JP4321625B2 - Digital filter processing method and digital filter apparatus - Google Patents

Description

  The present invention relates to a digital filter processing method and apparatus for performing digital filter operation intermittently for each of a plurality of samples.

  In recent years, various sound image localization apparatuses have been developed in order to realize realistic sound. For example, the 5.1 channel system has been standardized as a system that assumes movies and music concerts. In this system, a three-channel speaker is arranged in front of the viewer and a two-channel speaker is arranged behind the viewer, and the volume and phase generated by each speaker are adjusted according to the movement of the sound source. The

  As another method, a head related transfer function HRTF (Head Related Trausfer Function) from the sound source to the left and right ears is measured in advance in association with the coordinates of the sound source, and as the coordinates of the sound source change, A three-dimensional sound positioning system that dynamically changes the coefficient of the head-related transfer function HRTF is known. In this 3D sound positioning system, the speaker and the amplifier used for sound generation need only two channels corresponding to the left and right ears, so that the system can be configured simply.

  FIG. 7 is a functional block diagram of a general three-dimensional sound positioning system. In this example, three sound sources are used. In the figure, SG1 to SG3 are sound sources, and generate waveform data W1 to W3, respectively. F1 to F3 are FIR (finite impulse response) filters, and their characteristics represent the head-related transfer function HRTF from the virtual sound source to the left and right ears. Here, if the sound sources SG1 to SG3 are virtually moved, the coefficients of the FIR filters F1 to F3 are dynamically changed according to the coordinates.

  A1 and A2 are adders, which add data from the FIR filters F1 to F3. Thereby, the waveform data of the L and R channels can be obtained respectively. By the way, the waveform data output from the adders A1 and A2 is generated by the head related transfer function HRTF on the premise that it is directly input to the left and right ears. Therefore, there is no problem if the sound generation means is close to the left and right ears as in headphones, and the sound from one sound generation means is heard only by one ear, but the L channel and R channel speakers are connected to the viewer. If the speaker is installed in front of the speaker, sound is input to both ears due to the sound of one speaker, resulting in crosstalk. The crosstalk processing unit C is used to generate L channel waveform data DL and R channel waveform data DR that have been corrected in advance so as to eliminate the crosstalk in the sense of hearing. The processing up to the adders A1 and A2 is processing for each sound, while the crosstalk processing is processing common to each sound.

  Next, FIG. 8 is a block diagram of the FIR filter F1. If this apparatus is provided in, for example, a personal computer, the FIR filter F1 is connected to a host driver (not shown) via a PCI bus or the like, and controls the DSP 100 that performs a product-sum operation and the DSP 100. The CPU 200, the working memory 210 connected to the CPU 200, and the HRTF coefficient data memory 110 storing the HRTF coefficient data h are configured. With the above configuration, when the coordinate command P is transferred from the host driver to the FIR filter F1 via the bus, the CPU 200 uses the work memory 210 to store a storage area for HRTF coefficient data h corresponding to each tap coefficient of the filter. Is calculated, and the address is sent to the DSP 100. Thereafter, the DSP 100 reads out the HRTF coefficient data h from the HRTF coefficient data memory 110 based on the transferred address, performs a product-sum operation between the HRTF coefficient data h and the waveform data W1, and performs processing. Completed waveform data W1 ′ is generated.

  By the way, the direction accuracy for locating the sound source depends on the data amount of the HRTF coefficient data h. For this reason, if the accuracy of sound image localization is improved in order to produce sound with a sense of presence, it is necessary to provide a large-capacity HRTF coefficient data memory 110 inside the FIR filter F1.

  In the conventional system, it is necessary to interpret the coordinate command P transferred from the host driver and select the HRTF coefficient data h. Therefore, it is necessary to provide the CPU 200 as a controller inside the FIR filter F1. there were.

  On the other hand, on the host driver side, a CPU for controlling the entire personal computer, a large-capacity main memory, and the like are often provided. For this reason, the conventional system cannot effectively use the circuit configuration on the host driver side, and there is a problem that the circuit scale of the FIR filter becomes large.

  The present invention has been made in view of the above-mentioned points, and one of its purposes is a digital filter processing method and apparatus that operates by being connected to a computer bus without increasing the circuit scale. It is an object of the present invention to provide a digital filtering method and apparatus that can be operated efficiently.

  In order to achieve the above object, a digital filter processing method according to the present invention performs burst transfer of a plurality of samples of digital data to be filtered and coefficient sets used in the filter processing via a computer bus. And a second step of performing a filtering process on the plurality of samples of digital data that have been burst transferred using the coefficient group that has been burst transferred at a processing rate faster than a predetermined sampling period. And a third step of buffering the filtered digital data and outputting in accordance with the predetermined sampling period.

