TW580623B - Apparatus, methods, and computer program products for accurately determining the coefficients of a function - Google Patents

Abstract

The present invention provides apparatus, methods, and computer program products for providing coefficients of a function representative of a signal. In one embodiment, the apparatus, methods, and computer program products of the present invention, taking advantage of the independence of samples, provide complete sets of coefficients of the function as each sample is received and corrects the updated coefficients for accuracy by applying a rotating reference system. As such, the apparatus, methods, and computer program products of the present invention are capable of providing accurate coefficients with decreased latency. In another embodiment, the apparatus, methods, and computer program products of the present invention use learning models to derive complete sets of coefficients of functions as each sample is received. In another embodiment, the apparatus, methods, and computer program products of the present invention use learning models to determine apparatus, methods, and computer program products for generating coefficients of functions.

Description

580623 V. Description of the invention (0 The present invention is a provisional patent application with a serial number of 60/1/1 5346. The application was filed on February 26, 2000 and the name of the invention is " Processed information analysis and method and apparatus (mETHODS aND aPPARATUS FOR PROCESSING aND aNALYZING iNFORMATION) ", and application serial No. 60 / 201,609 of U.S. provisional Patent application (provisional patent application), which is filed on May 3 by the year 2000 and the inventors The name is "METHODS AND APPARATUS DERIVED FROM LEARNING MODELS FOR PROCESSING AND ANALYZING INFORMATION", which is derived from the learning model, and the priority is', and the content of the two applications is combined in in the present application, in order to present invention, there is mainly a function of the coefficients of a decision by reference use. in particular, the apparatus of the present invention, a method and computer program product is about upon receiving an input signal of each sampling, decisions Represents a function of the input signal to provide continuous updates Coefficients. ',' Signal processing is an important function of many electronic systems. In particular, in many electronic systems, data is transmitted in the form of signals. In addition, = 2 / system is by observing signal characteristics, such as vibration signals or other Class 1 ^ Series' are used to analyze and monitor the actions of mechanical or chemical systems. Some people have developed methods to characterize signals so that resources can be used here for data processing. For example, in many electronic systems, the time domain (t fork domain) signal is usually converted to a frequency ^ me

ΐΠ: strengthen calculated iC〇mPUtati〇nal intensive). This is the reason the ancient piano 'delay because some functions might require the use of a particular method at ΪΪ letter ί:;: capital: ί management. F: And more importantly, the use of multiple data processing shores may take an unreasonably long time to process the application. In fact, much information is available. … The first goal is the ability to process data signals in real time (without delay)

In terms of 12, the Fourier series is defined as: In other words: infinite sequence coefficients. ® For this, the use of -Fourier Sequences known times can be counted countless times. To remedy this problem, a number of generations of the ϋ = processing system based using * bit Fourier Transform (DFT), to take "^ leaf sequence. The digital Fourier transform (DFT) coefficient bit approximation of the Fu f columns, for processing digital Analogy information. Heavier flat U digital Fourier transform (DFT) uses the N-limit 有限 in a finite period to sample-a finite set to replace the benefit of the Fourier sequence.

: Take this L 'of the digital Fourier transform (DFT) to provide the computing system, the same number of two reception sampling factor, rather than the one desired sequence Fourier ..., limit the number of samples. Thus, use of the digital Fourier transform (DFT) means may be provided to present most satisfactory to handle 'II) so However, since the time required for reducing the weight of the processed signal = ways to further reduce the number of performing a one-bit signal The number of calculations required for Fourier transform (DFT). In particular, the digital converter (DFT) coefficient for each program is calculated based a similar process. A general coefficient processing system: normalizing each weight and independent variable by the angular velocity

Page 8 580623

In particular when the moving aperture Fourier transform (the SAFT) in 'thousand and as described in PCT application W0 / 67146 described the contribution of the process is started by subtracting the weight coefficients oldest shell taken together and offer the latest sampling factor, An error will be introduced in the calculation coefficient. This error occurs because the continuous processing of the coefficients causes a shift in the angles associated with each sample. For example, in digital Fourier transform (DFT) applications, a specific angle is related to the first sample, and the far angle is used to calculate the coefficient contribution of the first sample. Similarly, a specific angle is related to the second sample and every other sample in the "batch". Moving aperture Fourier transform (SAFT) maintains a batch of samples, and one of the samples is continuously changed. There are a total of eight samples, which means that the oldest sample (sample 1) is removed and the previous sample 2 will now become sample 1. This step will occur for each sample so that the latest sample becomes sample 8. However, while moving based coefficient update period (the SAFT) coincident with the aperture of Fourier transform coefficients previously sampled position angle of the contributions calculated deduced other words, will update coefficient comprises: using an error of the previous sample is applied to calculate the angular coefficient of the contribution An ideal moving aperture Fourier transform (SAFT) process. When receiving a new sample after receiving the original 8 samples (that is, N = 8), then sample 2 should occupy the position of sample 1 in the calculation, and The corresponding angle of sample 1 is used to generate the coefficient contribution, and so is each subsequent sample. However, 'this problem is not in the US patent Application 09/56221 and PCT application WO 0 0/67 1 46 are mentioned in the application of moving aperture Fourier transform (SAF Ding). Conversely, the moving aperture Fourier transform described in these patent applications ( SAFT) uses the previously calculated coefficients as the current system

V. Description of the invention (9) The update coefficient including new sampling is obtained. In other words, when the original N samples are processed and a new sample is received, the moving aperture Fourier transform (SAFT) of this ίμ is only subtracting the compensation of the first application = 0 and adding compensation for new samples. The moving aperture Fourier transform (SAFT) described in these two patents does not describe a hug: move the previously received samples 2-8 so that they occupy the position of the next one in the calculation and use the next corresponding to the angular position of the sampling to generate a new set of coefficients. In contrast, in the original collection of the sample, the second sample of the 2 coefficients contribution (which is now the first new set of sample) x is still connected to the angle of the second sample, rather than the first sample angle. ▲ In view of this, there is a need for a moving aperture Fourier transform (SAFT) leaf heterogeneity capability to quickly and accurately perform processing coefficients. In addition, 'there is still a general need for improvement methods, devices, and computer program products' for faster and more efficient processing of function coefficients. The apparatus, method, and computer program product of the present invention are provided to overcome many shortcomings of using signals such as Fourier transform to process signals. In particular, the present invention provides devices, methods, and computer program products to take advantage of reduced delay and / or improved accuracy to determine coefficients representing a function of an input signal. In addition, the present invention provides a device, a method, and a computer program product to determine a moving aperture coefficient representing a function of an input signal, such as a moving aperture Fourier transform (SAFT), so that these coefficients are both accurate and quickly obtained. Furthermore, the present invention provides a device, method, and computer program product that can be derived from the device, method, and computer program product to take advantage of reduced latency and / or improved accuracy

Page 14 580623

In order to make a decision, it is necessary to receive each device, so that these coefficients are used for each access. This signal indicates that a signal indicates that it is the method of delaying the sample extraction and using at least the previous angle input for the device 'to provide a 0 degree change obtained by the computer program. It also provides a device function to produce a computer program. Products to provide a signal, a method, and a signal that represent a signal, a product that expresses the accuracy of a signal, its coefficients, and the updated ones, by which, the method and the electrical coefficient. The coefficients of the coefficients of this generation device and square function are a function of the brain equation. The low-latency brain equations using the coefficients in the reception are also described and the electrical numbers. The number of products is complete. In particular, each sample of each sample is corrected to provide an accurate product. The loan device and brain program can be collected. The invention updates the relationship between compensation and production methods when it is independent. In one embodiment, the apparatus of the present invention includes a coefficient generator, so that each time a sample is received, and when the sample is received, the function coefficient is updated based on the sample without having to wait until the next sample is received. Therefore, an accurate, updated coefficient of a complete set can be obtained at the time of receiving each sample. In another embodiment, the device, method and computer program product of the present invention first store each of the samples when receiving the samples, and generate a first set of coefficients when receiving the last sample of the samples. . In addition, when a new sample of the input signal is received, after receiving the predetermined samples, the device, method and computer program product of the present invention apply a function related to the coefficients to the sample, so that according to the new sample, generating an item (Term) for each coefficient. To replace the new sampling

Page 15 580623 V. Description of the invention (11) The first sampling, the generation items of the new sampling are subtracted from the first sampling items previously stored in the memory device, which are related to the sampling. After this subtraction step, the coefficients are updated using the difference between the items corresponding to the first sample of the new sample and the scheduled samples. Then, the precision of the ^ coefficient is adjusted by adjusting the phase angle to the corresponding correct sampling value. ^ In another embodiment of the present invention, substituted at the latest sampling oldest sample such sampling apparatus of the present invention, a method and computer program product is subtracted corresponding to those sampled by each of the coefficients in the most The old sampled items are added to each coefficient. Then, the coefficient accuracy can be improved using a coefficient correction step, which is used to compensate for the change in the angles associated with these samples, so that the phase angle corresponds to the correct sample value. The phase angle correction step is accomplished by applying a rotating reference system that correctly matches the samples, phase angles, and coefficient contributions. Another feature of the system of the present invention may be substituted for the standard method and apparatus for electronic calculation of the mathematical information, such as: computing Fourier transform (F〇urier Transform), Laplace Transform (Laplace transform), polynomial equations (polynomial formula) and equation sequence (series formula). An embodiment of the present invention includes a method and a device for performing calculations using an algorithm derived from a learning model, such as: a Fourier transform, a Laplace transform, and a polynomial equation ) and sequence equation (series formuia). Suitable learning model comprises a neural network learning mode according to the (n e u r a 1 n e t w square r k) of

Page 16 580623 V. Description of the invention (12) type, learning model based on fuzzy logic network, learning model based on gene tica 1 gor i thm, and different types of learning The learning model that the model combines. Another feature of the system of the present invention comprises a method, apparatus and method so as to generate two alternative design choices, which functions to generate coefficients, such as: Fourier transform (Fourier transform), Laplace Transform (Laplace transform), Polynomial equation (pollynomial formula) and sequence equation (series formuia). The method of generating two alternative designs includes: using a learning model to derive the capital

At least one component of a Λ processor (compom seven). Examples of components of the information processor include: algorithms, circuits, and electronic devices that are relevant in calculations. [Simple description of the diagram] The first diagram A is a schematic diagram of a rotating reference system, which is used to sample one of the input and input signals and decide to represent a number of functions of the input signal. Among them, 'for each A set of received samples is output according to an embodiment of the present invention. Illustration. Figure-Figure B is a schematic diagram of the sample size introduced in Figure-Figure 8. Xixi: This is a schematic diagram of an image related to the adjustment of necessary coefficients according to an embodiment of the present invention. Of FIG: d; a block diagram according to which - the input signal, the; signals into the wheel - the exact coefficient function, which receives the sample, a set of system accuracy factor of the present invention

One embodiment of page 580623 and the third diagram is used to show the four diagrams of the input signal according to the input signal, which is used to determine the number of the input signals and one of the root function. The first number of the complex input letter set is shown in the sixth figure. In the sixth and fifth figures, the six figures are sampled according to one. FIG coefficient based on the apparatus shown in FIG nine lines for the wheel to be shown in FIG. According to an input signal based coefficient function of the present invention according to - one of the input signals in accordance with the present invention. Utilization factor the number of memory Pei apparatus according to a block of one of one of the input signal of the input signal according to the present invention, which performs the operation according to the present invention, one side of the 'memory function of the utilization factor, according to the present invention in accordance with the present invention administration set used with correction executed. Means for performing the correction coefficient in accordance with the present invention to make a block diagram of one embodiment of apparatus of the embodiment, sampled by the shutter is set to determine the number table PEI. In one embodiment, a sample of one of the blocks performing the action is determined to indicate that one of the devices in the embodiment of the input letter is a sample, a function map, which is placed in the gate, and an embodiment block diagram, the device, and the use of the coefficients. . It is based on a gate device for each and it is based on a gate device. For each block diagram, the memory device determines to receive the sampled output. The input signal indicates the input set to receive an embodiment and output a device plus one rotation reference step of the device. The input signal is used to represent the sample. A block diagram is used to compare with the first embodiment. The rotation reference step is applied when the action block is executed, so as to use 0

Page 18 580623 V. Description of the invention (14) — ___ The tenth figure is a neural network model that is used to map one of the last two examples of humans according to this report. The coefficient of Jukou and the latest sampling to the next set are the schematic diagrams of the examples. 笫 十 Used to introduce the image at the beginning of the training :: example. $ + _ Stomach Π Logic member function. • According to an image sound diagram according to an embodiment of the present invention, t is used to describe possible fuzzy logic member functions at the end of training. ..., the body of the second operating block diagram, the implementation of which is described in Gram for rapping the body of the free electron & Procedure to be performed into the time.

