MXPA99004654A - Adaptive sparse equalization filter - Google Patents

Abstract

Un ecualizador adaptable digital disperso que incluye filtros de Respuesta de Impulso Finito (FIR) directo (302) y de retroalimentación (304) exhibe operación mejorada de Mínima Media Cuadrática. Un conmutador (404) asigna uno de varios multiplicadores (414-1 a 414-5) de cada filtro a cada una de las ubicaciones de toma de ese nido. Cada multiplicados es asignado inicialmente a una ubicación de toma preseleccionada con un coeficiente de carga predeterminado. Después de completar ciclos de tiempo sucesivos que tiene una duración igual a un número dado de periodos de muestra de datos, para cada filtro, se determina (a) un primer grupo de coeficientes asociados con multiplicadores entonces asignados a ubicaciones de toma de valor no cero, y (b) un segundo grupo de coeficientes asociados con multiplicadores entonces asignados a ubicaciones de valor cero. Los multiplicadores asociados con los coeficientes del primer grupo durante un ciclo de tiempo recién completado conservan sus ubicaciones de toma durante el siguiente ciclo de tiempo. Los multiplicadores asociados con el segundo grupo de coeficientes durante un ciclo de tiempo recién completado son reasignados a nuevas ubicaciones de toma para su uso durante el siguiente ciclo de tiempo.

Description

ADAPTABLE DISPERSED EQUALIZATION FILTER The present invention relates to infinitely dispersed impedance response (IIR) equalizing filters and scattered decision re-equalization (DFE) equalization filters, such as those that can be used to remove phantom images ( multi-path equalization) in terrestrial television communications layers, and, more particularly, to a filter that is bi-dimensional, of least-squared-half (LMS), adaptive dispersed (TLLS) that, in response to a The signal received from multiple trajectories determines the particular location of the capture of the particular load factor for each non-zero-valued take-off of the filter that produces a maximum cancellation of phantoms that is achieved first and then maintained over time. Reference is made to the article "U n Manual on Cancellation of Phantoms in Television Systems," by Ciciora et. to the. , which appears in Transactions of the IEEE in Consumer Electronic Products, Vol. CE February 25, 1979. The structure and operation of an adaptive filter based on LMS used to remove the phantom images is described on pages 42 and 43 of this Article. Adaptive filters based on LMS have been widely used in digital communication systems for channel equalization. A desirable feature of the LMS-based adaptive filters is that the derivation of the appropriate device values from their load coefficient factors does not require a pre-calculated preparation signal. However, the number of mathematical operations elements implemented in an LMS-based adaptive filter is linearly proportional to the number of equalizer filter taps. In some applications, a large number of shots may be required. For example, in the transmission of digital terrestrial television, the range of multiple trajectories can reach up to 2 ms, resulting in the total number of takes of the equalizer filter required being of the order of 200, or in some cases I include more. An LMS-based digital equalizer filter for use with a continuous current of real-value data samples requires mathematical operations that include two multiplications per take for each period demonstrates successive T (where T is 0.2μs (5 million samples per second) or less in the case of a digitized NTSC television signal received). This produces a large number of mathematical operations per second, which results in a corresponding integrated circuit implementation cost for an adaptive equalizer filter based on LMS. This is a reason that prevents a large LMS-based adaptive equalizer filter from being used in many commercial applications. In the case of a filter to remove phantom images, only a small fraction of the 200 or more filter intakes have non-zero values. This allows the use of a sparse filter having pick-up locations corresponding only to these non-zero values. The respective load factor coefficient values corresponding to these filter sockets with non-zero value must be selected in such a way that a maximum cancellation of phantom images occurs. The problem is to determine (1) which tap locations at any time are the non-zero value filter sockets and (2) the required values of the respective load coefficient factors. The conventional manner known in the art to provide the required values of the respective load coefficient factors is to use an adaptive digital filter that responds to a preparation signal that has been previously derived by a computer program using prior information. By way of example, a preparative signal computer program of this type is disclosed in United States Patent Number Serial No. 5,065,242, issued to Dieterich and co-inventors on November 1, 1991. In addition, the Patent of the United States of North America Serial Number 5, 388,062 issued to Knutson on February 7, 1995, requires a computer signal preparation program for use in the reconfiguration of a filter to remove phantom images, selectively converting the actual tap locations of the reconfigured filter to only their non-zero value-taking locations and programmably provide values of selectable load-factor factors to only their non-zero value-taking locations. In many applications, including television, the communications channel is affected with scattered echoes. In this case, the adaptive filter of the receiver, after the time of the establishment of the adaptation, will have some non-zero value sockets and some zero value sockets. Only these zero-valued sockets contribute to channel echo cancellation. However, under slow time variation anal conditions, the echo delay may be time dependent, so that from time to time in the echo tracking mode, it may be necessary to re-locate the value capture locations no zero in the