  Since a plurality of samples of digital data to be filtered are collectively transferred via a computer bus, the computer does not need to send out digital data in synchronization with a predetermined sampling period, and is not time-constrained. Therefore, the computer and its bus can be efficiently used for other purposes. In addition, since the filter processing in the second step performs filter calculation processing according to a predetermined filter function at a processing rate faster than a predetermined sampling period, for example, to perform filter processing on digital data of a plurality of f samples. The filter operation is performed quickly without taking the time required for the f sampling period that is normally required. Thus, in the first step, for each block of digital data consisting of a plurality of samples, a plurality of samples of the digital data to be filtered can be intermittently transferred via the computer bus. Then, filter operation processing can be performed intermittently for each block of digital data composed of a plurality of samples. Data that has been subjected to the filter calculation process is successively stored in the buffer in the third step, and output according to a predetermined sampling period, whereby normal data that has been subjected to the filter calculation process according to the predetermined sampling period can be obtained.

  In the second step, since intermittent filter operations are performed, not only filter operations for one channel but also filter operations for a plurality of channels can be performed using empty time, and the circuit scale is increased. Independent filter operation processing can be performed for many channels.

  In the first step, a series of digital data to be filtered is divided into a plurality of sections, and a digital data block composed of a plurality of samples (for example, f) in each section is intermittently transferred. Samples larger than the number of samples (f) for one section by transferring a part of digital data (for example, n samples) belonging to the section adjacent to the section when transferring the digital data for the section. Time-series digital data consisting of a number (f + n) is transferred. Thus, in the second step, by using time-series digital data having a number of samples larger than the number of samples for one section, and performing filtering processing according to the predetermined filter function, at least one section is obtained. Filtered data corresponding to the number of samples (f) is obtained. In this way, intermittent filter calculation processing can be performed. That is, if the number of sets of filter coefficients is k, normally, a so-called k−1 order filter operation including k−1 unit delays is performed, and a product sum of k coefficients. Thus, filtered data for one sample is obtained. Therefore, in order to obtain filtered data for f samples, time-series digital data to be filtered should be input for at least f + k samples. Thus, by setting n ≧ k as the number of samples of the data in the adjacent section that is redundantly transferred in the above, it is possible to obtain filtered data corresponding to the number of samples (f) for at least one section. Can do. The number of samples n of the data in the adjacent section transferred in excess is used when the filter calculation process for the section is next performed. For this purpose, it may be stored appropriately at the stage of the second step, but instead, it may be retransmitted at the time of transfer processing at the first step for the section. Preserving this in the filter operation processing stage of the second step requires a corresponding register, so that the circuit scale is somewhat increased, but such retransmission processing is performed in the first step. This has the advantage that it can be easily performed without requiring an extra register as a computer process.

  As a part of the digital data belonging to the adjacent section transferred in the first step, digital data composed of a predetermined number of samples (for example, j) for interpolation may be further included. Thereby, in the second step, filtered data including the number of samples (for example, f + j) obtained by summing the number of samples for one section and the predetermined number of samples for interpolation is obtained. Further, a fourth step of buffering the filtered data corresponding to the predetermined number of samples for interpolation (j) for the next section, and the predetermined number of samples for interpolation (j in the previous section) And a fifth step of performing a predetermined interpolation process using the filtered data corresponding to the number of samples (f) corresponding to one section for the current section. .

  In this way, with respect to j sample points that overlap, the filter calculation result according to the filter coefficient used in the previous section and the filter calculation result according to the filter coefficient used in the current section are interpolated, and j data is obtained as the interpolation result. can get. These j interpolated data are output through the third step as filtered and interpolated data for a total of k samples together with other k−j filtered data. As a result, even if the content of the coefficient group used in the filter operation for the previous section is different from the content of the coefficient group used in the filter operation for the current section (that is, even if the filter coefficient varies over time), both It is possible to smoothly connect the data subjected to the filter calculation processing in the section by interpolation.

  Preferably, in the first step, coefficient sets used in the filtering process are also transferred through the bus of the computer. In the second step, the filtering process may be performed using the transferred coefficient set. As a result, it is not necessary to prepare a memory for storing a plurality of sets of filter coefficients in advance on the second step side, that is, the filter calculation processor (for example, DSP) side, and the circuit scale is increased. There is nothing. For example, when changing filter coefficients from time to time, a large capacity coefficient memory is required. However, if only a necessary coefficient set is transferred via a bus, such a large capacity memory is used for filter operation. It is not necessary to provide it on the processor side, and it is sufficient to provide it on the host computer side. Further, even when the filter coefficient is not changed over time, as in the case of sound image localization control, when analyzing the coordinate position of the sound image localization and performing a process of selectively reading out the filter coefficient set corresponding to the analyzed coordinate position, If the coefficient memory is provided on the host computer side, it can be easily performed on the complicated host computer side for analysis, and the configuration on the processor side for the filter operation can be simplified also in this respect.

