JP3659196B2 - Huffman code decoding method and apparatus - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a decoding method and apparatus for decoding a Huffman code at a high speed with a limited memory.
[0002]
[Prior art]
As a technique for decoding the Huffman code, there are generally techniques such as a binary tree search method and a direct method. In the binary tree search method, a variable-length Huffman code is inspected one bit at a time from the first bit, branched at 0/1, and a specified value is output when the last is reached. Is realized. In the direct method, a memory area having an address with the maximum bit length of a variable-length code is secured, and the value is written in an area to which a Huffman code is assigned, and the memory area is set with the Huffman code as an address. And the value of the accessed area is output.
[0003]
[Problems to be solved by the invention]
However, since the binary tree search method inspects the Huffman code bit by bit, there is a drawback that it takes time to decode. The direct method is fast because the value can be read directly by accessing the memory with the Huffman code. However, since the memory area having the address with the maximum bit length must be secured, the Huffman code is actually used. There is a drawback that a decoding table having an area much larger than the area to which is assigned is required.
[0004]
In particular, in the case of data that has a plurality of Huffman code tables and any one of them is selected depending on the case, providing a decoding table corresponding to all the Huffman code tables is realized because a very large memory is required. It is difficult to use a binary tree search method in an apparatus that decodes data in such a format. However, as described above, the binary tree search method has a problem that it takes time to decode.
[0005]
An object of the present invention is to provide a decoding method and apparatus capable of decoding data using a plurality of Huffman code tables at high speed with a small memory.
[0006]
[Means for Solving the Problems]
The present invention is a method for decoding time-series audio data such as MP3 standard encoded by switching and applying a plurality of Huffman code tables as appropriate in units of frames while one song is composed of a plurality of frames , For each frame of the time-series data, analyze which Huffman code table is applied to which the frame is encoded, and select some Huffman code tables having a short maximum code length from the plurality of Huffman code tables. Data is decoded by a direct method using a decoding table for frames encoded by application, and binary for frames encoded by applying most Huffman code tables other than the part of the Huffman code tables. Data is decrypted by a tree search method.
[0007]
The present invention is an apparatus for decoding time-series audio data such as the MP3 standard that is encoded by appropriately switching and applying a plurality of Huffman code tables in units of frames while one song is composed of a plurality of frames , Direct method decoding means for decoding data by a direct method using a decoding table for a frame encoded by applying a part of the Huffman code table having a short maximum code length among the plurality of Huffman code tables; Binary tree decoding means for decoding data encoded by applying a binary tree search method to a frame encoded by applying most of the Huffman code table other than the Huffman code table, and which Huffman code table to which the encoded data is applied Analyzing means that sequentially analyzes whether it is encoded from the beginning, and selecting a decoding unit based on the analysis result of the analyzing unit, and sequentially entering the encoded data into the selected decoding unit And a selection means for operating.
[0008]
The present invention is characterized in that, in the above invention, a code table having a high frequency applied when data is encoded among the plurality of Huffman code tables is the predetermined Huffman code table.
The present invention is characterized in that, in the above invention, a code table having a short maximum code length among the plurality of Huffman code tables is used as the predetermined Huffman code table.
[0009]
According to the present invention, “short codes often appear. Therefore, a code table having short codes is often specified. If only a code included in this well-specified code table can be decoded at high speed, the entire decoding process can be performed at high speed. Is based on the idea that Specifically, decoding is basically performed by the binary tree search method, but a code table that frequently appears (short maximum code length) is decoded by the direct method. As a result, it is possible to perform decoding processing with a small decoding table and to perform decoding at high speed.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments of the present invention will be described with reference to the drawings. For bit stream format data such as the MP3 standard, an optimum one of a plurality of Huffman code tables defined in the standard is selected and applied for each frame. Which Huffman code table is applied to each frame is written in the header of each frame. In the decoding apparatus, a Huffman code table is selected for each frame with reference to the header, and data is decoded using the Huffman code table. In the MP3 standard, 34 Huffman code tables are defined, but in the following, in order to simplify the description, a case where two code tables as shown in FIG. 2 are defined will be described as an example.
[0011]
Assume that the input bitstream is encoded using one of the two Huffman code tables shown in FIG. FIG. 1 is a block diagram of a decoding apparatus according to an embodiment of the present invention. In this decoding apparatus, in order to decode the bit stream as fast as possible without increasing the circuit scale, when code table # 0 having a small maximum code length is designated, decoding is performed by the direct method, and the maximum code length is determined. When a large code table # 1 is designated, the binary tree search method is used for decoding.
[0012]
In FIG. 1, the bit stream is input to the RAM 10. The Huffman code table analysis unit 11 analyzes the header of (each frame) of the input bit stream and analyzes which Huffman code table is used by the frame. In this embodiment, it is analyzed whether code table # 0 or code table # 1 is used. The analysis result is input to the decoding method selection unit 12, the acquired bit number calculation unit 13, and the two decoding units 15 and 16. The two decoding units are a direct method decoding unit 15 and a binary tree decoding unit 16, respectively.
[0013]
The decoding method selection unit 12 selects which decoding unit to use from the decoding units 15 and 16 corresponding to whether the input bitstream uses # 0 or # 1 code table, and the data Switch the flow. The obtained bit number calculation unit 13 is the number of bits (3 bits) of data read at a time when the direct method is applied to the code table # 0, and 1 when the binary tree search method is applied to the code table # 1. The number of bits (1 bit) of data to be read at one time is determined and output to the bit fetch / rewind unit 14. The acquisition bit calculation unit 13 is provided with a memory (ROM) that stores data shown in FIG.
