WO2014127663A1

WO2014127663A1 - Interpolation filtering method and interpolation filter

Info

Publication number: WO2014127663A1
Application number: PCT/CN2013/089334
Authority: WO
Inventors: 钟斌; 官华伯; 马艳玲
Original assignee: 中兴通讯股份有限公司
Priority date: 2013-02-20
Filing date: 2013-12-13
Publication date: 2014-08-28
Also published as: CN103997355A; CN103997355B

Abstract

Embodiment of the present invention discloses an interpolation filtering method and interpolation filter, which include setting N filter processing units, wherein the filter processing unit includes a filter data buffering unit, and the filter data buffering unit includes M data buffering sub-units; shifting and inputting source data points one by one to the filter data buffering unit of each filter processing unit in parallel and synchronously, processing filter operation to the source data points in each filter processing unit after the source data points are written to each data buffering sub-unit in each filter data buffering unit, and respectively outputting the operation result of the filter operation for each filter processing unit to an output data buffering unit; shifting and inputting next source data points to the filter data buffering unit of each filter processing unit until processing X times filter operation to each filter processing unit, obtaining X operation results and outputting the operation results, wherein X is the number of the source data points.

Description

Interpolation filtering method and interpolation filter

Technical field

The present invention relates to the field of communications technologies, and in particular, to an interpolation filtering method and an interpolation filter.

Background technique

In the WCDMA (Wideband Code Division Multiple Access) communication system, multipath search is needed for the uplink signal, and the descrambling and despreading of the uplink signal and subsequent symbol levels are performed on the basis of the multipath search. deal with. In order to support higher and higher transmission rates, physical layer modulation and demodulation technology is also constantly evolving, from BPSK (Binary Phase Shift Keying), QPSK (Quadrature Phase Shift Keying) to 16QAM (Quadrature Amplitude Modulation) and 64QAM, etc., especially the 3GPP R8 protocol proposes a large number of high-order modulation and demodulation techniques used in the BOOSTING mode. These techniques place higher and higher requirements on multipath search accuracy.

The traditional multipath search algorithm uses a 2x ADP (power delay) input, and searches for the peak point and phase adjustment to obtain the final multipath. The search accuracy error of 3/8chip (slice precision unit) was introduced in this process. High-order demodulation techniques require this error to be maintained at l/8chip, which is not possible with traditional searchers.

In order to support the performance of high-order demodulation, WCDMA multipath search supports higher search accuracy by performing up-and-down interpolation of the original 2 times ADP. Moreover, since the hardware implementation of the above-mentioned interpolation will greatly increase the bus bandwidth, software interpolation is an effective method.

The new EOL (early mid-late interpolation filter searcher) interpolates 2 times the ADP into 8 times the ADP by means of the upper-like interpolation method, and on this basis, achieves higher precision. The search satisfies the requirements of high-order demodulation.

The core of the EOL searcher is the upper-like interpolation, which is to increase the sampling rate by interpolation. In the implementation system of the related WCDMA, the upper-like interpolation part of the searcher needs to be implemented by software. Because the data rate is directly increased on the chip, the bandwidth demand of the system will rise sharply and exceed the capacity of the chip bus. However, software implementation of the interpolation filter requires very high computational power of the processor. It is estimated that the relevant software interpolation filter is only available on the 732MHz processor. It can provide 6 high-rate user EOL search results, which is difficult to meet the needs of the actual system.

Summary of the invention

Embodiments of the present invention provide an interpolation filtering method and an interpolation filter, which can improve the efficiency of interpolation filtering, and enable the processor to simultaneously provide multipath searching for more users.

An interpolation filtering method according to an embodiment of the present invention includes:

N filter processing units are disposed, the filter processing unit includes a filter data buffer unit, and the filter data buffer unit includes M data buffer subunits, where N is a filtered upper multiple, and M is L/N, L is the order of filtering, where N, M, and L are positive integers;

Shifting the source data points one by one to the filter data buffer unit of each filter processing unit in parallel, after each data buffer subunit in each filter data buffer unit writes the source data points, Performing a filtering operation on the source data points in each of the filter processing units, and outputting the operation results of the current filtering operations of each of the filter processing units to the operation results of the current filtering operations in the output data buffer unit Output buffer subunit; and

Shifting the next source data point to the filter data buffer unit of each filter processing unit, performing filtering operation on the source data points in each filter processing unit, and outputting the operation result of the current filter operation to the output data In the output buffer sub-unit corresponding to the operation result of the current filtering operation in the punch unit, until X times of filtering operations are performed for each filter processing unit, X operation results are obtained and the operation result is output, where X is the source data The number of points is a positive integer.