filter. Reference is made to the article "An adaptive multiple echo canceller for slow time variant echo trajectories," by Yip and Etter, which appeared in Transactions in the I EEE Communications, October 1990. The basic approach of Yip-Etter is that, where there is more than one echo to be canceled, the filter tracking performance for the time variant eco will be improved by using a separate filter from a plurality of smaller filters, each of which is designed to cancel a echo separated from the multiple echoes. The improvement in performance is due to the e-limitation of zero value shots of the filter to remove conventional large ghost images. The problem with the Yip-Etter approach is that it requires complex mathematical operations for the location of echoes and estimates of their duration. Therefore, the Yip-Etter approach will not reduce the cost of implementing integrated circuits. In view of the foregoing, there is a need for a filter to remove relatively inexpensive phantom images that operate continuously in real time to first determine, without the use of a preparation signal, the zero value of the filter to remove phantom images. conventional large and then effectively eliminate such zero-value shots. Additionally, such an inexpensive filter is required that is capable of canceling all the multiple echoes without suffering from the aforementioned problem of the Yip-Etter approach. The invention is directed to an equalizer filter system that improves the operation in accordance with the LMS algorithm of a dispersed digital adaptive filter DFE or I IR, wherein the adaptive filter includes an FI filter R comprising pickup locations L + 1 in number including a take location in the first entry of the same, M multipliers in number, where M <; L + 1, and switching means for assigning any of the multipliers M to any of the acquisition locations L + 1, where "L" is defined in equations (1) - (3) below. Each of the multipliers M is initially assigned to a previously selected separate take location of the pick locations L + 1 and each assigned multiplier is given its own load coefficient w previously? Selected. Then, at the end of each of the successive time cycles having a duration equal to a given plural number D of digital data sample periods T, a first group of individual load coefficients, associated with multipliers, is determined (a) which are then assigned to certain capture locations L + 1, which have a non-zero value (that is, they have an absolute magnitude that exceeds in value a given positive minimum value) and (b) a second group of load coefficients individual ws associated with multipliers then assigned to the certain pick locations, which have zero value (ie, have an absolute magnitude that does not exceed a given minimum positive value). The multipliers associated with the individual load coefficients w of the first group during a newly completed cycle of the successive time cycles retain their assignment to the certain pick-up locations of the pick-up locations L + 1 during the next cycle time of the cycles of successive times, and the multipliers with the individual load coefficients w of the second group during a newly completed time cycle of the successive time cycles are reallocated to take-up locations which are not the certain pick-up locations of the take-over locations L + 1 , for use during the next cycle of time that occurs from the aforementioned successive cycles of time. This reallocation preferably follows predetermined priority rules. BR EVE DESCR I PC I O N OF THE B UJOS Figure 1 is a block diagram of a preferred embodiment of an adaptive I I R filter based on LMS; Figure 2 is a block diagram of a preferred embodiment of an adaptive DFE Echo filter based on LMS; Figure 3 is a functional block diagram of the components of the complete system of an adaptive equalizer filter based on TLLS; Figure 4 is a block diagram of a preferred embodiment of the present invention that constitutes an adaptive IIR equalizer filter based on TLLS; and Figure 5 is a block diagram of a generic TLLS FI R filter that can be used specifically for each of the TLLS FIR direct and feedback filters of Figure 3. The preferred mode of the adaptive IIR filter based on LMS 100a shown in Figure 1 comprises the direct LR filter LR 102, the feedback LR filter LR 104, the adder 106, the adder 108 and the separator 1 10. Each of the digital data samples successively x (n) a continuous sample stream is applied to the input of the direct LR filter LR 102. The continuous digital data samples resulting in the output of the direct LMS filter 102 are applied to a first input of the adder 106 and the samples of continuous digital data in the output of the feedback filter FIR LMS 104 are applied to a second input of the adder 106. The samples of continuous digital data and (n) in the output of the adder 106 (which consti the filter output equalizer I IR adaptive) are applied to the feedback filter input LMS 104, the negative input of the adder 108 and the input of the separator 1 10. The separator 1 10, which separates a given number of bits Significant lower values of each digital data sample applied to the input of the same, directs the remaining most significant bits of each sample of digital data (y '(n) in Figure 1) to the positive input of the adder "| 08. Adder 108 drift, as an output, a current e (n) of digital error samples, each of which is applied to an error input of each of the FIR filters 102 and 104 (as shown in Figure 1 with separate contacts ). The preferred embodiment of the adaptive DFE equalizer filter based on LMS 100b of the prior art shown in Figure 2 also comprises the direct LRS filter LR 102, the feedback LR filter LR 104, the adder 106, the adder 108 and the separator 1 10. As shown in Figure 2, the continuous digital data samples at the output of the separator 1 10 (y '(n) in Figure 1) (as opposed to the continuous digital data samples and (n) at the output of the adder 106) are applied to the input