A fourteenth embodiment of an image diagram of FIG system according to one embodiment of the present invention, to introduce in the " sea transistor (seas of transist〇rs) " - A fragment "(fragment), which may be used to program a gene ( genetic programming). [detailed description] backbone of the present invention, one of the electrical model of the crystal (skelet〇n)

^ Hereinafter, the present invention will be described in more detail with reference to the accompanying drawings and preferred embodiments of the present invention (shown in the drawings). However, the present invention may also be combined with different types of embodiments, and should not be limited to the described embodiments; on the contrary, these embodiments are intended to make the disclosure description more sufficient and complete, and to extend the full scope of the present invention Completely passed to those familiar with this technology. In the text, similar symbols refer to similar elements. As mentioned above, data processing systems have been developed to determine the coefficients representing a function of a signal. However, because many of these systems are very complex and cannot calculate the coefficients before receiving all samples of the signal, these systems cannot provide real-time analysis of the signal. Moreover, although improved systems have been developed to take samples one by one

Page 19 5. The basis of the description of the invention (15) (sample by sample) ^ proved to be insufficient for some specific application coefficients, and the accuracy of the coefficients is the opposite. The present invention provides a device that can Provide a representation of Jd :: Yes, too heterogeneous to obtain these coefficients at low latency. The special kind of uniqueness is the use of the previous; two: i products, which use each to take a coefficient. The updated coefficients are corrected to compensate for the correlation: the angle t 'is used to obtain accurate coefficients with reduced delay. The invention also provides a device, a method and a computer program product for generating coefficients representing a function of a signal. The present invention also provides apparatus, method, and computer program product, which means for generating, methods and computer program products to provide for one of the functions represent the coefficients of a signal. S for the purposes of said introduction of the various devices, the present invention method, and computer-based product with the formula private Fourier sequence (Fourier series) Introduction and features described below. However, it is clear that the apparatus, method and computer program product of the present invention should be able to use many different types of functions. For example, 'the apparatus, method, and computer program product of the present invention can use the following functions', such as: Bessel function, Legendre polynomials, type 1 and type 2 Shebyshev polynomials (Tchebysheff polynomials of first and econd kind), Jacobi polynomial (Jacoby polynomials), El universal La polynomial (Generalized laguerre polynomials), Hermite polynomials (Hermite polynomials

P. 20 580623

) Two Bernoulli polynomial (Bernoulli polynomiais), Euler Polynomial (Euler polynomials) and provide less order mentioned by m columns of this table is definitely not exclusive In various quantum mechanics, a linear function analysis, and microwave classical geometry (Fractals) Yes, but it is only provided as an example. This method can be applied to any function as long as it can be expressed as a sequence of values. In addition, these and other functions that do not appear in the above table are quite common. This method provides a method to develop parallel computing devices and methods, and removes the redundancy of mechanical practices. It is also compatible with mechanical implementation.

Row. An important concept of the present invention is to determine the independence of each sample of a function coefficient. This concept is described and described in more detail in U.S. Patent Application 09 / 56,221 and PCT Patent Application WO 00/67 1 46 'and therefore only a brief explanation is provided below. The independence of each sample can be introduced by the connotation of Fourier transform. Fourier transform (Fourier transform) is performed according to the orthogonality principle based (principle of orthogonality). When the purposes of a Fourier transform (Fourier transform) in the digital Fourier transform (DFT) in

At the same time, it provides a device to judge the size of each frequency component of a signal completely independently. Baseband frequency coefficient (base freqUency) of consecutive integer multiples used in the calculation. The fundamental frequency is the time period required to acquire a set of samples. The sample of this signal is then multiplied with each component of the far-set orthogonal function and added after one or several fundamental frequency cycles have been corrupted. Each obtained coefficient is the size of the real and imaginary parts of a test frequency. The important thing is,

580 623

V. Description of the invention (17) The calculation of a coefficient is independent of the collection of other systems, which uses digital Fourier reed. In a sample with N samples, each sample is transformed according to the applicable angle. 1) Transform or cosine to compensate the sine (_) of the number, which is the mother coefficient of the function. In one embodiment, the present invention is' a new sampling line based on the use of compensating means $ f Yan two new methods and computer program based on - a set of samples of the current number of Kojima and remove more compensation oldest sample. Each; ^ dogwood solid Fourier coefficient, n-sampling of the classical Fourier included, this exhibition person called on # n Tea conversion (transform F〇urler

"Γ ί: Same as the old set, and the difference is only in the new connection, J 2 medium = subtraction of the sample and correction of the coefficient, which is used to compensate for the change in the degree of the sample. This update step can be performed at each coefficient This is done in the earlier update procedure. The coefficient correction of N samples is basically equivalent to changing the reference system of these samples. This step is also equivalent to rotating the reference system to match the appropriate samples to the proper angle. Therefore, the rotating reference The system redefines the position of the 51st solid sample. The position of the first sample must be redefined because

For the oldest sample (ie the original first sampling) system removed and added to the latest sampling system has a complete new set of samples of Ν. The rotating reference system causes sample 1 to be related to the angle of sample 1, sample 2 is related to the angle of sample 2, and so is each sample. This coefficient correction step includes: • Correcting the compensation of each sample for such coefficients, so that using the compensation of each sample can have the correct angle. Now refer to the first figure, which shows a diagram of a rotating reference system

Page 22 580623 V. Description of the invention (21) The 10th sampling system has a related vector, which is at the same angle with respect to the 2nd sampling, so it is marked as S2 '. After the arrival of the tenth sample S2 ', the updated coefficient system Cn is expressed by using equation (11).

Cn = C, + S2, -S2 (11) This calculation is also simultaneously applied to the update coefficient, A1 and B1 mapping. Equation 0 2) is a sample calculation of A and B. A = A + S9 * cos (45) -Siscos (1 * 45) (12) B = B + S9 * sin (45) -Sl * sin (1 * 45), however, although the coefficient for C, calculated above based sufficient and correct, but the calculation of the coefficient a and the coefficient B (which is a reference map based on the system) but needs to be rotated to adjust the reference system, if this adjustment step is missing, then its value (as in equation (12) as (Derived) will be inaccurate and will differ from the classical Fourier coefficients. In general, the coefficient is the first sample of the set, which is weighted using an appropriate 45-degree trigonometric function (cosine for coefficient a, sine for coefficient B), and the second sample ' utilizing (2 * 45) degree angle of two loading function, and a trigonometric function corresponding to a weighting of each sample, is defined. However, since a new collection (collection of samples comprising sampling from 2 to 9 of) the first set of original sample (sample containing 8} to the collection of samples) of the second sampling, and the system using (2 * 45) of the triangle function to calculate the weighted, so that an error will occur coefficient updating step. In view of this, the present invention corrects errors in this coefficient update step. Wherein, by providing the correct coefficient based a rotating reference system in order to get to. Image-wise expressed, which describes the first line in Figure VIII. The first picture "shows the use of the X axis 3a (which is also marked as χ45 in the first picture) and the corresponding

A set of five standard methods 580,623, the inventors described (25) using the acquired coefficients obtained one of the actual samples. _ Fine Fourier coefficient transform coefficients based, shellfish |] which may be utilized, such as a flash; li;: transducer (fast Fourier transform) method to unlock the digital Fourier Ping = change (the DFT), in U.S. Patent Application 〇9 / 56〇221 Patton and pcT application in the 00 ^ 67146 (Pelt〇n) method, or any conversion coefficient deriving the Fourier method to obtain. Alternatively, the initial set is a " dummy " set coefficient. In fact, it is also appropriate to derive the coefficients of a starting set using the complement 0 = method. Optional coefficient substantially beginning of the start set having great flexibility. The beginning of the set of coefficients may be stored in the open - in memory, such as: - a computer-readable memory. The next step based on the reception has to be represented using the coefficients of these - new samples, such as - signals - sampling (see step 11〇). The new sample is used to calculate the coefficient contribution of the new sample using j (refer to step 12). The coefficient contributions are calculated according to an equation representing the function, and the coefficient contributions used to convert the coefficient M to the Fourier transform are calculated using the sampling, the appropriate triangular Λ function, the number of samples, and the number of coefficients. The child action also includes the step of obtaining the coefficient of the oldest sample among the samples of the set (refer to step! 30). The coefficient of the oldest samples can contribute, with a variety of different techniques to get. | Embodiment, the coefficient of the oldest samples previously stored in the contribution of :: a memory accessible. Alternatively, the oldest sample can be stored in an accessible memory, and the coefficient contributions can be calculated using the oldest sample. Use the oldest sample of the contribution factor and the sampling coefficient latest contribution to this special factor will be updated (see Step 14〇). Updated coefficients can benefit

Page 30 580 623 V. invention is described in (26) by subtracting the contribution factor and the oldest sample of the latest sampling plus coefficient to calculate contributions. If the generated coefficient such a coefficient based Fourier transform coefficients is obtained in this step is identical to the system moving aperture Fourier transform coefficients (the SAFT) is provided. In particular, these coefficients are imprecise and do not equal the classical Fourier transform coefficients. In order to obtain accurate coefficients, the method, device and computer program product according to the embodiments of the present invention correct these updated coefficients so as to provide coefficients that can more accurately represent the signal (refer to step 150). The method, device and computer program product of the present invention apply the correction reference system to a set of samples including the new sample, so that the angle used to derive the coefficient contribution is the correct angle for each sample. For Fourier transform coefficients, the correction coefficient corresponding to the line Li Fu tea classical conversion coefficient. The bearing, the rotating reference system to correct the system provides each sampling angle. The rotating reference system may be applied to embodiments of the present invention, by utilizing the method of the present invention, apparatus and computer program product in Equation (Shu 4) to generate said Fourier coefficients. Application of the rotating reference system for other types of function coefficients of the equation has particular application may require that particular function. As an option 'like the formula may be derived based analysis using a formula based learning model may be deduced' '...