Filtering the source data points in each filter processing unit includes:

For the first filter processing unit, the source data point currently buffered by the filter data buffer unit of the first filter processing unit is the source data point that is first input into the first filter processing unit. The result of the operation of the sub-filtering operation.

Also includes:

For the second to Nth filter processing units, respectively, a filter coefficient buffer unit is provided, and the filter coefficient buffer unit includes M coefficient buffer subunits corresponding to the data buffer subunit of the filter data buffer unit. The M coefficient buffer subunits respectively store filter coefficients to be multiplied by the source data points in the corresponding data buffer subunit, where M and N are positive integers. Performing a filtering operation on the source data points in each of the filter processing units includes: dividing, for the second to Nth filter processing units, the data buffer subunits in the filter data buffer unit of the filter processing unit into a plurality of filters Group, each filter group includes a plurality of data buffer subunits, and serially transforms the source data points in the data buffer subunits in each filter group, corresponding to the data buffer subunits in the filter group The filter coefficients in the coefficient buffer subunit are serial-to-parallel transformed, and the source data points in the data buffer subunits in the filter group are multiplied in parallel with the filter coefficients in the corresponding coefficient buffer subunits, and the products are added. Obtaining the filtering result of the filter group;

The filtering results of each filter group are added, and the added result is subjected to parallel-to-serial conversion to obtain the operation result of the filtering operation.

The output data buffer unit includes X*N output buffer subunits;

Outputting the operation result of the current filtering operation of each filter processing unit to the output buffer subunit corresponding to the operation result of the current filtering operation in the output data buffer unit, including: For the first filtering process The number of the output buffer subunit corresponding to the operation result of the first filtering operation to the Xth filtering operation in the output data buffer unit is 0, N, 2N, 3N... (Xl)N, according to The number of filtering operations outputs the operation result to the corresponding output buffer subunit;

For the Kth filter processing unit, the sequence number of the output buffer subunit corresponding to the operation result of the Xth filter operation in the output data buffer unit is

K- 1 , Ν + Κ - 1 , 2N + K - 1 , 3 Ν + Κ - 1... (X - 1 ) Ν + Κ - 1 , output the operation result to the corresponding output buffer according to the number of filtering operations In the subunit, where Κ is 2~Ν and is a positive integer.

An interpolation filter includes: a source data buffer unit, a filter processing unit, and an output data buffer unit, each filter processing unit includes a filter data buffer unit, and the filter data buffer unit includes Data buffer subunits, where Ν is the upper multiple of the filtering, Μ is L/N, L is the order of filtering, and N, M, and L are positive integers;

The source data buffer unit is configured to buffer source data points, perform shift input to the filter data buffer unit of each filter processing unit in parallel with the source data points, and perform filtering once in the filter processing unit. After the operation, shifting the next source data point to the filter data buffer unit of each filter processing unit; The filter processing unit is configured to filter the source data points in the filter data buffer unit after each data buffer subunit in the included filter data buffer unit writes the source data point, and The operation result of the sub-filtering operation is output to the output buffer sub-unit corresponding to the operation result of the current filtering operation in the output data buffer unit, and after the source data buffer unit inputs the next source data point, the source is The data point performs a filtering operation, and outputs the operation result of the current filtering operation to the output buffer subunit corresponding to the operation result of the current filtering operation in the output data buffer unit, until X filtering operations are performed, and X pieces are obtained. The result of the operation is output and the result of the operation is output, where X is the number of source data points and is a positive integer.

The first filter processing unit of the N filter processing units is configured to input the first filter into the source data point currently buffered by the filter data buffer unit of the first filter processing unit. The source data point in the processing unit is used as the operation result of the current filtering operation.

The second to Nth filter processing units of the N filter processing units include a filter coefficient buffer unit corresponding to the filter data buffer unit, and the filter coefficient buffer unit includes data buffered with the filter data buffer unit a punch unit - a corresponding M coefficient buffer subunit, each of the M coefficient buffer subunits respectively holding a filter coefficient of a source data point in the corresponding data buffer subunit, wherein, M, N Both are positive integers.