of the feedback filter F1 R LMS 104. This is the only difference in structure between the adaptive DFE equalizer filter based on LMS 100b shown in Figure 2 and the adaptive IR I equalizer filter. based on LMS 100a shown in Figure 1. As is known in the art, each data sample is a binary number of multiple bits in which the most significant bit is indicative of the polarity of the number (i.e., positive or negative) and the remaining bits are indicative of the magnitude of the number. The most significant bits of each digital data sample that is sent from the output of the separator 1 10 to the positive input of the adder 108 may consist of only that most significant bit data sample (ie, the bit indicative of polarity) or it may consist of this most significant bit together with at least the next most significant bit (ie, the most significant bit indicative of the magnitude). In any case, the polarity of the number represented by each of the currents of the digital error samples e (n) is opposite to that of the data sample from which it is derived and its magnitude is equal to or less than that of the sample of data from which it is derived. The algorithm of a conventional LMS adaptive filter can be described by its input / output equation and its take value update equation. The output equation is = W ", aX" + W "? Xn-l + * • • + W "? Xn- (2) where the index j and n represent, respectively, the location of take and time; and the take value update equation is W, n + l, i wn, ^ e "xn, i = 0X..L, (3) where in is the error term defined as the difference of the filter output, and ", and the desired output, dn, in time = n. Thus, in = dn - yn (4) From the previous equations, it is clear that, during each sample period, the value x in each of the sampling locations of the filter L requires a first multiplication by the value of load coefficient wn during that sample period, according to equation 2, followed by a second multiplication by the value of its error term in during that sample period, according to equation 3. In a real example of a filter to remove conventional LMS digital ghost images for an NTSC television signal, the filter, such as that shown in Figure 1 or Figure 2, the direct LMS filter 102 R can have 100 filter locations and the filter FI R Feedback LMS 104 can have 150 filter locations, while each sample period lasts only about 0.2μs. Thus, the total number of reports per second is 1 .25 billion (or 5 million per second of each capture location of the 250 intake locations). A filter to remove phantom images for an NTSC television signal only uses samples of real digital data and real load coefficients. However, a filter to remove ghost images for high definition television (HDTV) uses complex digital data samples and complex load coefficients, while each sample period should have a duration of significantly less than 0.2μs to accommodate the width of Widest video band of a high definition television signal. This means that each tap location requires eight multiplications (instead of just two) during each sample period, causing the total number of multiplications per second for the 250 tap locations to be significantly greater than 1.25 billion (or significantly greater than 5 million per second for each shooting location of the 250 shooting locations). As is known, each capture location of each LMS FI R filter component of a LMS-based adaptive equalizer (such as FIR filters 102 and 104) is supplied with an initial load coefficient value, present at time ZERO. In the case of a filter to remove phantom images, a previously selected take location (for example, the middle take location) of each FIR filter is fed with an initial load coefficient value equal to +1, so as to correspond to the main television signal, and all other taps of each filter FI R are fed, respectively, with initial load coefficients having a minimum magnitude value. Both the LMS-based adaptive IQ-IR equalizer of Figure 1 or the LMS-based adaptive DFE EQ of Figure 2, operating from the ZERO time as a filter to remove phantom images according to the aforementioned LMS 1 and 3 equations, they will then produce a large number of sample periods in the minimization of the value of the error term in and the convergence of the load coefficients of each of the acquisition locations converging to a relatively stable value. Converged load coefficients that have nonzero magnitude values (ie, a value above a minimum absolute value) are associated with the minority of tap locations corresponding, respectively, to the main television signal and to each phantom, if There is, in the received television signal. Most tap locations are associated with converged load coefficients that have zero values and, therefore, are unnecessary, since they do not contribute anything to the filter output yn. The problem is that the identity of the particular take-up locations of the large total number of take-over locations of each of the FI filters R 102 and 104 that have zero value at any time remains undetermined until they have been determined only after the operation of the above-mentioned conventional LMS algorithm, with its need for a large number of multiplications. Referring now to Figure 3, a functional block diagram of the components of the complete system of a TLLS-based adaptive equalizer filter is shown which solves the aforementioned problem produced by the use of the conventional LMS algorithm. As shown, the complete TLLS-based adaptive equalizer filter system comprises the TLLS 200 equalizer filter, the energy index monitor 202, the tap control logic unit 204. A continuous stream of digital data samples x (n) is applied to a first input of the TLLS 200 equalizer filter; a series of initial take control signals determined by the user Tcf + (0) is applied to the respective second inputs of the equalizer filter TLLS 200 and to the respective first inputs of the tap control logic 204; a series of adaptive sample control signals Tc. + b (n) of the tap control