  Of course, the coefficient set used for the filter operation processing in the second step may be stored in a memory or the like on the processor side for the filter processing and may not be transferred via the bus. For example, if the filter coefficient does not change over time, it is sufficient that at least one set of such coefficients is stored, so that not much memory capacity may be required. Alternatively, when complicated coordinate analysis for sound image localization is unnecessary, such analysis processing is unnecessary. In these cases, the coefficient group used for the filter calculation process can be stored in advance in a memory or the like on the filter processing processor side.

  As an embodiment, the first step is executed by a program on the general-purpose processor side of the computer, and the second and third steps are executed by a program on the filter processing processor side connected to the bus. Thus, the present invention does not waste a high-capacity host computer for filtering, so that it can be used effectively, and the circuit scale and software scale of the filtering processor are increased. Therefore, it is possible to simplify and efficiently perform digital filter processing.

  According to another aspect, the digital filter device connected to a computer bus according to the present invention is subjected to filter calculation processing at any one or a plurality of points in a predetermined one frame period via the bus. An interface unit that receives burst transfer of a plurality of digital data samples and a filter coefficient set used in the filter calculation process, and the digital data received via the interface unit via the interface unit A filter operation unit that performs a predetermined filter operation process using the received filter coefficient group; and an output unit that buffers and outputs an output of the filter operation unit.

  In an embodiment, the filter calculation unit performs a filter calculation process asynchronously with a predetermined sampling period or at a high speed. On the other hand, the output unit outputs the buffered filtered data in synchronization with a predetermined sampling period. The interface unit receives digital data of a predetermined number of samples for each of a plurality of channels at different time points within the one frame period, and the filter operation unit receives a filter for each channel. Arithmetic processing is performed at high speed within the one frame period. In this case, the output unit buffers the filtered data for each channel while accumulating the data at the same sample time. Further, the interface unit also receives a filter coefficient set used in the filter calculation unit via the computer bus, and the filter calculation unit uses the received filter coefficient set to perform the predetermined filter calculation. Perform processing.

  The digital filter processing method and / or apparatus proposed in the present invention can be applied to various fields as well as processing of digital musical tone signals. That is, it is optimal as a method and / or apparatus for performing efficient filter processing by connecting to a computer, and exhibits excellent effects in terms of both time utilization efficiency of the computer and reduction of the circuit scale on the filter processor side. In the present invention, a computer does not necessarily indicate only a stand-alone type computer device / equipment / equipment such as a personal computer or a large host computer, but a CPU incorporated in various application devices / equipment / equipment. Are used in a broad sense, including microcomputer devices and chips whose main components are. For example, a microcomputer device is incorporated in an electronic game device, a specific application device / equipment, or some kind of multimedia device, and a filter processor (for example, DSP) is connected to the bus. Such forms are also included in the scope of the present invention.

  For example, by applying the present invention to sound image localization control, a high-quality sound image localization effect can be imparted to a musical sound signal without increasing the circuit scale. That is, the sound image localization apparatus according to the present invention is a sound image localization apparatus that divides waveform data into frames for each f sample, and intermittently gives a sound image localization effect to the waveform data on a frame-by-frame basis. A first storage unit that stores coefficient data corresponding to information in advance; a second storage unit that stores the waveform data; and a k-tap FIR that generates output data based on the coefficient data and the waveform data A filter operation unit, connected to the first storage unit, the second storage unit, and the FIR filter operation unit, a bus corresponding to burst transfer, and k number of the corresponding to the position information of the current frame Coefficient data is read from the first storage unit, and f + k or more waveform data is stored in the second storage unit so that k or more waveform data overlap between adjacent frames. Reading for each over arm unit, the waveform data and the coefficient data read out, characterized in that a control unit for controlling to burst transfer to the FIR filter operation unit via the bus. Here, the control unit may perform control so that the burst transfer is performed at a short cycle such that the movement of the sound source cannot be detected from an auditory sense. The position information may indicate the position of the sound generation at a timing near the center of the frame.

  The present invention can be configured and implemented not only as a method invention but also as a device invention. In addition, the present invention can be implemented in the form of a computer program or a microprogram for a DSP or the like, or in the form of a recording medium storing such a computer program or microprogram.

  As described above, according to the invention specific matter according to the present invention, since input data overlapping in adjacent frames is transferred to the digital filter, the digital filter can be operated intermittently.

  FIG. 1 is a block diagram of a three-dimensional sound positioning system which is an embodiment of a sound image localization apparatus according to the present invention. Reference numeral 1 denotes a CPU which controls the entire apparatus. M is a main memory of the system connected to the CPU 1, and is constituted by, for example, an EDODRAM (Extended Data Out DRAM). The storage area in which the main memory M is located functions as the waveform buffer memory M1 and the HRTF coefficient data memory M2. In addition, a driver program and an application program are stored in another storage area of the main memory M. These data and programs are stored in a secondary storage device such as a hard disk when the computer system is started or when a specific command is input. It is read out and stored in the main memory M.

