TW200404222A - Audio channel spatial translation - Google Patents

Abstract

Using an M:N variable matrix, M audio input signals, each associated with a direction, are translated to N audio output signals, each associated with a direction, wherein N is larger than M, M is two or more and N is a positive integer equal to three or more. The variable matrix is controlled in response to measures of: (1) the relative levels of the input signals, and (2) the cross-correlation of the input signals so that a soundfield generated by the output signals has a compact sound image in the direction of the spatial center of gravity of the input signals when the input signals are highly correlated, the image spreading from compact to broad as the correlation decreases and progressively splitting into multiple compact sound images, each in a direction associated with an input signal, as the correlation continues to decrease to highly uncorrelated.

Description

200404222 发明. Description of the invention: C. The technical field to which the invention belongs; j. The invention relates to audio signal processing. More specifically, the present invention relates to the conversion of M audio input channels representing a sound field into 1 ^ audio output channels representing the same 5 sound fields, where each channel is a single audio representing audio arriving from one direction Flow, M and N are positive full integers, M is at least 2 and N is at least 3, and N is greater than M. Typically, a spatial converter with N greater than M is called a "decoder". [Summary of the Invention 3 10 Summary of the Invention According to a first level of the present invention, a process is used to convert each of the M audio input signals provided with a direction into each of the N audio input and output signals provided with a direction, Where ν is greater than μ, M is a positive integer greater than 2 and N is a positive integer greater than 3, including providing a variable matrix of M × N, applying the 15 or so M audio input signals to the variable matrix, and the variable matrix Derive the N audio output signals, and control the variable matrix in response to the input signals, so that when the input signals are highly correlated, the sound field generated by the output signals has One of the dense sound images in the direction of the center of gravity of the space, the image is broadened by the tight diffusion 20 as the correlation decreases, and gradually divided into multiple tight sound images as the correlation continues to decrease to a highly uncorrelated, Each one is in one of the directions that is assigned an input signal. According to this first level of the present invention, a process is used to convert each of the M audio input signals assigned with 5 directions to each N audio input and output signals assigned with a direction, where ν is greater than μ, M Is 2 or more and ν is a positive integer of 3 or more. In this case, in order to have a cross correlation of the input signals within a first range bounded by a maximum value and a reference value, when the cross correlation is the maximum value, the sound field has a Tight sound images, and when the cross-correlation is the reference value, the sound field has a widely diffused image; and to have the intersection of the input signals in a second range bounded by a minimum value and a reference value Correlation measurement, when the cross-correlation is the minimum value, the sound field has several tight sound images, each in the direction of an input signal, and when the cross-correlation is the reference value, the sound The field has a wide spread image. According to a further aspect of the present invention, a process is used to convert each of the M audio input signals provided with a direction into each of the N audio input and output signals provided with a direction, where N is greater than M, and the percentage is 2 The above and N are positive integers greater than or equal to 3, including: providing several variable matrices of mxn, where m is a partial set of M and n is a partial set of N, applying each of these M audio input 仏 numbers A partial set, from which each of the variable matrices derives a respective partial set of the N signal rotation signals, and controls the partial set in response to a partial set of input signals applied to each of the variable matrices. Each of these variable matrices is such that when the input signals are highly correlated, the sound field produced by the outputs has a tight sound image in the direction of the spatial center of gravity of the input signals. As the correlation declines from tightly diffused to broad, and gradually cuts into multiple tight sound images as the correlation continues to decrease to a high degree of uncorrelation, each in one direction with an input signal, and Audio output channel portion of the other set of deriving the N audio output signals. According to this further aspect of the invention, the variable matrices can also be controlled in response to information in order to compensate for the effects of receiving one or more other variable matrices of the same input signal. Furthermore, deriving these N audio output signals from a partial set of N audio output channels can also compensate for multiple variable matrices that produce the same output signal. According to this further aspect of the invention, the variable matrix is controlled in response to (1) phase alignment of the input signals and (2) cross-correlation measurements of the input signals. According to a further aspect of the present invention, a process is used to convert each of the M audio input signals assigned with a direction into each of the N audio input and output signals assigned with a direction, where N is greater than M, M Is 2 or more and N is a positive integer greater than or equal to 3, including: providing a M × N variable matrix in response to a control matrix coefficient or a conversion factor of a control matrix output, applying the M audio input signals to the variable matrix, providing a number Mxn variable matrix conversion factor generators, where m is a partial set of M and 11 is a partial set of > 1, applying a separate set of these audio input signals to each of these variable matrix conversion factors A generator that derives a set of variable matrix conversion factors for each of the sets of N audio output signals from each of the variable matrix conversion factor generators, in response to being applied to each of the variable matrices Each of the variable matrix conversion factor generators is controlled under a partial set of input signals of the conversion factor generator, so that when the conversion factor generated by it is applied to a variable of 5 MH × χ At the time of the array, the sound field generated by the respective sets of parts of the output signal has a tight sound image in the direction of the center of gravity of the parts of the set of parts of the input signal to which the conversion factor is applied when such input signals are highly correlated. The image is gradually diffused from broad to broad as the correlation decreases, and progressively segmented into multiple tight sound images as the correlation continues to decrease to a highly uncorrelated, each of which is being formulated to produce the applied In one direction of an input signal of the scaling factor, N audio output signals are derived from the variable matrix. According to this further aspect of the invention, the variable matrix conversion factor generators can also be controlled in response to information in order to compensate for the effects of receiving one or more other variable matrix conversion factor generators. Furthermore, deriving these N audio output signals from a partial set of N audio output channels can also compensate for multiple variable matrix conversion factor generators that produce the same output signal. According to this further aspect of the invention, the variable matrix conversion factor generator is controlled in response to measurements of the phase alignment of the input signals and (2) the cross-correlation of the input signals. The invention and its various aspects can be implemented in an analog circuit, or more likely as a software function in a digital signal processor, a programmed general purpose digital computer, and / or a special purpose digital computer. The interface between analog and digital signal flow can be implemented by appropriate hardware and / or software functions and / or Weibo. This invention and its various aspects may involve analog or digital signals. In practical applications, most or all processing functions may be implemented in the digital domain on the digital signal stream, in which audio samples are presented. According to the present invention, the -audio converter or conversion function has an input of the M channel and an output of the N channel. Each channel has a matching direction (such as position and rising). Its goal is to generate the output of 200404222 from the M round signals. Although this can be done passively by deriving the M × N coefficient matrix from the spatial information, this configuration has poor channel isolation and cannot preserve power when the input signals are matched with correlation. The signal conversion according to the present invention can be applied to a wideband signal or each frequency band up to a 5-band processor, and is implemented once in each sample or in each block according to the operation. A multi-band embodiment may use a filter bank such as one of the discrete critical band filter banks, or a linear filter bank such as FFT (Fast Fourier Transform) or MDCT (Modified Discrete Cosine Transform) with related decoders. One of the frequency band structures of the capacitive filter bank. 10 In general, it is not necessary to check all possible combinations of signal commonality between these input channels. In a flat channel array (such as a channel that is arranged horizontally in the array direction), it is often appropriate to perform a pairwise similarity comparison of adjacent channels in space. For channels with canopy or spherical configuration, signal commonality can be extended to more than three channels. The use of signal commonality and 15 detection can also be used with additional signal information. For example, a vertical signal component can be presented by mapping to all five full-range channels of a horizontal five-channel array. Related to determining a combination of input channels with a preset input / output mapping matrix for commonality analysis, only 20 of each input / output channel converter or converter configuration is required in assembling the converter or converter function. once. The "initial image" (before processing) derives a passive "master" matrix that correlates the input / output channel grouping with the spatial orientation of the channels. As an alternative, the processor or processing section of the present invention can generate a time-varying conversion factor for each output channel, and the modification has instead been a simple output signal level cut by a 9-matrix or material coefficient. These conversion factors are again derived from (a) dominant, (i) uniformly dispersed (filled) and (c) residual (endpoint) signal components as described below. According to one aspect of the present invention, the converter or converter function of the present invention may be a converter module or a converter module function (hereinafter referred to as a converter module function) that connects input channels to each other and performs similarity analysis and directional mapping. "Module") is conceptualized. Since an input channel can involve several different similarity comparisons, some compensation for possible module interactions may be necessary. These interactions may depend on whether the two modules have the same or different number of inputs, so that these modules are arranged at a conceptually hierarchical level based on the number of inputs each has. There are two types of module interactions .... Modules that are at a common or lower level (that is, modules with the same or fewer inputs) are called "neighbors"; High-level (with more inputs) modules that share more than one common input are called "HO neighbors." Further details of the module interaction are set up below. The module interaction uses the common energy (average cross product of these signals) as a measure of similarity (rather than using cross correlation), which is better because cross correlation requires normalizing cross energy with individual energies. It will be affected by adjacent modules, but the common energy is generally not changed by adjacent modules when it is assumed that it processes independent signals. However, as explained further below, using smoothed levels to determine the spread of signal components between output channels requires module interaction compensation to reduce errors in the output of these modules. 200404222 Because any group of input channels is common to any set of these channels, the difference between this signal and other signals that are common to only some sets requires that the common signal component level must be The order of the reduced number of channels is evaluated and subtracted from the collective signal count of some collective channels. Some quantities are interdependent and difficult to solve simultaneously. In particular, each common input level (defined under the heading "Interaction Compensation" below) depends on the common input of the same input and the common input of each adjacent module in the class level . This calculation is most easily done using the previous value of the common 10 input levels for each adjacent module. The potential interdependence of common input levels of modules at different levels of the hierarchy can be addressed by using previous values as described above, or by performing calculations in a repeating order (ie, loops) from the highest level to the lowest level. Alternatively, simultaneous equation solutions are possible, although they may involve non-trivial costs. 15 The channel converter function according to the present invention may generally operate as follows: 1. Calculate the common signal energy for each module; 2. Calculate the effective input for each module after considering any interaction with other modules Input energy level; 3. Calculate the common (dominant) signal component and the spatial direction of the center of gravity of the effective input 20 energy level for each module; 4. Calculate the dominant signal component output conversion factor; 5. Calculate The non-dominant signal component output conversion factor (spread); and 6. Calculate the residual input signal level and image residual input signal to the output channel after subtracting the dominant and non-dominant signal components. 11 200404222 Interaction compensation As mentioned above, channel switching according to one aspect of the present invention can be considered as involving a "module" of a grid. Since multiple modules will share a certain input channel, interactions between modules are possible, and unless some compensation is applied, _ 5 will be inefficient. Although it is generally not possible to separate and separate signals based on the input to be made by the module, estimating the amount of turn-in signals used by each connected module can improve the results of correlation and direction estimation and lead to overall performance. Improvement results. As mentioned above, there are two types of module interactions: Modules that involve common or lower 10 levels (that is, modules with similar or fewer inputs) are called "neighbors", And modules that have a higher level than a certain module (have more inputs) but share more than one common input are called "high-order neighbors". Consider first neighbor compensation at a common level. In order to understand the problems caused by the interaction of the neighbors, consider two isolated input modules with one of the same L / R (left and right) input signals A isolated. This corresponds to a single dominant (common) signal of 0 midway between the inputs. Its common energy is A2 and its correlation is 1.0. Suppose a second two-input module has a common signal B2 at its L / R input and the correlation is also 1_0. If the two modules are connected at a common input, the signal of input 20 here will be A + B. Assuming that signals A and B are independent, the average product and product of AB will be 0, so the common energy of the first module will be A (A + B) = A2 + AB = A2, and the second module The common energy will be B (A + B) = B2 + AB = B2. Therefore, as long as the neighboring modules process independent signals, their common energy will not be affected. This is generally a valid assumption. If the signals such as 12 200404222 are not independent and are the same, or at least substantially share a common signal component, the system will respond to the human ear in a consistent manner, that is, their common turn will be large, causing the resulting audio signal to be oriented The common signal is pulled. In this case, because the common input has 5 more signal oscillations (A + B) than any outer input, the direction estimate is biased toward the common input, so that the L / R ratio of each mode is biased. . In this case, the correlation value of the two modules is now slightly less than 1 · 0 due to the different waveforms of the input and the two. Since this correlation value determines the spread of non-common signal components to the dominant (common signal component) to non-dominant (non-common signal component) energy ratio, the uncompensated common input signal 10 causes the The common signal distribution will spread. See the “Signal Distribution” section below for further details. 15 20

In order to compensate, the total energy of the internal output of the module that can be attributed to each input of the Lai group, that is, the "common input lexicons of each input of each module is estimated, and then each module uses the relevant At the same level of each module's input, the total amount of energy of the common input level with adjacent levels is notified. The analysis of the input money of the module will not allow the entry level of each _ input to each ... It should be directly solved, but should be the average of these common input power levels: What is the average. 共同 The common input power level of each input cannot be overshoot measured and the total power level of this input is known, The overall common energy is, = of: can be decomposed by factors into the estimated common-rounder level of the observed turn-in quasi-coffee.-The combined effect of the common-round renunciation level; all modules of the grid It is calculated that every common input level of adjacent modules ... In the input group of the mother-per-input, the "neighbor c knows" that it is called "at, Feng". Then the module loses> 13 404 222 5 10

The compensated wheel level is derived by subtracting the neighbor level from the input level, straight through to calculate the correlation and the direction (the spatial center of gravity of the input signals) /,, as for the example cited above In other words, the neighbor level starts at zero, so because the common input has more signals than the end-end input, the first module states that the input has an input power level exceeding A2, and The first module states that the input has an input power level exceeding B2. Due to the total available energy level of Mingda 2 here, these statements are limited to = and B, respectively. Since no other modules are connected to the common input, each common input level corresponds to the neighbor level of the other modules. The consequence is that the input power level after compensation as seen by the first module is: (Α2 + Β2) -B2 = Α2 and the input power level after compensation as seen by the second module is: (α2 + β2 ) -α2 = β2

However, these are only levels that will be observed when the module is isolated. As a consequence, the correlation value of the result will be 1.0 and the dominant center will be centered at the appropriate amplitude as desired. However, the recovered signal itself will not be completely isolated-the output of the first group will have some B signal components, and vice versa di i_ this is a limitation of the array system, and if the processing is on a multi-band basis The signal components after being 'mixed' will be of similar frequency, providing resolution with room for discussion during the 20th. In more complicated cases, the compensation is usually not so precise.> The experience of this system is that practical compensation will mitigate most of the effects of the neighbor module interaction. Based on the principles and signals used in neighbor level compensation, it is fairly straightforward to extend to higher-order neighbor level compensation. This application is applied to more than one 14 5 ίο is the case where the modules of the same k-level share more than one common input channel3. For example, it may have a three-input module and a two-input module. The common signal component for all three inputs will also be common to one of the two input groups and will be provided by each module at the same position without compensation. More generally, it may have—the signal component 榼 7 has three inputs in common, and—the second component is only for those two inputs, and the inputs are common and need to have as much influence as possible Many are separated so that Θ 仏 properly outputs the sound field. The consequence is that the common-signal effect of the two inputs, such as the common input level described above, must be subtracted by those inputs before the two-input calculations can be properly implemented. In fact, when performing low-order calculations, these high-order common signal elements should not only be reduced by the input level T of the low-order module from the observed common energy level. This is different from the common input level of modules in the same m-level that do not affect the common energy levels of neighbors and enemies. Therefore, these higher order neighbor levels should be considered and applied separately from the same order neighbor levels. While the high-level neighbor level is transmitted down to the lower module in the Hierarchy, the remaining common levels of the low-level module should also be transmitted in the upper layer. The reason is that as described above, the low-level module system It acts like a neighbor of a higher-order module. In order to avoid resource sensitivity and complex simultaneous solving, the value calculated before S can be transmitted to the relevant