logic 204 are applied to the respective third inputs of the TLLS 200 equalizer filter; a series of load coefficient values w. + b (n) of the equalizer filter TLLS 200 is applied to the respective inputs of the energy index monitor 202, a series of load coefficient energy indices Pf + monitor index of energy index 202 are applied to the respective second inputs of the tap control logic 204, and a continuous stream of digital data samples and (n) constitutes the output of the TLLS 200 equalizer filter. As shown in Figure 3, the TLLS 200 equalizer filter, by way of example, may be an IIL equalizer filter TLLS 200a having a structure similar to that of the adaptive equalizer filter II R based on TLLS 100a shown in Figure 1. More specifically, the structure of Figure 3 differs from that of Figure 1 only in that the direct LMS FIR and feedback filters 102 and 104 of Figure 1 are replaced in Figure 3 with the LMS FI R direct and feedback filters. n 302 and 304. In all other respects, the filter I IR of equalizer filter TLLS 200a is identical in structure and function to the adaptive I IR equalizer filter based on TLLS 100a. alternatively, the TLLS 200 EQ filter can take the structural form of a TLFS 200b equalizer filter DFE filter (not shown) that has a structure identical to that of the filter l | R LMS-based adaptive filter equalizer 100b shown in Figure 2, except for the replacement of only LMS FI R direct and feedback filters 102 and 104 of Figure 2 by LMS FIR direct and feedback filters 302 and 304. As indicated in Figure 4, the direct filter TLLS FI R 302 has the respective substrings Tc. (0) and Tc. (N) of the series TCf + b (0) and Tcf + b (n) applied as additional inputs it has the substring w. (n) of the series wf + b (n) constituting an additional output from it. Similarly, the TLLS FI R 304 feedback filter has the respective substrings Tcb (0) and Tcb (n) of the series Tcb + f (0) and Tcb + f (n) applied as additional entries to it and has the substring wb (n) of the series wf + b (n) constituting an additional output from it. Figure 5 shows a block diagram of a generic TLLS FI R 400 filter that can be used specifically for each of the direct and feedback LMS FIR filters 302 and 304 of Figure 4. The TLLS FI R 400 filter comprises the digital change register of the take L 402, the means of commutation of the multiplier M 404 and the adder 406. A current of digital data samples x (n) having unsp T is applied to the entry of the change register 402 and the shift register is normally changed at a rate of change equal to 1 / T. The tap locations L + 1 are defined by the tap change register L 402, including an additional tap 0 at the input of the shift register 402. The switching means 404 is initially able to assign each of the multipliers M to any of the L + 1 tap locations according to the series of tap control signals Tc (0) applied to them at the ZERO time and, subsequently, according to the series of adaptive tap control signals Tc (n) applied to them. The respective outputs of the assigned multipliers are applied as inputs to the adder 406, and the adder 406 derives an output equal to the sum of the assigned multiplier outputs that are applied to it (the output of the adder 406 constitutes the output of the TLLS FIR filter 400).

In general, the present invention only requires that the number of multipliers M, which is determined by the user, be less than the number of locations of the change register L + 1 (ei M <L + 1). However, when the TLLS FIR 400 filter is used in an equalizer filter to remove scattered phantom images, the M / L ratio is generally never larger than 50% and is substantially less than 50%. In the next . In a detailed description of the principles of the present invention, it is assumed that the TLLS FI filter R 400 is being used in an equalizer filter for removing scattered phantom images in which the L / 2 tap location corresponds to the main signal 408 of a signal received television and that at time ZERO there is no information on whether the received television signal includes phantom images or not, in spite of the unknown facts that the location of shot a .T corresponds to the position of a single previous ghost image 41 0 in the received television signal and the pickup location a2T corresponds to the position of a single rear phantom image 412 in the received television signal. In accordance with the principles of the present invention, at time C ERO, the series of tap control signals determined by the user To (0) are such that the switching means 404 allocates one of the multipliers M, with a coefficient of load + 1, to the main signal tap location L / 2 and assigns the remaining M-1 multipliers, each with a positive minimum load coefficient, to the tap locations that are substantially symmetrically arranged in the main signal tap location L / 2 and are, to the extent possible, equally spaced apart in the following predetermined order (where the main signal socket location L / 2 is number 1) by an amount that allows a maximum coverage of the entire range of L + 1 tap locations by the multipliers M: (M-1) - 5-3- 1 -2-4- -M, where - represents the separation between the adjacent multipliers. It is so, by way of example, where M / L = 50%, there is an unassigned take location between the adjacent assigned take locations; where M / L = 33%, there are two unassigned tap locations between the adjacent tap tapped locations, and where M / L = 25%, there are three unassigned tap locations between adjacent tap tap locations. Then, the TLLS F1 R 400 filter is allowed to operate in an LMS manner according to the aforementioned equations 1 and 3 for a first of successive time cycles, each of which has a duration DT, wherein D is a a relatively large predetermined number (eg, 3000) that produces a sufficiently long DT cycle duration to allow the convergence of the load coefficients of all initially assigned Joma locations. At the end of this first cycle of time, the respective load coefficients