  The waveform buffer memory M1 stores each waveform data W generated by a plurality of sound sources. The sampling rate of the waveform data W is 48 KHz, for example. The HRTF coefficient data memory M2 stores HRTF coefficient data h in association with the coordinate command P for designating the coordinates of the sound source, and the HRTF coefficient data h corresponding to the coordinate command P is read in accordance with the read command of the CPU1. It is supposed to be.

  Thus, the waveform buffer memory M1 and the HRTF coefficient data memory M2 are configured as a part of the main memory M of the system. Therefore, in this system, it is not necessary to provide a special memory in advance on the digital filter computer 3 side in order to store the waveform data W and the HRTF coefficient data h.

  An example of the waveform data W is shown in FIG. In this example, continuous waveform data W is divided into f sample frames, and the effect imparting process is performed in units of frames. If the sampling rate is 48 KHz and one frame is composed of 256 samples, one frame period is approximately 5.3 ms as shown in the figure.

  Next, reference numeral 2 denotes a bus capable of high-speed data transfer. For example, a PCI bus is preferably used. The PCI bus has a data / address transfer of 32 bits (or 64 bits). The operation clock of the PCI bus is 33 MHz, and the theoretical maximum data transfer rate is 132 Mbytes / second (or 264 Mbytes / second). Furthermore, with the supported bus master function, it is possible to reduce the load on the high-speed DMA and CPU that do not use the general-purpose DMA controller.

  The PCI bus also has a burst transfer mode that is not supported by the ISA bus. The burst transfer mode is a mode in which a plurality of data is transferred continuously in a single address specification. By using this burst transfer mode, a DRAM (Dynamic RAM) having a burst transfer mode described later is used. Reading of continuous data from can be speeded up. Further, the PCI bus has an advantage that an interface for connecting to the bus can be manufactured at a low cost. The above is the reason for using the PCI bus. If it has such characteristics, an expansion bus other than the PCI bus may be used.

  Next, reference numeral 3 denotes a digital filter computer connected to the bus 2, which performs a high-speed product-sum operation. For example, a simple FIR filter operation is performed. The digital filter calculator 3 is constituted by a DSP (Digital Signal Processor). A functional block of the digital filter calculator 3 is shown in FIG. In the digital filter computer 3 shown in FIG. 3, data is exchanged with the bus 2 via the bus interface 301. The wave input buffer 302 buffers and stores waveform sample data (for example, composed of f + k + j samples) for one transfer unit among the data transferred in burst via the bus 2. The coefficient input buffer 303 stores filter coefficient data (for example, k pieces of coefficient data) for one transfer unit among data transferred in burst via the bus 2. The control unit 304. The digital filter calculator 3 controls the exchange of data between the units and other controls. The data stored in the wave input buffer 302 and the coefficient input buffer 303 is supplied to the filter operation unit 305, and the supplied waveform is supplied. Based on the sample data and the coefficient data, a filter operation according to a predetermined algorithm is executed by the filter operation unit 305 at high speed.

  FIG. 4 is a functional block diagram showing a typical example of the filter calculation algorithm executed by the filter calculation unit 305, and performs simple FIR (finite impulse response) filter calculation. In this figure, 311 to 31k-1 are D flip-flops that operate in synchronization with the sample clock. Reference numerals 320 to 32k-1 denote multipliers, in which HRTF coefficient data h0 to hk-1 are set, and multiplication with the waveform data W is performed. An adder 33 adds the output data of the multipliers 320 to 32k-1. With the above configuration, an FIR filter having k taps is configured. By providing two systems of these FIR filters based on HRTF coefficient data corresponding to the left and right, the sound effect processing is performed on the waveform data W. Thereby, it is possible to give the waveform data W an effect as if the sound source has moved, and to produce a three-dimensional presence.

  Reference numeral 34 denotes a cross-fade processing unit, which performs cross-fade processing on the output data of the adder 33. The cross fade process will be described later. The filtered data (and the crossfade processed data) output from the crossfade processing unit 34 is cumulatively taken into one of the output buffers 306 and 307 formed of a double buffer shown in FIG. That is, the output buffers 306 and 307 each have the number of address locations corresponding to the number f of waveform data samples for one frame, and store the filter output data for each sample in each address location. Here, the filter calculation unit 305 in the digital filter calculator 3 performs the filter calculation for one frame for a plurality of channels independently and at high speed, and the filter output data for each sample for each channel. Is added (accumulated) to each corresponding address location in one of the output buffers 306 and 307 formed of a double buffer. Thus, when the high-speed filter operation for one frame is completed for all channels, the filter operation result for each channel for the number of f samples is stored in each address location in one of the output buffers 306 and 307 in the write mode. That is, the accumulation of the same sample times is accumulated. The output buffers 306 and 307 formed of a double buffer are controlled so that the write mode and the read mode are alternately switched every frame. That is, one of the output buffers 306 and 307 that was in the write mode in the previous frame period is in the read mode in the current frame period, and the accumulated data of the filter operation results of each channel for the number of f samples is set to a predetermined value. Read sequentially according to the sampling frequency (for example, 48 kHz). In order to generate two left and right filter processing outputs, output buffers 308 and 309 made up of another double buffer are provided separately from the output buffers 306 and 307 made up of a double buffer. The filter calculation unit 305 may be shared by the left and right two-line filter processing, or may be provided in another system.