Kernel Group of 20 Although the interaction compensation techniques described above only produce approximately correct values for complex signal assignments, they are believed to provide improvements in lattice configurations that cannot consider modulo-interaction. 15 200404222 As mentioned above, cross-correlation measurement will determine the ratio of dominant (common signal component) to non-dominant (non-common signal component) energy in a module and these non-dominant signals Distribution of components between the output channels of the module-5 degrees. This can be better understood by considering the signal assignment of the two input modules under different signal conditions to the output channels in one module. Unless stated, the principles established extend directly to higher-level modules. The problem with signal distribution is that the information for recovering the amplitude distribution of the original signal is too much less than the signal itself. The basic information available is the average cross product of the signal levels of the 10 inputs in each module and these input signals (referred to above as the common energy level). The intersection of the zero time offsets is related to the ratio of the geometric average of the common energy level to the energy levels of the input signals. The importance of cross-correlation lies in its net amplitude, which is the common belief of all inputs. If there is a single signal 15 (-"internal" or "intermediate" signal) anywhere between the module's inputs, all inputs will have the same waveform (although there may be different amplitudes). In this case, The correlation will be I · 0. In the other extreme case, if all input signals are independent, it means that there is a common signal component in Lu /, and the correlation will be 0. Correlation values between 0 and 10 can be regarded as corresponding to some single-, common signal components in these losers, the middle balance level of 20 independent signal components. Consequences of any input signal; Brother 7 is divided into a common signal, the "dominant" signal and the "remaining" signal component ("non-dominant" or residual signal) that is left after subtracting the common signal cause Energy) of the input signal component. As noted, the common "j sex" signal amplitude may not be greater than the "nondominant" or residual signal bits = 16 200404222

For example, consider being mapped to five channels (1, MidL, C,

MidR, R) arc case. If all five channels have the same # 5 standalone # number, the amplitudes of Lt and Rt will be equal, with an intermediate value of the common energy corresponding to the intermediate value of the cross correlation between 0 and 1 (the reason is that ^^ and Seems to be an independent signal). The same level can be achieved with the appropriate selection of L, c, R levels without k number from MidL and MidR. Therefore, a two-input and five-output module can feed only the output channel corresponding to the 10 dominant direction (C in this case) and the input signal residual (L 'after removing the c energy from the Lt and Rt inputs. R) output channels without giving a signal to this and tear-shaped output channels. This result is undesirably an option that unnecessarily cuts off the channel almost forever, because the small confusion in the signal condition can cause the "off" channel to jump between on and off, causing annoying electrical noise (electricity) The trembling noise is quickly turned on and off by channel 15), especially when the "off" channel is being listened to in isolation. The consequence is that when the input signal value of a certain group of modules may have multiple output signal allocations, the conservative approach from the quality perspective of each channel is to spread these non-dominant as even as possible among the channels of the module's rotation. The signal components are consistent with 20 of these signal conditions. One aspect of the present invention is to uniformly disperse the available signal energy based on the three-way segmentation rather than the "dominant" two-way segmentation of "all remaining" under these signal conditions. Preferably, the three-way segmentation includes a dominant (common) signal component, a filling (uniformly dispersed) signal component, and a residual L 旒 component. Unfortunately, there is only enough information to perform a double 17-way segmentation (energy is missing from the ^ component and all other signal components). It is desirable to realize the three-way segmentation of the Y 雔 text, where the correlation value higher than a specific value 'qianxiang segmentation uses the non-difficult signal component of money and dispersal; for the correlation below this value, the two-way segmentation uses dissemination. The non-dominant signal component and the energy of ^ Λ /, the same number are between "dominant" and "uniform spread". The "uniform distribution of people" component includes one of "common" and "residual". Therefore, "dispersion" involves a mixture of common (correlated) and residual (non-correlated) signal components. Before processing 'For a specific input / output combination of a module, a correlation value is calculated corresponding to all output channels receiving the same processing amplitude. This correlation value can be referred to as the "randam-xcor" value. For a single, centrally derived intermediate output channel and two input channels, the randam_xcor can be calculated as 0.333. For three equally spaced intermediate channels and two input channels, the randam-xcor can be calculated as 0.483. Although these time values have been found to provide satisfactory results, they are not critical. For example, values of about 0.3 and 0.5, respectively, are available. In other words, for a module with μ input and N output, it has a certain degree of correlation with M input, which can be regarded as representing equal energy at all N outputs. This can be achieved by considering the μ inputs as if they had been derived using a passive N-to-M matrix that received N independent signals of equal energy, although the real inputs could of course be derived using other methods. The relevant value of this threshold is “randam_xcor” and can represent one of the dividing lines between the two operation areas. Then during processing, if the input signal of a module is greater than or equal to randam-xcor, it is set to a range of 1.0 to 0. 200404222 scaled-xcor = (correlated -randam-xcor) / (l-randam-xcor ) The "scaled-xcor" value represents the amount of dominant signal above the uniformly dispersed level. No matter what is left behind, it can be equally assigned to the other output channels of the module. — 5 However, there are additional factors that should be considered, that is, as the center of gravity of the input signal gradually moves away from the center, if the equal distribution and allocation of all output channels is maintained, the amount of dispersed energy is gradually reduced. , Or alternatively, the amount of scattered energy should be maintained, but the energy allocated to the output channel should be reduced relative to the "off-center" of the apparent energy quantity. In other words, the energy decreases obliquely along the output channel along _10. In the latter case, additional processing complexity would be required to maintain the output power equal to the input power. On the other hand, if the current correlation value is less than the randam_xcor value, the apparent energy is regarded as 0, the uniformly dispersed energy is gradually reduced, and no matter what residual signal is left to be allowed to accumulate at such inputs. At correlation = 0, 15 it has no internal signal, only independent input signal is directly mapped to the output channel. The operation at this level of the present invention can be further explained as follows: Lu (a) When the true correlation is greater than randam_xc0r, this is considered to have enough common moons to be-the dominant signal is manipulated between two adjacent outputs ( (Determined) (or, '20 occurs in a direction consistent with an output, of course being fed to an output "is accused of-the offender is subtracted from these inputs to give distribution across all outputs (better (The ground is uniform). The true correlation is exactly randam__xcor, and the input energy ,,,, and all the residuals are evenly distributed among all the outputs (this is the definition of 19 randam-Xcor). (C) When the real relevant small dam—the clutter does not have enough common energy ', so the energy of this round of people will be matched by the ratio of how much is reduced between rounds. This seems like someone has processed the relevant part as the Residues that are evenly distributed among all outputs; and if not, will be sent to the uncorrelated parts of several explicit signals corresponding to the outputs of these inputs. In the extreme form where the correlation is 0, each-input People are only fed by-output location (-general One of the outputs, but it can be the position between the two). Therefore, in the complete correlation, the entire randam xcor has inputs that are evenly distributed among all the outputs, and a single energy appears according to the relative energy of those inputs. There is a continuous closed set between the two outputs and the zero correlations that are independently fed to the M output position. Pairwise calculation of common energy For example, suppose the input conversion pair A / B contains uncorrelated signals along the respective The common signal X between Υ and Z is: Α = 0.707χ + Υ Β = 〇 · 707Χ + ZZ where the conversion factor 0.707 = λ / 5Ι provides a power reserve image for the nearest input channel. RMSEnergy (A) = j * A2 at = Α2 = (· 7〇7Χ + γ) 2 = (0.5X2 + 0.707XY + Y2) = 0.5X2 + 0.707 χΫ + Ψ Since X is not related to γ, χΫ = 〇200404222 Α2 = 0 · 5 × 2 + Y2 That is, because X and Υ are uncorrelated, the total energy in the input channel A is the sum of the energy of the signals X and Y. Similarly: 5 ^ = 0.5 ^ + Z7 Since X and Y are not related, A and The average cross product of B is ^ = 0.5 ^. Therefore, an output signal can also contain independent, uncorrelated signals. In the case where adjacent input channels are equally shared, the average cross product of the signals 10 is equal to the energy of the common signal component of each channel. If the common signals are not equally shared, that is, appearing towards one of the inputs, the The average cross product will be the geometric mean between the energies of the common components in A and B. The estimates of the common energies of the individual channels can be derived by normalizing the square roots of the amplitude ratios of these channels. The true time average is calculated using a leaky integrator (first-order low-pass filter) with a suitable decay time constant of 15 to reflect ongoing activity. This time constant smoothing is done as best as possible with the non-linear attack and decay time options, and frequency scaling is available in a multiband system. High-order calculation of common energy 20 In order to derive the common energy with more than three input decoding modules, it is necessary to form the average cross product of all input signals. Input-only pairing cannot distinguish between each pair of inputs and a separate output signal for all signals that are common to each other. For example, consider the three input channels 21 200404222 consisting of uncorrelated signals W, Y, Z and common signal X:

A = X + WB = X + YC = X + Z 5 If the average cross product is calculated, all terms involving the combination of W, Y, and Z are eliminated as in the second-order calculation, leaving the average of X3: ABC = Unfortunately, if X is a zero-averaged time signal as expected, its cubic mean is zero. Unlike the average x2 (this is positive 10 for any non-zero value), X3 has the same sign as X, so positive and negative causes will tend to cancel out. Obviously, the same holds for X of odd powers, corresponding to the module input of odd numbers, but even exponents greater than 2 will also lead to erroneous results; The four inputs will have the same product / average as (X, X, X, X). 15 This problem can be solved by using the deformation of the average product technique. Before being averaged, the sign of each product is discarded by taking the absolute value of the product. The sign of each term of the product is checked. If they are all the same, the absolute value of the product is applied to the averager. If any sign is different from others, the negative value of the absolute value of the product is averaged. Since the number of possible combinations of the same symbols may not be the same as the number of possible combinations of different symbols, a weighting factor containing the ratio of the same pair of different symbol combinations is applied to the absolute value product with a negative sign to compensate. For example, a three-input module has two methods out of eight possibilities to make them have the same sign, and the other six methods have different signs. As a result, the conversion factor is 2/6 = 1/3. If and only if it has a signal component to a solution 22 200404222 All inputs of the code module are the same, this compensation will cause the integral or total product to grow in a positive direction. However, in order for the averages of modules of different orders to be comparable, they must all have the same dimensions. A customary second-order correlation involves the average of the two-input multiplication 5 and thus has a quantity related to the dimension of energy or power. Therefore, the terms to be averaged at higher-order correlations must be modified to have the power dimension. For the k-th order correlation, the absolute value of each product must be raised to a power of 2 / k before being averaged. Brief Description of the Drawings 10 Figure 1 is a plan view schematically showing the use of a 16-channel horizontal array on a wall around a room, a 6-channel array placed in a circle above the horizontal array, and a single suspended channel One of the test arrangements is an ideal decoding configuration. Figure 2 is a functional block diagram that provides a multi-band conversion embodiment of several modules operated by the central 15 supervisor implementing the example of Figure 1. Figure 3 is a functional block diagram useful in understanding the method by which the supervisor 201 of Figure 2 can determine an endpoint conversion factor. Figures 4A-4C show a functional block diagram of a module according to one aspect of the present invention. 20 Figure 5 is a schematic diagram showing a hypothetical configuration of a triangle made of input channels, three internal output channels, and a three-input module fed in a dominant direction. Figures 6A and 6B are functional block diagrams respectively showing a suitable configuration for (1) generating a total estimated energy for each input of a module in response to the total energy at each input, and (2) An additional endpoint energy conversion factor component is generated for the endpoints of each module in response to the cross-correlation measurements of the input signals. Fig. 7 is a functional block diagram showing a preferred function of the "summing and / or taking the 5 largest" block 367 of Fig. 4c. FIG. 8 is an idealized diagram of a method for generating a conversion factor component under a measurement in response to a cross correlation in one aspect of the present invention. Figures 9A and 9B to 16A and 16B show the output conversion factors of a module caused by the results of various examples of input signal conditions. 10 [Embodiment] Detailed description of the preferred embodiment In order to test the various aspects of the present invention, a configuration is developed that has a horizontal array of 5 loudspeakers on each wall of a four-walled room (each corner 1 loudspeaker, with 3 evenly spaced between each corner), a total of 16 15 loudspeakers, forming a common corner loudspeaker, plus about 45 in the central listener. Above the vertical angle-6 loudspeakers in the ring, and a single loudspeaker directly above, reaching 23 loudspeakers, plus a subwoofer (SUbw0fer) / LFE (low frequency effect) channel 'total Hunan loudspeakers were all fed with a personal computer set up for 24 channel playback. Although the current

Pardonably, this system may be referred to as a 23-channel system, but it will be referred to herein as a 24-channel system for simplicity. S Figure 1 is a -plan view 'which schematically shows an ideal decoding configuration in the manner of the test configuration just described. The five wide-area horizontal turn-in channels are displayed as four orthodox circles on a peripheral circle called the morpho 1, 3, 5, 5, 9, and 24. Four wide-area turners such as 4 pass through the butterfly. Or the generated echoes are supplied from the ground (as shown in Figure 2), or the sword straight ship is cut into the center of the dashed line = 23. The 23 wide-area rotation channels are displayed as circles filled with 填 numbers. The sound circle of the 16 output channels is in the -horizontal plane, and the inner circle of the 6 output channels is 45. , _ Λ45 is above the horizontal plane. The output channel 23 is directly above the one or more listeners. The output is decoded by using parentheses Γ and the layer circle material. The five-side, two-input, and two-mode analogue signals are aligned with 29 · 33 to connect the channel to each horizontal input. The center rear channel (output channel 21) of the horse being lifted is shown by an arrow between the output channel 21 and the input channels 9 '13 and 23-a three-input decoding module 34. Therefore, the two-input decoding module 34 is one level higher than its lower-level neighbor modules 27, 32, and 33. In this example, each module is paired with each pair or three spatially closest adjacent input channels. 15 per module

In this example there are at least three neighbors of the same rank. For example, modules 25, 28, and 29 are neighbors of module 24.