of all the initially assigned tap locations are sent to the energy index monitor 202. The power index monitor 202 performs a first function of measuring the energy of each of these take locations by squaring each of the respective load coefficients sent to it (ie, Pj = | wn, il 2 and then performing a second function of indicating power to each P? by assigning a binary ONE to each Pi that exceeds in magnitude a given minimum value (which is indicative of a non-zero value charge coefficient) and assigns a binary ZERO to each Pi that does not exceed in magnitude a given minimum value (which is indicative of a coefficient zero value charge) The energy index resulting from all the evaluated load coefficients is sent as a P-index input to the take control logic unit 204 which, in response to it, reads to energy index, and then performs logical functions with the following rules to derive the adaptive tap control signal output: 1. The tap locations assigned to the multipliers associated with a binary ONE energy index (ie, a non-zero value acquisition location) during the newly completed DT cycle retain their assignment to these same multipliers for the duration of the next DT cycle that occurs. 2. The take-over locations assigned to the multipliers associated with a binary ZERO-ZERO energy index (ie, a zero-take location) during the newly completed DT cycle lose their assignment to any of the multipliers for the duration of the next DT cycle. what happens and these multipliers are available for reassignment to other shooting locations according to the following criteria: (1) An unassigned take location during the newly completed DT cycle becomes a member of a first group of candidates with a higher priority to assign an available multiplier during the next DT cycle that will only occur if at least one of its immediate neighboring take locations during the newly completed DT cycle was assigned a multiplier associated with a binary ONE energy index. (2) An unassigned take location during the newly completed DT cycle becomes a member of a second group of candidates with a lower priority to assign an available multiplier during the next DT cycle which occurs only if both of its take-over locations Immediate neighbors during the newly completed DT cycle were not assigned to a multiplier. (3) Any intake location that does not meet criterion (1) or (2) is not a candidate for allocation of a multiplier available during the next DT cycle that occurs (so, the criter? (3) includes any tap location that during the newly completed DT cycle was assigned a multiplier associated with a binary ZERO energy index). (4) The intake locations that are members of the first group of candidates are assigned multipliers available to a given order of importance, in which any candidate intake location closest to the average intake location L / 2 is more important than any candidate pick-up location away from the average L / 2 takeoff location, until all pickup locations that are members of the first group have been assigned multipliers or all available multipliers have been assigned. (5) If any of the available multipliers remain after all members of the first group of candidates have been assigned multipliers, available multipliers are assigned to the intake locations that are members of the second group of candidates, in the same order given of importance, until multipliers have been assigned to all intake locations that are members of the second group or all available multipliers have been assigned. The take control logic unit performs the aforementioned logic functions at the end of each successive cycle of the DT cycles, which makes the M multipliers perform an efficient search to first find and, once they find them, to preserve locations of non-zero value taking (ie, those associated with a non-binary U-energy index) in successive continuous DT cycles, while each non-zero finding location yalor persists e? weather. At the same time, the search continues for additional non-zero value-taking locations during each of the successive continuous DT cycles, using, in each of the successive cycles DT, those of the available M multipliers during the DT cycle which immediately preceded that DT cycle. Returning to Figure 5, after a number of DT cycles, the search will have found non-zero value-taking locations LT72, a .T and a2T (indicated in Figure 5 by solid lines) that contribute to the output of adder 406 , and multipliers 414-1 to 414-3 then assigned to these three non-zero value-taking locations will be retained by them during future DT requests. However, the M-3 multipliers are then associated with zero value-taking locations (such as the multipliers 414-4 and 414-5 associated with the 0 and 100 tap locations)., which are indicated in Figure 5 by interrupted lines) continue to be involved in the search during each subsequent Dt cycle. Therefore, if in time, the non-zero value-taking location corresponding to one or both of the ghost images mentioned above had to be changed, or if another ghost image appeared, the search would continue to find them relatively quickly. It is evident, statistically speaking, that, in practice, a larger M / L + 1 ratio of multipliers to intake locations results, on average, in a smaller number of DT cycles that are required for the search in order to find all the ghost images that may exist. However, a larger ratio M / L + 1 increases the cost by requiring a greater number of multipliers. Therefore, the optimal M / L + 1 ratio depends on the particular environment in which the TLLS adaptive filter is to be used: Although only a preferred TLLS filter has been described specifically in the present one, which is implemented with a TLLS filter FI R direct output load and a filter TLLS FI R output load feedback, the present invention is intended to cover an adaptive filter TLLS that is implemented with one or more filters TLLS FIR input load, or with only a TLLS FI R direct filter or with only a TLLS FI R feedback filter.

