amplitude response Ha(ω) comprising a plurality of pulses, each with a dominant central lobe and weaker side lobes, or an amplitude response being the sum or difference of the said plurality of pulses.
Another significant aspect of this invention is using an auxiliary impulse response ha[n] producing an auxiliary amplitude response Ha(ω) having pulses with dominant central lobes that interfere with the windowed amplitude response Hw(ω) in the transition band regions, increasing the magnitude of the transition slopes and inducing faster transitions between the filter bands, thereby reducing the bandwidth of the transition bands.
Another significant aspect of this invention is using of an auxiliary impulse
response ha[n] producing an auxiliary amplitude response Ha(ω) having pulses with weak side lobes that interfere destructively with the stopband ripples of the windowed amplitude response Hw(ω) close to the transition band edge, thereby increasing stopband attenuation.
A preferred embodiment of this invention is using an auxiliary impulse response ha[n] having an even-symmetric structure, and having the functional profile of one of the window functions known in the art. Such a specification yields an amplitude response having the said multi-lobed pulse forms. The pulse height, width and position on the frequency axis are controllable by independent parameters in the mathematical description of the said function. Moreover, being even symmetric, ha[n] can produce a linear phase response.
Another preferred embodiment of this invention is using an auxiliary impulse response ha[n] that is based on the rectangular window (Dirichlet window) defined by
and generating an auxiliary transfer function (ejω) having a zero-phase amplitude response Ha_dirichlet(ω), known as a Dirichlet Kernel, which is given by
This amplitude response has the required multi-lobe pulse form of the embodiments above. The Dirichlet Kernel has zeroes on the frequency axis located at
Another significant aspect of this invention is prescribing a mathematical means for frequency shifting the pulses of the amplitude response Ha(ω) of the auxiliary impulse response ha[n] on the frequency axis so that they are positioned strategically to have the desired interference effects mentioned in [0022] and [0023]. This is achieved by using the known procedure of applying a phase modulation to the discrete time domain impulse response ha[n], resulting in a frequency shift of the transfer function (ejω) in the frequency domain, as given by the transformation equation
with ωo being the value of the frequency shift.
Another aspect of this invention that is necessary for any real-valued impulse response, is having the pulses of the auxiliary amplitude response Ha(ω) to occur in pairs, wherein the pair members are symmetrically shifted to opposite sides of the symmetry axis centered at zero frequency by equal magnitude frequency shifts ±ωo. This is achieved by applying the method in to create the said pair of pulses, with the auxiliary impulse response set to be even-symmetric as
A preferred embodiment of this invention is producing the pulse pair of [0027] by using the Dirichlet impulse response of [0025]. The auxiliary impulse response ha[n] becomes
This gives an auxiliary transfer function (ejω) defining a pair of frequency shifted Dirichlet pulses in its zero-phase amplitude response Ha(ω) given by
The factor εo serves to scale symmetrically the amplitudes of the pulse pair.
One embodiment of this invention is using an auxiliary impulse response ha[n] providing one frequency shifted Dirichlet pulse for each transition band in the windowed amplitude response Hw(ω), where this situation is henceforth referred to as a “singlet”. The windowed amplitude response Hw(ω) must be even-symmetric for a real-valued impulse response, resulting in that the transition bands occur in pairs. Therefore, a “singlet” situation involves using an auxiliary amplitude response Ha(ω) that includes the terms
with one of the pulse pair members being used for each member of the said transition band pairs. For a one-pair transition band filter, such as a low-pass or high-pass filter, the auxiliary impulse response that produces a singlet becomes,
Another embodiment of this invention is using an auxiliary impulse response ha[n] providing the difference of two adjacent frequency shifted Dirichlet pulses for each transition band in the windowed amplitude response Hw(ω), this situation is henceforth referred to as a “doublet”. The auxiliary amplitude response Ha(ω) in this case will include the terms
with ω1 being the shift of the second pulse pair, and the factor ε1 serves to scale symmetrically their amplitudes. Therefore, a “doublet” situation involves using the two pulses inside each bracket of the expression above for each of the said transition band pair members. For a one-pair transition band filter, such as a low-pass or high-pass filter, the auxiliary impulse response that produces a doublet becomes,
Another preferred embodiment of this invention is using an auxiliary impulse response ha[n] producing an amplitude response Ha(ω) that is designed to have one of its zeroes match the frequency of each of the cutoff frequencies in the amplitude response of the windowed impulse response Hw(ω).