Although the decoding module presented in Figure 1 has three, four, or five output channels in various ways, a decoding module may have any reasonable number of output channels. An output channel can be located in the middle of two or three input channels or the same position as an input channel. Therefore, in the example of Fig. 1, the bit 20 of each input channel is also an output channel. Two or three solution modules share each input channel. Although the configuration in Figure 1 uses five modules (each with two inputs) and five inputs (1, 3, 5, 9, 9, and 13) to derive 16 horizontal outputs that represent positions around the room (Μ6), similar results are available with a minimum of three inputs and three fork groups (each with two inputs, each module sharing one input with another module 25 200404222 input). Turn the channels into an arc or a straight line (as shown in the example in Figures 1 and 2). The multi-modules encountered by the calcite horses of the known art

The decoding ambiguity can be avoided, where ㈣ less than 0 is decoded into the backward direction of the table ^ 5. X7F Although the input and output channels can be characterized by their physical location or at least in their direction, it is useful to characterize them with a matrix because it provides a well-defined signal relationship. Each—matrix element (column 丨, the first channel i and the output channel j are associated with one of the conversion functions. Matrix element pass = 10 is a signed multiplication coefficient, but can also include phase or delay terms (mainly any Filter), and can be a function of frequency (in discrete frequency terms, every 2 frequencies—different matrix). This is straightforward in the case of a dynamic scaling factor applied to the output of a fixed matrix, but it also borrows It is adapted to form a variable by having a separate conversion factor of 1 for each matrix element or a matrix that works harder than a simple scalar conversion factor of 15 elements, where the matrix elements themselves are variable (such as a variable delay). Matrix. &Amp; There is some flexibility in the mapping entity position to the matrix element, basically this embodiment of the present invention can handle mapping_input channel to any number of 20: out channel 'and vice versa' but The most common case is to assume that the signal is only mapped to the nearest round-out channel via the pure Kig sub of Jan =, which is to save work = square sum is L0. This mapping is often via a sine / cosine steering function元 成 For example, When the two input channels and the three internal output channels are always in the middle, plus the second output terminal channels are combined with these input positions (that is, 26—MxN modules, where M is 2 and N is 5), We can assume that the span represents an arc of 90 ° (the range of the sine or cosine from 0 to 1 and vice versa), so that the parent channel is 90/4 interval = 22.5., And we get (cos (angle) , Sin (angle)) conversion matrix coefficients:

Lout coefficient = cos (0), sin (〇) = (l, 〇)

MidLout coefficient = cos (22.5), sin (22.5) = (0.92, 0.38), Cout coefficient = cos (45), sin (45) = (0.7, 0.71) MidRout coefficient = cos (67.5), sin ( 67_5) = (0.38, 〇 · 92) Rout coefficient = cos (90), sin (90) = (0,1) Therefore, it has a fixed coefficient and a variable controlled by a conversion factor at each matrix output In the case of gain, the signal output in each of the five output channels is:

Lout = Lt (SF ()

MidLout = ((0.92) Lt + (0.38) Rt)) (SFMidt)

Cout = ((〇.45) Lt + (0.45) Rt)) (SFc)

MidRout = ((〇.38) Lt + (0.92) Lt)) (SFMidR)

Rout = Rt (SFK) (The above "SF" is the conversion factor for the specific output using the subscript.) Generally, given an array of input channels, we can conceptually connect the nearest input with a straight line (which is " "Potential" because if no output channel needs to be exported by a module, the module is not required). In a typical configuration, any output channel on the straight line between the two input channels can be exported by a two-input module (if the source and the transmission channel are in the same plane, any source appears on the two input channels at most. In this case, it is not good to use 200404222 with more than two turns.) An output channel at the same position as an input channel is an endpoint channel and may belong to more than one module. An output channel that is not on the same line or at the same position as an input (such as the inside or outside of a polygon formed by two input channels) needs to have one of two or more modules. — A decoding module with two or more inputs is useful when a common signal occupies more than two input channels. This can occur, for example, when the source channels are not on the same plane as the input channels: one source channel can be mapped to more than two input channels. In the example in Figure 1, when 24 channels # 10 (16 channels on the horizontal ring, 6 channels on the elevated ring, i vertical channels plus LFE) are mapped to channel 6.1 (including A composite vertical channel). In this case, the center rear channel in the ascending circle is not on a direct straight line between the two source channels, it is in the middle of the triangle formed by Ls (13), RS (9) and the top (23) channel, so A three-input module is required to extract it. One way to map a 15-channel raised channel to a horizontal array is to map it to two or more input channels. This allows the 24 channels in the example in Figure 1 to be mapped into a conventional 5.1 channel array. In this alternative, several three-round input modules can extract the elevated channels, and the remaining signal component can be processed by the two-input module to extract the main horizontal ring channel. _ 2 From the position or direction of the image to the conversion factor, we can obtain a "master" matrix for a certain configuration of the input and output channels. This main matrix is useful in assembling a module configuration as shown in the example in Figure 1, and is further described in relation to Figure 2 below. By examining the main matrix, we can, for example, summarize how many decoding modules are needed, how they are connected, each with more than 28 200404222 fewer input and output channels, and how these matrix coefficients combine the input and output of each module. The output is correlated. These coefficients can be obtained from the main matrix; unless an input channel is also an output channel (ie, an endpoint), only non-zero values are required. 5 Each module preferably has a "local" matrix, which is part of the main matrix applicable to that particular module. In the case of a multiple module configuration (such as the examples in Figures 1 and 2), the module generates a conversion factor (or matrix coefficient) for controlling the main matrix as described in the relevant Figures 2 and 4A-4C below. ), Or for the purpose of generating a set of output signal parts that process the combined output signal centrally using one of the supervisors as described in the related Figure 2. The local matrix can be used. In the latter case, this supervisor compensation has a common output signal, similar to the one in which the supervisor 201 of FIG. 2 decides a final conversion factor to replace the initial conversion factor that is generated for the same output channel as the initial conversion factor. 15 versions of the same output signal generated in mode. In the case of multiple modules that generate scaling factors instead of output signals, such modules can continuously obtain matrix information about themselves from a main matrix via a supervisor instead of having a local matrix. However, if the module has its own local matrix, less computational cost is required. In the case of a single group of 20 independent modules, the module has a local matrix, which is only the required matrix (in fact, the local matrix is the main matrix), and the local matrix is used to generate the output signal. Unless specified, embodiments of the present invention with multiple modules are described as an alternative approach in which modules generate conversion factors. 29 200404222 Any decoding module output channel with only one non-zero coefficient in the local matrix of the module (this coefficient is 丨 .0, because the sum of the squares of these coefficients is 丨. ⑴ is an endpoint channel. It has more than one Non-coefficient output channels are internal output channels. Consider a simple example. If the rotation channels 〇1 and 〇2 are both derived from the input channels η and 12 (but have different coefficient values), then we need to produce output. A 2-input module that is connected between II and 12 of 1 and 02 (possibly among others). In more complicated cases, if there are 5 inputs and 16 outputs, and _ one of these decoding modules Has inputs: ^ and] ^ and feeds outputs 〇1 and 〇2, so that: _ 1〇01 = All + BI2 + 013 + 014 + 015 (note: input channels 13, 14, or 15 did not contribute), and 02 = Cl 1 + DI2 + 013 + 014 + 015 (note: the input channel 13, 14 or 15 has no contribution), then the decoder can have two inputs (II and 12) and two outputs, and the conversion factor related to 15 is: 〇 1 = ΑΙ1 + ΒΙ2, and 〇2 = CIl + DI2. In the case of a single independent module, the master Either a matrix or a local matrix may have matrix elements that provide more than multiplication. For example, the 20 elements mentioned above may include a filter function (such as a phase or delay term, and / or a filter), It is a function of frequency. An example of a filter that can be applied is a matrix that can provide a pure delay of a photon projected image. In practice, this main or local matrix can be divided into two functions, for example, one uses coefficients to derive the Equal output channels, and the second applies a filter function. 30 200404222 Figure 2 is a functional block diagram that provides an overview of a multi-band conversion embodiment implementing the example shown in Figure}. For example, with multiple interleaved audio One of the signals, the PCM audio input, is applied to a supervisor or supervisor function 201 (hereinafter referred to as "supervisor 201"), which includes an interleaving canceller that restores 6 of the 5 inputs carried by the interleaving Separate streams of each of the audio signal channels (1 ', 3, 5, 9, 9, 13, and 23,) and apply each to the time domain to the frequency domain transfer_transform or conversion function (hereinafter referred to as "Forward"). Or These audio channels can be received in separate streams, in which case no interleaving canceller is required. As noted above, the signal conversion according to the present invention can be applied to a wide Lu 10-band signal or a multi-band For each frequency band of the processor, it can use, for example, a discrete critical band waver bank or a filter bank with a band structure compatible with the relevant decoder, or such as FFT (Fast Fourier Transform) * MDCT (Modified Discrete Cosine) Transformation) of the linear filter bank. The 2nd, 4A_4C and other diagrams are described in terms of multi-band transformation. 15 For the sake of brevity, the optional LFE input channels (potentially 7th input channels in Figures 1 and 2) and output channels (Numbers 1 and 2) are not shown in Figures 1, 2 and the others. Potential 24th output channel). This LFE can generally be processed in the same way as other input and output channels, but its own conversion factor is fixed at 1 and its own matrix coefficient is also fixed at i. 2 In the case where the source channel does not have LFE but the output channel has (such as 2:51 upmixing), an LFE channel may use a low-pass filter (for example, with a 120Hz corner) applied to the sum of these channels The fifth order Butterworth low-pass filter is derived, or in order to avoid cancellation when adding channels, the phase sum of one of these channels can be modified. In this input 31 200404222 the input has LFE but In the case that the output does not have, the LFE channel can be added to more than one rotation channel. Following the description of Figure 2, modules 24-34 receive 6 inputs in the same manner as shown in Figure 1, 5, 9, 13, 13, and 23, each one of the 5 modules generates a preliminary transmutation factor (PSF) output for each audio output channel shown in the relevant Figure 1. Therefore, for example, module 24 Receive input μ, and 3, and generate preliminary conversion factor outputs PSF, pSF2 and PSR3. Or as mentioned above, each module can generate a preliminary set of audio outputs for each audio output channel associated with it. Each module Can also be explained with 10 supervisors as explained further below 201 communication. The information sent to the various modules by the supervisor 201 may include neighbor-level information and higher-order neighbor-level information (if any). The information sent by the supervisor to each module may include Due to the total estimated energy inside the output of each module input. These modules can be considered as part of the control signal generating part of the overall system in Figure 2. 15 The monitor 201 as shown in Figure 2 can implement the data A variety of functions.-For example, the supervisor can determine whether more than one module is in use. If not, the supervisor does not need to implement any functions related to the neighbor level. When it is initialized, it can notify the loose group. It has the number of inputs and outputs, the matrix coefficients associated with it, and the sampling rate of the signal. As already mentioned, the blocks of the interleaved PCM samples can be read and deinterleaved into separate ones. Channels. It may, for example, apply an unrestricted action in the time domain in response to additional information indicating the source signal's amplitude limitation and degree of limitation. If the system operates in a multi-band mode, it may apply windowing With wave filter Arrange groups (such as FFT'MDCT, etc.) to each channel (so that multiple modes do not implement a redundant transformation that substantially increases processing costs), and send a stream of transformed values to each mode 15 for processing. Each mode A two-dimensional array of conversion factors is transmitted back to the supervisor. A conversion factor is used for all conversion baskets of each sub-band of each output channel (when multi-band conversion is assembled, otherwise there is one conversion factor per output channel). ), Or a two-dimensional array of output signals ... The comprehensive effect of the complex transform basket is used for each sub-band of each output channel (when multi-frequency W transform is assembled, otherwise there is one output signal per output channel). The supervisor can smooth the scaling factor and apply it to the signal path matrix (matrix 203 described below) to obtain the composite spectrum of the output channel (in the case of a multi-band transform assembly). Alternatively, when these modes generate output signals, the supervisor can derive the output channels (output channel composite spectrum, in the case of multi-band transform and assembly) to compensate for a local matrix that will produce the same output signal. Then in the case of MDCT, it can perform inverse transform plus windowing and overlapping addition for each output channel, and interpolate the output samples to form a synthetic multi-channel output stream (if it is alternatively, it can be omitted Interleave to provide multiple output streams) and send it to an output file, audio card, or other final destination. 20