The data buffer subunits in the filter data buffer unit of the second to Nth filter processing units are divided into a plurality of filter groups, and each filter group includes a plurality of data buffer subunits;

The second to Nth filter processing units include a serial-to-parallel transform unit connected to the data buffer sub-unit in the filter group, and a string buffer sub-unit connected to the data buffer sub-unit in the filter group. a transform unit, an arithmetic unit corresponding to the filter group, and a parallel-to-serial transform unit, wherein:

The serial-to-parallel transform unit connected to the data buffer subunit in the filter group is configured to perform serial-to-parallel conversion on the source data points in the connected data buffer sub-unit;

And the serial-to-parallel transform unit connected to the coefficient buffer subunit corresponding to the data buffer subunit in the filter group is configured to perform serial-to-parallel transformation on the filter coefficients in the connected coefficient buffer subunit;

The operation unit corresponding to the filter group is respectively connected to the serial-to-parallel transform unit connected to the data buffer sub-unit in the filter group and the coefficient buffer sub-unit corresponding to the data buffer sub-unit in the filter group. a serial-to-parallel conversion unit connection, which is arranged in parallel to filter the data buffer subunits in the group The source data points are multiplied by the filter coefficients in the corresponding coefficient buffer subunits, the products are added, the filtering result of the filter group is obtained, and the filtering of the filter group sent by the operation unit corresponding to the adjacent previous filter group is received. As a result, the filtering result of the received filter group is added to the filtering result of the filter group obtained by itself, and the added result is sent to the operation unit corresponding to the adjacent next filter group; and the parallel string conversion unit is set. The result of adding the filtering results of each filter group in pairs is subjected to parallel-to-serial conversion, and the operation result of the current filtering operation is obtained.

The output data buffer unit includes X*N output buffer subunits;

The number of the output buffer subunit corresponding to the operation result of the first filtering operation unit of the first filter processing unit to the Xth filtering operation is 0 in the output data buffer unit. N, 2N, 3N...(X-1)N;

The number of the output buffer subunit corresponding to the operation result of the Kth filter processing unit of the Kth filter processing unit to the Xth filter operation unit in the output data buffer unit is K- 1 , Ν + Κ - 1 , 2N + K - 1 , 3 Ν + Κ - 1... (X-1) Ν + Κ - 1 , where Κ is 2 Ν and is a positive integer.

In summary, the embodiments of the present invention have the following beneficial effects:

(1) The insertion 0 operation before the upper sample filtering is omitted, and the operation of multiplying and adding with the 0 data is omitted;

(2) By arranging the filter coefficients and classifying and outputting the operation results of the filtering operations, the source data points and the filter coefficients are naturally aligned during the filtering operation, and the output data index calculation and output are omitted. The link of the coefficient index of the input data corresponding to the data forms a parallel operation structure of the first level;

(3) Since the source data points and the filter coefficients are naturally aligned, when performing the filtering operation, the scattered source data points and the filter coefficients can be assembled into higher bit width data and coefficients by serial-to-parallel conversion to utilize The parallel processing capability of the processor itself forms a parallelism of the second level;

(4) When performing filtering operations, it is possible to wrap continuous multiply-accumulate operations, which will double the efficiency of the processor's pipeline unit, avoid processor pipeline bubble, improve processor execution efficiency, and form the third and fourth levels. Parallel

(5) For the source data points in the first filter processing unit, the flight with the filter coefficients is not performed. Calculate, reducing unnecessary filtering operations;

According to the parallel top-order interpolation filtering method of the embodiment of the present invention, according to the bit width of the data, when the processor uses 2 times parallel execution, the filtering efficiency is improved by 9 times; in the processor, 4 times of parallel execution is performed. In the case, the filtering efficiency is increased by 17 times. In the actual WCDMA system, the number of high-precision search users that the system can support is increased by 10 times.

BRIEF abstract

1 is a flowchart of an interpolation filtering method according to an embodiment of the present invention;

2 is a block diagram of an interpolation filter according to an embodiment of the present invention.

Preferred embodiment of the invention

Due to the software implementation of the above-mentioned interpolation, a large number of multiply-and-accumulate operations require a huge amount of processor calculation time. In addition, due to the complex operation, the source data points and the filter coefficients are complicated in alignment. Each time an interpolated data point is calculated, The index of the interpolation coefficient corresponding to each source data point needs to be calculated first, and a large amount of additional auxiliary computing resources are consumed; and since the correspondence between the filter coefficient and the source data point is unknown during calculation, parallelization and pipelining of the multiply-add operation cannot be realized. , processor utilization is low. On modern high-performance processors, traditional software implementations can only support a very small number of users and become a bottleneck in the system.