After generating these correction coefficients, such steps may be selectively output a coefficient (see step 160). Coefficient based collection of samples in this step, including including the new sample may be applied, such as: not significant, further signal processing and analysis and other fields. Another step after generating the coefficients is to include the newly obtained

580 623

Fifth, the invention is described in (27) a set of coefficient samples, including samples, replace the previous set of coefficients people step 170). If this step is to be continued, the operation step "collect" a new sampling step (that is, return to step 丨 丨 0). (Refer to step by step. As described earlier, the method, device and computer of the present invention: use various techniques to generate the coefficients of the initial set. ": Also, the method, device and computer program product of the present invention also has an elastic use different kinds ^ coefficient contribution of these techniques to produce a sampled. Xing Fourier transform in terms of η, these coefficients may be utilized fast Fourier Ping i = / FFT) to give. "who can utilize these coefficients American

1 Pelton method as described in application 09/560221 and PCT application 00/67146. In addition, the coefficient update can also benefit as Patton (Pe 11 on) mentioned technology to get. The third and fourth diagrams illustrate an embodiment of the architecture of the Pelton method, which is suitable for determining coefficient contributions and / or coefficients for use in the methods, devices, and computer program products of the present invention. These drawings are similar to those in the U.S. patent application 009/5 60 22 1 and the PCT application WO 00 /

Commentators 1 and 4 6 Zhou. Referring now to FIG third, which are displayed according to a number of input samples s left foot, so as to determine the beginning of the coefficients represents an architecture of the set start one of the functions of the input signal. The apparatus of this embodiment of the present invention includes a unitary number generator 10. The violator generator has a receiver 12 by which it receives a sample of the input signal. The coefficient generator also has a first gate 1 4 'which is digitally communicated to the receiver, and a second gate 6 which is digitally communicated to the first gate. It should be understood that the so-called digital communication systems to allow digital transmission refers to one information link. Various technologies can be used to achieve

Page 32 Lake 623

= Number of communications purposes. An example of digital communication is the use of electronic signals such as electronic signals. Another example of digital communication is the use of optical signals, such as optical signals of optical communication.考, 侗 2 2 Let us consider the fourth figure, which introduces the action of the coefficient generation and generation. In this; 1, 5 Hai coefficient generator system in accordance with the signal Ν = 8 samples, by

Corresponding coefficient for 2 students. For each sample and t, the receiver is provided for receiving a sample of Hu te k, and inputs the sample to the first shutter (refer to step 20 square γ). For each coefficient, the second gate system receives two values, one of which 18 represents the number of the coefficient, and the other 20 represents the number of the sample. According to the coefficient number and the sampling number, the second gate system generates an orthogonal function part of the signal (refer to step 21) for the zeroth coefficient A 0, the sampling value is added to the coefficient For the first sample and coefficient AL, the orthogonal function system is: cos (2 氺 7Γ 氺 Cn ^ Sn / N) where Cn is the number of coefficients; Sn is the number of samples; and N is the number of samples. In order to calculate the first coefficient and the first sampling item, the second gate system receives the coefficient number 18 and the sampling number 20. According to this, for the first sample Sn and the first coefficient Cn, the first two gate lines produce cos (2 * 7τ * 1 Shui 1/8) or cos (2 * 7Γ / 8) and outputs this value to the first shutter (see step 210).

Then, the first gate receives the value from the second gate and the sampled value from the receiver. Based on these values, the first gate generates an item to represent the sample's compensation for the coefficient (ie: S

Page 33 580623 V. Description of the invention (29) " cos (2 7Γ / 8)) (refer to step 2 20). This item is then added to the coefficient A1 (i.e., A1 + AL) (refer to step 230). This U action is repeatedly performed for each coefficient (refer to step 240 and step 25 ^). The above discussion is about updating ~ coefficients with each sample. Furthermore, it should be understood that the coefficient generator is capable of updating all coefficients simultaneously with each. For example, the coefficient generator can scoop two first and second plurality of gates respectively connected to the receiver. In this example, the sampling is supplied to each group of gates at the same time, and every two: Then each item of coefficients is generated at the same time, which represents the compensation of each sampling number and the coefficients are updated simultaneously. . In particular, the coefficient production system as embodied in the coefficient update all. Furthermore, in some examples [the coefficient generator is structured to receive turns from several channels, the = generator generates a set of coefficients for each channel. However, in the above-mentioned embodiment, the coefficient generator described above sequentially updates each coefficient and has only one channel (for clarity). ',-The coefficient generator described in the figure and its action are suitable for use in the embodiments of the present invention to generate the start or coefficient contribution. If the coefficients of the mouth and / third system described in FIG. Contribution of these coefficients and a coefficient based on the shutter, ·· such as multipliers, adders, using the method to obtain function ^. The fifth figure, which also discusses: = The determination method of 2 and the contribution of the coefficient, which is to use the coefficient of ^ to record the use of S 己 memory body ^ (compared to the use of the gate) can have many advantages,

580,623 V. invention is described in (30) --- as: Many of these values must be calculated to determine the coefficients can be pre-storage in the memory device. Therefore, this method can save time decide ^. Please refer to FIG fifth 'coefficient generator according to the present embodiment has a system a receiver 12, whereby a signal of the reception sampling. The coefficient generator also having a first memory device 22, to which the digital communication receiver, and two second memorized body apparatus' which digital communication to the first memory device. Alternatively, the second memory device may have an array of cell element, wherein ^ each cell having a pre-computed value of 'orthogonal function used to represent the signal portions and for each sampling number system. For example, one cell of the orthogonal function part of the first sample and coefficient of the signal in the second memory device is equal to cos (2 * 7T * Cn * Sn / N) or cos (2 * 7Γ / 8 ). In this architecture, the first memory device may be a multiplier. Please refer to the fourth and fifth figures. In operation, for each sample, the receiver receives a sample of the signal and inputs the sample to the first memory device 22 (refer to step 20 0). . For each coefficient, the second memory device receives a system flag (token), which indicates that contains the address signal bit orthogonal function in the extracellular portion of the element and the sampling coefficient. This flag can be provided by inputs 18 and 20, where one part of the flag is the coefficient number Cn and the other part is the sampling number sn. According to this flag, the second memory device captures a value related to the coefficient and the sample, and outputs this value to the first memory device (refer to step 2 10). Subsequently, the first memory device receives the value from the second memory device and the sampled value from the receiver. According to this

Page 35 580,623 V. invention is described in (31) - These values, It means the first memory system to generate a program for showing the compensation coefficients for the sample (i.e.: Si * c〇s (2 ^ / 8), for the first - and the sampling coefficient terms) (refer to step {22)). This item is then added to the coefficient (i.e. A1 + A1!) (See step 230). This action is repeatedly performed for each coefficient (refer to step 24 () and step 25). In some embodiments of the invention +, the sampling received by the receiver is one of a plurality of finite values. Since the orthogonal functions and a coefficient for each sample portion prior knowledge-based (i.e. :), and the sample may be only a finite number of values of a ~, indicates each sampling value and the sampling value of the number of coefficients numeral average number It may be calculated in advance and stored in advance. In this way, when sampling & _ sampling, the item indicating the compensation for each coefficient of the sampling can use the table lookup method, and find the corresponding value in the memory device to determine based on the sampling value, sampling number and coefficient number. . In view of this, in the next embodiment, the first memory device is a memory device including an array of cells. Each element of the first cell of the memory device having a pre-calculated values representing each sampling value and the sampling number count coefficient number. For each coefficient and sample, the memory device includes a group of cells, each having an orthogonal function related to the coefficient and sample multiplied by one of the possible values of the sample. For example, for the first sample and a coefficient, namely, a group of cells having cell SJcos one value (2 * 7Γ / 8), wherein each value represents in cell lines may correspond to a different one of S1 value. Further, the second line memory device having a cell element arrays, each having a flag indicating orthogonal function portion for each of the samples and coefficients.

580623 V. Description of the invention (32) Operation 'Please refer to the fourth figure. For each sample, the receiver receives a sample of the signal and inputs the sample to the first memory device 22 (see step 200). For each coefficient, the second memory device receives values from inputs 18 and 20, which represent the address of the cell that includes the signal in the sample and the orthogonal function portion of the coefficient. The second memory means based on the correlation coefficient and grab sampling flag, and the flag is output to the first memory device (see step 2 Shu square). The first memory device then receives a tag from the second memory device and the sampled value from the receiver. According to the marker and the value of the sample, the first memory device queries the cells corresponding to the values in the array, and outputs an item to represent the compensation of the sample for the coefficient (ie, SJcos (2 * 7Γ / 8), for the first sample and coefficient) (refer to step 2 2 0). This term is then added to the coefficient (ie: A1 + A1 i) (see step 2 3 0). This action is repeated for each coefficient (refer to step 240 and step 250). Once again, it should be understood that a device of this architecture can be operated in a coupler architecture, by using a plurality of first and second memory devices (all of which are connected to the receiver) to each coefficient, thereby updating at the same time Each coefficient. In this embodiment, each set of first and second memory devices receives the samples simultaneously and is appropriately addressed so that each set of first and second memory devices can remember the address values of different coefficients. In this way, the compensation of the sampling for each coefficient can be determined in parallel and simultaneously. Please refer to the fifth figure. In another architecture for providing the coefficients of the initial set and / or providing the coefficient contribution, the coefficient generator may have a counter

Page 37 580 623 V. Description of the Invention (33) Μ, which digital communication to the second memory means 24. The counter may vary, when a clock is incremented, not shown, which allow the system to perform a calculation timing operation. The counter may include two outputs 18 and 20, which indicate the coefficient number and the sampling number, thereby addressing the second memory device. In operation, for each sample, the number of the sampling number is fixed, but for each cycle or multiple periods of the clock, the counter is incremented by the coefficient number. Accordingly, this step may be the second addressable memory device, so as to determine the compensation factor for each sample. After all the coefficients corresponding to the sampling are calculated, the sampling number of the counter is incremented, and the coefficient number is reset to zero, so that the next sampling can be recalculated for each coefficient. In addressing the above embodiment using the system will need to use the system is the need assessment requires a minimum shelf application WO above numerical values and the standard used, the sixth operation, to detailed above, and the like for the memory device to the memory means for The off-the-estimated design requirement number of components and the number mentioned in Yunwen and the number in 00/67146 can be expressed in bit numbers. Introduction to the introduction and introduction of a coefficient to reduce the calculation time. One implementation stores pre-calculated values and labels, which are used. An important consideration in many electronic designs is to minimize the number of components and minimize the need for shelf components. In view of this, the need aspect determines the design scheme that allows the use of standard components. As discussed in US Patent Application No. 09/560221 and the PCT, it must be stored to calculate the function coefficients. Remove the extra values, store only the size of the value, the value zero or the sign of the value, and reduce it. For example, each coefficient of each received sample is updated. The generator ’uses the increase of the gate to reduce