Although various functions can be implemented by a supervisor or multiple supervisors as described here, those skilled in the art will understand all of them; can or one of them-can be in the models themselves rather than all Or some mode 3 common supervisors are implemented. For example, there is a "single-independent" module = no need to distinguish between module functions or supervisor functions. Although in multiple modules, the common supervisor can reduce or eliminate redundant processing tasks by reducing the overall processing power required, the elimination of the common supervisor or its permissible set can be easily achieved. Add to each other for example to upgrade to more output steps 33 200404222. Returning to the description in FIG. 2, the six inputs 1, 3, 5, 5, 9, 13, and 23 'are also applied to a variable matrix or a variable matrix function 203 (hereinafter referred to as Matrix 203). The matrix 203 can be considered as one part of the signal path of the system of FIG. The matrix also receives as input a set of final scaling factors SF1 to SF23 for each of the 23 output channels of the first illustration by the supervisor 201. These final conversion factors can also be considered as the rotation of the control signal part of the system in Figure 2. As explained further below, the supervisor 200 preferably transmits the special preliminary conversion factor for each "in" output channel as the final 10 conversion factor for the use, but the supervisor outputs a channel for each endpoint The final conversion factor is determined in response to the information it receives from the module. An "internal" output channel is in the middle of more than two "endpoint" output channels of each module. Alternatively, if the modules generate output signals instead of conversion factors, then there is no need for a matrix 203; the supervisor itself can generate the output signals. 15 In the example in Figure 1, it is assumed that the endpoint output channel coincides with the input channel location, although it does not have to coincide as discussed further elsewhere. Therefore, output channels 2, 4, 6-8, 10-12, 14-16, 17, 18, 19, 20, 21, and 22 are internal output channels. The internal output channels are between or enclosed by the three input channels (input channels 9, 13, 13, and 23), while the other internal 20 channels are between the two input channels (that is, between or enclosed by them). live). Since these endpoint output channels have multiple preliminary conversion factors that are shared among the modules (ie, output channels 3, 5 '9, 13 and 23), the supervisor 201 determines the final endpoint between the conversion factors sfi to SF3. Conversion factor _, yang, etc.). The final internal output conversion factors (SF2, SF4, SF6, etc.) are the same as these initial conversion factors. 34 苐 3 is a functional block diagram which is useful in understanding the way in which supervised zero can determine an endpoint conversion factor as shown in Figure 2. The supervisor does not sum up all the outputs of a mode that share one input to obtain an endpoint conversion factor. Instead, as in a combiner 301, the modules 9 and 27 share inputs 9 shared by the module, as shown in Figure 2. Each of the inputs—the module is an input—additively combined to estimate the interior. Energy ^ This sum represents the total energy level declared by the internal outputs of all connected modules. Then as in the combination of 303, the input of the module (in this example, module% or module 2 is any of the inputs (such as the output of the smoother 425 or 427 in Figure 4B described below) ) After the smoothing, the energy level of the input loses this sum. In the common input, the smooth input of these modules is selected—for full, because each module is independent of each other. It is also slightly different between the filament groups. The output of the combiner 3 is the desired output signal energy input here, and this energy conversion is allowed to be lower than zero. By using this smoothing after the input The input level of is the same as that divided by 5 in the division 除 output signal level and the square root is taken as in block 3G7, and the final conversion factor (nano in this example) of the output is obtained. Note Regardless of how many modules share the input, the "supervisor" is a final conversion factor that derives a single-for each input. This is used to determine the total estimate of the internal output attributable to the input of the material module. The energy will be described in the related subordinate figure. Since the scales are pure and (the first order number Subtracted Energy Level (Second-Order Quantity) 'After the division operation, the square-square calculation is omitted to obtain the final conversion factor (when the conversion is related to the first-order quantity). The addition of the internal levels and the The subtraction of the total input level is all implemented in the sense of pure energy. 200404222 The reason is that the outputs of different modules are assumed to be independent (irrelevant). If this assumption is not true under unusual circumstances, , It calculates that more input signal should be obtained at this input button, which will cause a slight spatial distortion in the reproduced sound ¥ (such as a slight pull of other nearby internal images toward this input 5 but in the same situation the human ear The response may be the same. ⑹ · The internal output channel conversion factors of PSF6 to PSF8 of module 26 are transmitted by the supervisor as the final conversion factor (which has not been modified). For the sake of brevity, item 3 only shows these The generation of one of the final conversion factors of the points. The other end-final conversion factors can be derived in a similar way. _ 10 The description to Figure 2 is as mentioned above. In the variable matrix 203, its Variability possible Complex (all coefficients are variable) or simple (coefficients are changed in groups, such as those applied to fixed matrix inputs or outputs). Although either approach can be used to produce essentially the same result A simpler method, that is, a fixed matrix with a variable gain of 15 for each output (the gain of each output is controlled by the conversion factor) has been found to produce satisfactory results and is described in the embodiment described here. Although each of the matrix coefficients is variable-the variable matrix is usable, its disadvantages are that it is more and requires more processing power. The supervisor 201 is also applied to the variable matrix at the final conversion factor. Before 203 · 20 Optional time-domain smoothing is performed on it. In a variable matrix, the output channel _ will never be "electrically shocked", and its coefficients are arranged to strengthen some signals and eliminate others. However, in a fixed matrix, a variable gain system, as in the embodiment described in the present invention, does turn channels on or off, and is more suspicious of unwanted "switching" artifacts. This will happen regardless of the two-stage smoothing described below (such as smoother 419/425, etc.). For example, when a conversion factor is close to 0, since only a small change is required to change from "small" to "none" and back again, a zero-to-zero shift will cause audible electrical shock. The smoothing performed by the supervisor 201 is preferably determined by the absolute difference (abs-diff) between the newly converted instantaneous conversion factor value and the smoothed conversion factor. A variable time constant smoothes these output transmutation factors. For example, if abs_diff is greater than 0.4 (and of course less than 10), little or no smoothing is applied; a small amount of additional smoothing is applied to an abs-diff value between 0.2 and 0.4; and low At a value of 0.2, the time constant is the continuous inverse function of the Πabs-diff. Although these values are not critical, they have been found to reduce audible electrical shock artifacts. As an option, in the multi-band version of the module, the time constant of the conversion factor smoother is also adjusted by frequency and time in the manner of the frequency smoothers 413, 415, and 417 of the following figure. 15 As described above, the variable matrix 203 is preferably a decoding matrix having -㈣ at a matrix output having a variable scaling factor (gain). Each-matrix output channel can have (mosquito's) matrix coefficients. Wei had the remaining encoding. ㈣ The channel used to be down-mixed encoding coefficients (instead of directly mixing the source channels into down-mixed arrays.) It avoids the need for discrete encoders). The number of 20 is preferably a square sum of 10 for each output channel. The matrix coefficients are fixed when it is known that the output channels are there (as discussed above with respect to the "main" matrix), and the conversion factors that control the output gain of each channel are dynamic. Contains the frequency domain transformation of module 2 wood 34 that is applied to Figure 2. The number of inputs is 200404222. The initial amount of energy and the common energy are used as explained further below. Modules are grouped into frequency sub-bands. Therefore, each frequency sub-band has a preliminary conversion factor (PSF in Figure 2) and a final conversion factor (SF in Figure 2). Each of the 5 frequency domain output channels 1-23 generated by the matrix 203 contains a set of transform baskets (the group of transform baskets made into sub-band sizes is processed by the same conversion factor). Each set of frequency domain transform baskets is transformed into a set of PCM output channels 1 _23 by a frequency versus time domain transform or transform function (hereinafter referred to as "inverse transform") 205. The inverse transform can be a function of the supervisor 201, but it is clear It is displayed separately for the sake of convenience. Supervisor 201 can interleave the resulting PCM channels 1-23 to provide a single interleaved pCM output stream or leave these PCM output channels as separate streams. Figures 4A-4C show functional block diagrams of a module according to one aspect of the present invention. This module receives more than two input signal streams by the supervisor 2001 as shown in Figure 2. Each input contains the synthesis results of a complex-valued frequency domain transform basket. Each input from 1 to 111 is applied to a function or a device (such as input 1 is a function or device 401 and input m is a function or device 403), which calculates the month b of each basket, which is each Transform the sum of the squares of the real and imaginary values of the basket (only two input paths of 1 and m are shown in the figure for simplicity). Each input is also applied to a function or device 405 which calculates the common energy of each basket passing through the input channel 20 of the module. In the case of the FFT embodiment, this can be calculated by taking the product of the intersection of the input samples such as 忒 (for example, in the case of 1 and 11 bis inputs, the value of the complex L basket value and the complex R basket value The real part of the product of the complex number of the complex number in total). Embodiments using real values need only intersect and multiply the real parts for each input. As far as two or more inputs are concerned, the special intersection and multiplication technique described in 38 200404222 3 above must be profitable, that is, if all signs are the same, the soil product is a positive sign, otherwise it is ㈣, and the possible The number of positive results (always 2 · It is either all positive or all negative) is converted to the ratio of the number of possible negative results. 5 Each function of the block conversion basket can be grouped into sub-bands by using separate functions or devices 407, 409 and Chuan. These sub-bands can, for example, approximate the critical solution of the human ear. 4AH Module Implementation_The rest operate separately and independently for each sub-band. For clarity, only operations on a sub-band are shown. 10. From the block technique, each of the sub-bands 411 and 411 is applied to a frequency smoother or a frequency smoothing function 413, 415, and 417 (hereinafter referred to as "frequency smoothers". The purpose of these frequency smoothers is as follows. Explanation: The frequency-smoothed subbands from a frequency balancer are applied to optional "fast" frequency smoothers or frequency smoothing functions 419, 421, and 423 (hereinafter referred to as 15 "fast smoothing ϋ"), This provides _smoothing. Gu fast smoothers are better, but when the time constants of the fast smoothers are close to produce the forward transformation of the input baskets (such as the forward transformation in the supervisor 201 in Figure 1) It can be omitted when the block time length. These fast smoothers are relative to "slow" variable time constant smoothers or smoother function milks, 427 and 429 (^ 20 will be referred to as "slow smoothers"), It receives the individual outputs of these fast smoothers) fast. Examples of fast and slow smoother time f-numbers are given below. Therefore, whether it is the inherent operation of forward transformation or the fast smoothing provided by a fast smoother, a two-stage smoothing action is preferred, in which the second and slower stages are variable. However, single-stage smoothing will provide acceptable results. The time constant of the slow smoother is preferably synchronized with others in a module. This can be achieved, for example, by applying the same control information to each slow smoother and by combining each slow smoother in the same way to react to the applied control information 5. The derivation of the information used to control these slow smoothers is described below. 10 15 20 Preferably, each pair of smoothers is connected in series in pairs 419/425, 421/427 and 423/429 shown in Figures 4A and 4B, including a fast smoother fed to a slow smoother . The advantage of the series configuration is that the second stage is impedance to the short and rapid signal surge of the pair of inputs. However, a similar result can be obtained by assembling these pairs in parallel. For example, in a parallel configuration, the short and rapid 彳 § surge of the second stage of the series configuration can be processed by a logic circuit in a time constant controller. Each -P of the two-stage smoother can be used as an Rc low-pass filter (in the analog embodiment) or equivalently as a first-order low-pass filter (in the digital embodiment). An extremely low-pass filter is implemented. For example, in a digital embodiment, the H-th skinner can be implemented with a _ "twisted pair" filter (a general second-order nR filter), where certain coefficients are set to 0, which makes the waver function Become the first waver. Alternatively, the two smoothers can be combined into a single-second stage "twisted group" stage, 'however if it is separated from the-(fixed) stage, it is calculated for the second (variable) stage The coefficient value is simpler. It should be noted that in the embodiments of Figures 4A, 4B, and 4C, all signal levels are expressed as energy (after squared) levels, unless the amplitude is leveled by

40 200404222 A square root is required. Smoothing is applied to the energy level of the applied signal, so that the smoother has cents instead of the average number cents (Ping Cong is fed with linear amplitude when in operation). Since the signal applied to the smoothers is the squared level, these smoothers respond faster than the average smoother with a sudden increase in the letter-5 level, and the increase is amplified by the squared hearing. -The P white-band smoother thus provides time-averaged and per-subband energy for each input channel (the first channel is provided by k flat m H 425 and the m channel is provided by slow smoother 427) The average energy per input channel (provided by slow smoother 429) is the average number of sub-bands. # 1〇 The average energy output of the slow smoother (425 '427, thing) is applied to the groups σ 43, 433, and 435, respectively: ⑴The energy levels of these neighbors (if any', such as from the second Supervisor 2 in the figure 1) The smoothed energy level of each input channel is subtracted 'and (2) The energy levels of these higher-order neighbors (if any) are, for example, the supervisors from Figure 2 20.1) The b average energy output of each slow smoother is subtracted. For example, each module receiving input 3 (pictured here) has two neighbor modules and receives neighbor output level information, which compensates for the effects of these two neighbor modules. However, none of these two modules are "high-level" «modules (that is, all modules of shared turn channel 3 are two-input ·). To Zhaozhi) the next module 28 (Figures 1 and 2) is an example of a high-order module with a shared input of-. Thus, for example, in module 28, input 3 and the average energy output from the slow smoother receives higher order neighbor level compensation. / Is the energy level of “after neighbor compensation” obtained by the result of each sub-band in turn of the module. The directional gravity center function or wave set 437 of these energy levels is calculated by the button. This direction indication can be calculated as the input life 41, ^ ϋ after these energy weightings 41. In the case of a two-input module, this is simplified to the L / R ratio of the input energy level after such smoothing and neighbor compensation. For example, false δ and the plane in which the channel is located are two inputs around the array, and the month opening ν is & is the second layer representing the x, y coordinates. The listener in the center assumes (0, 〇). The left-front channel has (1,1) coordinates in the normalized space. Right front 'member I is (1 1). If the left input amplitude (Lt) is 4 and the right input amplitude (Rt) is 3 ', the unitary amplitude is used as the weighting factor, and the center of gravity and direction are ... (4 * (1,1) +3 * (-1,1) ) / (4 + 3) = (0.143, 1) or slightly to the left of the center of the horizontal line connecting left and right. Or, once the main matrix is defined, the entity direction can be expressed in matrix coordinates instead of physical coordinates. In this case, the sum of squares is normalized to the effective matrix coordinates with the input amplitude as the direction. In the above example, the left and right levels are 4 and 3, which are normalized to 0.8 and 0.6, and the "direction" is (0.8, 0.6). In other words, the directional center of gravity normalizes the square root of the smoothed input energy level after neighbor compensation to the normalized form. Block 437 produces the same number of outputs, indicating the spatial direction, because it has input to the module (2 in this example). The smoothed input energy level after each subband neighbor compensation applied to each module input of the direction determination function or device 437 is also applied to the function or device 439, which calculates the cross correlation (1 ^ §1 ^ 〇1 ^ 11 ^ 61 ^ {6 heart \ (: 〇1〇. Block 439 is the average energy of the common input of these modules after each sub-band from the slow-variable smoother 429 receives the common energy as an input. , Which has been compensated in the combiner 335 with higher-order neighbor energy levels (if any). The neighbors are compensated for cross-correlation at block 439. The input channels of each module in the 200404222 group are calculated as high-order compensation to smooth the common energy. Divide by the M root of the product of the smoothed energy levels after these neighbor compensations (here M is the number of rounds) to derive true mathematical correlation values in the range of 1.0 to -1.0. Neighbor-cmpensated —Xc〇r provides an estimate of the 5 cross correlations that exist when other modules are absent. The neighbor-cmpensated—xcor from block 439 is applied to a weighting function or device 441 that uses the neighbor compensation direction information to neighbor the neighbors. -cmpensated xcor plus The weights are compensated for the direction-weighted-xcor by the neighbors that generate the direction-weighted neighbors. The weighting increases as the directionality 10 center of gravity moves away from the center condition. In other words, unequal input amplitudes (and thus energy) cause directions -weighted_xcor is proportionally increased. direction-weighted—xcor provides an estimate of the compactness of the image. Therefore, for example, in the case of a two-input module with left L and right R inputs, the weighting increases toward left or right. The deviation direction increases (that is, the weight is the same as the deviation from the 15 center in either direction is the same). For example, in the case of a two-input module, the neighbor-cmpensated-xcor is L / R or r / l 'Weighted' makes the uneven signal distribution force the direction-weighted xcor towards 1.0. For the two input modules,