Based on the shortcomings of the related software filtering method, this paper proposes a highly parallel upper-order interpolation filtering method and interpolation filter suitable for RISC processor, with highly parallel computing structure and high computational efficiency, suitable for WCDMA. The interpolation efficiency is greatly improved, and more high-precision search users are supported on the software-implemented WCDMA system.

In the present embodiment, according to the characteristics of the filter coefficient of the interpolation and the characteristics of the filtering operation, a new operation structure is designed, which is especially suitable for software to implement interpolation filtering. The interpolation filter of the present embodiment minimizes unnecessary interpolation operations and unnecessary auxiliary operations for coefficient alignment, and also makes parallel operations possible, making full use of the parallel processing capability of the RISC processor itself. The parallel calculation of interpolation filtering and the improvement of interpolation efficiency are very effective. It should be noted that the embodiments in the present application and the features in the embodiments may be combined with each other without conflict. Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

In this embodiment, it is assumed that the multiplication factor of the interpolation filter is N, and the number of source data points is X, then the total number of data points obtained by interpolation filtering is X*N, and the order L of the filter is N*M, that is, each source data point is filtered with M filter coefficients, where N, M, L, and X are positive integers.

As shown in FIG. 1, the interpolation filtering method of this embodiment includes:

Step 101: Set N filter processing units, each filter processing unit includes a filter data buffer unit, the filter data buffer unit includes M data buffer subunits, and each data buffer subunit is used to cache a source. Data point D;

Step 102: For the N*M-order filter, a total of N*M filter coefficients C are included, and N*M filter coefficients are divided into N groups, wherein the filter coefficients in the group 1 are in the first filter processing unit. The filter coefficient to be multiplied by the source data point, the sequence number of the filter coefficient in group 1 is 0, N, 2N, 3N (Ml)N, and the filter coefficient in group 2 is the second filter processing unit. The filter coefficient of the source data point to be multiplied, the sequence number of the filter coefficient in group 2 is 1, N+1, 2N+1, 3N+1 ...

(Μ-1)Ν+1, and so on to the filter coefficients in the group;

Step 103: Set a filter coefficient buffer unit for the second filter processing unit to the second filter processing unit, where the filter coefficient buffer unit includes one coefficient buffer subunit, and the coefficient buffer unit of the filter coefficient buffer unit The subunit is corresponding to the data buffer subunits of the corresponding filter data buffer unit, and is used for buffering the filter coefficients to be multiplied by the source data points in the corresponding data buffer subunit;

Since the filter coefficient in group 1 is 0 except that ^_ is 1, the filter coefficient buffer unit is not set for the first filter processing unit.

Step 104: Save the filter coefficients corresponding to the filter coefficients in the group 2 to the group 系数 to the coefficient buffer subunit of the corresponding filter coefficient buffer unit;

Step 105: Perform shift input to the filter data buffer unit of each filter processing unit in parallel with the source data points one by one, and write source data in each data buffer subunit in each filter data buffer unit. After the point, filtering the source data points in each filter processing unit; Referring to FIG. 2, a shift of the parallel synchronization of the source data points D ₀ to D _{X is} input to the data buffer subunit of the filter data buffer unit of each filter processing unit, and one source is buffered in each data buffer subunit. data point.

Filtering the source data points in each filter processing unit includes:

For the first filter processing unit, the source data point currently buffered into the first filter processing unit among the source data points currently buffered by the filter data buffer unit of the first filter processing unit is used as the current filter operation. The result of the operation. In the instantaneous state in FIG. 2, the source data point currently input into the first filter processing unit is D _Q , and the source data input into the first filter processing unit at the earliest after the next shift input is performed. The point D _l Do has been removed from the first filter processing unit.

For the second to Nth filter processing units, dividing the data buffer subunit of the filter data buffer unit in the filter processing unit into a plurality of filter groups, each filter group including a plurality of data buffer subunits, for each The source data points in the data buffer subunits in the filter group are serial-to-parallel transformed, and the filter coefficients in the coefficient buffer sub-unit corresponding to the data buffer sub-units in the filter group are serial-to-parallel converted, in parallel The source data points in the data buffer subunit in the filter group are multiplied by the filter coefficients in the corresponding coefficient buffer subunits, and the products are added to obtain the filtering result of the filter group; and the filtering result of each filter group is Add, and perform parallel and parallel conversion on the result of the addition, and obtain the operation result of this filtering operation. Each filter bank in Figure 2 contains two data buffer subunits.