580623 V. Description of the invention (34) "--The number of values that must be stored. In particular, the sixth diagram provides a framework for utilizing the -first-memory device X (which has the least number of stored values), a second memory device (which has a mark representing a sign or a bit of zero) , And a signed input value to determine the coefficient contribution and / or coefficient of a function. The coefficient generator 10 of the sixth figure has first and second memory devices 23 and 24. Both memory devices have arrayed cells to store values. The second memory means having a flag to indicate that the sample number and the number of coefficients, and the first memory device having all possible unique values and coefficients for each sample, the sample bind orthogonal function portion. While any system or device may be used in addressing the second memory means, for example, describes a line counter 26 in addressing the second memory device of the present embodiment. Coefficient generator according to the present embodiment further has an adder 44, which outputs digital communications in the first memory means. A line connected to the adder latch (crossbar) 44 or the selector 46, and a third memory device 48. The coefficient generator also has a nuU or pull down device 50, which digitally communicates with the latch, and a NAND gate 52 and an AND gate 54 to connect To the output of this counter. In addition, the coefficient generator of this embodiment also has a first AND gate 28, a second and gate 30, and a XOR gate 32. The output connected to the second AND gate 30 is an adder 36, which is usually implemented to obtain a signal's binary complement (2, s complement). The XOR gate (x〇r gate) 32 in communications with the coefficient bits and a gate output of the second memory means

Page 39 580623 V. Description of the invention (35) ----(gate combinatl0n) 40. The coefficient generator of this embodiment also has an invalidation device or pull-down device 34, which digitally communicates with the output of the first memory device. The coefficient generator also has a code converter 38, which is digitally communicated with the receiver 12 to convert the input signal (if necessary) to a 12-bit value, where n bits are Used to indicate its value, and 1 bit is used to indicate its sign. The coefficient generator of this embodiment also uses a 6-bit tag, which is stored in the second memory device, thereby locating the value stored in the first memory device. The tag line) 36 and the combined operation of the adder 30, so as to obtain the output of the first memory means is negative and the second AND gate (ANI) gate, if the marker bit symbols imply that a negative value is value. Worse '4 is invalid or pull-down devices 34 and the operation of the zero bit line labeled with an invalid or zero output of the first memory means, this implies that if the tag value of the zero wide. Furthermore, the coefficient generator also has a first latch 6 0 for flashing or inputting "quote" and a second flash 6 2 for flashing out the value of the output coefficient. The coefficient generator also has an input 44 and a gate combination 40 to use the coefficient of the inverse function of the signal of the coefficient generator. In addition, the coefficient generator also has a reset device (reset dev ce) 64, which is used to reset the * self situation and outputs 56 and 68 to zero, so as to output the coefficient and channel number = please refer to the seventh FIG, operation in this embodiment, similar to the previous embodiment, for each sample, the system receives the receivers as the number ^

580623 V. Description of the invention (36) A sample is input to the first memory device 22 (refer to step 300). For each coefficient, inputs 18 and 20 are provided to the second memory device to indicate the number of samples and the number of coefficients (see step 3 0), and according to these inputs, the second memory The body device outputs a mark (refer to step 320). According to the mark and the sample, the first memory device generates an item to represent the compensation of the sample for the coefficient (refer to step 330). If the sign of the signal is negative or the sign bit of the mark is set ', the second AND gate outputs a carry bit (carry bit) to the adder 36. The adder will turn the output of the first memory device into a negative number (refer to step 340 and step 3 50). Similarly, if A " icr 5 or 4 shows that there is no zero value, a zero value will be output by the first AND gate 28. This zero value is output to the first memory device, which further disables (d i sab 丨 e) the first memory device. Since the first memory means catching rabbits disability, the system 34 or the pull-down means for outputting an invalid value to indicate that the value ten of the coefficients (see step 36 billion and 37 billion

° For examples with zero values, the symbols are omitted. Subsequently, the project department of the coefficients to be output. A This item is then provided to the adder 44, which is increased from the third memory device 48 receives the value of the previous coefficient. The third memory device is also connected to the second memory device. The markers from the second memory device address the coefficients stored in the third memory device, which are then output to the $ method to be added to the item. This item uses the adder to add the coefficient to ^ (refer to step 380). This step continues until the receiving agency

Page 41 580623 V. Description of the invention (37) After sampling and calculating all the coefficients based on these samples, the coefficients are provided according to the sampling of a first set. As in the embodiment described in the above embodiment, means, methods, processes and computer-based product formula was sampled and processed in accordance with a plurality of such samples to generate the coefficient function. In the embodiment, a plurality of samples to be received in a portion of the embodiment of the present invention, the coefficient of the apparatus of the present invention, a method and computer program product is generated in the output, such re-zero coefficients, and obtains the re-sampled signal. However, in some embodiments, it may be advantageous to generate and output a complete set of coefficients as each sample is received and processed. This is the moving aperture Fourier transform (SAFT). 'In this embodiment, the device, method and computer program product of the present invention do not reset each coefficient to zero after receiving the last sample of a plurality of samples and outputting the coefficients. In contrast, the apparatus of the present invention, a method and computer program product will replace the previous plurality of samples received first sampled sampled followed. With this new sampling, the apparatus, method and computer program product of the present invention will output the coefficients for the next set. If so, the device, method, and computer program product of the present invention are not specific to each batch, batch, and sample generation-set coefficient, but instead generate-set coefficients each time a new sample is received. In order to receive a new sample each time a new set of coefficients. The present invention provides several embodiments production apparatus each time it receives a new sample, the apparatus of the present invention, a plurality of samples before the first sample, the method and computer program product, a set of coefficients so as green. In each of these methods and computer program products, the sample is compensated before the next sample is received.

Page 42 580623 V. Explanation of the invention (38) Compensation, and then the medium, the invention, the device, mathematical function invented to the item when a new sample is generated when each sample is stored. For the benefit of the sample, it is the first one of the samples that are stored in a new system, which is previously stored in this subtraction, and is collected after the new method and the new sample are used. The number of samples is output. The coefficients are received by the method and the electricity. The pre-computer program is received and the sampling device is replaced according to the sampling.) The items of these coefficients are sampled, and the number is subtracted and added to the new application. In the example, the remainder is first compared with the latest routine experience, and the sixth map is communicated with the logarithm of the sample produced by the waiting plural program. As the first implementation project of the project, the first item introduced in the memory of the present invention is that the items in the sample should be in the example of each other, and then add two real devices of the sixth figure. The first one of each department is a subtraction of one reality. These are related to the above-mentioned examples. In the example, the number of examples is a plurality of brain programs. In the case of a predetermined duplicate product, the new ones are taken. The number of multiple samples minus the utilization value is used to calculate the number of items that are newly added to the sample of the minus samples in the multiple sampling stations of the present invention. The new sample is updated every several samples. Use the device to take a number of corresponding samples and equal coefficients to connect the most eyes to the calculation. A real sample is generated after the last input letter sample is connected to the first term of the coefficient samples and the example is received. while samples No. ~), this coefficient is a coefficient generating an admission order (which projects et generator further comprises a plurality of a fourth of the third memory device 48, thereby replacing the new sample, method, and computer-like The first takes the item into the oldest sampled term of each department; the coefficients in the other old sampled term are equal. Its (drift) and

IM Page 43 580623 V. Description of the Invention (39) Store the item corresponding to each of the plurality of samples. The coefficient generator further has an adder 68, which is digitally communicated to the third and fourth memory devices' for subtracting from the items corresponding to the latest received samples the first sample corresponding to the previous plurality of samples. project. The fourth memory device is also connected to the second memory device. In this embodiment, after the initial coefficients are determined by the sampling of the original set '§ When each new sample is received, the coefficients are updated using the information of the new sample, and the compensation of the oldest sample is subtracted, whereby A moving aperture Fourier transform (SAFT) is provided. In particular, please refer to FIG. 6 and FIG. 7. The items of the tag memory from the second memory device stored in the fourth memory device represent the first sample of the plurality of samples (also That is, the oldest sample): the corresponding item. This item is provided to the adder 6 8 in which ′ wide items are subtracted from the updated coefficients (refer to step 3 90). The items related to the latest sampling are stored in the fourth memory device (t:; 00), and the updated coefficients are stored in the third memory device (1:!: Γ1〇) and output (refer to step 420). The above steps are repeated for each coefficient. Ji Zhao Ru Shi Yu ▲ ^ —A f 夂 赵 牛. Until all the coefficients have been updated according to the sampling (refer to step 430 and step 44). Apparatus of the present invention, in ancient rain - Series to Woo channel ,, ί computer program product may be used in parallel to the channel shown in a "two Shifa: Sheet is capable of producing in the invention, respectively receiving each channel sampling gj = . the coefficients are as hereinbefore 'this mathematical function of the coefficients # &amp.; each sample and the sample and the number of connected worse, expressed as a function of these lines and the floor velocity signals' samples such dETAILED DESCRIPTION And various coefficients

44580623 five page, description of the invention (40) Π ,,,. The sum can be calculated in advance and stored in an addressable memory device, or it can be calculated by a gate function with reference to the value associated with the tag. Generally speaking, the label of a cell addressing the memory device is derived from a state inf or matrix, which can be obtained by a counter representing the coefficient and the sampling. When in normal operation, these standards. Have been. Hexyl, memory and means for determining the coefficient line represents a signal of one function, those labeled 2 and memory means can also by increasing the bit counter adapted to: re-use position 'and is thus used as a plurality of channels . In this architecture, extra coefficient memory cells are used to maintain extra coefficients, and a channel number can be output for user identification. In addition, it is also possible to use an electronic signal to determine the forward coefficient (f0rward coefficient) or to determine the inverse function of the signal ^ ⑽ "^ 化" ^ ⑽). When in normal operation, it indicates a signal or a function. Furthermore, the device used to determine the table can also be used, so as to s memorize the cell to output for the channel, so). These transactions generate two-dimensional transfers in the present structure to use equal coefficients to increase the utilization to complex numbers. They are used in the conversion of possible production replacements. A clear matching pair is used to correct the participation bits to the channel. Maintain additional identification . The alternate life cycle can be exemplified later. Exceptions embodiment electronic test system element (when such a marking device and a coefficient of the coefficient memory, and memory of these markers in place of the counter to cause repeated. In this architecture, an additional factor coefficient, and a The channel number can also be treated as a different frequency conversion by continuous sampling (interleaved transform is transmitted to a second similar program, so that the device system described in the sixth diagram (not shown in the sixth diagram) converting the rotation of the reference system in particular), by