When R-L 20 direction-weighted_xcor = (1-((1 -neighbor-

cmpensated_xcor) * (L / R)), and when R < L direction-weighted—xcor = (1 _ ((1 -neighb〇r_ cmpensated_xcor) * (R / L)) 43 200404222 has more than two inputs In terms of the module, directi0n_weighted is calculated by neighbor_cmpensated—xCor—for example, it needs to be between 1.0 and 0—the “uniformity” measurement replaces the above L / R or R / L ratio. For example, for any The number of inputs is used to calculate the uniformity measurement. The input signal level can be normalized with a total of 5 input powers. As a result, the normalized input level is obtained after the energy (squared) meaning sum is 1.0. Used in The center of the array is similarly normalized input level divided by each normalized input level. The smallest ratio is the uniformity measurement. Therefore, for example, there is an input with 0 level For a three-input module, the uniformity 10 is measured as 0 'and the direction_weighted xcor is equal to 1 (in this case, the signal is at the boundary of the three-input module, that is, between the two inputs Online 'and a two-input module (lower level in the hierarchy Determine where the directional center of gravity is on the line and how wide the output signal will spread along the line.) 15 Returning to the description of Figure 4B, direction-weighted-xcor is further applied to a function or device 443 and weighting, this applies a rand0m_xc sink weighting to produce an effective_xCor. Effective_xcor provides an estimate of the distribution shape of the input signals. Random_xcor is the average cross product of the input amplitudes divided by the 20 The average root of the average input energy. The value of random-xcor can be obtained by assuming that the output channels are the original module input channels and calculating the result of passively downmixing the signals with independent but equal levels of signals. Sink value is calculated. According to this method, for the case of two-input three-output module, random-xcor is calculated as 0_333, and for the case of 44 200404222 with three-input five-output module (three are internal Output), random__xcor is calculated as 0.483. The random_xcor value only needs to be calculated once for each module. Although these random_xcor values have been found to be satisfactory results, these values are not The key, and other values can be used at the discretion of the system designer. 5 The change of the random-xcor value will affect the two lines of operation of the signal distribution system as described below. The precise location of this line is not Pivotal. The random_xcor weighting implemented by the function or device 443 can be regarded as a renormalization of the direction-weighted_xcor value, so that effective-10 xcor is obtained: effective_xcor = (direction-weighted_xcor-random_xcor) / (1- random-xcor), if direction-weighted-xcor — random-xcor effective_xcor = 0, the other 15 Random_xcor weightings decrease with direction · weighted_xcor below 0 and accelerate the decrease of direction-weighted-xcor, so that when direction-weighted-xcor When equal to random_xcor, the effective_xcor value is zero. Since the output of a module represents a direction along a solitary line or a straight line, an effective_xcor value less than 0 is considered to be equal to 0. 20 The information used to control the slow smoothers 425, 427, and 429 is derived by the non-neighbors to compensate the energy of the slow and fast smoothed input channels and the energy of the slow and fast smoothed input channels. Specifically, the function or device 445 calculates a fast non-neighbor compensated cross-correlation in response to the energy of the input channels after the fast smoothing and the common energy of the input channels after the fast smoothing. The function or device 447 calculates a fast non-neighbor compensated direction in response to the fast smoothed input channel energy (as discussed in the description of the relevant box above as a ratio or vector). The function or device 449 calculates a slow non-neighbor-compensated cross-correlation in response to the joint energy of the de-smoothed input channels and the slow-smoothed input channels. The function or device 451 calculates the direction of a slow non-neighbor complement after responding to these slow smoothed input channel energies (as discussed in the description of the relevant box above as a ratio or a vector). The fast non-neighbor compensated cross-correlation from block 441, the fast non-neighbor compensated cross-correlation direction, the slow non-neighbor compensated cross-correlation, the slow non-neighbor compensated post direction, and direction-weighted-xcor are applied to the function or device 453 ' Provides information for controlling variable slow smoothers 425, 427, and 429 to adjust their time constants (hereinafter referred to as adjusting time constants). Preferably, the same control signal is applied to each variable slow smoother. Unlike other values that are fed to the time constant selection box that is faster and more speculative, the direct_weighted_xcor is preferably used without reference to any fast value, so that if the directi.weighted_xc〇r If the absolute value is greater than a threshold value, it will cause the adjustment of the time constant 453 to select a faster time constant. The operating rules for "adjusting the time constant" are established below. Generally speaking, in a dynamic audio system, it wants to use the slow time $ as much as possible and keep it at a static value so as to minimize the audible interruption of the reproduced sound field unless there is a "new event" in the audio Signal occurs, in which case it wants to make the control signal quickly change to a new static value, and then retain that value until another "new event" occurs for this. Typically, audio processing systems have equal changes in the amplitude of the "new event." However, when dealing with cross-correlation cross-correlation, the new 200404222 event and amplitude will not always be equal: the new event may cause a decrease in cross-correlation. By induction, changes in the parameters related to the operation of the module (ie, cross-correlation and direction measurement), the time constant of the board group will accelerate and quickly adopt the new control state as desired. Work-5 $ Appropriate dynamic red effects include: image loss, electric shock, "L quickly turns on and off" system (unnatural changes in level), and the sound in the multi-band module (in accordance with Band-based electrical shock and pumping). Some such effects and the quality of the isolated channels are particularly critical. Such as the embodiment of Figures 1 and 2-the decoding module of the grid. This group · 1〇 The results of two types of dynamic problems: intra-module and inter-module dynamics. In addition, various methods of audio processing (such as broadband, multi-band using chat or MDCT linear filter banks, or discrete filtering) Device, bank, critical band, or other) need to optimize its own dynamic behavior. The basic decoding process in each branch depends on the measurement of the input signal's energy ratio of 15 and the intersection of these input signals Relevant measurements (especially the output of block 441 in Figure 4B above, that is, directm-weighted-xcor), which together control the signal and signal allocation among the outputs of a module. These The derivation of basic values needs to be smooth, which is in the time domain The time-weighted average of the _ values of these values needs to be calculated. The range of the ㈣ · 20 constant is very large: the change from the rapid transition of the signal condition is very short (for example, 1 msec) to the very low value of the correlation (very long) For example, 15〇_0, where the instantaneous variation may be much larger than the true average. The common method of implementing variable time constant behavior is in analogy terms & with-"acceleration" diode. When the instant Level—When the threshold number exceeds 47 average levels, the diode will conduct electricity and form a short effective time constant. The disadvantage of this technique is that a temporary spike in the steady state input instead will cause a smoothed level. Large changes, and then it decays very slowly to provide an unnaturally unnatural emphasis on isolation that would have less audible consequences instead. The calculation of the correlation values described in Figures 4A_4C speeds up the diode (Or its DSP equivalent) becomes problematic. For example, all smoothers in a particular module preferably have an isochronous time constant so that their smoothed levels are comparable. Therefore, a comprehensive (joint) time constant switching mechanism is better. In addition, rapid changes in signal conditions are not necessarily related to an increase in the common signal level. The use of diodes of this level may produce deviations in related values, Inaccurate estimation. Therefore, the embodiments of the present invention at various levels preferably use two-stage smoothing that does not require diode equivalent acceleration. Correlation and direction estimation can be performed at least by the first and second stages of these smoothers. It is derived to set the time constant of the second stage. For each pair of smoothers (such as 419/425), the time constant of the first stage (fixed fast stage) can be set to a fixed value such as lmsec. The time constant of the second stage (variable slow stage) can be selected, for example, from 10 insec (fast), 30msec (medium), and 150msec (slow). Although such time constants have been found to provide satisfactory results, their values are not critical and other values can be used with the system designer's consideration. In addition, the second stage time constant may be a continuous variable rather than a discrete one. The selection of the time constant is not only based on the signal conditions described above, but also a hysteresis mechanism using the "fast flag", which is used to ensure that it should be maintained in fast mode once it encounters a true fast transfer, avoiding the use of medium time Constant until the t further contributes to the slow time f number depending on the situation. This helps to ensure rapid adaptation to new salt production conditions. The “choose to use three possible second-stage time constants—available adjusted time constants” 453 is completed for the two input cases according to the following rules: If the absolute value of direction-weighted—xcor is less than a first reference value (for example, 0.5) and the absolute difference between fast non-neighbor-compensated-xc〇r and slow non-neighbor-compensated-xcor and the ratio of fast and slow direction (each has a range of +1 to _1) is less than The same first reference value, the second stage time constant is used, and the fast flag is set to true, which facilitates the subsequent selection of the medium time constant. Otherwise, if the fast flag is true, fast and slow non -neighbor-compensated The absolute difference between xcor is greater than the first reference value and less than a second reference value (for example, 0.75), and the absolute difference between the fast and slow temporary L / R ratio values is greater than the first reference value and less than If a second reference value and the absolute value of direction-neighted_xcor is greater than the first reference value and less than a second reference value, the medium time constant is selected. Otherwise, the fast second-stage time constant is used. And the fast flag is set to false, so that the subsequent use of the medium time constant is invalidated until the slow time constant is selected again. In other words, when all three conditions are less than a first reference value, the slow time constant Selected; when all conditions are between a first reference value and a reference value and the previous condition is a slow time constant, the medium time number is selected, and when any condition is greater than the second reference value, The fast time constant is selected. — 5, * Although the rules and reference values just mentioned have been found to be satisfactory, 'is not critical and adopts rules that take into account fast and slow cross-correlation and fast and dagger direction deformation or other The rules can be applied at the discretion of the system designer. In addition, other changes can be made. For example, using a diode ^ catch type of processing but with a joint operation such that if any one of the modules is smooth ⑺ = is in & Speed mode ^ 'All other smoothing ϋ Φ is switched to fast mode ^, fortunately, it is simple but equally effective. It can also be decided by the time constant

The Uu distribution has a separate smoother. The time constant is used to determine the smoother used. It is maintained at S1. The number of hours and the time constant of the financial information distribution will change. ° Even in the fast mode, the smoothed signal level needs a few milliseconds to adapt '-Time delay can be built into the system to allow the control signal to adapt before it is applied to the path ^. In a wideband embodiment, this delay can be implemented as a discrete delay in the signal path (eg, plus ㈣. In a multiband (transform) type, the delay is a natural consequence of block processing, and if the block analysis is in The signal path of this block is implemented before it is formed into a matrix, and no significant delay of 20 is required. The multi-band embodiments of the layers of the present invention can use the same time constants and rules as the wide-band type, except for such smoothers. The sampling rate can be set as the signal sampling rate divided by the block size (such as the block rate), so that the coefficients used in the smoother are appropriately adjusted. 50 200404222 For frequencies below 400 Hz, in a multi-band embodiment In time, these time constants are preferably adjusted in inverse proportion to the frequency. In the wide band type, it is impossible to do so because there are separate smoothers with different frequencies, so as a partial compensation, a mild bandpass / A pre-emphasis filter can be applied to the 5-input signal to control the path while emphasizing the middle and upper middle frequencies. This filter can have, for example, two poles at a corner frequency of 200 Hz High-pass characteristics, plus two extremely low-pass characteristics at a corner frequency of 8000Hz, plus a pre-emphasis network that applies a 6dB boost from 400Hz to 800Hz and a 6dB boost from 160Hz to 3200Hz Although this filter has been found to be suitable, its filter characteristics are not critical, and other parameters can be used at the discretion of the system designer. In addition to time domain smoothing, there are many aspects of the invention The frequency band pattern preferably also uses frequency domain smoothing (frequency smoothers 413, 415, and 417) as described in the related FIG. 4A above. For each square, the level of these non-smoothers after compensation can be 15 Before applying to the above-mentioned subsequent time-domain processing, a sliding frequency is averaged and adjusted to about 1/3 Octave (critical frequency band) bandwidth. Since the transform-based filter bank has a linear frequency resolution in nature, this The width of the window (in terms of the number of transform coefficients) increases with increasing frequency, and is only the width of one transform coefficient at low frequencies (below about 400 Hz). Therefore, the total smoothness of 20 applied to multi-band processing is low. Frequency dependent Inter-domain and high frequency are more dependent on the frequency domain, where a quick time response is often necessary. Turning to the description in Figure 4C, the initial conversion factor that affects dominant / filling / endpoint signal allocation (in the first (Figure 2 is shown as PSF) Function or device 455, 457, and 459 can be calculated by calculating 51 5 _Page 丨 "conversion factor score," fill "conversion factor component, and" extra endpoint volume "conversion factor component. The maximum number of combinations of individual normalizers or normalizer functions 461, 463, and 465, and the summation of the summation and filling conversion factor components and / or the addition of the filling and additional endpoint energy conversion factor components A function or combination of devices 467 is generated. The preliminary conversion factors may be transmitted to the supervisor 201 as shown in FIG. 2 when the module is one of several modules. Each of the preliminary conversion factors may have a range from 0 to 1.

Explicit conversion factor component 15

20 Function or device 355 ("Calculate explicit conversion factor component") Divided by an effectives, by Qian 43 7 Receive the direction information after the compensation of the dwelling ^ By -local_to receive the information of _local matrix Wei, so that the decision can be determined. The N nearest output channels (where, for example, the number of coefficients) are applied to the -weighted sum to obtain the center of gravity coordinates, and the "dominant" conversion factor component is corrected to its simple physical coordinates. If the center of gravity's direction coincides with the -output direction, the output of the square pole is suppressed "-the conversion factor component (every sub-band), or 4 for multiple conversion factors (every number of inputs for each sub-band- A) Envelope the directional direction of the center of gravity and apply it at a ratio of = to determine or reflect the dominant signal _ leave the meaning to modify the virtual position (ie, n = 2, these two are referred to; The sum of the squares of the explicit channel conversion factor components should be equal to effecti.) T :: For the two groups, 'all the rotation channels are within-the line is thin, so: there is a natural ordering (from left to right), and the report understands What is the side road is adjacent to each other. As for the hypothetical situation discussed above-there is a round-in channel and a five-lost 52 200404222 channel and the sin / cos coefficient as shown, the directional center of gravity can be assumed Is (0.8, 0.6) between the center left ML channel (0.92, 0.38) and the center C channel (0.71, 0.71). This can be obtained by finding the L-coordinate of the directional center of the L coefficient greater than the center of gravity and The channel on the right has two consecutive channels with a [coefficient greater than the dominant L 5 coordinate and is completed. Such dominant conversion factors The amount is allocated to the two closest channels in a fixed power sense. To do so, the system of the second equation and the two unknowns is solved, and the unknowns are the dominant transversals of the channels to the left of the dominant direction (SFL) The factor component and the conversion to the right of the directional center (SFR) corresponding to the center of gravity are modified by 10 factor components (these equations are solved for SFL and SFR). First-dominant—coord = SFL * left channel matrix value + SFR * right channel Matrix value 1 second—dominant—coord = SFL * left channel matrix value 2 + SFR * right channel matrix value 2 15 Note that the left and right channels mean the channels that surround the directional center of the center of gravity, not the L of the board group Input the channel with R. The solution is the inverse dominant level calculation of each channel, which is normalized so that the squared sum is 1.0, and each channel is used as an explicit allocation conversion factor component (SFL, SFR). In other words, for signals with coordinates C and D, the inverse dominant value of the output channel with 20 coefficients A and B is AD-BC. As far as the numerical value under consideration _ For example:

Antidom (ML channel) = abs (0.92 * 〇.6_〇 · 38 * 〇8) = 〇248 Antidom (C channel) = abs (0.71 * 0.6-0.71 * 〇8) = 〇142 (here abs means take Absolute value) 53 200404222 Normalize the last two digits so that the sum of squares is 0 to get the values of 0.8678 and 0.4968, respectively. Therefore, to switch these values to the opposite channel, the explicit conversion factor components are (note that before direction weighting, the value of the explicit conversion factor is the square root of effective_xcor): 5 ML dom sf = 0.4969 * sqrt (effective_xcor) C dom sf = 0.8678 * sqrt (effective-xcor) (the dominant signal is closer to Cout than MidLout). For example, the use of the anti-dominant component of a channel after the normalization of the dominant conversion factor component of other channels can be considered by If the directional center of gravity does indeed indicate what happens to Xin 10 when one of the temples is chosen, it is better understood. Assuming that the coefficient of one channel is [A, B] and the coefficient of another channel is [C, D], and the directional center of the center of gravity coordinate is [A, B] (pointing to the first channel), then:

Antidom (channel 1) = abs (AB-BA)

Antidom (second channel) = abs (CB-DA) 15 Note that the first antidominant value is zero. When the two anti-dominant signals are normalized such that the sum of squares is 1.0, the second anti-dominant value is 10. After being switched, the first channel receives one of the dominant conversion factor components (multiplied by the effective square root) and the second channel receives 0.0 as desired. When this approach was extended to modules with more than two inputs, the frequency-20 channels no longer appear in a natural order when they are on a line or arc. Again, for example, block 437 of FIG. 4B, calculates the directional center of the center of gravity by taking the input amplitudes and normalizing them to make the sum of squares after the neighbor compensation. Then, for example, block 455 in FIG. 4B determines the ^^ closest channels (of which ^^ 2 input numbers) can be applied to the weighted sum to obtain the dominant seats 54 404222 檩. (Note that distance or proximity can be calculated as the sum of the squares of the coordinate differences, as if _ appears to be 0, y, z) spatial coordinates). Therefore, we don't always have to pick out the N closest channels, because they must be weighted and summed to get the directional center of gravity. 5 For example, suppose we have a channel with Ls, Rs and τ0ρ as shown in Figure 5: a three-input module fed by an angle. Assume that there are three internal turn-out channels ~ from the bottom of the triangle, the module's local matrix coefficients are [0.7, 0.69, 0.01], [0 · 70, 0 · 70, 0 · 〇1], and [0.69, 0 · 7ι, 〇〇1]. Assume that the directional center of the center of gravity is slightly lower than the center of the triangle and the coordinates are Lu 10 [0.6, 0.6, 0.53] (note that the center of the triangle is [〇5, 〇5, 0.707]) closest to the center of gravity The three channels at the center of directionality are the three channels at the bottom, but they do not add up to the dominant coordinates using a conversion factor between 0 and 1. Therefore, the bottom and top end channels are used to select two. To assign the dominant signal and solve the three equations for three weighting factors to complete the dominant 15 calculation and proceed to the filling and endpoint calculations. In the example of Figures 1 and 2, it has only one three-input module, and it is used to derive only one internal channel, which simplifies the calculation. In addition to the effective conversion factor component, the function or device 457 ("Calculate filling conversion_ 20 factor component") receives random_xcor, direction--weighted-xcor, "EQUIAMPL," ("EqUIAMPL," in It is defined and explained below) and information about the local matrix coefficients from the local matrix (in the case where the same filling conversion factor component is not applied to all the outputs as explained in the relevant Figure 14B below). The output of block 457 is a conversion factor component (for each sub-band) used by each 55 200404222 module output. As explained above, effective-xcor is 0 when direction-weighted_xcor is less than or equal to random_xcor. When direction-weighted— xcor is greater than or equal to random_xcor, the filling factor 5 for all output channels is:

Filling conversion factor component two sqrt (l-effective_xcor) * EQUIAMPL Therefore, when direction-weighted_xcor = random_xcor, effective-xcor is 0, so (1-effective-xcor) is 1.0, so the filling vibration is 10 conversions The factor component is equal to EQUIAMPL (make sure output power = input power in this case). This point is the maximum value reached by these filling conversion factor components. When weighted-xcor is less than random _xcor, the explicit conversion factor component is 0 'and the filling conversion factor component is reduced to 0 as the direction · weighted_xcor approaches 15 to 0.

Filling conversion factor component = sqrt (direction_weighted_xcor / random—xcor) 氺 EQUIAMPL Therefore, at the perimeter, direction-weighted_xcor = random_xcor, the filling initial conversion factor component is equal to EQUIAMPL again, ensuring that the situation where the 20 directions-weighted—xcor is greater than random_xcor The result of the equation. Associated with each decoding module is not only the value of random_xcor, but also the value of EQUIAMPL, which is the conversion factor value that all conversion factors should have if the signals are equally distributed so that power is saved, that is: 56 200404222 EQUIAMPL : Square-root-of (Number of decoding module turn-in channels / Number of decoding module output channels) For example, for a two-input module with three outputs: EQUIAMPL = Sqrt (2/3) = 0.8165 5 Here Sqrt () means Square-root-of () For a module with two inputs with four outputs: EQUIAMPL = Sqrt (2/4) = 0.7071 For a module with two inputs with five outputs:-EQUIAMPL = Sqrt (2/5) = 0.6325 ❿ 10 Although these EQUIAMPL values have been found to provide satisfactory results, these values are not critical and other values are applicable at the discretion of the system designer. Changes in the EQUIAMPL value will affect the level of these output channels with respect to the "filling" condition (the middle correlation of these input signals) against the "dominant" condition (the minimum correlation of the input signals). 15 In addition to the endpoint conversion factor component, neighbor_Compensated—xcor (from the block diagram of the first offense graph), the function or device 459 (“Calculate additional endpoint energy conversion factor components”). Lu receives each of the 1st to mth inputs. Smoothed non-neighbor compensation energy (from blocks 325 and 327), and optional information about the local matrix system 20 numbers from the local matrix (such as the endpoint output and one input of the module discussed further below) Non-coincidence and when the module applies extra endpoint energy to both or any of the directions that have the direction closest to the input). The output of the block is rounded off by one conversion factor component for each endpoint if the directions coincide with the input direction, otherwise it is the second conversion factor component as explained below, one 57 200404222 for each closest to the The output of the endpoint. However, the additional endpoint energy conversion factor component generated by block 459 is not the only "endpoint" conversion factor component. There are three other sources of endpoint conversion factor components (two are in the case of a single independent module): 5 First, in the preliminary conversion factor calculation of a specific module, these endpoints are block 435 (and the normalizer 461 ) Possible candidates for dominant signal scaling factor components. Second, in the “filling” calculation of block 457 (and the normalizer 463) of FIG. 4C, these endpoints are considered as possible filling candidates with all internal channels. A non-zero filling conversion factor component can be applied to all outputs, even the endpoints and the selected explicit outputs. Third, if there are multiple modules in a grid, a supervisor (such as supervisor 201 in Fig. 2) implements the final and fourth assignment of the 15 "endpoint" channels as described in the related Figs. 2 and 3 above. In order for block 459 to calculate the "extra endpoint energy" conversion factor component, the total energy in all internal outputs is reflected back to the input of the module according to neighbor-compensated xcor to estimate how much internal output energy is Input contribution (internal energy at input (η)), and this energy is used to calculate the extra endpoint energy conversion factor component at each module output (ie, an endpoint) consistent with an input. It is also necessary to reflect the internal energy back to these inputs to provide the information required by the supervisor 201 as shown in Figure 2 to calculate the neighbor level and higher-order neighbor level. The method of calculating the internal energy of the input in each module is provided for each endpoint. 58 200404222 The method for determining the components of the additional endpoint conversion factors is shown in Figures 6A and 6B. Figures 6A and 6B are functional block diagrams respectively showing a suitable configuration of any one of the modules 24_3 * of Figure 2 for a module in response to the total energy of each input 5 丨 to 111 Each input from 1 to m produces the total estimated iris' and (2) in response to neighbor-compensated_xcor (see the output of block 439 in Figure 4B), generating additional endpoint energy conversions for the endpoints of each module Factor component. The total estimated internal energy of each coefficient of a module (Figure 6A) is required by the supervisor in the case of a multi-module configuration, and in any case by the module itself to generate the additional Endpoint energy conversion factor component. Using the conversion factor components and other information derived in block 455 or 457 of Figure 4C, the configuration of Figure 6A calculates the total estimated energy at each internal output (but not its endpoint output). Use these calculated internal output energy levels' which multiplies each output level to match the output with each input with the associated matrix coefficients [m inputs, m multipliers], which provides the input pair The energy contribution of the round. For each input, it sums all the internal outputs · All energy contributions of the channel to get the total internal energy of that input. The total internal energy contribution of each input is reported to the supervisor and used by the module to calculate the additional endpoint conversion factor component for each endpoint output. -Referring to Figure 6A in detail, the smoothed I level (preferably non-neighbor compensation for the latter) for each module input is applied to a set of multipliers, one multiplier is used for each module Internal output. For the sake of simplicity, Figure 6A shows—turn in “1” and “m”, and two internal outputs “X” and “Y”. 59 200404222 for each module input. Multiply by a matrix coefficient (belonging to the local moment of the module): match the specific rotation of the port with the internal output of these modules (Note-the second and fourth matrix coefficients are their inverses, because the matrix coefficients are之 平 # is equal to 丨). This is done for each combination of turn-in and internal output. As shown in Figures 6 and 8, the total energy level after smoothing the input (the {j The input of the slow smoother 425 of the graph is obtained) is applied to a multiplier 601 'which multiplies this energy level so that the internal output X and the input lg are correlated by a matrix factor, providing a conversion at one of the outputs χ The output energy level component X1. Similarly, multipliers 603, 005, and 607 provide the converted energy _ 10-quantity level component Xm, 21, and 2111. Each internal output energy level component Xm, Zl, and Zm According to neighbor-compensated an xcor is added to combiners 611 and 613 in an amplitude / power manner. If these inputs of a combiner use a neighbor-weighted cross-correlation of 1.0 to indicate the same phase, their linear amplitudes add up. If they use a neighbor-weighted cross-correlation of 015 to indicate that they are uncorrelated, their energy The levels are summed. If the cross-correlation is between 1 and 0, the sum is part of the amplitude sum and part of the power sum. In order to sum the inputs to each combiner appropriately, the amplitude sum and power sum are calculated , And weighted with neighbor-compensated_xcor and (1 -neighbor-weighted_xcor) respectively. In order to obtain a weighted sum of 20, the weighted sum is reduced to a linear amplitude and squared to obtain an amplitude of equal value, or the linear amplitude The sum is squared to obtain its power level. For example, in the latter approach (weighted sum of power), if the amplitude levels are 3 and 4, and neighbor-weighted_xcor is the amplitude and 3 + 4 = 7, or the power level is 49 and the power energy sum is 9 + 16 = 25. So the weighted sum is 0 · 7 * 49 + (1-0.7) 60 200404222 氺 25 = 41.8 (power energy level) or the square root is 6 and the addition result (Xi + Xm; Zi + Zm) X and Z are multiplied by the conversion factor components in multipliers 613 and 615 to generate the total energy level in the internal output of the mother, which can be determined as X, and Z. Each-to-plus ± conversion of the internal output of the mother The factor 5 component is obtained from block 467 (Figure 4C). The phonetic τ Jiangxin, the additional end point energy conversion factor component from the factory of Block 459 (Figure 4C, Bhutan, Mouth) does not affect internal output and does not involve Calculations implemented by Figure 6A configuration. 1〇 15 2〇

The total energy X, and z, of every-such as "multiplying-the matrix coefficient (belonging to the local matrix of the module) of this module with the specific output and each module input-a collision reflection_each The input of the sequential group. This is done for each combination of internal output and input. Therefore, as shown in FIG. 6A, the total energy level X 'of the internal output X is applied to a multiplier 617' which applies the energy The level is multiplied so that the internal output χ and the rotation are correlated ^-the matrix coefficient (which is the same as its inverse as described above) and the energy level component X / after the magic input.