Step 106: Output the operation result of the current filtering operation of each filter processing unit to the output buffer subunit corresponding to the operation result of the current filtering operation in the output data buffer unit;

The output data buffer unit includes X*N output buffer subunits;

For the first filter processing unit, the sequence of the output buffer unit corresponding to the operation result of the first filter operation to the Xth filter operation in the output data buffer unit is 0, N, 2N, 3N... (Xl)N, the operation result is output to the corresponding output buffer sub-unit according to the number of filtering operations; and the filtered first path data is directly outputted according to the filtered position. That is, the 0th (D ₀ ) of the source data point is output to the output buffer subunit O ₀ , and the 1st source data point (D output to the output buffer subunit 0 _N , the 2nd output of the source data point to Output buffer subunit 0 _2N ..... The X-1th (D _x-1 ) of the source data point is output to the output buffer subunit 0( _X-1 ) _N . For the Kth filter processing unit, the serial number of the output buffer subunit corresponding to the operation result of the first filtering operation to the Xth filtering operation is in the order of the output data buffer unit.

Κ - 1 , Ν + Κ - 1 , 2N + K - 1 , 3 Ν + Κ - 1... (X - 1 ) Ν + Κ - 1 , output the operation result to the corresponding output buffer according to the number of filtering operations In the subunit, where Κ is 2~Ν and is a positive integer.

In this embodiment, the filtered data points are divided into parallel loop data according to the characteristics of the filter coefficients used by the source data points. A first channel number corresponding output buffer subunit is 0, Ν, 2Ν, 3Ν ... ... (Xl) N, i.e., in FIG. _{_{1 O 0, 0 N ... ...}} 0 (Χ _- ι _)Ν (partial not shown), the filtered data point of the output is the source data point and the coefficients Co, C _N , C _2N ... C (^ multiply and add; the output corresponding to the second way The serial number of the buffer subunit is 1, N+1, 2N+1, 3N+1 ... (X-1)N+1, that is, Οι , 0 _{Ν +} ι ... in Fig. 1 . ..0 _(Χ- ι _)Ν+ ι (partial not shown), the filtered data point of the output is the source data point and the coefficient d, the coefficient C _N+1 , C _2N+1 ... ... C (M-1) N+1 multiply and add; and so on to N-way data.

Step 107: Shifting and inputting the next source data point to the filter data buffer unit of each filter processing unit, repeating the filtering operation on the source data point in each filter processing unit, and performing the operation result of the current filter operation Outputted to the output buffer sub-unit corresponding to the operation result of the current filtering operation in the output data buffer unit, until X times of filtering operations are performed for each filter processing unit, X calculation results are obtained, and the operation result is outputted and then ended. .

In the present embodiment, for the second to Nth filter processing units, when filtering the source data points in each of the filter processing units, the first filtering operation is to take the source data points D _Q ~D _M ^ and The corresponding M filter coefficients are multiplied, D _Q corresponds to C( _)N+1 , ... , D _M4 corresponds to the first data point, and the second filter operation takes 0 to D _M , corresponding to The M filter coefficients are multiplied correspondingly, corresponding to C( _M-1)N+1 , ..., D _M corresponds to d, the second data point is obtained, and the third to Xth data points are analogously.

When a single filtering operation is performed on the source data points in each filtering processing unit, serial data conversion is performed by serial-to-parallel conversion of the source data points and the filter coefficients, thereby making full use of the parallel processing capability of the CPU. At the time of output, the data points are serially outputted to the output buffer sub-units with interval N, that is, the input and output are serial, but the parallel-parallel conversion and parallel-to-serial conversion are performed in the middle to realize the parallel multiplication operation.

When implementing the method of the present embodiment, it is necessary to construct a filter according to the parameters of the filter, including the order of the filter, the filter coefficient, the multiple of the upper sample, the parallel calculation coefficient, etc., according to the structure shown in FIG. The source data buffer unit, the filter processing unit and the output data buffer unit of the wave device are set with a multiple of N, and N filter processing units are set, and for the second to Nth filter processing units, respectively, the second to The Nth filter processing unit sets a filter coefficient buffer unit to perform parallel filtering on a channel by way. When the filtering operation of all the paths is completed, the data in the data buffer unit is output, which is the result of the interpolation filtering.

The present embodiment realizes a high degree of parallelism by analyzing the characteristics of the upper sample interpolation filter, through the target filter data classification, the filter coefficient rearrangement, the parallel multiplication and addition operation, the special operation omission, the serial-to-parallel conversion, and the parallel-serial input and output. Filter operation structure.