  Here, the plurality of channels correspond to channels that use different filter coefficient groups (so-called filter processing channels). For example, in a three-dimensional sound positioning system, when waveform data from three sound sources (3-channel waveform data) is filtered with the corresponding head-related transfer function HRTF, 3 channels for the left system and 3 for the right system Channel filtering is performed using the respective HRTF coefficient data. The buffers 302, 303, and 306 to 309 each need only the number of addresses required to buffer-store data for one frame for one channel, and therefore the total capacity is extremely small. That is, even when the filter processing for a plurality of channels is performed, the digital filter calculator 3 performs the filter calculation processing separately for each channel at a high speed (that is, at a speed higher than the output sampling frequency) within one frame time. Therefore, each of the input buffers 302 and 303 need only have a capacity for buffer storing data for one channel necessary for the filter calculation for one frame at that time. Further, each of the output buffers 306 to 309 only needs to have a capacity for buffer storing data of the number of samples (f) for one frame.

  Here, data transferred from the main memory M to the digital filter computer 3 is transmitted intermittently via the bus 2. FIG. 5A shows a state of burst transfer corresponding to processing of one sound. As shown in the figure, in one burst transfer, HRTF coefficient data h corresponding to the number of taps k of the digital filter calculator 3 and waveform data W of a predetermined number of samples (f + k + j) larger than k are transmitted via the bus 2. It is transferred to the digital filter calculator 3.

  As described above, in the present embodiment, since the data is transferred by burst transfer, the bus 2 is not always occupied to give the acoustic effect. In the example shown in FIG. 5 (a), the waveform data W corresponding to one sound source and one system of HRTF coefficient data h0 to hk-1 are transferred. For one sound source, waveform data W (common) and two systems of HRTF coefficient data corresponding to the left and right ears are transferred. Furthermore, since the bus 2 can transfer data at a high speed like the PCI bus, when the effect imparting process is performed on a plurality of sounds, the transfer data can be time-division multiplexed and transferred. Thereby, it is possible to perform sound image localization processing on sounds from a plurality of sound sources.

  By the way, since the HRTF coefficient data h0 to hk-1 is determined by the head related transfer function HRTF, it continuously changes as the sound source moves. Therefore, it is desirable to update the HRTF coefficient data h0 to hk-1 every sampling period. However, in order to update every sample period, it is necessary to read out the HRTF coefficient data h0 to hk-1 from the HRTF coefficient data memory M2 in a short time, and to transfer the read data via the bus 2. is there. In this case, since the bus 2 is occupied for a long time, there is often a contention for the bus 2 between the transfer processing of the HRTF coefficient data h and other processing. On the other hand, human hearing has a characteristic that even if the sound source moves discontinuously at short time intervals, it cannot be detected and the sound source feels moved continuously. In other words, when the sound source moves discontinuously over a certain period of time, the human ear first detects that it is discontinuous.

  Therefore, in the present embodiment, the number of coefficients necessary for data transfer is reduced by dividing the filter processing into fixed intervals (frame periods) and treating the head-related transfer function HRTF as not changing within a fixed period. In addition, the occupied time of the bus 2 is reduced. That is, the head-related transfer function HRTF is updated at a time interval such that no change in audibility is recognized. This time interval is preferably about 10 ms or less.

  In this example, the update period of the HRTF coefficient data h0 to hk-1 in the digital filter computer 3 is set to 10 ms or less so as to coincide with the burst transfer period (that is, the frame period). For example, if the sampling frequency is set to 48 KHz and the number of samples in one frame is set to 256 samples, the update cycle is approximately 5.3 ms. When the sound source moves at high speed, the change in the head-related transfer function HRTF becomes large, and it may happen that the sound source is felt by the human ear so that the movement of the sound source becomes discontinuous. For this reason, in the present embodiment, a cross-fade process, which will be described later, is performed so that the sound source can be felt to move continuously with human ears.

Next, the operation of this embodiment will be described with reference to the drawings. First, burst transfer processing will be described. As shown in FIG. 5B, the transfer data group Fn transmitted from the driver side via the bus 2 in the nth frame is as follows.
Fn = (hn, 0, hn, 1,... Hn, k-1, df (n-1), df (n-1) +1, ... dfn-1 + k + j)
Here, f is the number of samples per frame, and j is the number of overlaps.