Claims (23)

1. CLAIMS 1. In an equalizer filter system comprising a dispersed digital adaptive filter (200) and means for controlling the operation of such adaptive filter (202 and 204) wherein (1) such an adaptive filter includes a pulse response filter finite multi-tap digital (FI R) (302 or 304 and Figure 4) responding to a stream of successive data samples x (n) having a sample period T applied to a first input thereof; (2) said means for controlling the operation of such adaptive filter includes first means for deriving at least one error term e (n) of minimum root mean square (LMS) in response to each successive output sample and (n) of such adaptive filter (306 and 308) and applying each error term e (n) derived LMS to a second input of such FIR filter to perform the operation of said FI filter R according to the LMS algorithm; and (3) said multi-tap filter FI R comprises take-up locations L + 1 in number including a take-up location in said first input thereof (402), multipliers M in number (414- 1 to 414-5 of the multipliers M) where M < L + 1, and switching means for assigning any multiplier of said multipliers M to any of such tap locations L + 1 (404); the improvement wherein said means for controlling the operation of said additional adaptive filter includes: Seconds means for initially assigning each of said multipliers M to a previously selected separate take location of the said tap locations L + 1 and initially supply each multiplier assigned its own load coefficient w previously selected (Tc (0)); third means, which respond to the completion of each of the successive time cycles, to determine (a) a first group of individual load coefficients w, associated with multipliers that are then assigned to certain L + 1 tap locations, which have a absolute magnitude exceeding in value a given positive minimum value and (b) a second group of individual load coefficients w associated with multipliers then assigned to certain tap locations, which have an absolute magnitude that does not exceed a positive minimum value. given, wherein each of said suesive time cycles has a duration equal to a given plural number D of sample periods T (202); middle rooms, which respond to the individual load coefficients w of such a first group during a newly completed cycle of one of the successive cycles of time, to preserve the allocation of the multipliers associated with these individual load coefficients w of such a first group to certain take-over locations of the take-over locations L + 1 during the next cycle of time occurring from said successive time cycles (204 and Ct (n)); and fifth means, which respond to the individual load coefficients w of said second group during the newly completed cycle of the aforementioned successive time cycles for reassigning multipliers associated with these load and individual coefficients of said second group of intake locations. which are not the certain locations of taking the locations from to to L + 1, for use during the next cycle of time occurring from the aforementioned successive cycles of time (204 and Tc (n)).
2. 2. The equalizer filter system defined in claim 1, wherein: said fifth means follows predetermined priority rules when reassigning multipliers to such take-up locations that are not true of said take-up locations L + 1.
3. 3. The equalizer filter system defined in claim 2, wherein said predetermined priority rules comprise the following rules: any such take-up locations that are not true of the said take-over locations L + 1 has a higher priority than that assign a multiplier for its use during the next cycle of time that occurs from the successive time cycles only if at least one of its immediate neighboring intake locations was a member of such first group during the newly completed time cycle of such successive time cycles; any of the aforementioned take-up locations that are not certain of the said take-over locations L + 1 has a lower priority than being assigned a multiplier for use during the next cycle occurring from the said successive time cycles only if none from their immediate neighboring take-off locations were a member of such first or second groups during the newly completed cycle of such successive cycles of time; and any of the aforementioned take-up locations that are not true of the above-mentioned L + 1 tap locations that do not have a higher or lower priority take location priority may be assigned a multiplier for use during the next cycle that occurs from the aforementioned successive time cycles.
4. 4. The equalizer filter system defined in claim 3, wherein said predetermined priority rules comprise the following additional rules: first, the higher priority take locations are assigned multipliers in a given order of importance, in which any higher priority take location that is closest to a previously selected take location member of such a first group is more important than any higher priority take location of the take location until at all higher priority take locations they have been assigned multipliers or all available multipliers have been assigned; and second, if multipliers are still available after all the higher priority take locations have been assigned multipliers, multipliers are assigned to the lower priority take place locations in a predetermined order of importance, in which any location The lower priority take location that is closer than another lower priority take location to that previously selected take location is more important than the said lower priority take location, until all the lower priority take locations are have assigned multipliers or all available multipliers have been assigned.
5. 5. The equalizer filter system defined in claim 4, wherein: said previously selected tap location is the middle tap location, L / 2, of said FI filter R (Figure 4, tap L / 2).
6. 6. The equalizer filter system defined by claim 1, wherein said dispersed digital adaptive filter is suitable for use as a filter to remove phantom images for a television signal that has been received on a terrestrial communications channel; and wherein: such second means initially allocate one of said multipliers M (41 4-2) to a previously selected particular pick location of said pick locations L + 1 and initially supplies a load coefficient w (wL / _) that it has a value + 1 to that multiplier (414-2) which is assigned to said particular pick location previously selected from said Take Locations L + 1 and initially assigns each of the remaining M-1 multipliers (e.g. 414- 1) and 414-5) to other previously selected tap locations of such tap locations L + 1 and initially supplies a load coefficient w (eg, w. and w5) that has a minimum positive value to each of the aforementioned M-1 remaining amplifiers, wherein said particular pick location previously selected from said L + 1 tap locations corresponds to the main television signal.