As shown in FIG. 2, the interpolation filter provided in this embodiment includes: a source data buffer unit 21, N filter processing units 22, and an output data buffer unit 23, each of which includes a filter data buffer. Unit 221, the filter data buffer unit includes M data buffer subunits, where N is a filtered upper multiple, M is L/N, L is a filtered order, and N, M, and L are positive Integer; where:

The source data buffer unit 21 is configured to buffer the source data points, perform the shift input to the filter data buffer unit of each filter processing unit in parallel by the source data points, and perform a filtering operation after the filter processing unit completes the filter operation. Transmitting the next source data point to the filter data buffer unit of each filter processing unit;

The filter processing unit 22 is configured to filter the source data points in the filter data buffer unit after each data buffer subunit in the included filter data buffer unit 221 writes the source data points, and The operation result of the filtering operation is output to the output buffer subunit corresponding to the operation result of the current filtering operation in the output data buffer unit, and after the source data buffer unit inputs the next source data point, the source data is The filtering operation is performed on the point, and the operation result of the current filtering operation is output to the output buffer subunit corresponding to the operation result of the current filtering operation in the output data buffer unit, until X filtering operations are performed, and X operations are obtained. The result is output and the result is output, where X is the number of source data points.

The first filter processing unit of the N filter processing units is specifically configured to input the source data point currently buffered by the filter data buffer unit 221 of the first filter processing unit into the first filter processing unit. The source data point is used as the result of the operation of this filtering operation.

The second to Nth filter processing units of the N filter processing units are included with the filter data buffer The filter coefficient buffer unit 222 corresponding to the unit 221, the filter coefficient buffer unit 222 includes M coefficient buffer sub-units corresponding to the data buffer sub-unit of the filter data buffer unit 221, and M coefficient buffer sub-units. Each of the filter coefficients of the source data points in the corresponding data buffer subunit is multiplied.

The data buffer subunits in the filter data buffer unit 221 of the second to Nth filter processing units are divided into a plurality of filter groups, each filter group including a plurality of data buffer subunits;

The second to Nth filter processing units include a serial-to-parallel conversion unit 223 connected to the data buffer sub-unit in the filter group, and a serial-to-parallel connection of the coefficient buffer sub-units corresponding to the data buffer sub-units in the filter group Unit 224, an operation unit 225 corresponding to the filter group, and a parallel-to-serial conversion unit 226, where:

And a serial-to-parallel transform unit 223 connected to the data buffer sub-unit in the filter group, configured to perform serial-to-parallel transformation on the source data points in the connected data buffer sub-unit;

a serial-to-parallel transform unit 224 connected to the coefficient buffer sub-unit corresponding to the data buffer sub-unit in the filter group, configured to perform serial-to-parallel conversion on the filter coefficients in the connected coefficient buffer sub-unit;

The operation unit 225 corresponding to the filter group is respectively connected to the serial-to-parallel conversion unit 223 connected to the data buffer sub-unit in the filter group and the coefficient-buffer sub-unit connected to the coefficient buffer sub-unit corresponding to the data buffer sub-unit in the filter group. 224 connection, for parallelly multiplying the source data points in the data buffer subunit in the filter group by the filter coefficients in the corresponding coefficient buffer subunit, adding the products, obtaining the filtering result of the filter group, and receiving Filtering result of the filter group sent by the operation unit corresponding to the adjacent previous filter group, adding the filter result of the received filter group to the filter result of the filter group obtained by itself, and transmitting the added result to the adjacent one The operation unit corresponding to the next filter group; the parallel-to-serial conversion unit 226 is configured to perform parallel-to-serial conversion on the result of adding the filter results of each filter group to obtain the operation result of the current filter operation.

The output data buffer unit includes X*N output buffer subunits;

The number of the output buffer subunit corresponding to the operation result of the first filter processing unit of the first filter processing unit to the Xth filter operation unit in the output data buffer unit is 0, N, respectively. 2N, 3N...(X-1)N;

The first filtering operation of the Kth filtering processing unit of the N filtering processing units to the Xth time The sequence number of the output buffer subunit corresponding to the operation result of the filter operation in the output data buffer unit is K-1, N+K-1, 2N+K-1, 3N+K-1...(X- 1) N+K-1, where K is 2~N and is a positive integer.

In the filter shown in Fig. 2, the leftmost one is the source data buffer unit 21, and the far right is the output data buffer unit 23. The filter of the present embodiment includes a two-level parallel operation structure. The first layer parallelism is divided into parallel parallel N paths according to the characteristics of the coefficients; the second layer parallel is in each calculation, for each output data point, the multiplication and addition of the source data points and the filter coefficients The parallel operation can be performed by serial-parallel conversion, and the result of the parallel operation is output by serial-parallel conversion.