Paragraphs [13]-[31] provide disclosure to produce the FIR impulse response of the invention, h[n]. For h[n] to be effective in obtaining filters of required performance, suitable choices must be made for the values of the independent parameters in its mathematical expression of h[n]. An aspect of the present invention is providing a method for setting the values of the independent parameters to achieve the required filter performance.
An aspect of this invention is to provide a method for determining the characteristics of an FIR digital filter with coefficients set to the impulse response of the invention h[n]. This is done by carrying out a plurality of numerical computer simulations to generate the transfer function of the filter {tilde over (H)}(ejω)=Σn=0Mh[n]e−jnω, from which the magnitude response |{tilde over (H)}(ejω)| can be determined.
Another aspect of this invention is to perform the said simulations with input parameters set to be the independent parameters defining the mathematical description of h[n]. These are identified to be the length of the windowed impulse response M+1, the cutoff frequencies of the windowed amplitude response ωc, the pulse frequency shifts ωo and ω1, the pulse amplitude parameters εo and ε1, and the input parameters of an adopted windowing design method, wherein the Kaiser window design method serves as a non-limiting preferred embodiment, with transition bandwidth Δ, stopband attenuation SB, and passband ripple PB, being the input design parameters.
Another aspect of this invention is producing the said plurality of transfer function simulations, wherein the input parameter values are stepped iteratively, in discretized selected ranges that are unique to each parameter, whereby the simulations cover the chosen input parameter combinations. The magnitude response |{tilde over (H)}(ejω)| for each transfer function is calculated, and the set of the output filter characteristics for each amplitude response are determined. The output filter characteristics include the magnitude of the band gain, the peak to peak ripple, and the transition bandwidth, of all bands and transition regions of the magnitude response |{tilde over (H)}(ejω)|. The output filter characteristic sets and the input parameters sets that produce them are recorded in computer data files which are used to plot families of multi-variable graphs of the interdependence between the input parameters and output characteristics. Graphical analysis is performed on the said graphs by applying a mathematical technique of multi-variable curve fitting, whereby generating a plurality of interpolation formulae that model the mapping of the input parameters to the output characteristics;
Paragraphs [33]-[35] provide disclosure of the process for obtaining the necessary filter characteristics data to be used for designing an FIR digital filter based on the aspects and embodiments of this invention.
Another aspect of this invention is a design process for an FIR digital filter having a target set of filter specifications. The design process depends on the type of filter sought which might include, but is not limited to the low-pass, high-pass, band-pass and band-reject varieties. The design process also depends on the choice of the embodiments of the auxiliary impulse response ha[n], including the choice of the window function, whether a singlet design or a doublet design is used, and whether zero alignment to cutoff frequency is implemented ([0031]).
Another aspect of this invention is a design process, notwithstanding the particularities mentioned in [0037], comprising the generalized steps of:
Another aspect of the invention is the implementation of the filter characterization and design method [0038] of the preferred embodiment of this invention by a computer program, whether being custom-written or implemented on a commercial software platform, in any computer language.
Another aspect of this invention is a computer program to emulate numerically the filtering action of an FIR digital filter, by calculating the transfer function of the filter using filter coefficients designed by the methods of this invention, and then using this transfer function to filter a discrete time input signal, thereby obtaining the output signal.