7. The equalizer filter system defined in claim 6, wherein: said means follow predetermined priority rules for reassigning multipliers to such tap locations that are not certain tap locations of said tap locations L + 1.
8. 8. The equalizer filter system defined in claim 7, wherein said predetermined priority rules comprise the following rules: any of those tap locations that are not true of the said tap locations L + 1 has a higher priority than that assign you a multiplier for use during the next cycle of time that occurs from the successive time cycles only if at least one of your immediate neighboring take locations was a member of such first group during the newly completed time cycle of such successive time cycles; any of the above-mentioned take-up locations that are not true of the said take-over locations L + 1 has a lower priority than being assigned a multiplier for use during the next cycle occurring from the said successive time cycles only if none from their immediate neighboring take-off locations were a member of such first or second groups during the newly completed cycle of such successive cycles of time; and any of the aforementioned take-up locations that are not true of the above-mentioned L + 1 tap locations that do not have a higher or lower priority take location priority may be assigned a multiplier for use during the next cycle that occurs from the aforementioned successive time cycles.
9. 9. The equalizer filter system defined in claim 8, wherein said predetermined priority rules comprise the following additional rules: first, the higher priority take locations are assigned multipliers in a given order of importance, in which any higher priority take location that is closer than another priority take location greater than said previously selected take location is more important than the aforementioned other higher priority take location, until either all the take-over locations are Higher priority have been assigned multipliers or all available multipliers have been assigned; and second, if there are still multipliers available after all of the higher priority take locations have been assigned multipliers, multipliers are assigned to the lower priority take locations in a predetermined order of importance, in which any location of lower priority take that is closer than another lower priority take location to that particular previously selected take location is more important than said other lower priority take location, until all the lower priority take locations are given priority. have assigned multipliers or all available multipliers have been assigned.
10. The equalizer filter system defined in claim 9, wherein: said particular previously selected tap location is the average tap location, L / 2, of said filtrp FI R (Figure 4, tap L / 2). eleven .
11. The equalizer filter system defined in claim 1, wherein: the ratio of the number of multipliers M to the number of tap locations L + 1 is not greater than 50%.
12. The equalizer filter system defined in claim 1, wherein said third means comprises an energy index monitor (202) including: sixth means for calculating the square of each of said individual load coefficients w associated with multipliers then assigned to certain picking locations of said picking locations L + 1; and seventh means for assigning a binary NO U energy index to each square individual w load coefficient having an absolute magnitude exceeding that given positive minimum value and assigning a binary ZERO energy index to each load coefficient square individual w that has an absolute magnitude that does not exceed in value said given positive minimum value; wherein said first group comprises each of the individual square load coefficients w to which an energy index of binary NO U has been assigned and said second group comprises each of the individual load coefficients w added to those that are they have been assigned an energy index of binary ZERO.
13. The equalizer filter system defined in claim 12, wherein said system comprises (A) a tap control logic unit (204) incorporating both fourth and fifth media and (B) eighth media (Tc (0) ) to supply such a take control logic unit with the identity of each of said previously selected take-up locations separate from said take-up locations L + 1 initially assigned by the second means, and wherein: said fourth means determine which of said multipliers will conserve their take-up location assignment during the next cycle that occurs from said successive time cycles according to said binary UNO energy indexes of said cycle. newly completed of the aforementioned successive time cycles; and such fifth means determine which of said multipliers are to be reassigned during the next cycle occurring from said successive cycles of time according to such binary ZERO energy indices of said newly completed cycle of such successive time cycles, determines the identity (Tc (n)) of the allocation locations assigned to multipliers reassigned during the second of said successive cycles of time according to the identity (Tc (0)) of each of the initially assigned tap locations of said locations of takes L + 1 during the first completed cycle of said successive time cycles, and, subsequently, determines the identity (Tc (n)) of the allocation locations assigned to multipliers reassigned during each subsequent cycle of said successive time cycles of according to the identity (Tc (n)) of each of the assigned take-up locations of said take-over locations L + 1 during the duty cycle mpo completed immediate previous of such successive time cycles.
14. 14. The equalizer filter system defined in claim 1, wherein each successive output sample y (n) is a binary number of multiple bits, and said first means comprise: an algebraic adder (306) having each successive output sample and (n) applied to a negative entry thereof; and, a separator (308) responsive to each successive output sample and (n) applied to an input thereof, to separate one or more of the less significant bits of that sample to thereby derive an output of such a separator comprising at least the most significant bit in each successive output sample and (n), such a separator output is sent to the positive input of such an algebraic adder; where the error term e (n) is derived at the output of said algebraic adder.
