KR19990081316A - Table structure of variable length code and parallel decoding device and method - Google Patents

Abstract

The present invention relates to a structure of a code table for decoding variable length code (VLC) coded digital data such as MPEG, a method of creating the code table, an apparatus for parallel decoding of variable length code using the code table, and a method thereof. It is about.

Since the conventional variable length code decoder can decode only one code at a time, it is difficult to overcome the difference in the operation speed with the peripheral block (IDCT / motion compensator) when applied to an HDTV decoder, etc., and the operation speed can not be lowered accordingly. .

The present invention rearranges the frequency of occurrence of the variable length code in order of high frequency, and represents the group code of the variable length code. When the group code appears, the group code is included (represented). Provided are a table structure, a parallel decoding apparatus, and a method of variable length code, which enable high-speed parallel processing by decoding at least two or more variable length codes simultaneously.

Description

Table structure of variable length code and parallel decoding device and method

The present invention relates to a structure of a code table for decoding variable length code (VLC) coded digital data such as MPEG, a method of creating the code table, an apparatus for parallel decoding of variable length code using the code table, and a method thereof. It is about.

Conventionally, variable length codes are widely used in high-speed data transmission systems, and are generally called half-man codes, and are one of optimal coding techniques using statistical characteristics of data.

In other words, when the statistical characteristics of the data are known, shorter datawords are assigned as the data is frequently generated, and longer codewords are assigned as the frequency of occurrence is small.Average codeword is allocated when appropriate codewords are allocated in this way to all possible data. Since the length is shorter than the average codeword length of the original data, it is widely used for effective coding without loss when the statistical properties of the data are known.

In particular, it is adopted to increase the compression rate in MPEG1, MPEG2 or HDTV field, and it is required to decode bit by bit at the same time as a variable speed code decoder of high speed is needed. SLAMBOL-SERIAL programmable logic array-based architecture (PLA-BASED ARCHITECTURE) that decodes one code at a time is proposed.

Fig. 1 shows a simple example of a variable length code decoder using this conventional symbol unit coding / decoding technique.

The input buffer 100 stores a serial bit stream of a variable length word received through a data channel, and the interface unit 101 includes a latch and a barrel shifter to interface between the input buffer 100 and the decoder unit 102. do.

The decoder unit 102 includes a latch and a barrel shifter and includes a VLC table 103 for decoding although illustrated externally.

The VLC table 103 includes a codeword table 105, a codeword length table 106, and a decodeword table 107 to output a codeword.

The decoder unit 102 decodes the continuous variable length codeword into a continuous stream through the code division according to the carry output of the adder 104 and correspondingly outputs a fixed length codeword fixed according to the read signal.

However, when the conventional symbol string variable length code decoder is applied to a high performance system such as HDTV, the following problems are caused.

HDTV video decoder is composed of variable length decoder, motion compensator and inverse discrete cosine transform block (IDCT Block). There are various parallel processing techniques for IDCT block and motion compensator. Although it can be easily lowered, the variable length code decoder cannot lower the operation speed because the conventional technology can decode only one code at a time.

Especially in the case of a downconverter that receives HDTV signals and outputs them in accordance with NTSC TV or SDTV format, the difference in the operation speed becomes larger.

That is, both the IDCT block and the motion compensator perform the operation on the image reduced to NTSC TV resolution or SDTV resolution, so the variable length code decoder needs to process the HDTV signal directly, requiring much higher operation speed.

Therefore, when the HDTV decoder or downconverter is implemented as a single chip, a high clock must be used specifically for the variable length code decoder, resulting in high power consumption, difficulty in implementation, and poor chip operation stability.

The present invention bundles several short variable length codes into a single code and names them as group codes and adds them to an existing decoding table so that multiple codes are decoded at once.

That is, the present invention rearranges the frequency of occurrence of the variable length code in the order of high frequency of occurrence and represents it as a group code of the variable length code, and when the group code appears, the group code is included. The present invention provides a table structure, a parallel decoding apparatus, and a method of variable length code, which enable high-speed parallel processing by simultaneously decoding at least two or more variable length codes.

1 is a block diagram of a conventional variable length code decoder

2 is a block diagram of a variable length code decoder of the present invention.

3 is a view showing an example of a variable length code table structure of the present invention;

In the present invention, a short variable length code generally takes shorter decoding time than a long variable length code to construct a new variable length code (group code) that is redefined into several short codes. If the length is shorter than the length of the longest variable length code that can be decoded at one time, the decoding performance can be improved without affecting the existing operation speed.

In other words, by searching several strings of short variable length codes, experimentally / statistically preliminarily obtain the frequency of repeated occurrences of the variable length code columns, and grouping the obtained variable length code columns into a new code form called group code. Define a code representing the column of length codes, compare the length of this redefined group code with the length of the longest code that can be decoded at the same time, and decide whether to group. The variable length code is decoded from the recoded decoding table in the state of adding to.

When the group code is added in this way, the size of the decoding table increases and the number of codes that can be newly created when a plurality of existing variable length codes are bundled into one code becomes very large. Adding to the table is not appropriate.

Therefore, in the present invention, given a limitation on the size of the decoding table, a combination of codes that can bring the maximum performance improvement within the given constraint is found and added to the decoding table.

Fig. 2 is a block diagram of a variable length code decoder using the group code of the present invention, and Fig. 3 shows an example of the code table structure of the present invention.