An important aspect of the present invention aims to design an FIR filter circuit having the direct form delayed line structure of
The present invention is described further as being a non-iterative method for
designing an FIR filter claiming advantages of both the windowing method and the Parks-McClellan equiripple method. The filter of the invention, compared to two other filters of the mentioned types having the same set of filter specifications, provides: (i) a passband lacking the equiripple artefact, being as flat as the windowing filter with only half of a ripple cycle to one ripple cycle depending on the number of auxiliary pulses used, (ii) a transition response slope almost identical to that of the equiripple filter, (iii) a larger attenuation in the deep stopband compared to the equiripple filter, and (iv) unlike the equiripple method which relies on convergence conditions for achieving the set specifications where convergence is not guaranteed, precise values of the specifications can be implemented with the method of the invention. The cost of achieving these advantages is a relatively small increase in the number of filter taps over to the optimal Parks-McClellan equiripple filter, when compared to the number of taps required by the windowing filter. Generally, the filter of the invention requires about only one third of the extra taps needed by the windowing filter, but can be even less depending on the specifications.
So that the manner in which the above recited features of the present disclosure can be understood in detail, a more particular description of the disclosure, may have been referred by embodiments, some of which are illustrated in the appended drawings. It is to be noted, however, that the appended drawings illustrate only typical embodiments of this disclosure and are therefore not to be considered limiting of its scope, for the disclosure may admit other equally effective embodiments.
These and other features, benefits, and advantages of the present disclosure will become clearer by reference to the following figures , wherein:
rectangular window impulse response, the doublet auxiliary impulse response and the compensated impulse response;
embodiment for a low-pass filter and that of a Kaiser windowing filter;
Detailed embodiments of the preferred mode are described herein; however, it is to be understood that such embodiments are exemplary of the present disclosure, which may be embodied in various alternative forms. Specific process details disclosed herein are not to be interpreted as limiting, but merely as a representative basis for teaching one skilled in the art to variously employ the present disclosure in any appropriate process.
The terms used herein are for the purpose of describing exemplary preferred
embodiments only and are not intended to be limiting. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the described methods and mathematical forms do not preclude the presence or addition of one or more steps, terms or operations other than a mentioned step, term or operation.
The embodiments of the present disclosure will now be described more fully hereinafter with reference to the accompanying drawings, which form a part hereof, and which show, by way of illustration, specific example embodiments.
The detailed description of the preferred embodiment includes the design of a low-pass filter. But it is understood that the methods and mathematical formulae of the invention are not limited to this filter type only but can be applied as effectively to other types of band selective filters such as high-pass, band-pass band-stop and other designs.
A preferred mode for a low-pass filter is to assign hw[n] to be the rectangular window impulse response applied to equation (4), which becomes
with n running from 0 to M, and fc being the normalized cutoff frequency (in Hz),
Also, a preferred mode is to assign the auxiliary impulse response ha[n] to be the “doublet” impulse response given in equation (5), and rewritten here as
with n running from 0 to M. This impulse response gives an auxiliary amplitude response having the two pairs of Dirichlet pulses for each of the pairs of transition bands, as explained in [0030].
The frequency shifts fo and f1 appearing in equation (6) are assigned the values
to make both the first zero of the first pulse, and the second zero of the inverted second pulse to coincide in frequency with the cutoff frequency fc.
To exemplify the action of the auxiliary impulse response embodiment, reference is made to
The next crucial step in the method is the application of the discrete time modulation m[n], acting as a frequency domain averaging that diminishes both the stopband and the amplified passband ripples of the amplitude response of the filter.
Also, a preferred mode is to assign the discrete time modulating function m[n] to be defined as a zeroth-order Bessel function of the first kind (also known as Kaiser window) given in normalized form as:
with n running from 0 to M, and with Io being the symbol for the zeroth order Bessel function, β is a bandwidth parameter, and M is the filter order. Reference is made to the equation h[n]=m[n]×(hw[n]+ha[n]) in [0014] which defines the impulse response of the filter h[n] as
h[n]=m[n]×h
w
[n]+m[n]×h
a
[n].