15. 15. The equalizer filter system defined in claim 14, wherein: said FI filter R is a direct IF filter R (302) of such adaptive filter.
16. 16. The equalizer filter system defined in claim 14, wherein: said FI filter R is a feedback filter FI R (304 of the adaptive filter 200a in Figure 3) of such adaptive filter, in which each output sample successive and (n) is applied to said first input thereof, wherein said adaptive filter operates as an I IR filter.
17. The equalizer filter system defined in claim 16, wherein said additionally adaptive filter includes a direct digital multi-tap FI filter (302) that responds to a stream of successive data samples x (n) 'having such a period of sample T applied to a first input thereof and said error term e (n) applied to a second input thereof to effect the operation of said direct FI filter R according to the LMS algorithm, and wherein said direct FIR filter it comprises a structure similar to that of the feedback filter FI R, and wherein said system additionally comprises: a second adder (304) having the successive output samples of said direct FIR filter applied to a first input thereof and the successive output samples of said feedback FIR filter applied to a second input to derive, as an output thereof, such successive output samples and (n) that apply to said first input of said feedback filter FIR and said negative input of said algebraic adder.
18. 18. The equalizer filter system defined in claim 14, wherein: said FIR filter is a feedback FIR filter (TLLS feedback filter of the adaptive filter 200b, not shown, similar to the LMS feedback filter FIR of FIG. 1) b) of said adaptive filter, in which each successive output sample of said separator is applied to said first input thereof, wherein said adaptive filter operates as a DFE filter. 9. The filter filter system defined in claim 18, wherein additionally adaptive filter includes a digital direct multiple-filter FI.R filter (direct FIR filter.
19. TLLS of the adaptive filter 200b, not shown, similar to the direct FIR filter LMS 102 of Figure 1 b) which responds to a stream of successive data samples x (n) 'having a sample period T applied to a first input of the same and said error term e (n) applied to a second input thereof to perform the operation of such a direct FI filter according to the LMS algorithm, and wherein said direct FIR filter comprises a structure similar to that of the filter FI R of feedback, and wherein such a system additionally comprises: a second adder (adder of adaptive filter 200b, not shown, similar to sudor 306 of Figure 2a) having the successive output samples of such a filter FI R applied directly to a first input of the same and the successive output samples of said filtering FI R of retroal imentación applied to a second input of the same to derive, as an entry thereof, said successive output samples and (n) that are applied to said input of said separator and to said entry of the algebraic adder tat.
20. 20. A method for assigning tap locations for a dispersed equalizer filter of sampled data having delay means with L + 1 taps and having M less than L multipliers for loading samples at selected tap locations, and circuits for combining respective loaded samples to provide a filter output signal, said method comprises the steps of: assigning such multipliers M to said predetermined tap locations M; apply predetermined multiplication coefficients to the respective multipliers of said multipliers; operate the dispersed equalizer filter of sampled data in a manner to produce updated multiplication coefficients that tend to produce convergence of the dispersed equalizer scatter of sampled data; examine the resulting multiplication coefficients according to a predetermined criterion; exclude intake assignments associated with multiplication coefficients that do not meet this criterion; and assigning new shooting locations according to an algorithm that polarizes the selection of a different shot depending on its previous selection and the current selection of close shots. twenty-one .
21. A method according to claim 20, comprising the additional step of operating dispersed equalizer filter of sampled data with new acquisition locations in a manner to produce updated multiplication coefficients that tend to produce convergence of the dispersed equalizer filter of sampled data. .
22. 22. A method according to claim 20, comprising the additional step of assigning a multiplication coefficient that meets said criterion to a predetermined tap location corresponding to a television signal received from a terrestrial communications channel.
23. 23. A method according to claim 22, wherein said predetermined tap location is the middle tap location of such a filter.

    RESU MEN A dispersed digital adaptive equalizer that includes filters of Direct Finite Impulse (FI R) (302) and feedback (304) filters exhibits improved operation of Minimum Mean Square. A switch (404) assigns one of several multipliers (414-1 to 414-5) of each filter to each of the tap locations of that filter. Each multiplied is initially assigned to a preselected take location with a predetermined load coefficient. After completing successive cycles of time having a duration equal to a given number of data sample periods, for each filter a first group of coefficients associated with multipliers then assigned to non-zero value-taking locations is determined. , and (b) a second group of coefficients associated with multipliers then assigned to locations of zero value. The multipliers associated with the coefficients of the first group during a newly completed time cycle retain their take-over locations during the next cycle of time. The multipliers associated with the second group of coefficients during a newly completed time cycle are reassigned to new take-up locations for use during the next time cycle.