First, a method of creating the code table of FIG. 3 to implement the decoder of FIG. 2 will be described.

The first step is to select candidate codes that can make a new code based on the statistical characteristics of the existing codes, name them individual candidate codes, and arrange them in order of high frequency.

Since this is based on statistical characteristics, it means that the frequency of occurrence is determined by the demonstration and rearranged in the order of appearance frequency.

Step 2: Set the number of codes that can be bundled together (Ncode) as '2'.

Step 3: A new group candidate code is generated by selecting Ncode codes from the list of individual candidate codes.

This is because the group candidate code is a group code arranged before verifying whether decoding with a group code takes a shorter time than decoding with an individual code.

Fourth step: comparing the length of the group candidate code generated in the third step with the maximum code length that can be decoded at one time, and if the length of the group candidate code does not exceed the maximum code length that can be decoded at one time, it is a valid group code. Define and add it to the existing decoding table to measure the degree of performance improvement in actual decoding.

Step 5: Measure the performance improvement for all code combinations that can be made with Ncode codes, and if this measurement is finished, go to the next step, otherwise repeat the above steps.

Step 6: A group code list is created by rearranging the valid group codes in the order of the highest performance improvement of the valid group codes based on the result of the execution up to the fifth step.

Step 7: When the group code in the highest rank of the group code list created in step 6 is added to the decoding table, the size and operation speed of the decoding table are predicted.

Step 8: When the group code is added to the decoding table, it is checked whether it satisfies the constraints such as the size and operation speed of the decoding table. Removes the group code added in the current step and then proceeds to the ninth step.

Step 9: If there is any remaining code in the group code list, if there is any cryptic code, the process is repeated from step 7 to verify the next group code, and if there is no remaining code, Ncode ' 1 'increase.

If the newly generated Ncode is less than or equal to the total number of individual candidate codes, the process is repeated from the third step and the next (Ncode = 3) consecutive columns of individual candidate codes, that is, the number of codes that can be bundled together are 3 Select individual individual candidate codes to generate new group candidate codes and verify them.

However, if the newly created Ncode is larger than the total number of individual candidate codes, no further grouping is required, so the decoding table thus produced is output and all processes are completed.

Through the above series of steps, a variable length code decoding table having a new structure as shown in Fig. 3 was created.

The decoding table of FIG. 3 is a form of a string (bitstream) of a high frequency code, that is, a group code table in which at least two symbols are represented by one group code and a group of a conventional symbol unit decoding table are merged. .

In the method of decoding variable length codes using the decoding table, the input bitstream is searched, and when the group code appears, the decoded codeword is output by decoding the representative code of the group code table, otherwise the existing symbol unit decoding table is output. The codeword decoded in symbol units is output with reference to FIG.

Such a group / symbol coded decoding apparatus can replace the existing decoding table with a newly created decoding table (FIG. 3) in the existing symbol-serial decoder as shown in FIG. The parallel variable-length decoder is configured by adding a multi-input port buffer (206) that can accept the maximum number of codes as an output buffer.

An input buffer 201 stores a serial bit stream of a variable length word received through a data channel, and a second shifter 202 interfaces between the input buffer 201 and the first shifter 203.

The output register 204 latches the shifter output and inputs it to the group code symbol decoder 205 and the group code length decoder 207, and the group code length information is calculated through the operation of the adder 208 and the accumulator (acc: 209). Used for code slices from the input bitstream.

The variable length code table structure and parallel decoding apparatus and method provide a technique for easily converting a conventional symbol string variable length decoder into a symbol-parallel variable length decoder.

In addition, a method for efficiently constructing a new decoding table has been presented, and the present invention reduces the operation speed of a conventional symbol string variable length code decoder because the decoding time difference between codes having different lengths in the existing decoding table is used. Higher decoding performance can be attained without making a difference.

In addition, compared to the conventional symbol string variable length code decoders, they have an equivalent longest delay time and have higher performance, and thus, a lower clock can be used if the same performance is required, which is advantageous in terms of power consumption and chip operation stability.

In addition, the decoding performance of the variable-length code can be adjusted according to the clock speeds of blocks connected to the surroundings, which is effective when implementing multiple functional blocks in one chip.

Claims (4)

In order to decode variable-length coded data, the variable-length code is characterized by arranging the at least two variable-length codes in one representative code and arranged in order of successive frequency of at least two variable-length codes. Table structure. The variable length code table structure of claim 1, wherein the variable length code table has a structure in which a decoding table in a symbol unit is merged. Retrieving the frequency of occurrence of the variable length code based on the statistical characteristics of the variable length code, arranging the variable length codes in the order of high frequency, and combining at least two repeating of the arranged variable length codes into one code Rearranging the group codes into a group code, adding the group codes to the variable length code decoding table, measuring the decoding performance thereof, and rearranging the group codes in the order of high decoding performance; And measuring the decoding performance of the variable length code table for the group codes satisfying a predetermined condition. Means for storing a serial bitstream of a variable length word received through a data channel, means for slicing the bitstream stored in the storage means in units of variable length codewords, and decoding the sliced variable length codewords. Means for storing decoding information as a group code by grouping at least two of the symbols according to successive frequency numbers, and searching for whether the slicing variable length codeword corresponds to the group code of the storage means. And a means for decoding as a group code if not and if not, as a symbol unit code.