The computational structure of each channel includes coefficient buffering, data buffering, serial-to-parallel transformation of coefficients and data, parallel computation and accumulation computation, parallel-to-serial transformation, and output. The filter coefficient buffer unit saves the filter coefficients associated with the current path, and is obtained by the previous coefficient classification operation; the data buffer stores the source data points that are sequentially shifted into the filter processing unit from right to left. When the buffer is full, the source data point on the right is shifted out, the source data point on the left is moved into one, and each time the output data point is calculated, a shift is made.

Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general-purpose computing device, which can be concentrated on a single computing device or distributed over a network composed of multiple computing devices. Alternatively, they may be implemented by program code executable by the computing device so that they may be stored in the storage device by the computing device, or they may be separately fabricated into individual integrated circuit modules, or Multiple modules or steps are made into a single integrated circuit module. Thus, the invention is not limited to any particular combination of hardware and software.

The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes can be made to the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the present invention.

Industrial applicability

Claims

claims

1. An interpolation filtering method, including:

N filtering processing units are provided. The filtering processing unit includes a filtering data buffer unit. The filtering data buffering unit includes M data buffering subunits, where N is the sampling multiple of filtering, and M is L/N, L is the order of filtering, where N, M, and L are all positive integers;

The source data points are shifted and input to the filtered data buffer unit of each filtering processing unit one by one in parallel and synchronously. After each data buffer sub-unit in each filtered data buffer unit writes the source data points, Perform a filtering operation on the source data points in each filtering processing unit, and output the operation results of this filtering operation for each filtering processing unit to the corresponding operation results of this filtering operation in the output data buffer unit. in the output buffer subunit; and

Shift input of the next source data point into the filter data buffer unit of each filter processing unit, perform a filter operation on the source data points in each filter processing unit, and output the operation result of this filter operation to the output data buffer. In the output buffer subunit corresponding to the operation result of this filtering operation in the buffer unit, until X filtering operations are performed on each filter processing unit, X operation results are obtained and the operation results are output, where X is the source data The number of points is a positive integer.

2. The method of claim 1, wherein performing filtering operations on the source data points in each filtering processing unit includes:

For the first filter processing unit, among the source data points currently cached by the filter data buffer unit of the first filter processing unit, the earliest source data point input into the first filter processing unit is used as the current source data point. The result of the filtering operation.

3. The method of claim 1 or 2, further comprising:

For the second to Nth filter processing units, filter coefficient buffer units are respectively provided. The filter coefficient buffer unit includes M coefficient buffer sub-units corresponding to the data buffer sub-unit of the filter data buffer unit. , the M coefficient buffer subunits each store the filter coefficient to be multiplied by the source data point in the corresponding data buffer subunit, where M and N are both positive integers.

4. The method according to claim 3, wherein performing a filtering operation on the source data points in each filtering processing unit includes: For the second to Nth filter processing units, the data buffer sub-units in the filter data buffer unit of the filter processing unit are divided into multiple filter groups, and each filter group contains multiple data buffer sub-units. The source data points in the data buffer sub-units in the filter group are subjected to serial-to-parallel transformation, and the filter coefficients in the coefficient buffer sub-unit corresponding to the data buffer sub-unit in the filter group are subjected to serial-to-parallel transformation. In parallel, The source data points in the data buffer subunit of the filter group are multiplied by the filter coefficients in the corresponding coefficient buffer subunit, and the products are added to obtain the filtering result of the filter group; and

Add the filtering results of each filter group, and perform parallel-to-serial conversion on the added results to obtain the operation result of this filtering operation.

5. The method of claim 1, wherein:

The output data buffer unit includes X*N output buffer sub-units;

The operation result of this filtering operation of each filtering processing unit is output to the output buffer subunit corresponding to the operation result of this filtering operation in the output data buffer unit, including: For the first filtering process unit, the sequence numbers of the output buffer subunits corresponding to the operation results from the first filtering operation to the Xth filtering operation in the output data buffer unit are 0, N, 2N, 3N... (X-l)N, according to The number of filtering operations outputs the operation results to the corresponding output buffer subunit; and

For the K-th filter processing unit, the sequence numbers of the output buffer sub-units corresponding to the operation results from the first filter operation to the X-th filter operation in the output data buffer unit are in sequence:

K- 1 ,N+K- 1 ,2N+K- 1 ,3N+K- 1... (X- 1 )N+K- 1 , output the operation result to the corresponding output buffer according to the number of filtering operations In the subunit, where K is 2~N and is a positive integer.