  As shown in this figure, the transfer data Fn is composed of k pieces of HTRF coefficient data hn, 0 to hn, k-1 and f + k + j pieces of waveform data df (n-1) to dfn-1 + k + j. It is configured. Here, the reason why the number of waveform data to be transferred is f + k + j, which is larger than the number of samples f in one frame, is as follows. First, in the FIR filter processing, output data of a certain sample is generated by multiply-adding the waveform data W and the HRTF coefficient data h corresponding to the number of taps k of the filter. Therefore, in order to generate output data for one frame period (f samples), f + k pieces of waveform data W are required. Furthermore, j pieces of waveform data are used for the overlap processing for the cross-fade processing. Therefore, the number of waveform data to be transferred is f + k + j. That is, in this example, the simple FIR filter calculator 3 can perform intermittent processing by retransmitting the waveform data W having the number of samples f or more in one frame period.

Next, an example of FIR filter processing in the calculation unit 305 will be described. In a certain frame, the arithmetic unit 305 processes f + j, which is obtained by adding the number of samples f per frame and the number of samples j for overlap processing. Here, in the n-th frame, when a vector having each output data of the adder shown in the figure as an element is Xn, the vector Xn is expressed by the following expression.
Xn = (Hn.D [f (n-1): f (n-1) + k-1], Hn.D [f (n-1) +1: f (n-1) + k], ... Hn ・ D [fn + j: fn-1 + k + j])
= (Xn, 0, xn, 1, ... xn, f + j-1)
However, “·” represents an inner product of vectors.

Here, the relationship between the sample order and the data is defined. In the n-th frame, HRTF coefficient data h0 to hk-1 having k taps are represented by (hn, 0, hn, 1, hn, 2,... hn, k-1), and each coefficient data is an element. The coefficient vector Hn is assumed to be given by the following equation:
Hn = (hn, 0, hn, 1, hn, 2,... Hn, k-1)

Further, a vector D having the waveform data of successive samples as an element is defined by the following equation. However, d0, d1, d2,... Dm are waveform data in each sample.
D = (d0, d1, d2, ... dm)

Further, a vector having only the a-th waveform data da as an element is represented by D [a], and a vector having elements a to b as waveform data da to db is represented by D [a: b]. Assume that D [a] and D [a: b] are given by the following equations.
D [a] = daD [a: b] = (da, da + 1,... Db)

Using the above definition, the vector Xn is specifically considered. First, when the transfer data shown in FIG. 5B is supplied to the simple FIR filter 3 in order from hn, 0, the HRTF coefficient data hn, 0, hn, 1,... Hn, k-1 are shown in FIG. The multipliers 320 to 32k-1 are set in order. Thereafter, the inner product is obtained between the k waveform data df (n-1) to df (n-1) + k-1 and the HRTF coefficient data hn, 0, hn, 1,... Hn, k-1. The output data xn, 0 is generated by calculation. Therefore, xn, 0 is given by the following equation.
xn, 0 = hn, 0 df (n-1) + hn, 1 df (n-1) +1 + ... + hn, k-1 df (n-1) + k-1

In the next sample period, the waveform data is shifted by one sample and the calculation is performed to generate output data xn, 1 represented by the following equation.
xn, 1 = hn, 0 df (n-1) + 1 + hn, 1 df (n-1) +2+... + hn, k-1df (n-1) + k, and similarly, the waveform data is sequentially shifted while calculating As a result, output data xn, 0 to xn, f + j−1 is output from the adder 33.

  Next, the crossfade process performed by the crossfade processing unit 34 of the digital filter computer 3 will be described. FIG. 6 is a conceptual diagram for explaining the cross-fade process. The nth frame output data xn, 0 to xn, f + j-1 and the (n + 1) th frame output data xn + 1,0 to xn + 1, f + j-1 are as shown in FIG. Are generated based on the waveform data df (n-1) to df (n + 1) -1 + k + j. As can be seen from this figure, the waveform data dfn to dfn-1 + k + j are used redundantly in the nth frame and the n + 1th frame. As a result, j pieces of output data xn, f to xn, f + j-1 and output data xn + 1,0 to xn + 1, j-1 are generated in accordance with different HRTF coefficient data. .

Based on the output data xn, f to xn, f + j-1 and the output data xn + 1,0 to xn + 1, j-1, crossfade processing between the nth frame and the n + 1th frame Is done. Here, if a vector having the output data of the crossfade processing unit 34 in the (n + 1) th frame as an element is Yn, Yn + 1 is given by the following equation.
Yn + 1 = ((1-aβ) Xn [f + a-1] + aβXn + 1 [a-1], Xn + 1 [j: f])
However, β = 1 / j. Moreover, a represents the order of data to be overlapped, and is a natural number equal to or less than j.