V. Description of the invention (41) Coefficients updated by positive cross. These additional electronic components, such as: components that use gates (eg, multipliers, additional gate functions). Or, the device described in Figure t of the present invention can use memory and store; rotate the reference system. For Fourier transform calculations: The memory will be based on factors such as equation (丨 4) and coefficients. ^ As mentioned earlier, the coefficient generator shoulder Fourier transform (SAFT) architecture of the sixth graph allows continuous updates to be used when sampling such gifts, which uses the effect of subtracting and adding the latest sampling. As mentioned above, this can be achieved in two ways. First, the oldest sample is calculated, subtracted from these coefficients, and the latest sampled number is added. Similarly, the effect is the oldest sample can be calculated by subtracting the effect, then added to the remainder of these coefficients. Such update coefficients to provide a moving aperture Fourier transform (the SAFT) receiving each new sampling technique (such as the latest sampling sampled), but these methods and solution samples helmet method starts at 2 * is 45 degrees (45 degrees instead of ) 'under the conditions asked. In view of this, in one embodiment, the generator of the present invention, with reference to the sampling system utilizing rotation such as to thereby provide a classical digital Fourier transform (knock d) In the sixth embodiment described in FIG fourth memory Example thereof may include the following attachments element adder, the divider, or the sixth function to provide a deposit ^ 'or additional elements (17) providing a derivation of the corrected moving aperture k; the number of each of the receiving New: The effect of removing the oldest sample. This effect can be re-calculated and added to the department. Although these methods are the latest sampling method, it allows the following method to sample only the oldest new set. One question, in the case of N = 8, a coefficient is provided to correct these coefficients. 3 In particular, this implementation is 6 6 and the adder 5 0 580623. 5. Description of the invention (42) Only the effect of subtracting the oldest sample, And the embodiment of the present invention further reduces the time by using the rotation of the reference system so as to have a correct angle value for use in the drawing, which is not according to the present invention. The use of the rotation reference system to correct the displayed device may FIG notably 'eighth FIG lines showed additional positive-based system to generate a sixth device of FIG completed Equation (14) is calculated, the rotational angle difference. In particular, the apparatus according to the present embodiment so as to connect to the latch coefficients 62 in number of the memory device having an array, " number and sampling related sinusoidal (s Shu ne value For example, if the coefficient of the coefficient is. The fifth memory device sampling combination corresponds to the calculated value of the compensation equation ((cosine) part. The farming device of the present invention is further connected to the fifth memory device as shown in the sixth figure. The first chord portion of the latch 62 is a coefficient equation (14) is coupled with the effect of the latest sampled sinusoidal portion However, to the oldest sample and adding the latest sample of such compensation coefficients, whereby the calculation to take such in particular, please refer to the architecture block diagram of an eighth embodiment of the apparatus, by means of Fourier transform coefficients. in an eighth apparatus described in FIG used with. Laid logic to attach a number of correction coefficients importantly Yes, the device in the eighth figure is used to contribute to these coefficients. The samples have a fifth memory device 70, a number 72 output and a phase number 74 output. It includes pre-calculated, each department) and cosine ( cosine) The compensated number generator uses eight samples to determine that the eight arrays will include the sine and cosine of each coefficient number and 14). There are two multipliers 76 and 78, two of which are set and the latch 6 2 8 billion output coefficient multiplier 76 multiplies the Department of equation (14) is a remainder value, and the second multiplier and the coefficient values output from latch 62 with

580623 V. Description of the invention (43) Multiplication. Also connected to these multipliers 76 and 78 is an adder 82 having an output 84 having a correction coefficient. Referring to Fig. 9, the operation of the device for correcting the Fourier transform coefficient using the rotating reference system is described. In particular, as described in the sixth drawing, the first coefficient generator according to a first set of samples to generate a set of coefficients ί (see step 500). After receiving the sampling coefficient τττ3 of the first set and receiving the next sample, the fourth memory Pei 66 and the adder 68 of the coefficient generator will subtract the compensation of the oldest sample and add the latest sample. Compensation (see step 510). As previously described, the fourth memory Wherever body apparatus 66 and the adder 68 does not compensate for the angle of such sampling and sampling by discarding the oldest added latest sampling caused by the rotation. In view of this, the device of the invention described in the eighth figure is used to provide this compensation. In particular, the latch for the coefficient generator 62, the receiving system of the fifth coefficient destroyed numeral 72 and digital 74 phase, respectively, and the coefficients of the table ^ numeral sample (see step 52〇). Based on these values, the fifth memory device stores the values of the cosine and sine portions of the array grabbing equation (14) related to the coefficients and sample numbers, and sends these values To these multipliers 76 and 78 (refer to step mo). In particular, when the number of the coefficients of the sample No. 7 and 5, the device will provide two-body equation (14) should wait for the cosine and sine values of the portion. The material multiplier receives the cosine and sine values of the fifth memory device, respectively, and also receives the coefficient value 80 by the latch 62. The numerical value is pre-compensated by using the fourth memory device 66 and the adder 68 in the form of a subtraction compensation and a compensation by adding the latest sampling. Take advantage of this

580623

These values, such an adder 82 adding the reference system provides a system according to the invention will be to select every sampling, when a tracking system which is previously generated as coefficients and / described above, the present invention, the movable aperture raw Fourier techniques and hardware Can be counted from these schools JL 傅立 # Another feature affects the number of each other coefficient is not necessary, so it can be converted using a simpler embodiment (SAFT can also be used (SAFT)) other coefficient conversion coefficient whereby the coefficient of multipliers based. compensation calculation unselected sample. to calculate the presence of a single hardware using various applications further improved) shift values is provided to update the formula (14), by (see step independence., characterized in that there is a demand for other systems, its structure. coefficients are of a different technology. Fu movable aperture value of the coefficients, and the step 540 to take advantage of this rotation). In particular, since the coefficient of interest can allow the user to countable. Allows faster, as described above to produce on-the-fly improved coefficients. Liye transformation and updating and correction.

In response to this demand, other embodiments of the present invention use & automatic decision to generate a representation-letter function. Some real-life techniques use learning techniques to determine these models. In general, transformations (such as: Fourier transformations) are variables, and mappings. The main definition of a mapping is usually parsed. Some models used for mapping calculations can be predetermined; others can be more efficient, based entirely on the computing technology used. The definition of the Fourier transform relates to variables in the time domain (frequency domain) and variables in the frequency domain (frequency domain). In the case of Digital Fourier Transform (DFT), the mapping is between continuous time sampling and specific frequency coefficients. The coefficient can be changed on each sample, and

V. Description of the invention (45) _ & includes the latest access + method system including mapping = :: group 2 = sampling. Calculate these coefficients between calculations. In particular, the values of any parameters and coefficients of ^ two spoons are actually a mapping relationship between the coefficients of the new set and the coefficients of the new set. a There are several such models. An embodiment of the present invention is a method having a mapping relationship between the Fourier coefficients in which a model is automatically determined using sampling and Fu. Description; ^ During training, j is the latest sample of the learning model applied to any mapping relationship and the new set =: = = ”coefficient plus the same as the standard technology, the embodiment of the present invention. Not such as: Fourier transform, pull Plath's switch to the embodiment of the present invention will result in: earlier, faster or more parallel hardware implementations on the computer ^ ^ Examples may also have (but not limited to) similar :: partial implementation (SIFT) And moving aperture Fourier transform (SAFT) 丄 = as described in US patent application 09/560221 and substituting applications w0, 1, minus 671 46. '/ These solutions: the case model may be exact or approximate Although these models are based on different methods, these methods are similar to the methods of applying error assessment and selecting the required parameters of 0 °. These implementations are provided: J 3

Page 50 580,623 V. invention is described in (46) ---- exhaustive, but are merely used to show, for example, a range of application of the method. For those persons skilled in this art, various embodiments of the present invention may also be in a similar calculation and decision making by stacking portion, thereby deduced different points and has the same functionality of the method can be achieved. A variety of learning models and approximate model techniques can be applied. The following sections discuss examples of the two model technologies. The first technology is related to the use of special structures as general function proximity devices. Here, neural networks and fuzzy logic circuits are used as examples. A second technique based on the use of a general approximation (pattern recognition) method, which can be used in various structures, and each portion of the search model may be a synthesized, gene-based algorithm here (genetic algorithm) ° Example For example, for a neural approximator, the element block with a special structure is a neuron; and for a fuzzy system approximator, the element block with a special structure is a fuzzy rule (fuzzyru 1 e). For both kinds of approximators, the universal function approximation has been formally confirmed. Fuzzy logic processors are known in the art and are commercially available. For example, fuzzy logic processors can be implemented in the form of computer executable code on a computer or other electronic device. Similarly, neural networks can be obtained commercially. Neural network computer executable code can form, as a software implementation of a camel. In particular, neural network chipsets are commercially available for use as neural network processors. Said embodiment of the present invention, a ^ and a sample by using a predetermined number of training neural networks established based wafer, whereby the input sample is mapped to the coefficients. As a further embodiment, the neural network chip is connected to a receiver

580623 V. Description of the invention (48) Practice / learning / searching procedures. In this example, a set of input-output pairs (pa i r) obtained by the system to be used by the model is used for training. In the example of determining the coefficients of the Fourier sequence, the training pair includes a pair of vectors, for example, sampling and coefficients. For a moving aperture Fourier transform (SAFT), the pairs (p a i r) can be samples and coefficient updates. The selection of the correction training set has a great impact on the feasibility of the model. The following example uses 8 samples coded in 5 bits. Consider, for example, that the model to be determined by the mapping relationship is spoken as: " At each new sample time, the value of the sample together with the value of the current set of coefficients (input + current state) determines the new set of coefficients · coefficient (t + 1) = function (sample (t + 1), coefficient (t)). "first, generating a random vector, which is the input samples in the time domain. the size of the vector can vary according to the model It should be understood that, generally, there will be a training pair corresponding to each sampling index (index); this training pair can be considered as the second equation of linking input and output. The number of equations that may be needed is based on a plurality of factors; for example, the number of equations may be related to the number of variables used in the model can be used when a linear model ^, then a linear conversion system calculation, the number of equations will == number of unknowns in order. get the exact answer fewer equations

580623 Fifth, the description of the invention (49) (that is, the training pair) will not be enough to produce the general formula. Typically, if the information is contingent (probabilistic), it will need more equality. The next step of the system using a variety of convenient methods to produce samples for the Fourier coefficient of each group (one of eight in consideration of a set of samples, the number of samples may be a group number and the like 1-8,2 -9,3_1〇 ). Examples of finite Fourier transform of the Fourier find convenient method (FFT), butterfly side 'method (butterfly), and the product set sampling Fourier Transform (SIFT), as described in U.S. Patent: Please Case 09 /% and the application 56〇221 in w〇〇〇 / 67146 described. This means that, for each sampling index (丨 to), there will be a set = coefficients to highlight the samples in the window. The coefficients ending at this index and f f are referred to as the desired target output. Used for training. In 2 embodiments, this problem may be decomposed in the model, which calculates only-㈣ number, 纟 In this example, the vector containing the coefficient value is the target. In this example, the 'training output consists of the combined set of oldest samples and the set of coefficients corresponding to the samples in the previous window (ie: the samples before index sampling; the set of coefficients). More precisely, for an 8-sample window, words, the input - output may be a training index corresponding to each pair (pair) · input comprises sampling time (t + 1) and the value of a coefficient value of seven times eight, and the output time comprising (t +1) of one or more coefficients. ', table 1 describes the system used to generate code data of the model of learning and training 0

Page 54 580,623 V. invention is described in (50) Table 1. Learning and training information generated model to map the Fourier transform coefficients M a 11 ab code% we prepare a larger collection of samples, not all may be necessary for training nsamp = 200;

Samples = (round (rand (l, nsamp + 8) * 31)) / 31; ^ initialize coefficient matrix on the columns will be the 8 coef% to preserve the same index it is made of size nsamp + 8, but the first 8 will not h 'occupied or used in training, the proper training set being from 8 to nsamp + 8 Coef = zeros (8, nsamp + 8); for i = 8: nsamp + 8

SampleWindow = Samples (i-7: i);% here we use SIFT to create the outputs, buy any method giving the Fourier

Coefficient can be used [otto, OutT] = sift (SampleWindow);

Coef (:, i) = OutT; end% The input-output training pairs are thus defined:% initialize matrices first Input = zeros (8, nsamp);

Output = zeros (8, nsamp); for j = l: nsamp

Input (: j) = [Samples (j + 8); Coef (:, j + 7)];