The first term m[n]×hw[n] , being expanded as
is exactly the impulse response of a low-pass filter of the Kaiser windowing method. The second term m[n]×ha[n] expanded as
gives an amplitude response having two pulses each of which corresponds to a frequency shifted DTFT of the Kaiser window.
It is concluded from the description in [0069] that the preferred embodiment under consideration is tantamount to a Kaiser windowing method for a low-pass filter that is compensated by two frequency shifted Kaiser window DTFT pulses in each transition band of the filter. A design method can therefore be pursued based on the Kaiser window design method, which is well known in the art. This method stipulates that a low-pass filter having a stopband attenuation SB and a transition bandwidth Δ, can be realized by a filter of order M and window bandwidth parameter β, being given by the two equations
The Kaiser window filter passband ripple is designated PBK, and is given by the expression, PBK=20 log(1+δ) with
The inclusion of the auxiliary doublet pulses in the amplitude response of the Kaiser window filter has the effect of narrowing the transition bandwidth Δ, and increasing the stopband attenuation SB for a wide range of design values. This occurs at the expense of introducing exactly one prominent passband ripple cycle at the transition band edge, wherein the ripple is larger in magnitude than the characteristic Kaiser window filter passband ripples. This is something that can be tolerated to a certain extent as explained in [0009].
Also, the preferred embodiment outlines a method for quantifying the effect of the doublet pulse amplitude parameters εo and ε1 on the filter characteristics. First, it is emphasized that εo, the amplitude of the first pulse, is treated as an independent parameter, while ε1, being the amplitude of the second pulse, is treated as a dependent parameter related to εo by the relation ε1=μεo. Here, μ is called the symmetry parameter, and has a typical value in the range 0.3-0.7, being also dependent on SB. According to the present embodiment, μ is adjusted such that the two opposite peaks of the said prominent single passband ripple are symmetrical about their mean passband value.
Also, the preferred embodiment includes performing computer simulations for calculating the filter transfer function {tilde over (H)}(ejω) from the filter impulse response h[n] by applying the DTFT, from which the magnitude response |{tilde over (H)}(ejω)| is calculated. The filter characteristics comprises the filter's actual passband ripple peak PBout, the actual transition bandwidth Δout and the actual stopband attenuation SBout. These characteristics are determined from the magnitude response |{tilde over (H)}(ejω)| by using numerical techniques. The magnitude response |{tilde over (H)}(ejω)|, being equal to |H(ω)|, is preferred over the zero-frequency amplitude response H(ω) for the purpose of calculating the characteristics because |{tilde over (H)}(ejω)| is positive definite, thus enabling the use of the standard dB logarithmic scale.
In this preferred embodiment, a broad set of filter characteristics is obtained by iterating the Kaiser window design parameter values SB, Δ and the auxiliary parameter εo, while running the said simulations for the transfer function, and calculating and recording the actual (output) filter characteristics PBout, Δout and SBout. The results quantifying the change from the set window filter values to the output values upon the insertion of the auxiliary pulses are given via the fractional change parameters. These are defined as
While for PBout, it is convenient to give the output value additively via the change in the passband ripple peak PB as
PB
out
=PB+PB
K, or alternatively PB=PBout−PBK.
SBout, Δout and PBout converge to SB, Δ and PBK, respectively, for the case εo=0, i.e. when the auxiliary impulse response is zero. The parameters SB, Δ and εo are stepped in discretized selected ranges that are unique to each parameter. SB is stepped from 35 to 90 dB, Δ is stepped from 0.025 to 0.15 Hz, and εo is stepped from zero to a value εmax that produces a maximum of PBout=0.5 dB (1 dB passband peak to peak ripple). A parameter εF is defined as
Being directly proportional to εo, εF is instrumental in relating the first pulse amplitude parameter εo to the change in the peak passband ripple PB. εF ranges from 0 to 1, corresponding to a change in PBout from PBK to 0.5 dB, wherein PBK is a small quantity for most simulations, but still non-negligible in general.