It should be noted that if the total energy level X, a second order value is weighted with a first order value such as a matrix coefficient, a second order weight is required. This is equivalent to taking the square root of the energy to obtain an amplitude, multiplying the amplitude by the matrix coefficient, and then squaring the result to return to an energy value. Similarly, the multipliers 619, 621, and 623 provide the changed energy bits "xm''Z! 'And Zm'. The energy components (such as X, and Zm, Xm, and Zm,) associated with each output are added at combiner 625 and 627 in amplitude / power according to neighbor_compensated_xcor. The outputs of the combiners 625 and 627 v the total estimated internal energy of inputs 1 and m, respectively. In the case of multiple modules, the information is sent to the supervisor 201 as shown in Fig. 2 so that the supervisor can calculate the neighborhood level. The supervisor requests the total internal energy contribution of each input from all modules connected to the input, and then, for each input, notifies 5 to Every module. This result is the neighbor level of the input of this module. The generation of neighbour level negatives is described further below. The total estimated internal energy contributed by each input 1 and m is also required by the module to calculate the additional energy conversion factor component for each endpoint output. Figure 6β shows how this conversion factor component is calculated. For the sake of simplicity, only the calculation of the conversion factor component information of 10 endpoints is shown, and it is understood that similar calculations are performed for each endpoint output. The total estimated internal energy contributed by the input 丨 is smoothed by the same input (input 1 in this example) in a combiner or combination function 629. The total input energy is subtracted (total energy after the same smoothing at the input 丨). The level is obtained, for example, in the slow flat 15 slider 425 of FIG. 43 applied to the multiplier 601). The result of this subtraction is divided by the same input 1 in the divider or division function by the smoothed total energy level. The square root of the division result is obtained in the square rooter or the square root function 631. It should be noted that the operation of the square root in the square root divider or square root function 631 (and other dividers described herein) should include a test of the zero denominator. In this case, its quotient = 20 is set to zero. If it has only -single-independent modules, the initial conversion factor component of the endpoint is therefore determined by the nature of the explicit, filling, and additional endpoint energy conversions that have been determined. Therefore, all channels including the endpoints have been assigned conversion factors, and we can use them to implement a production matrix using signal paths. However, if it has multiple modules with a grid, each assigns an endpoint conversion factor to each input fed to it, so each input has more than one module connected to it and has multiple conversion factor assignments, each There are 5 connected modules. In this case, the supervisor (such as supervisor 201 of FIG. 2) implements the final fourth assignment of these “endpoint” channels as described above in relation to FIGS. 2 and 3, and the supervisor determines the final conversion factor , It will reject the conversion factor assignment completed by each module as the endpoint conversion factor. In a practical configuration, there is no certainty for a real output channel direction corresponding to a 10-end position, although this is often the case. If it has no physical endpoint channel, but at least one physical channel is behind the endpoint, the endpoint energy is set to the physical channel closest to the end as if it were a dominant signal component. In a horizontal array, this preferably uses a fixed energy allocation to the two channels closest to the endpoint (the sum of the squares of the conversion factors is 15 1.0). In other words, when a sound direction does not correspond to the position of the real sound channel, even if this direction is an endpoint signal, it is better to place it on the closest available pair of the real channel, because if the sound is slowly Move, it will suddenly jump from one output channel to another. Therefore, when there is no physical endpoint sound channel, it is not appropriate to place an endpoint signal at one of the sound channels closest to the endpoint position 20, unless there is no physical channel behind the endpoint, in which case it is close to the endpoint. There is no choice other than a sound channel at the location. Another method to implement this kind of movement is as shown in the supervisor 201 of FIG. 2 according to each input also has a corresponding output channel (ie, each corresponding input 63 200404222 coincides with the output, representing the same location) to generate Γ final "conversion factor. -Then the variable matrix as shown in Figure 2 can map an output channel to more than one appropriate output channel if there is no actual output channel directly corresponding to an input channel. 5 As described above, the output of each "calculate conversion factor component" function or device 455, 457 and 459 is applied to a respective normalization function or device 461, 463 and 465. These normalizers are desirable because the conversion factor components that differ by blocks 455, 457, and 459 are based on the neighbor-compensated levels, and the final signal path is made into a matrix (in the case of multiple modules) The main matrix, which is a local matrix in the case of a Lu 10 single independent module, involves the level after non-neighbor compensation (the input signals applied to the matrix are not compensated by the neighbors). Typically, the value of the scaling factor component is reduced by a normalizer. Appropriate methods for applying a regularizer are as follows. Each normalizer receives the input of the neighbor-compensated smoothed input energy for the input of each module (such as from the I5 combiner 431 and 433), and receives the non-applied normalizer 425 and 427 for the input of each module The local matrix coefficients and the respective outputs of blocks 455, 457 and Te are received by the local matrix. Each normalizer calculates the desired output for each output channel Φ and calculates a true output level for each output channel '. What is false is a conversion factor of 然后. Then, it calculates each calculated — — ^ Divide the desired output of the output channel by the calculated true output and output level of the output channel, and take the square root of its quotient to provide a potential preliminary conversion factor for applying to "addition and / or comparison Big "operation 467. Consider the following example. Assume that the non-neighbor compensation input energy level of a two-input module after smoothing is 6 and 8, and the corresponding neighbor compensation input energy level is tear. It is also assumed that the matrix coefficient of the central internal input channel of 64 200404222 j = (〇 · 7ι, ο ″) or after squared is (0 · 5 '0.5). If the module is selected for this channel (based on the position after the neighbor compensation) The standard conversion factor is 0.5, or squared. Θα. The desired output level of this channel (for simplicity, the level after pure energy addition and compensation using neighbor 5) is: 0.25 氺 ( 3 氺 0.5 + 4 氺 〇 · 5) = 〇.875. Since the real input levels are 6 and 8, if the above conversion factor (after squared) 0.25 is used for the final signal path making matrix, the output level is : 0.25 氺 (6 氺 0.5 + 8 ^ 0.5) = 175 10 instead of the desired output level of 0 · 875. The normalizer adjusts the conversion factor so that after the non-neighbor compensation, the level is made to obtain the desired output level. True output (assume SF = 1) = (6 * 〇5 + 8 * 0 9 = 7. Desired output level / true output (assume SF = 0 = 0.875 / 7.0 = 0.125 = final conversion factor after squared 15 The final conversion factor for this output channel = Sqrt (0.125) = 0.354, replacing the originally calculated value of 0.5. The "addition and / or take The "computer" operation 467 preferably adds the corresponding filling and endpoint conversion factor components for each output of each sub-band, and selects a larger 20 of the dominant and filling conversion factor components for each output of each sub-band. The preferred form of the "add and / or take" function of block 467 can be characterized as shown in Figure 7. That is, the explicit conversion factor component and the filling conversion factor component are applied to the function or device 701. , Which selects the larger of these conversion factor components for each output (the "bigger one" 701) and applies it to the additive combiner or combination function 703, which will be derived from the comparison of 701 The conversion of the larger one is 65 200404222. The factor component and the additional end point energy conversion factor are added for each round. Or, when the "addition and / or whichever is greater" 467 is performed in area 丨 and added to area 2, (2 ) Obtain the larger of Area 1 and Area 2, or (3) Select the largest of Area 丨 and add up in Area 2. Acceptable results can be obtained. 5 Figure 8 is one of the inventions The level generates the conversion factor component in response to one of the cross-correlation measurements. The ideal representation of the formula. This figure is particularly useful when following the examples of Figures 9A and 9B to Figures 16A to 16B. As mentioned above, the generation of the conversion factor component can be considered to have two regions or areas Operation: a first area (area 1) is bounded by "all dominance" and "uniform filling 10", where the available conversion factor component is a mixture of dominant and filling conversion factor components; a second area ( Area 2) The boundary between "uniform filling" and "all endpoints", where the available conversion factor component is a mixture of filling and additional endpoint energy conversion factor components. This "all dominant" boundary condition occurs when direction-weighted-xcoAi. Area 丨 (explicitly filling 15 fills) expands from this limit to the direction-weighted—xcor is equal to random xcor, which is the "uniform filling" condition. This "all endpoints" boundary condition occurs when direction-weighted-xCor is zero. Area 2 (filling plus endpoints) Lu extended from "even filling" condition to "all endpoints" condition. This "uniform filling" status point can be considered in area 1 or area 2. As mentioned above, the precise boundaries are not critical. As shown in Figure 8, as the dominant conversion factor component value decreases, the filling conversion factor component value increases. 'The dominant conversion factor component reaches a maximum value when it reaches 0. At this point, the dominant conversion factor component value decreases. The value of the additional endpoint energy conversion factor component increases. As a result, when applied to the -appropriate matrix of input signals of the 66 module, the output signal is assigned, which provides f dense sound and images when these input signals are highly ground phase, and as Phase_low consists of tightly spread (widening) Qian Guang, and Qian's correlation continues to decrease to a high degree of non-phase_ from wide to gradually split into multiple sound images at each-at-end. Although it wants to have a single spatially tight sound image in the fully correlated case (look for the directional heart of the input field L) and in the complete unrelated case there are several spatially tight sound images (each one in the One end point), the sound and image scattered in the space between these two extremes can be achieved by a method not shown in FIG. 8. For example, in the case of · d〇m_xc〇r = direction-weighted—xc〇r, it is not critical that the value of the three conversion factor components is filled to reach a maximum value, and changes linearly as shown. Nor is it critical. The modification of the relationship in Figure 8 (with the formula expressed here that is the basis of the figure) and an appropriate cross-correlation measurement and the ability to cross-correlate the measurement from highly correlated to highly uncorrelated produce widespread dissemination to be closely distributed Other relationships among the conversion factor values of the end point signal assignments are also planned by the present invention. For example, in lieu of the wide spread to tight endpoint signal distributions obtained by applying the two-region approach as described above, these results can be obtained using mathematical methods such as solving using a virtual inverse-based formula. Example of Output Conversion Factors A series of idealized representations of one of the series of Figures 9A and 9B to 16A to 16B shows the output conversion factor of a module that is an example of various input signal conditions. For simplicity, a single independent module is used so that the conversion factor generated for a variable matrix is the final scale conversion factor. Cheng group and phase_variable matrix have an input channel (such as left L and right R), which coincides with two internal output channels (also called L and R). In the examples of these series, there are three internal output channels (e.g. left middle Lm, center c and right * Rm). Examples of "All Dominance", "Mixed Dominance and Filling", "Homogeneous Filling", "Mixed Filling and Endpoints" and "All Endpoints" are shown in Figures 9A and 98 to 16A and 16B. Further explained. In each pair of graphs (for example, graphs 9A and 9B), graph A shows the energy levels of two inputs (left L and right R); and graph B shows the conversion factor components of five channels (left L, left, center c) , Right middle RM and right R). The figures are not drawn to scale. In Figure 9A, the input energy levels shown as two vertical arrows are specific. The other direction-weighted-xcor (and effective xcor) is 1 · 0 (completely related). In this example, only a non-zero conversion factor is shown in Figure 9B as a single vertical arrow C, which is applied to the central internal channel c output, resulting in a tightly dominant signal in space. In this example it is centered (l / r = 1) and thus coincides with the central internal channel c. If there is no coincident output channel, the dominant signal is applied to the nearest output channel at an appropriate ratio and the 4 dominant signal is placed in the correct virtual position in between. For example, if there is no center output channel C, the left middle LM and right middle RM output channels will have a non-zero conversion factor, so that the dominant signal is equally applied to the LM and RM outputs. In the case of this complete correlation (all dominant signals), it has no padding and no end signal components. Therefore, the preliminary conversion factor generated by block 467 (Figure 4C) is the same as the conventional post-explicit conversion factor component generated by block 461. In Figure 10A, the input energy levels are equal, but 200404222 direction-weighted—XC0r is less than ΐ · 〇 and greater than rand〇m-xc〇r. As a consequence, the conversion factor components are a combination of dominant and filling conversion factor components belonging to the region. The larger of the normalized explicit conversion factor component (from block 461) and the normalized fill conversion factor component (from block 463) is applied to each output channel (using block 467), making explicit conversion The factors are placed in the same central output channel as in Fig. 10B, but are relatively small, and the filling conversion factor appears in each of the other output channels L, LM, ruler ... (the ends include L and R). -In Figure 11A, the input energy levels remain the same, but _ 10 direction_weighted-Xcor = random-xcor. The consequence is that the conversion factor of the first graph is a condition where the boundary conditions of areas 1 and 2 are uniformly filled. There is no explicit or endpoint conversion factor, only if each output is indicated by the same arrow in each Output a fill conversion factor with the same value (hence the "fill evenly"). The filling conversion factor components reach their highest value in this example. As discussed below, the fill conversion factor may be applied unevenly, such as in a progressively tilted manner depending on the input signal conditions. In Fig. 12A, the input energy levels remain equal, but the direction-weighted-xcor is smaller than the random-xcor and larger than 0 (area 2). The consequence is that as shown in Figure 12B, it has a filling and endpoint conversion factor, but no 20 dominant conversion factor. In Fig. 13A ', the input energy levels remain equal, but the direction-weighted-xcor is less than 0. The consequence is that the nose change factor shown in Figure 13B is one that belongs to all endpoint limits. It has no internal output conversion factor, only the endpoint conversion factor. 69 200404222 In the examples of Figures 9A / 9B to 13A / 13B, because the energy levels of the two inputs are equal, direction-weighted_xcor (as generated by block 441 in Figure 4B) and neighbor-compensated_xcor (as shown in Figure 4B) The producer of block 439 in the figure) is the same. However, in Figure 14A, the input energy levels 5 are not equal (L is greater than R). Although neighbor-weighted_xcor is equal to random_xcor in this example, the conversion factor results shown in Fig. 14B are not uniformly applied to the filling conversion factors of all channels (such as the examples in Figs. 11A and 11B). Instead, the unequal input energy levels cause a proportional increase in direction-weighted_xcor (proportional to the degree that the center of gravity's direction is offset from its _ 10 center position), making it larger than neighbor-compensated_xcor, causing the The equal conversion factors are weighted more explicitly toward all (as shown in Figure 8). Since strongly L and R weighted signals should not have a large width, they should have a tight width near the end of the L or R channel, so this is the desired result. The output result shown in Figure 14B is located at 15 and is closer to the L output than the R output (in this case, the direction information after the neighbor compensation happens to place the dominant component exactly at the left-middle LM position), reducing the filling conversion factor amplitude. A non-zero dominant conversion factor that has no endpoint conversion factor (this direction weighting pushes the operation into region 1 (mixed dominant and filling) in Figure 8). _ 20 For the five outputs corresponding to the conversion factors in Figure 14B, these outputs can be expressed as:

Lout = Lt (SFL)

MidLout = ((.92) Lt + (.38) Rt)) (SFMidL) Cout = ((. 45) Lt + (. 45) Rt)) (SFc) 70 200404222

MidRout = ((.38) Lt + (.92) Lt)) (SFMidR)

Rout = Rt (SFR) Therefore, in the example in Figure 14B, even if the conversion factor (SF) of each of the four outputs of non-MidLout is equal (filling), since Lt is greater than Rt (there is more than 5 signal outputs to the left Result) and the dominant output in the middle left is greater than that represented by the conversion factor, so the corresponding signal outputs are not equal. Since the directional center of the center of gravity coincides with the center-left output channel, the ratio of L0 ^ Rt is the same as the matrix coefficient of the center-left output channel, that is, 0.992 to 0.38. Assume these are the actual amplitudes for Lt and Rt. To calculate these output levels, we multiply these levels by 10 corresponding matrix coefficients, add them, and use the respective conversion factors: Output amplitude (out_channel_sub_i) = sf (i) * (Lt—Coeff (i ) * Lt + Rt—Coeff (i) * Rt) Although we better consider the mixture between amplitude and energy addition (as calculated in Figure 6A correlation), the cross correlation is quite high in this example (large dominant Change 15 calculation factors) and general addition can be implemented:

Lout = 0.1 氺 (1 氺 0.92 + 0 氺 0.38) = 0.092 MidLout = 0.9 氺 (0.92 * 0.92 +0.38 氺 0.38) = 0.900 Cout = 0.1 * (0.71 * 0.92 + 0.71 * 0.38) = 0.092 MidRout = 0_l * ( 0.38 * 0.92 + 0.92 ^ 0.38) = 0.070 20 Rout = 0.1 氺 (0 wood 0.92+ 1 book 0.38) = 0.038 Therefore, this example shows that because Lt is greater than Rt, even if the conversion factors of these outputs are equal, in Lout, MidLout , Cout, MidRout and Rout signal output are not equal. The filling conversion factors can be assigned to the output channels as shown in the 10B, 11B, 12B, and 14B. 71 = etc. Alternatively, the non-uniform filling conversion factor can vary from location to location as a function of explicit (correlated) and / or endpoint (irrelevant) input signal tillers (or equivalently as a function of direction_weighted_xcor value). For example, ^ dkeetiQn_weighted_x (^ For moderately high values, the amplitudes of these filling conversion factor components can be convexly curved so that the output channel in the direction near the center of gravity receives more signal levels than the channel far away. As for direction_weighted_xcor = dom-xcor, the filling components can be flat to a uniform distribution, and in the direction-weighted-xc0r < rand〇m-ca, the vibration axes can be curved concavely, while the The channel of the direction is favorable.