6. An interpolation filter, including: a source data buffer unit, N filter processing units and an output data buffer unit. Each filter processing unit includes a filter data buffer unit, and the filter data buffer unit contains Contains M data buffer subunits, where N is the sampling multiple of filtering, M is L/N, L is the order of filtering, and N, M, and L are all positive integers; where:

The source data buffer unit is configured to buffer source data points, perform shift input of the source data points one by one in parallel and synchronously to the filter data buffer unit of each filter processing unit, and complete one filtering in the filter processing unit. After the operation, the next source data point is shifted and input to the filtered data buffer unit of each filtering processing unit; and The filtering processing unit is configured to perform a filtering operation on the source data points in the filtered data buffer unit after each data buffer subunit in the included filtered data buffer unit writes the source data point, and then The operation result of this filter operation is output to the output buffer sub-unit corresponding to the operation result of this filter operation in the output data buffer unit. After the source data buffer unit inputs the next source data point, the source data buffer unit The data points are filtered, and the result of this filtering operation is output to the output buffer subunit corresponding to the result of this filtering operation in the output data buffer unit until X filtering operations are performed and X The operation result is calculated and output, where X is the number of source data points and is a positive integer.

7. The filter of claim 6, wherein:

The first filter processing unit among the N filter processing units is configured to input the earliest source data points currently cached by the filter data buffer unit of the first filter processing unit into the first filter The source data points in the processing unit are used as the results of this filtering operation.

8. The filter according to claim 6 or 7, wherein:

The second to Nth filter processing units among the N filter processing units include a filter coefficient buffer unit corresponding to the filter data buffer unit, and the filter coefficient buffer unit includes a data buffer unit corresponding to the filter data buffer unit. Crush subunit—corresponding M coefficient buffer subunits, the M coefficient buffer subunits each store the filter coefficients to be multiplied by the source data points in the corresponding data buffer subunit, where, M, N are all positive integers.

9. The filter of claim 8, wherein:

The data buffering subunits in the filtered data buffering unit of the second to Nth filtering processing units are divided into multiple filtering groups, and each filtering group contains multiple data buffering subunits;

The second to Nth filter processing units include a serial-to-parallel conversion unit connected to the data buffer sub-unit in the filter group, and a serial-to-parallel conversion unit connected to the coefficient buffer sub-unit corresponding to the data buffer sub-unit in the filter group. Transformation unit, the arithmetic unit corresponding to the filter group and the parallel-to-serial transformation unit, where:

The serial-to-parallel conversion unit connected to the data buffer subunit in the filter group is configured to perform serial-to-parallel transformation on the source data points in the connected data buffer subunit;

The series-parallel transformer connected to the coefficient buffer subunit corresponding to the data buffer subunit in the filter group The conversion unit is configured to perform serial-to-parallel conversion on the filter coefficients in the connected coefficient buffer sub-units; the operation unit corresponding to the filter group is respectively connected to the serial-to-parallel conversion unit and the data buffer sub-unit in the filter group. The serial-parallel transformation unit connected to the coefficient buffer subunit corresponding to the data buffer subunit in the filter group is configured to parallelly convert the source data points in the data buffer subunit in the filter group to the corresponding coefficients. Multiply the filter coefficients in the buffer subunit, add the products, and obtain the filtering result of the filtering group, and receive the filtering result of the filtering group sent by the adjacent operation unit corresponding to the previous filtering group, and add the received filtering group. The filtering result of is added to the filtering result of the filter group obtained by itself, and the added result is sent to the operation unit corresponding to the adjacent next filter group; and the parallel-to-serial conversion unit is set to filter each filter group. The result after the addition is subjected to parallel-to-serial transformation to obtain the result of this filtering operation.

10. The filter of claim 6, wherein:

The output data buffer unit includes X*N output buffer sub-units;

The serial numbers of the output buffer sub-units corresponding to the operation results of the first filter operation to the X-th filter operation of the first filter processing unit among the N filter processing units in the output data buffer unit are 0 in sequence, N,2N,3N...(X-1)N; and

The serial numbers of the output buffer sub-units corresponding to the operation results of the first filter operation to the X-th filter operation of the K-th filter processing unit among the N filter processing units in the output data buffer unit are in sequence K- 1 ,N+K- 1 ,2N+K- 1 ,3N+K- 1... (X- 1 )N+K- 1 , where K is 2~N and is a positive integer.