For example, if j = 4 and f = 256, the output data yn + 1,0 to output data yn + 1, j-1 corresponding to the overlap portion of the (n + 1) th frame are as shown in FIG. It can be represented by the following formula.
1st (a = 1); yn + 1,0 = 0.8xn, 256 + 0.2xn + 1,0 2nd (a = 2); yn + 1,1 = 0.6xn, 257 + 0.4xn + 1 , 1 3rd (a = 3); yn + 1,2 = 0.4xn, 258 + 0.6xn + 1,2 4th (a = 4); yn + 1,4 = 0.2xn, 259 + 0.8xn +1,3

  That is, as the order of data in the overlap portion advances, the influence of the output data of the nth frame decreases, and conversely, the influence of the output data of the (n + 1) th frame increases. Thereby, even if the sound source moves at high speed, it can be felt that the sound source has moved naturally in terms of hearing. Of the data generated by the overlap for the cross-fade process, the data corresponding to the overlap of the preceding frame (for example, the j pieces of filter output data xn, f to xn, f + j-1 of the nth frame) ) Is stored in a buffer as appropriate so that it can be used in the cross-fade calculation performed in the next frame period. A buffer storage area for this purpose may be provided inside the cross-fade processing unit 34 (there should be a buffer area of “number of channels × j”) or in the main memory M via the bus 2. It may be transferred once and stored in the buffer in the main memory M, and burst transferred to the filter computer 3 side together with other data in the next frame period. That is, the filter output data xn, f to xn, f + j-1 for crossfading processing (that is, for overlap) obtained at the nth frame is appropriately buffered and stored until the next n + 1th frame. Crossfade synthesis is performed with the first j pieces of filter output data xn + 1,0 to xn + 1, j-1 of the (n + 1) th frame. This crossfade result is the output data for the first j of the (n + 1) th frame that has been filtered (and crossfade processed), and the subsequent fj filter output data xn + 1, j to xn + 1, Along with f-1, a total of f filtered (and crossfade processed) output data for the (n + 1) th frame is sent to the output buffer 306 (or 307) and added to each of the f address locations. .

  As described above, according to the present embodiment, in order to perform the filtering process in units of frames obtained by dividing the waveform data W into a predetermined number of samples, waveform data having the number of filter taps K or more is retransmitted in the next frame. Thus, although the data is burst transferred to the digital filter calculator 3 via the bus 2, the filter processing can be performed on the continuous waveform data.

  Further, the waveform data W and the HRTF coefficient data h are stored and held on the driver side, and are divided and transferred as necessary via the bus 2 for each frame. However, it is not necessary to previously provide a memory for holding all the waveform data W and all the HRTF coefficient data h that fluctuates over time. Further, the digital filter computer 3 only needs to perform normal filter processing for executing product-sum calculation at a predetermined timing, which accompanies the filter processing. Therefore, control means for interpreting and executing coordinate commands for sound image localization control There is no need to provide specially. For this reason, for example, when the digital filter computer 3 is housed in a computer expansion board as a sound source device, the configuration of the sound source device, that is, the expansion board can be simplified, and the manufacturing cost can be greatly reduced.

  In the embodiment described above, the waveform data W and the HRTF coefficient data h are burst-transferred for each frame period, and the FIR filter process is performed using the same HRTF coefficient data h in the same frame. In this case, since the HRTF coefficient data h is updated in the burst transfer cycle, theoretically, the sound source has moved discontinuously in the burst transfer cycle. However, in the human ear, even if the sound source moves discontinuously at a certain speed, it is detected as if it moved continuously. Therefore, as described above, the burst transfer cycle (frame cycle) is set to a short cycle so that when the sound source moves at a normal speed, it is felt that the sound source has moved continuously. Can be obtained.

  Furthermore, in the above-described embodiment, in burst transfer, in addition to waveform data (f pieces) for one frame and waveform data for retransmission (k pieces) for FIR filter processing, waveforms for overlap processing are used. Since the data (j pieces) are burst transferred, the FIR filter processing using different HRTF coefficient data is performed on the j pieces of waveform data in adjacent frames. Since the filtered data is subjected to crossfading, it is felt that the sound source has moved more continuously in the human ear. Thereby, even if the sound source moves at a high speed, it can be felt as if it has moved continuously.

  The embodiment according to the present invention has been described above. However, the present invention is not limited to the above-described embodiment, and various modifications described below are possible. In the above-described embodiment, the HRTF coefficient data h is updated in the frame period in the simple FIR filter calculator 3 by burst-transferring the HRTF coefficient data h in units of frames. In this case, if the HRTF coefficient data h in a certain frame is generated at any timing during the frame period, it can be handled as a representative value of the frame.

  By the way, since the HRTF coefficient data h is generated based on the coordinates P of the sound source, it accurately changes at each sample timing. Here, if the HRTF coefficient data h representing the frame is generated at a timing near the center of the frame period, the error can be minimized, and even if the sound source moves at high speed, the movement of the sound source is audible. Can be felt more continuous. Therefore, the HRTF coefficient data h is desirably generated based on the coordinates P of the sound source in the vicinity of the center of the frame period.

  Further, in the above-described embodiment, in order to intermittently perform the sound image localization processing on the continuous waveform data W, the waveform of the number of samples corresponding to the number of taps of the FIR filter (that is, the number of coefficients in one set) k. Data is retransmitted, but the present invention is not limited to this. That is, the present invention proposes a novel intermittent filter processing method in a general digital filter. Therefore, the present invention can be applied regardless of the purpose of use of the digital filter and the type of digital data to be processed. In this case, the filter coefficient may be fixed without changing with time. In such a case, the filter coefficient may be stored and held in advance on the digital computer 3 side without burst transfer via the bus 2. Further, the present invention may be applied not only to the FIR filter but also to an IIR (infinite impulse response) filter.