Output (:, j) = Coef (:, j + 8); end% divide in a training set and a test set (half each) sizesample = size (Input, 2); sizetrain = 100;% this model will calculate coefficients separately, use 9 inputs and 1 output, input is (sample and 8 current coefficients), output is one coefficient% prepare a training set and a test set, not used in training but used to verify the model

j = l; p = Input (:, l: sizetrain); t = Output (j, 1: sizetrain); ptest = Input (:, sizetrain + l: sizesample); ttest = Output (j, sizetrain + l: sizesample ); One embodiment of the present invention relates to a neural model. The neural model system uses neurons as the main building blocks. The neurons are mapped to a collection

Page 55 580623 V. Description of the invention (51) Input to an output calculation block. The aggregated full compensation (a, two, t? Is usually all input gggated total contribution) of the = function 4 is usually non-linear. Each compensation system is multiplied by a weight _ μ ί ^ ^, the 70 series is connected by some technology interactively, so that some of the nerves are used as input for other neurons. Can these neurons be woven? Laminate (Layer) group and limited by alternately stacking the web Onter ayer network) (interconnection system in ^ the output layer neurons is coupled to the library Γ:. Λ, each connecting line between neurons. -A = ㈣, and each neuron usually has a corresponding offset term line technique '⑸ ：::: The learning of the Xuan neural model is usually related to the interconnection technique, weight value, offset The determination of the value. The topology is usually administered by the trainer. Please refer to the tenth figure, where JJ shows an embodiment of the present invention. The tenth item introduces a model of a neural network, alpha mapping the coefficient sum of sets and the latest sample to the next set of coefficients abbreviations as: the latest sampling; Cci is the coefficient i (i = 1: 8) present values; and NCh are coefficients Shang new value in training a neural. In the example of the network learning the mapping relationship between (pa 1 r), firstly, the output of the infusion output based on the values of t + i and eight values at time t is eight at time t + ι New coefficient. A simple gradient descent path /, ^ gradient descent rout INE) is typically adjusted to a suitable working system weights the difference between the amount of membership to the neural model provides a helmet and the head ## measured I ^^ p 56,580,623

To minimize. A suitable example of the difference between the measurement-based mean squared difference (mean square error) of such measurement. An example of the Mat 1 ab code for this problem is as follows% coef (t + l) = function (sample (t + l) + coef (t)), range = minmax (p);% create a network with 3 neurons in first layer and one in out layer,% with pure linear characteristics, training by Levenberg-Marquardt% backpropagation net3 = newff (range, [3 l], {'purelin' 'purelin'L'trainlm'); net3 = train (net3, p, t); After a few round trips, a neural model can be obtained using this program. This particular problem may utilize a laminate practical solution, as follows 7] ^ · net3 = newf f (range, [1], { 'purelin'}, 'training'); net3 = train (net3, p, t) as A special example, this mapping relationship can be a line 彳 生 # change. In this example, the set of equations is A * X = B, where A is the input value, X is the transformation matrix, and B is the output set. In the special case with nine training pairs, nine equations, and nine unknowns, A is 9 by 9, X is 9 by 8, and B is 9 by 8. The first index is for nine training pairs. This answer can be obtained by X = A \ B. Table 2 shows the Mat lab code used for this example and results.

Page 57 580623 V. Description of the invention (53) Table 2. Examples of neural models M a 11 ab code and results ^ determine a linear mapping between% (new sample +8 Fourier coef) to (8 new Fourier coef)% this uses a linear model% A is the set of inputs from a training set% these are 9 raw vectors / examples, each with 9 terms% the first is the new sample and the remaining 8 are the ^ coefficients of the 8 samples window before this new sample% X is the unknown model (a 9x8 matrix)% B is the output sizesample = size (Input, 2); sizetrain = 9;% initialize a matrix (works faster) X = zeros (9,8);% calculate each column of the X matrix by solving a sys of linear eq for j = l: 8 p = Input (:, l isizetrain);. t = Output (j, l: sizetrain); A = pf; B = tf; X (: j) = mldivide (A, B);

;% this is the same command as% X (:, j) = A \ B end

% Display it X% test on other pairs t = Output (:, l isizetrain); ptest = Input (:, sizetrain + l: sizesample); ttest = Output (:, sizetrain + l: sizesample);% calculate for ptest inputs strange = (p test '* X)';% and see is equal to desired values sis = ttest; llllgl page 58 580623 V. Description of the invention (54) 0 /, the result is • • / 〇J% \ ~% Col 1 Col 2 Col 3 Col 4 Col 5 Col 6 Col 7 Col 8% .1.0000 1.0000 1.0000 0.5000 0.0000 -0.0000 -0.0000 0.1250% 0.5303 -0.1768 -0.1768 -0.0884 -0.7071 0 -0.0000 -0.0221% -0.0000 0.0000 0.0000 0.0000 0.0000 -1.0000 0.0000 -0.0000% 0.1768 0.1768 -0.5303 0.0884 0.0000 -0.0000 -0.7071 0.0221% 0.2500 0.2500 0.2500 -0.8750 -0.0000 0.0000 0.0000 0.0312% 0.5303 -0.1768 -0.1768 -0.0884 0.7071 0.0000 -0.0000 -0.0221% -0.2500 0.7500 -0.2500 -0.1250 -0.0000 0.0000 -0.0000 -0.0313% -0.1768 -0.1768 0.5303 -0.0884 0.0000 0.0000 -0.7071 -0.0221% 〇 乂 -1.0000 -1.0000 -1.0000 -0.5000 -0.0000 0.0000 -0.0000 0.8750 pause% more verifications % a first set of 8 samples SampleValid = [l 0 0.2 0.7 0.3 0.8 0.1 0.5]% coefsvcolumn are their F coef [coefsv, coefsvcolumn] = sift (SampleValid)% the new 8 samples (slided, new one is 0.4)

SampleValid2 = [0 0.2 0.7 0.3 0.8 0.1 0.5 0.4]% coef for samplevalid2 calculated by SIFT (rococo) [roc, rococo] = sift (SampleValid2)% coef 1 for samplevalid2 calculated by matrix multiplication with X% new vector input (sample + coef) vec2test = [0.4; coefsvcolumn]

% NEXT LINE IS THE ONLY COMPUTATION

ansbyinv = (vec2test,) * X rococo1% more tests

SampleValid = [0 0 0 1 0 0 0 0] [coefsv, coefsvcolumn] = sift (SampleValid)

SampleValid2 = [0 0 1 0 0 0 0 0]% coef for samplevalid2 calculated by sift [roc, rococo] = sift (SampleValid2)% coef 1 for samplevalid2 calculated by nn vec2test = [0; coefsvcolumn] ansbyinv = (vec2test,) another p * X rococo * 59580623 V. invention is described in (55) system according to the present embodiment of the invention having a fuzzy logic model. A fuzzy model is characterized in that mapping input lines to the output of a set of criteria (ru i e). Hereinafter, this example directly to a Fourier-based mapping has eight sampled input system which includes a sample value at time (t +. 1) and eight coefficient values at time t, and the output line comprises at time (t + i) coefficient. The following labels are used to simplify writing, where LS is the latest sample; '' CCi is the current value of the coefficient 1 (i = l: 8); Nci is the new value of the coefficient i; NB is the negative negative (negative big); NM Negative medium (negative medium); NS is small negative (negat 丄 ve smaji); ZR is about zero; PS is small positive (pos iti ve small), PM is positive medium (positive medium); PB is large positive (positive big) ° for the embodiment of the present invention, a first coefficient {^ (: possible model 1 may be:, w • 'If lS 1S NS & CCI is NM & CC2 is PM & … CC8 is PB then NCI is PM, f CC8: If LS ls PL & CCI is NS & CC2 is ZR & is PB then NCI is PM 丨 丨, the number is related to the problem itself and the desired approximation Cheng Cheng. What is shown in the tenth figure is defined in the regular domain

60580623 Page V. invention is described in (56) (normal ized domain) Bu 1,1], and Pragmatic off port connected to the fuzzy set (fuzzy set) set (1 ingu ist ic set) NS-PB of sound and meaning. , I think the main criteria fuzzy system variables and model of fuzzy sets. Other variables can also be used for modeling, such as the type of interference, the full interpretation of the fuzzy connection in the criterion (that is, the logical AND's & can be fully translated to the minimum value mi N, the product Product, or more generally such as Parameter t-norm; logical OR can be fully translated to MAX, statistical sum (Probabilistic sum), parameter s — n〇rm

For example, consider here ... Basically all parameters are fixed, except for quasi-tests and membership functions / fuzzy sets (f u z z y s e t). These variables will be changed by a learning mechanism 'to minimize the distance between the model output and the target. Examples of suitable learning mechanisms are gradient-based mechanisms and genetic algorithms. _ Training (model recognition) can have various designs. For example, ^ LS = -0 · 8 can be evaluated as 0.8NS and 0.2 legs and then enter the quasi-j calculation. Additional measurements can also increase the quasi; Measurement of the registration items may be part of: ,, _ after nine or variable to be trained 'member function (m e m b e r s h i p function) may appear as shown in the twelfth FIG. ^ ^ The evolutionary algorithm is repeated multiple times (mul ti-iteration, generation and test (generate —and_test), parallel search and evolution, and the mechanism of selecting the appropriate one in a set of candidate solutions, ^, st ΐInterplanting (breeding) method of the appropriate to generate ,, and / or, in particular, in the embodiments of the present invention, the candidate sought

^ 0623 V. invention is described (57) Fourier-based mapping case - a calculation scheme. This calculation may be square slaughter, such as in - a given mechanical terms can - A kind of expressions. In Γ ίϊί species in a set of allowed operations: promoter, such as based on - set conditions :: f also may be a program electrode, the resistance 4 m of the shutter open circuit, and the like candidate electric crystal may be an electronic hardware description scheme , Then the example of the Spice net_list); °: ship, VerilOg, *. In general, the waiting Ϊ = to;:;: number = any of the components in the arrangement of the component list is not in the number of 4 'component type, the application of evolutionary algorithms U start * 3 * is the technology of the gene program (GP) computer program space The purpose of the fruit. U # I m It can reach a series of instructions to generate a predetermined soil factor program (GP) based on searching for a program that can be used in any language 'from a very high-order power with a high degree of expressiveness 0: language PU. ) High-order scheme may need to be compiled or translated ^ groups - one was Optional "7 easy to do a search program only. - Another choices k choose tied with a direct search on a program implemented in hardware (program implementati〇 n), such as: a = Π circuit configuration praise Qian side opposite already has sufficient elasticity to allow it about the structure and configuration of the timber again said n # - field programmable gate array (Fieid new

Pr〇5rammable Gate Array (FPGA) function into the order of the South order, and converted into the architecture command (configuration * T; ma there are Λ's can be implemented in the architecture command this layer. These devices go through many tentative architectures as after five portions 580,623, the inventors described (58) the possibility of search. another option is to program the decision algorithm, the search data corresponding to the search processing algorithms and hardware implemented separately from (de-coupled). Alternatively, also possible to perform Unique search to find one of the hardware solutions to this problem. Although separate options have a smaller (and possibly better) permutation search space, because of its more powerful notation, the same thing may be counterproductive, if This notation cannot be applied to this problem, in this case unified search is beneficial. The idea behind the synthesis or design of evolutionary circuits is to apply a genetic search to an optimization algorithm that operates in all possible circuit spaces. , And decided to have a solution circuit with a imaginary, functional response. The gene search system is closely linked to one of the circuits, such as a coding table Each circuit is related to a "base code" (genetic code) or chromosome. The simplest of a chromosome is a binary string, that is, a series of 0 bits that encode a circuit. And 1 bit. Synthesis is a search in the chromosomal space to find a corresponding solution with a desired functional response. The gene search is based on a " generate and test strategy, where each time Maintain the candidate systems of a group & the corresponding circuit systems are then evaluated, and the best candidate solution is selected and reproduced in the next generation. This process is continued, straight to a performance goal far (performance g〇al). in the present non iiiertrinsic evolution), the search based on a circuit architecture manner member 2, intended to facilitate the downloading by the final solution, or to estimate it. ', hardware blueprint. such candidate The solution is evaluated by hardware software models, such as: VHDL, Verilog, Spice net-list, etc.