the parameter εF (introduced in [0066]) are plotted for different values of SB. The numbers (35-90) on the curve lines identify the curve line for the indicated value of SB. It is observed that the larger εF the larger the value of PB. The curve lines can be identified in the lower left half of the graph through their linear extension. The data is plotted in transposed form (i.e. εF vs. PB), being convenient for obtaining suitable interpolation equations necessary for the design process. The interpolation equations for εF are obtained in a two variable curve fitting process of εF (PB, SB) as a function of the two variables PB and SB. It was found that εF depends minimally on Δ.
The description of the preferred embodiment includes a non-iterative design method that is based on the archived and plotted characteristics for the Kaiser-window-based doublet displayed in
The referenced interpolation equations in steps d) to h) are included in the computer program description that implements the design method above, where the program being presented in the APPENDIX.
The preferred embodiment of the invention also includes a computer program for the design of a doublet type impulse response of a low-pass filter. The executable source code is written in MATHCAD software syntax and is listed in full detail in the APPENDIX.
To exemplify the characteristics of the filter and the benefits obtained by the method of this embodiment, the results of a typical design process for a low-pass filter are presented. A typical set of specifications for a low-pass filter are assigned as:
It is instructive to undertake a comparison of the exemplified filter of the preferred embodiment designed above in with filters designed by the two conventional methods referred to in this disclosure, being the Parks-McClellan equiripple and the windowing methods. To achieve this, the two other filter types are designed by their well-known prior art techniques using the same set of specifications listed in [0082].
The table herein provides a comparative summary of the numerical properties of the three filters of figures
In order to make a meaningful comparison between the filter tap requirements of the various filters, a reference should be made to the minimum number of taps NPM needed by the optimal Parks-McClellan equiripple filter that satisfy the specification requirements. We define the savings fraction as
where NKW is the number of taps required by the uncompensated Kaiser window filter, and NINV is the number required by the invention filter embodiment. The larger the savings fraction, the more filter taps that are saved. It was found that the savings depend on the passband ripple and stopband attenuation.
The terms and descriptions used herein are set forth by way of illustration only and are not meant as limitations. Examples and limitations disclosed herein are intended to be not limiting in any manner, and modifications may be made without departing from the spirit of the present disclosure. Those skilled in the art will recognize that many variations are possible within the spirit and scope of the disclosure, and their equivalents, in which all terms are to be understood in their broadest possible sense unless otherwise indicated. In the following paragraphs, we set several examples of alternative embodiments.
The method of the invention is not limited to using only the Kaiser window but can be applied to any conventional window function. Embodiments were successfully developed for several conventional window functions, wherein the optimal discrete prolate spheroidal sequence (DPSS) window showed the best filter tap efficiency for the same set of specifications. However, the absence of an analytic closed form expression for the DPSS window complicates the design process and makes its utilization more demanding computationally.
The design method of the preferred embodiment is not limited to the linear phase low-pass filter type only. Indeed, the method of this embodiment can be thought of a design process for the transition band itself between the bands of a filter, and thus being generalizable to other filter types. By changing the positions and amplitudes of the compensating pulses, it is straightforward for those skilled in the art to adapt the method for designing other filter types. Demonstrations of a band-pass filter and a band-stop filter are shown in FIG.13 and
As disclosed in the invention disclosure, the auxiliary pulses are not limited to the doublet scheme of the preferred embodiments, other quantities and sequences of compensating pulses are possible.
The appendix lists a computer program code in PTC Mathcad syntax. The program implements the design procedure of the preferred embodiment disclosed in [0089], wherein the corresponding steps are listed in the comments. The program determines the FIR filter coefficients of a linear phase low-pass filter having target specifications. The complete program code comprises of:
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/IB2020/061125 | 11/25/2020 | WO |