15B and 16B are erected by ax. Fig. 15B outputs the result from the same input (Fig. 15A) as the above hidden image. The first picture is output by one of the inputs i2A_Xiao (picture 16A). Communication between the board group and the governor—the communication between the neighboring office and the county level as shown in Figures 1 and 2. Each module in the multi-module configuration requires two institutions to support it and the supervisor as shown in Figure 2. Communication between 001: (The magic one is used to select and report the information required by the supervisor. The ten-neighbor neighbors compensate and the south-order neighbor compensation (if any). The ring required by the supervisor It is the total estimated internal energy attributable to the input of each module as generated by the configuration of FIG. 6A. ⑼ In addition, it receives and applies the neighbor level (the one on the right) and the station from the supervisor. Order neighbor level (if any). In Figure 4B] the transit neighbor compensation is smoothed by the mother-round energy level in the respective combiners 431 and 433, and the higher order neighbors are reduced. The compensation (if present on the right) is smoothed by each round in the respective combiners 431, 433, and 435. This level and the common energy in these channels are subtracted. Once a supervisor knows each The total estimated internal energy contribution of each input of the module: (1) It determines the total estimated internal energy contribution of each input (by all connected The modules to the input are added) Whether the total available signal level input here is exceeded. If the sum exceeds the total available signal level, the supervisor recalls each defendant reported to the input The internal energy is added to the total input level. (2) It notifies each module of its neighbor level at each input, and, as the sum of all other internal energy contributions of the input (if (If any). The higher-order (H0) neighbor level is the neighbor level of more than one high-two module sharing a lower-order module. The above calculation of the neighbor level is only for specific rotations with the same p white layer Related to the modules: all three-input modules (if any are alive) 'and then all two-input modules, etc. The H0 neighbor level of a module is all neighbor levels of all higher-order modules entered here (That is, in a second-round module—the rounded H0 neighbor level is the sum of all third, fourth, and higher-order modules (if any) that share the two-input module. The Yila group knows the level of the H0 neighbor in its particular rotation, and the total loss of the neighbor The energy level is subtracted from the neighbor level of the same level to obtain the level after the neighbor compensation of this input node. This is shown in 2004β222 in Figure 4β, where the neighbor levels of input 1 and input m The outputs of these variable slow smoothers 425 and 427 are subtracted in combiners 431 and 433, respectively, and the high-order neighbor levels of round 1 and input m are combined by these in combiners 431, 433, and 435, respectively. The output of variable slow smoothers 425 '427 and 429 is subtracted. 5 The difference between the used neighbor level and the HO neighbor level is that the HO neighbor level is also used to compensate the common energy of the entire input channel ( (For example, in the combiner 435, it is completed by subtracting the HO neighbor level.) The reason for this difference is that the common level of a module will not be affected by the use of adjacent modules of a level, but it will be shared by a A high-level module effect of all inputs of the module. 10 For example, suppose the input channels Ls (left surround), Rs (right surround), and top have an internal output channel (behind the raised ring) in the middle of the triangle in between, plus the line between Ls and Rs. An internal input channel (behind the main horizontal ring). The former output channel requires a three-input module to restore the common signal to all three inputs. Then, the output channel of the latter 15 on the straight line between the two inputs (Ls and Rs) requires a one-input module. However, the total common signal level observed by the two input modules includes the common elements of the three input modules that really belong to the latter's output channel, so I subtract the HO neighbor compensation from the common energy of the two input modules. The square root of the pairwise product to determine how much common energy is solely due to its internal channel (I mention the latter). Thus in Figure 4B, the smoothed common energy level (from block 429) has been subtracted from its derived HO common level to obtain a neighbor-compensated common energy level (from combiner 435). , Which is used by the module to calculate (at block 439) the neighbor-compensated xcor. It should be understood that the invention and other variations and modifications of its various aspects 74 200404222 will be apparent to those skilled in the art, and the invention is limited to the specific examples described. Therefore, it is intended to cover any and all aspects of the present invention. Modification of 4 liters or equivalent matters, which falls within the true spirit and realm of the principle disclosed and declared here. Soil foundation 5 [Schematic simple surface ^ Ming] Figure 1 is a plan view schematically showing a 16-channel horizontal array on a wall around a room, and a 6-channel array is placed on the horizontal array. One of the ideal test layouts for internal and a single suspended channel — the decoding configuration. 10 Figure 2 is a functional block diagram providing a multi-band conversion embodiment of several modules operated by a central supervisor implementing the example of Figure 1. Fig. 3 is a functional block diagram, which is useful in understanding the method by which the supervisor 201 of Fig. 2 can determine an endpoint conversion factor. Figures 4A-4C show a functional block diagram of a module according to one aspect of the present invention. Fig. 5 is a schematic diagram showing the Lu Yi hypothesis configuration of a triangle made of input channels, three internal output channels, and a three-input module fed in a dominant direction. Figures 6A and 6B are functional block diagrams, respectively showing a suitable configuration, 20 for (1) each input of a module in response to the total energy at each input, and the input generates a total estimated energy, And (2) generating an additional endpoint energy conversion factor component for the endpoints of each module in response to the cross-correlation measurements of the input signals. Fig. 7 is a functional block diagram showing a preferred function of the "summing and / or taking 75 200404222" box 367 of Fig. 4C. FIG. 8 is an idealized diagram of a method for generating a conversion factor component under a measurement in response to a cross correlation in one aspect of the present invention. Figures 9A and 9B to 16A and 16B show the output conversion factors of a module caused by the results of various input signal 5 status examples. [Representative symbol table of main components of the drawing] 1 ... output channel 15 ... output channel Γ ... input channel 16 ... output channel 2 ... output channel 17 ... output channel 3 ... output channel 18 ... output channel 3 '... input channel 19 ... output channel 4 ... output channel 20 ... output channel 5 ... output channel 21 ... output channel 5' ... input channel 22 ... Output channel 6 .. Output channel 23 ... Output channel 7 ... Output channel 23 '... Vertical channel 8 ... Output channel 24-Input decoding module 9 ... Output channel 25 ... Two input decoding module 10 ... output channel 26—input decoding module 11 ... output channel 27—input decoding module 12 ... output channel 28 ... two input decoding module 13 ... output channel 29 ... two input vertical decoding module 13 '... input channel 30 ... two-input vertical decoding module 14 ... output channel 31 ... two-input vertical decoding module 76 200404222 32 ... two-input vertical decoding module 431 ... combiner 33 .. Two-input vertical decoding module 433 ... combiner 201 ... supervisor 435 ... combiner 203 ... matrix 437 ... function or device 205 ... inverse transform 439 ... function Number or device 301 ... combiner 441 ... function or device 303 ... combiner 443 ... function or device 305 ... divider 445 ... function or device 307 ... block, take root 447 ... function Or device 401 ... function or device, block 449 ... function or device 403 ... function or device, block 451 ... function or device 405 ... function or device, block 453 ... function or device 407 ... function or device, block 455 ... function OR device 409 ... function or device, block 457 ... function or device 411 ... smoother 459 ... function or device 413 ... smoother 461 ... normalizer 415 ... smoother 463 ... normalizer 417 ... smoothing 465 ... Normalizer 419 ... Smoother 467 ... Function or device 421 ... Smoother 469 ... Matrix 423 ... Smoother 601 ... Multiplier 425 ... Smoother 603 ... Multiplier 427 ... Smoother 605 ... multiplier 429 ... smoother 607 ... multiplier 77 200404222 609 ... multiplier 611 ... combiner 613 ... combiner 615 ... multiplier 617 ... Multiplier 619 .. Multiplier 621 .. Multiplier 623 .. Multiplier 625 .. Combiner 627 .. Combiner Multiplier 629 ... Combiner 631 .. The divider 633 ... 701 ... square root .. 703, whichever is greater. Combiner

78

Claims (1)

200404222 Scope of patent application: 1. A process for converting each of the M audio input signals assigned in one direction to each of the N audio input and output signals assigned in one direction, where N is greater than M and M is 2 or more and N are positive integers greater than or equal to 3, including 5: providing a variable matrix of M × N, applying the M audio input signals to the variable matrix, and deriving the N audio outputs from the variable matrix Signals, and controlling the variable matrix in response to the input signals such that the sound field produced by the output signals when the input signals are highly correlated has a direction in the spatial center of gravity of the input signals One of the tight sound images, which gradually spreads from tight to broad as the correlation decreases, and gradually divides into multiple tight sound images as the correlation continues to decrease to a highly uncorrelated, each of which is being matched with An input signal 15 in one direction. 2. The process described in item 1 of the scope of patent application, wherein the M × N matrix is a variable matrix with variable coefficients or a variable matrix with fixed coefficients and variable outputs, and the variable matrix is changed by changing the The variable coefficients are controlled by changing these variable outputs. 20 3. The process as described in item 1 of the scope of patent application, wherein the variable matrix is controlled in response to the following measurements: (1) the relative levels of the input signals; and (2) the inputs Cross correlation of signals. 4. The process as described in item 3 of the scope of patent application, wherein in order to measure the cross-correlation of the input signals within a first range bounded by a maximum of 79 200404222 and a reference value, when the cross- When the correlation is the maximum value, the sound field has a tight sound image, and when the cross correlation is the reference value, the sound field has a widely diffused image; and in order to have a boundary bounded by a minimum value of 5 and a reference value A measurement of the cross-correlation of the input signals in a second range. When the cross-correlation is the minimum value, the sound field has several tight sound images, each in the direction of an input signal, and When the cross-correlation is the reference value, the sound field has a widely diffused image. 10 5. The process as described in item 4 of the scope of the patent application, wherein the reference value is approximately a cross-correlated measured value of the input signals in the case of equal energy in each output. 6. The process as described in item 3 of the scope of patent application, wherein a measurement of the relative levels of the input signals is a smoothed energy level in response to one of each input signal. 7. The process as described in item 3 or 6 of the scope of patent application, wherein a measurement of the relative level of the input signals is the spatial center of gravity of the input signals. 8. The process as described in item 3 of the scope of patent application, wherein a measure of the cross-correlation of the input signals is a common energy smoothed in response to one of the input signals divided by the input energy of each input signal. The smoothed M level root of the energy level, where M is the number of inputs. 9. The process as described in claim 6, 7, or 8, wherein the smoothed energy level of each input signal is obtained using time domain smoothing with a variable time constant. 80 200404222 10. The process as described in item 6, 7 or 8 of the scope of patent application, wherein the smoothed energy level of each input signal is obtained using frequency domain smoothing and time domain smoothing with a variable time constant. 11. The process described in item 8 of the scope of patent application, wherein the common energy of the input signals 5 is obtained by cross-multiplying the input amplitude levels. 12. The process described in item 11 of the scope of patent application, wherein the smoothed common energy of the input signals is obtained by time-domain smoothing of the common energy of the input signals by a variable time constant. 10 13. The process according to item 12 of the scope of patent application, wherein the smoothed energy level of each input signal is obtained by smoothing in a time domain with a variable time constant. 14. The process as described in item 11 of the scope of the patent application, wherein the smoothed common energy of the input signals is a time domain with frequency domain smoothing and variable time constants of the common energy of the input signals. Smoothness is obtained. 15. The process described in item 14 of the scope of the patent application, wherein the smoothed energy level of each input signal is obtained by smoothing in the frequency domain and the time domain with a variable time constant. 20 16. The process as described in claim 9, 10, 12, 13, 14, or 15 in which the time domain smoothing of the variable time constant is performed by having a fixed time constant and a variable time constant. The two are implemented by smoothing the time domain with a variable time constant. 17. As described in item 9, 10, 12, 13, 14, or 15 of the scope of patent application 81 200404222, wherein the time-domain smoothing of the variable time constant is made variable by only having a variable time constant. The time domain of the time constant is implemented smoothly. 18. The process as described in claim 16 or 17, wherein the variable time constant is variable in steps. 18. The process as described in claim 16 or 17, wherein the variable time constant is continuously variable. 19. The process as described in claim 16 or 17, wherein the variable time constant is controlled under a measurement that is related to the relative levels of the input signals and their cross-correlation. 19. The process according to item 6 of the scope of patent application, wherein the smoothed energy level of each input signal is smoothed by time domain of a variable time constant for each input signal having substantially the same time constant. given. 15 20. The process as described in item 3 of the scope of patent application, wherein the relative levels of the input signals and their cross-correlation measurements are each obtained by smoothing the time domain with a variable time constant, where the same time constant is Apply to each smooth. 21. The process as described in item 8 of the scope of patent application, wherein the cross-correlation amount 20 is measured as a first measurement of the cross-correlation of the input signals and one of the additional measurements is applied by applying A measurement of the relative level of the input signal and the first measurement to the cross-correlation are obtained by generating a weighted measurement of the direction of the cross-correlation. 22. The process described in item 21 of the scope of patent application, in which cross-correlation has an additional measurement of 82 200404222. In the case where each output has equal energy, a cross-correlation of approximately equal to the input signals is applied. A conversion factor of the measured value is obtained. 23. —A kind of processing is used to convert each of the M audio input 5 signals in one direction to each of the N audio input and output signals in one direction, where N is greater than M, M is 2 or more and N Is a positive integer greater than or equal to 3, including: providing a number of mxn variable matrices, where m is a partial set of M and η is a partial set of N, 10 applying the respective partial sets of these M audio input signals, A respective partial set of the N audio output signals is derived from each of the variable matrices, and each of the variable sets is controlled in response to the partial set of input signals applied to each of the variable matrices. Change the matrix so that when the input signals are highly correlated, the sound field produced by the output signals has a tight sound image in the direction of the spatial center of gravity of the input signals, and the image decreases with the correlation From tightly diffused to broad, and gradually divided into multiple tight sound images as the correlation continues to decrease to a high degree of uncorrelation, each of which is in one of the 20 directions with an input signal, and by N A partial set of audio output channels derives the N audio output signals. 24. The process as described in item 23 of the scope of the patent application, wherein the variable matrices are also controlled in response to information to compensate for the effects of more than one variable matrix that can receive the same input signal. 25. The process as described in item 23 or 24 of the scope of patent application, wherein deriving the N audio output signals from a partial set of ^ audio output channels includes compensating for generating multiple variable matrices of the same output signal. 26. The process as described in claim 23, 24, or 25, wherein each of these variable matrices is controlled in response to the following measurements: (1) the relative bit of the input signal applied to it Standards; and (2) cross-correlation of these input signals. 27 · —A kind of processing is used to convert each of the M audio input signals provided with one direction into each audio input and output signal provided with one direction, where N is greater than M, M is greater than 2 and N is greater than or equal to 3 A positive integer, including: providing a MxN variable matrix in response to a control matrix coefficient or a conversion factor of a control matrix output, applying the M audio input signals to the variable matrix, and providing several mxn variable matrix conversion factor generators Here, 111 is a partial set of M and 11 is a partial set of ^, applying a separate set of the audio input signals to each of the variable matrix conversion factor generators, and each of the variable matrix The conversion factor generator derives a set of variable matrix conversion factors for the respective sets of the 1 ^ audio output signals, and responds to a set of parts of the input signal generated in response to each of the variable matrix conversion factors. Control each of these variable matrix conversion factor generators so that when the conversion factor generated by it is applied to the variable matrix, each part of the output signal is output When the combined field is highly correlated, it has a compact sound image in the direction of the spatial center of gravity of the partial set of input signals that are converted by the conversion factor. The image is closely related as the correlation decreases. Diffusion is broad 'and progressively segmented into multiple tight sound images as the phase continues to decrease to a high degree of irrelevance, each of which is one of a round of incoming signals that are configured to generate the conversion factor applied by the child Up, and N audio output signals are derived from the variable matrix. 28. The process described in item 27 of the scope of the patent application, wherein the variable matrix conversion factor generators are also controlled in response to information in order to compensate for the effects of receiving more than one variable matrix of the same input signal. 29. The process according to item 27 or 28 of the scope of patent application, wherein the N audio output signals derived from the variable matrix include compensation for a multiple variable matrix conversion factor generator that generates a conversion factor for the same output signal . 30. The process described in item 27, 28, or 29 of the scope of patent application, wherein each such variable matrix is controlled in response to the following measurements: (1) the relative bit of the input signal applied to it Standards; and (2) cross-correlation of these input signals.