63580623 Page five, made clear ⑽) '~ 1 ----- movable, with hung, complete system using the simulator. In essential evolution, the waiting = scheme is presented in the form of an actual hardware architecture on a programmable device or structure, and it is evaluated by the hardware in the loop using a test or evaluation device. The twelfth figure shows the main steps of evolutionary synthesis. First, a dyeing system ^ group of randomly generated. These dyeing systems are converted into circuit models or bit strings for download to programmable hardware. The circuit response is compared with the specifications of a target response, and each dyeing system is arranged according to how close it is to meet the needs. ° Considering the separation model identification of each individual coefficient of the Fourier transform, the candidate circuits obtain the input value determined by the previous coefficient and the latest sampling, and the goal is a new coefficient. Prepare a new loop over and over again (iterati〇n l〇〇p), a new population of individuals (individual) lines generated by the previous generations of the best individual collection. This step can be achieved by random selection of common individuals, random exchange of their chromosome parts (mixed action), and random flipping of chromosomal bits (mutation action) to achieve this. This process is repeated over several generations to get better and better individuals. This random step can help avoid the optimum value limited regionalization. Monotonic convergence can be forced to use the unchanged individual of the best individual from the previous generation to the next generation. For the great height of the unknown in terms of the search space, this evolution and gene hunting system considered to be the best choice. The search procedure will usually end after a few generations, or after reaching the end of a degree close to the target response. You may find one or several individual solutions in the last generation.

Page 64 580 623

# 1 "=, consider searching for an electronic side with only CM0S transistor '. The basic structure may be based on a transistor arrangement interconnected by a switch sw 11ch, as shown in Figure 14. The transistor arrangement in the fourteenth figure can also be expanded by adding a layer) to the right and additional layers between V + and V- to form a "sea of transistors". Figure shows a part of the Department of the fourteenth transistor sea, the basin can be one of the backbone of a crystal model of electricity. For example, consider nine input signals, corresponding to the input training set and one output. These inputs are connected to different points in the structure using programmable switches, such as the gates of transistors N5, N6, etc. Similarly, the output line by the switch connected to the intermediate point of the probe (intermediate probing point). Each switch, including those of such transistors interconnected endpoint, that te # output to interconnect these transistors endpoint, endpoints that interconnect these transistors to the output (for example, a model of the output determined by each system independently coefficient) switch, each having two states based billion and 1 to indicate an open circuit (open) and closed (closed). As such, the crystal is equivalent to the electrical model based topology (Topology), and may be a square and a defined sequence, which may be a candidate gene described θ model. For one sample and eight coefficients, one embodiment of the present invention has the following steps. First, a chromosome of a group, such as: a color bodies 10 billion. Next, generate a corresponding Spice net - list, wherein a square-based and substituted by the voltage, so as to determine the switching transistor. Each signal source has two signal patterns, which is the time spent by the input training set of a plurality of samples

Page 65 580623 V. Description of Invention (61) Dominate. The first input signal corresponds to this sample. Eight times eight input lines respectively corresponding to the coefficients. The circuit provides an output based on a particular response. The circuit is responsive to the target system in response to comparison. The target response may include a parameter, such as the coefficient value. S. Several circuits were tested. How close each circuit will be to the target to get a suitable value. Suitable values way of example may be a plurality of squared error of how small a point measurement. Best electricity π: to match and include in the next generation. Preferably, the final solution is outside the circuit with a minimum distance of ^ t and the input. For any combination of these post-sampling and Fourier coefficients, the final solution = can provide the target output within a predetermined acceptable error. . The read case is a device that has a computer-accessible media, and is encoded with a computer-executable Fourier transform (DFT) coefficient ^ Ray:? 10 calculations and processing numbers. Derived using a learning model. The X electronic executable steps are in another embodiment. 'The device has the same as: a special application integrated circuit (2 body circuit, steps for calculating and processing digital Fu, the encoding is executable') The computer-executable steps are based on the use of coefficients. In another embodiment, the "derivation is obtained. For example: a special application product circuit has a product circuit, the coefficients of the product circuit. Design C) is used to derive a function. Another embodiment of the present invention is to learn the model to generate. Transformation (DFT) and inverse digital Fourier transform " line such as digital Fourier transform (lnVerse DFT) calculation to use 66 580 623

One of the products derived from the information derivation is stored in the media. What the software side has done is to provide a data-based system that is computer-executable, the computer's address method and installation method, or the number of input signals. The computer can read and store the 50% coefficient, which can be read and stored for control. The circuit, the method and device for generating a series of brain program reading media and storage media, and a method and device or algorithm for making the system, and the present invention also samples to determine that the product has a code device to replace the coefficient contribution system. generator may be provided with a number. Has a learning model circuit. It indicates that the computer program provides the input signal producing a computer readable storage medium described in the number generator, the generator functions via. And in particular to determine the coefficient line, - the computer readable program code device having a first computer instruction means for receiving a sample at a time. Furthermore, the computer-readable code device has a second computer instruction device for updating the function coefficients according to the received samples when receiving each new sample (without waiting for the next sample), and the correction The updated coefficients, so that the coefficient contributions are paired with the correct angle using appropriate sampling, reduce the time delay required to determine the coefficient of the function. In a further embodiment, each coefficient system includes at least one term, which is at least partially related to a combination of a sample and a mathematical function. In this embodiment, the computer-readable code device further includes a third computer instruction device that, when receiving a mathematical function related to the samples and coefficients, uses a combination of each sample to determine a corresponding item for each coefficient. The first to fourteenth drawings, and Tables 1 and 2 are diagrams, block diagrams, learning models, flowcharts, and control flows of the method, system, and program product according to the present invention. It should be understood that in the block diagram, flowchart and control

Page 67

V. Description of the Invention (63) each manufactured by the process described in FIG., The control flow in the manner described, and ^ in block diagram, the flow to reach. These computer program instructions can be set using computer program instructions to generate a machine or computer or other programmable device. These instructions can be used to generate a device or other programmable or control flow blocks or steps. 、 Flow chart order can also be stored in a computer with J = function. These computer programs refer to other programmable devices in a special way, which can instruct the computer and the instructions to read the memory to be duplicated, so that the computer can be stored, which is used to achieve the box circle: including a command device The functions defined in the manufacturing process 1 = brain control flow or other programmable devices, ^ -type private cards can also be loaded into a computer, computer or other programmable steps ^ ^ resulting in a series of operating steps, which are performed by cause the computer to execute: 2 may by his students Da - computer implemented program for realizing the block diagram of the step 3 = such instructions on a device-type = the functionality provided. FIG Yun μ or control flow block defining step or steps performed based support 5 \: flowchart or control flow block or step described in a combination, and means for performing the Μ engagement Laid == 'step # line is a specific function, in box @ earth work programs designated for this device. You should also understand the steps, and ^ blocks :, or each block or step combination in the control flow introduction can be a special = or control block or step in the flow introduction ~ 1 special, hardware-based computer system means; a combination of: I = or step), or special purpose hardware and computer page 68 580623 £, the invention described (64) " " " the 'a reference to the present invention previously described and related drawings, Those skilled in the art will be able to consider many variations and other embodiments of the present invention. Therefore, it should be understood that the present invention is not limited to those disclosed in specific embodiments, and the scope of the present invention should include various possible variations and embodiments, as listed in the scope of the attached patent application. Although Fu eight of his real language, but it is only for a generic and descriptive purposes and not with a specific range of y patent use. Not intended to limit the invention

Page 69 580623 Simple explanation of the diagram 72: Coefficient number output 7 4: Phase number 76, 78: Multiplier 8 0, 8 4: Coefficient output

Claims (1)