  In short, when the input data is divided into frames for every f samples, this input data is transferred to a k-tap filter (that is, using k coefficients), and FIR filter processing is performed intermittently. Input data is transferred to the filter by f + n samples during one frame period, and n + f pieces of n + f input data of the latter half of the f + n sample input data transferred in one frame during the next frame period are overlapped. Is transferred to the filter and set so that n ≧ k. In this case, the input data (n pieces) having the tap number k or more is retransmitted to the filter, so that the processing can be continued.

  In the above-described embodiment, the cross-fade processing unit 34 performs the cross-fade processing. However, this processing may be omitted as appropriate. Alternatively, the cross-fade processing unit 34 may not be provided inside the filter calculation unit 305, but a cross-fade synthesis processing unit (interpolation unit) may be provided outside, that is, on the output buffer output side. In that case, a cross-fade (interpolation) calculation is performed at a timing synchronized with a predetermined sampling period.

  In the above-described embodiment, the frame period, which is a processing unit of the filter, is handled as a fixed one, but it may be variable. For example, if the number of sounds to be processed simultaneously is four, one frame is composed of 256 samples, and if the number of sounds to be processed simultaneously is two, one frame is composed of 128 samples. Depending on the number, the number of samples constituting the frame may be varied. In this case, when the number of sounds is reduced, the burst transfer cycle is shortened, so that the update interval of the HRTF coefficient data h can be shortened, and the direction accuracy of sound image localization can be improved. In the above-described embodiment, data for one frame is transmitted by one burst transfer. However, the data is not limited to one time, and the data for one frame is divided into several times and transferred in batches. You may do it. In the above embodiment, the output buffers 306 to 309 are configured to accumulate and store the filtered data for a plurality of channels for each sampling. However, the present invention is not limited to this, and an output buffer is provided for each channel. Instead of accumulating, one channel of filtered data may be simply buffered. Further, reading of data from the output buffers 306 to 309 of the digital filter computer 3 is not necessarily performed in synchronization with a predetermined sampling period. For example, the filtered data is read out from the output buffers 306 to 309 of the digital filter computer 3 at a high speed and transferred separately to an output device or a utilization device. Data may be output in synchronization with the sampling period.

  In the above-described embodiment, the number of taps (that is, the number of coefficients in one set) k of the digital filter calculator 3 is described as being fixed. However, this may be varied. For example, when the number of sounds is 4, the tap number k is set to 90 as described above, and when the number of sounds is 14, the tap number is set to 25. Thus, the tap number k may be varied. In this case, when the number of sounds decreases, the number of taps k of the FIR filter increases, so that a filter that more faithfully reproduces the head related transfer function HRTF can be configured. On the other hand, when the number of sounds increases, the number of data to be transferred can be reduced, so that the bus 2 occupying time per sound can be reduced and the bus 2 contention can be avoided.

1 is a block diagram of a 3D sound positioning system which is an embodiment of a sound image localization apparatus according to the present invention. It is a figure which shows an example of the waveform data concerning the embodiment. It is a functional block diagram of the digital filter computer concerning the embodiment. It is a functional block diagram which shows the typical example of the filter calculation algorithm which the filter calculation part concerning the embodiment performs. In the same embodiment, it is the figure which showed the mode of the burst transfer corresponding to the process of 1 sound. It is a conceptual diagram for demonstrating the cross fade process concerning the embodiment. It is a functional block diagram of a general 3D sound positioning system. It is a block diagram of FIR filter F1 shown in FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... CPU (control part), 2 ... Bus, 3 ... Digital filter computer (arithmetic part), 34 ... Cross fade processing part, W ... Waveform data (input data), M1 ... Waveform buffer memory (1st memory | storage part) , M2 ... HRTF coefficient data memory (second storage unit), h ... HRTF coefficient data (coefficient data), P ... coordinate command (position information).

Claims (2)

A first step of burst-transferring a plurality of samples of digital data to be filtered and coefficient sets used in the filtering via a computer bus;
A second step of performing a filtering process on the plurality of samples of digital data that have been burst-transferred using the coefficient set that has been burst-transferred at a processing rate faster than a predetermined sampling period;
And a third step of buffering the filtered digital data and outputting the buffered data according to the predetermined sampling period. A digital filter device connected to a computer bus comprising:
A burst transfer of a plurality of samples of digital data to be filtered and a filter coefficient set used in the filtering operation is received via the bus at any one or a plurality of points in a predetermined one frame period. An interface part;
A filter operation unit that performs a predetermined filter operation process on the digital data received through the interface unit using the filter coefficient group received through the interface unit;
An output unit for buffering and outputting the output of the filter operation unit.

Images