580 623 Jj 虎 9011Q17R __ ^ year ^ 0 month repair ". ^, Each time you enter a serial number of ti I, 忐 / 士 ^ ~ ΐίχ 豕 'number transfer device for continuous sampling, 1 φ generator, i total · ^ 1Γ / 、, the device includes numbers, and the complex = is to update at least one of these coefficient tables when receiving each sample: the coefficient generator is to correct the coefficients so that the 2 π — Correct sampling sequence. Sentence-based packet ^ f special range of item 1 set Xiang 'wherein each coefficient Ϊ one composition (which is at least partially based on a two head J - sampling and a mathematical function - when the number of the mathematical function, fine man The spores are connected to one of the corresponding items of the sampling and distance, σb7, to determine each coefficient ... Add the phase library item to ς in the coefficient of the mother, the coefficient generator is: a value immediately before turn -2 mesh to be members of the coefficients, so as to update the system in each of two opposing receiving eye application according to Patent Fanchen Sheng is based? 'wherein the sampling coefficient S production, wherein, when each of:' Update each of these coefficients with each of these ... The solid coefficient is updated with the last sample of the sample; 2 May 丨 No, the titanium ... function is based on the predetermined-coefficient annoying j. 2 'λ \ 修 所when, and wherein each "J f to take the" last - samples be updated on a per sample | wrong to make the contribution factor can be number 4 field angle, (iv) the number of output lines of the generator 4. the line. The patent application range is 0. The generator is connected to two devices: two and two devices, in which the coefficient produces Coefficients, and where: Volume: 丄 Based on the parent sample, to simultaneously update the coefficients and output the coefficients. When the coefficients are updated and corrected, Section 5. According to the number of patents applied for, j, " The life generator also includes a second pass of a memory chipset: in the middle, the coefficient production-—_ Λ coefficient generator is more stored Page 72 580623 ___ No. 90119178____ (June / June) Date: Zheng Zheng 6. Scope of patent application Each sample ’is received when it is received in the memory device. 6. The device according to item 5 of the scope of the patent application, wherein, after the coefficient generator receives and processes the predetermined plurality of samples, the coefficient is generated as' the coefficients are output, and when the coefficient is received, After receiving a new sample of one of the input signals, after receiving the predetermined plurality of samples, the coefficient generator is subtracted from the items based on the new samples, based on the previously stored in the memory device, the predetermined One of the plurality of samples is the first sampled item, and wherein the coefficient generator updates the coefficients based on the difference between the new sample and the first sampled item of the predetermined plurality of samples. ^ Mock apparatus recited in item 5 patents range based 'wherein, until the line ^ -2: governance ΐ generator receives and processes after such predetermined of a plurality of samples, the coefficient generating " " train output the Equal coefficients, and when one of the two input signals is newly received and the predetermined plurality of samples are received, the coefficient generator is subtracted from the coefficients such as f based on the previous storage in the memory device 2. One of the first sampling items of the plurality of pre-sampling samples, and among which, should the number of employees be 4 or 7? ', Wang Wei added items based on the new sampling to each of these coefficients. Middle 8. · ^ Root of the apparatus according to item 2 chilly Patent application range, wherein, of the input signal f, a Fourier transform-based mathematical function of the previous coefficient value, and a gear train which ^ Hai new coefficients according to the following equation: (CH2 + Si2—2 * CM * Si * cos (Pi-i + 180-45 * i)) 05 Second, 丨 is the new coefficient, C i -1 is the previous coefficient, S i is the new sampling, and p 1- 1 系 弁 治 # The phase of the coefficients. 9. According to the device described in φ 上 太 宙 ll 夂 τ # patent scope item 2, at least one Lian ^ page 73 lines have a predetermined coefficient for each of the received, to update the coefficients of the sample function is independent of the piece goods •, according to the patent Fan Jie each update system 1. The function of the patented system a signal range Fourier's mathematical function is based on the following: The correction coefficient of the rotation reference system is based on the following: 'Ai' = Ai * c〇s ((i * 2 * gBi '= Bi * cos ((i * 2 * < wherein Ai 'And Bi, Department 12. 12. One step to represent the input signal based on an input: (A) Take for a set (B) Receive a new sample; (C) Calculate the new sample (D) Get the oldest (E) Sampling: C-the oldest sampling complement; (F) Replaces the six corrections in the trigonometric function ϋ ::. It is affected by 10 £ each time, and the coefficient generator is sampling every class and receiving The sampling is based on each of the pre-selected coefficients so that the pre-selected coefficients are calculated in advance of the other coefficients. The device described in item 9, wherein, when the coefficients are the coefficients, the pre-selected coefficients are output. The device of item 2, wherein the input leaf is transformed, and wherein each coefficient is related Trust the diagonal function, and where the coefficient generator is to update the correction coefficients, and where the equation is related to the update coefficients: π) / Ν) + Βί * 3ίη ((i * 2 * 7f ) / N) ^) / N) -Ai ^ sin ((i ^ 2 ^ 7r) / N) Positive update coefficients, and A i and B i are update coefficients. A method for continuously determining the coefficients of a function by a plurality of predetermined samplings of a signal. The method includes a small sample to obtain a complete set of coefficients C; coefficient contributions; coefficient contributions; compensation + new sampling compensation to calculate the angle of the updated system At an angle corresponding to the appropriate sampling number, P.74 623 VI-^^ 90119178 'Scope of patent application: shaving ϊ 本 10 P Calculate the corrected coefficient from these update coefficients; M _ 1 3 Repeat steps (β) to (G). The second method is based on the method described in item 12 of the scope of the patent application, and further includes the step of storing the sample in a memory device when receiving "these samples."): According to the third scope of the patent application The method described, wherein 'the coefficients are output after the step ^ has been received and the update step has processed the new samples, and the meter 15 has corrected the coefficients. According to the scope of the patent application The method described in item 2, wherein the input is a Fourier transform of the function, and wherein the mathematical function related to each coefficient = signal is a trigonometric function. The method described in item 2 further includes the step of outputting the predetermined coefficient each time ΑΑ 本 I fetches ^ and) WU Rangu solid spear Z item corpse / T geocentric | 17 new coefficient, according to an input signal one sample for determining the given input signal train including at least a factor of G. · an apparatus for one of the functions of the decision in accordance with one foot on a sampled input signal, wherein each coefficient item 'according to at least partially the sampling and a mathematical function of In the device, the device includes: a combination, a first memory device for storing a numerical value representing the mathematical function related to the sampling and each coefficient; and a coefficient generator connected to the first memory device digital communication, wherein the coefficient generator of the sampling system receives the input signals, and 'the coefficient generator accesses the first memory-based counter PEI for each coefficient, and the sample and is multiplied by the sample indicates that the associated The numerical value of the mathematical function of the coefficient, dream to define an item 'and then add the item to the previous value of the coefficient to update the coefficient, wherein the coefficient generator corrects the coefficients. 580623 «2. Application for amendment 18. the patentable scope - x * teeth will to the two input signals η app :: brain-based product into a plurality of sampling signals indicated by the number of coefficients comprising computer program product, Shu, the power of a computer with I to the recording medium, the electrical: 4 ::,: computer-readable code means receiving the first - to the computer means may: take; Sigma attack comprises: mock ^ # computer instructions to receive a second sample of the signal And, among them, at least one of the coefficients is updated and corrected each time the sample is taken, more = including the computer instruction device 1 for receiving, the number to make these departments at least-and update the 1q other habitat-based -Γ different number of tables in the correct sequence of samples',;. item 18 of the patentable scope of ΐ computer program product, a straight line is set to receive each such sample, which sample reception I protection means ㉔ based on the received samples do not wait for the next decision to reduce the data for each sampled ^^ update coefficients of the function of such 'by - the exact time required for (ii) the number of coefficients 2 billion · the delay range of the patent Shu r g in said computer system comprising at least one item of which at least a portion ;;; Sampling - # = = ί - a combination, which the computer can read the code means further comprises two power means unscrewed $, in which upon receipt of the sample and associated with the coefficients of the mathematical function, a combination of each sample to determine each coefficient corresponding to one item 'and where' for each coefficient, the second computer instruction means corresponding to the line items added to the one of the previous coefficient value So as to update the coefficient. 21. The computer program product of the square of the second term chilly Patent application range, in which 'only one item for each sample of each compensation coefficient, and wherein the said sixth > Page 76 is the scope of patent application in 2 2 computer instructions, without the need for • according to the application, the second Based on updating every 10-2 r samples of each such sample 23 · According to the application, the computer can take 2 4 each of them · According to the application, the predetermined coefficient has been processed in the first, wherein after receiving the sample and then the computer-based instruction means placed in contact patentable scope of the brain were then stored instructions Installing these lines, the program reads the sample patentable scope second when it is patentable scope of computer instructions and the like after sampling 'when the input receiving when the 'counter for the number of samples in a line from the first use of the sampling head of such a difference, by the predetermined plurality of received update to each item of the sample based on the number of counter 20, and computer 20 refers to the item The code device is stored in the 23rd device. The first input computer can take the new item. When sampling the medium coefficient, the electric power mentioned above is received by each of them. When the electric power is connected to the electric power device, it includes the first memory. The electric receiving and the two computer-referenced newly acquired reading program items and wherein each of the new sampling brain program samples outputs the brain program in the four-command device of the brain program when sampling. Second electrical means to make the sample based on the code means and said second subtracting the first product, which is updated to update the coefficients of each coefficient. Product, its device, for its product, its brain instruction device, which outputs the order, etc., and includes a fifth such order, which includes a computer command, and a sampling item 25. a sampling unit for determining a sampling unit based on an input signal means one of the functions of the coefficients of representation of the input signal, wherein the apparatus includes a coefficient generator system, which provide such received coefficients after each sampling, the coefficient generator comprising: a receiver for receiving a New sampling; and a learning model processor that uses predetermined coefficients and Page 77 580623 Amends h years. June 6th, the scope of patent application '' 一 " '~' 疋 Pick up new 26. Model 27. Model 28. Model 29. Signal product coefficient pattern to a series of $ 焴: 广 、 〆 、. the learning learning samples according to the application processor is based on the application of the wafer to one kind of processor-based system comprises one of a function 'of the new coefficient-based sample using a pre-processor system according to the application. ’By mapping the new system model processor, it learns the model processor. Item 25 of the patent scope includes a neural network At least one of the patent scope 25 includes a neural network. The scope of patent No. 25 & includes a fuzzy logic number in Sun Xin sampling. _..., the receiver is capable of providing the described device, wherein the learning processor. The apparatus of claim, wherein the integrated circuit chip and learning means of the fuzzy logic a, wherein the processor _ study. The coefficients of the input decision represented by a sampled input signal according to the computer program product, wherein the coefficient generator computer program, which comprises a generator providing such executable code after receiving a learning model for each sample, the study a predetermined fixed coefficient mode and sampled training 'whereby a new mapping coefficients 30. The computer program product according to item 29 of the scope of the patent application, wherein the learning model code includes a neural network code. 31 · The computer program product of the second range of nine patent applications, wherein the system comprises a learning model code Fuzzy logic code. 3 2 · — A method for determining the coefficients of a function of an input signal represented by sampling of an input signal, wherein for each sampling system a complete set of coefficients is generated, the method comprising the following steps: providing a predetermined Sampling of sets and coefficients of predetermined sets; 580623 No. 6i 90mm_7 working years / ° month "^ said to amend it patent application scope to train a learning model to use the coefficients taken from these predetermined sets to map coefficients to sampling; provide a new sampling to the learning model; And or allow the learning model to derive J3 3 for the new sample. * The square number system according to item 32 of the scope of the patent application includes a Fourier transform coefficient. A method for deriving shows a method of designing a system of one of the coefficients as a function of the input signal, and hard, and at least one of the software, which comprises reacting, for a predetermined set of sampling lines and ^ in the predetermined set of predetermined The sampling of the set is an input, and the set is trained with one of ^ as the output; the projection algorithm determines the chromosome population that generates the code ^ p ^ on these inputs (A); (B) generates the candidate mapper form; Nocha Describe the advantages of using hardware description or soft (C) to connect the complete members, such as the members of f, etc .; and the candidate mappers (E) choose the best chromosomes to test for f combinations. The best (F) repeats step (A) ... to the remote solution; and the lower error scheme mapping. (F), up to the sample and the predetermined set coefficients. Method, the sampling of which includes the number of steps; in the order, the input letter of the department, the system includes at least one case of the coefficient of the set and the image description between the outputs, the chromosome judges the knots to obtain Found under new dye conditions 580623 Xiu / J No. 9011917S Patent Application Range 35 · —A method for determining the Fu value and the most recently received sample from a sub-set coefficient to a sub-set value. The method includes the following steps: The new value of the sub-set of the coefficient Provide a learning model processor; w train the learning model processor into the next set of Fourier coefficients, and the input model of a predetermined subset of the Fourier coefficients can produce a Fourier system :: 2 outputs, where the study • Input of a subset of coefficients. A sub-market, which maps to Fourier. A computer-implemented method, using # > like, to derive the input letter c. The design of an input signal based on the selection of an input signal includes the following steps: & the coefficient of a function The device ~ (A) generates-randomly selects the dyeing cells and compares each chromosome & one: generation; (C) specifies the turn-in value to each electric = type; it consists of a predetermined sampling and a pre = type, Among them, at least one of the input numerical update is determined by the coefficient of the set and a predetermined set of fD) the specification of the target response; E) the specification of the response and ratio of the derived circuit model; & the special model response And the target response is similar to the circuit model according to the circuit response;, the degree of μ private response, rank each ^ ρ \) to generate the circuit model including the circuit model with the highest = in the previous generation -new Ethnic group,) Ik machine father changed the components of these chromosomes and randomly reversed the bit six of the chromosome 580623 amendment number 90119178 6. Application for patent scope yuan; (I) Repeat the process of several generations to achieve better Circuit model; and (J) at a predetermined number - or when the generation of the similarity between the model response and the target response circuit becomes close enough to stop the operation. 37. The method according to item 36 of the scope of patent application, wherein the coefficients are used for a Fourier sequence. 0.-IA p. 81