[0014]
The bit extraction / rewinding unit 14 reads out the data of the number of bits input from the acquired bit number calculation unit 13 from the RAM 10 and supplies the data to the decoding unit selected by the decoding method selection unit 12. Further, it receives the rewind bit number returned from the direct decoding unit 15 and rewinds the read pointer of the RAM 10 by the rewind bit number. Note that in the case of decoding a Huffman code by the binary tree search method, decoding is performed when one bit is read out and hit any of the Huffman codes, so the number of rewind bits is always 0.
[0015]
The direct decoding unit 15 includes a decoding table, inputs bits for the maximum code length, accesses the decoding table, and outputs a corresponding value (decoded data) to the next stage (not shown). Also, the actual number of bits of the current Huffman code is determined by this decoding process, and the number of rewind bits, which is the number of extra bits read (the number of bits of the maximum code length−the actual number of bits), is extracted / rewinded. 14 for output.
[0016]
Further, the binary tree decoding unit 16 inputs the bit stream bit by bit, and searches the binary tree for decoding the code table # 1 to decode the Huffman code. The decoded data is output to the next stage.
[0017]
When the bit stream is input to the RAM 10 and the Huffman code table number analysis unit 11 determines the code table used for the data, the decoding method selection unit 12 uses the code table number information, and the decoding method selection unit 12 One of them is selected, and the number of bits to be acquired is calculated by the acquisition bit calculation unit 13 and the bit number information is input to the bit extraction / rewinding unit 14.
[0018]
The bit fetch / rewind unit 14 reads the data of the number of bits from the RAM 10 and outputs the data to the selected decoding unit 15 or 16. The selected decoding unit decodes the Huffman code of the input data and outputs decoded data. When the Huffman code is decoded by the direct method decoding unit 15, the number of rewind bits is output together with the decoded data, and this is returned to the bit extraction / rewind unit 14. The bit fetch / rewind unit 14 rewinds the read pointer of the RAM 10 by this number of bits, and reads data for the next decoding process from the rewind position.
[0019]
It is assumed that, for example, bit stream frame data (Huffman code string) as shown in FIG. The Huffman code table used as the head position of the frame data has been analyzed by the Huffman code table number analysis unit 11. Here, in order to see that the amount of calculation is reduced by switching between two or more processing methods, (1) the number of accesses to the decoding unit when decoding only with the binary tree search method, and (2) the direct method And the binary tree search method are compared by the code table number. In the binary tree search method, access occurs every bit, whereas in the direct method, one Huffman code can be decoded by one access.
[0020]
Therefore, a total of 14 accesses occur in (1) and a total of 11 accesses in (2). This indicates that the number of accesses has been reduced by 21% by switching between the direct method and the binary tree search method by the table number.
[0021]
A table used in the binary tree method is generally composed of 2n-1 words, and a decoded value and an in-table address are stored in one word. In the case of table number # 0, since n = 4, the number of words in the table is 7. The address field for accessing this 7-word table requires log (2) 7≈2.8 bits or more, and the number of bits is an integer, so the number of necessary bits is 3 bits. Since 2 bits are required to express the decoded value, one word of this table is 3 + 2 = 5 bits. Therefore, in order to express all the words, a memory area of 5 × 7 = 35 bits is required. Similarly, in the case of table number # 1, similarly, log (2) 31≈4.95 and the necessary number of bits is 5, and a memory area of (5 + 4) × 31 = 279 bits is required.
[0022]
On the other hand, a decoded value and a code length are stored in one word of the table used in the direct method. If the number of codes defined in the code table is n, the number of words in the table that realizes the number = ²ⁿ words. In the case of the decoding table of table number # 0, the number of bits necessary for expressing the decoded value is 2 bits, and the number of bits required for expressing the code length is 2 bits, so 1 word is 2 + 2 = 4 bits It's okay. Since 2 ³ = 8 words are required, a memory area of 4 × 8 = 32 bits is required to represent all the words. Similarly, in the case of the decoding table of table number # 1, a memory area of (4 + 4) × 2 ⁸ = 2048 bits is required.
[0023]
As described above, when both table number # 0 and table number # 1 are decoded by the binary tree method, a ROM of 35 + 279 = 314 bits is required. If this is all replaced by the direct method, a ROM of 32 + 2048 = 2080 bits is required. In this case, the amount of calculation is certainly drastically reduced, but the amount of memory is rapidly increasing, which is not practical in an actual application.
[0024]
On the other hand, when only the table number # 0 is replaced by the direct method, the required memory is 32 + 279 = 311 bits, and the amount of memory is slightly increased as compared with the case where both are decoded by the binary tree method as described above. It is possible to greatly reduce the amount of calculation simply by doing.
[0025]
Hereinafter, with reference to FIG. 5 and FIG. 6, experimental results when MP3 encoding of actual PCM audio data are shown.
First, two types of encoders (Encoder 1 and Encoder 2) were used in the experiment. Two types of PCM audio data were input to these two types of encoders to create a total of four types of bit streams (stereo, 128 bps). The first PCM audio data (music source A) is a pop song with a large volume as a whole, and the second PCM audio data (music source B) is a quiet piano solo song. Which code table is designated for these four types of bitstreams was counted for each codeword. The result is shown in FIG.
[0026]
From FIG. 5, in both Encoders 1 and 2, a code table with a small number tends to be selected for source A, and a code table with a large number tends to be selected for B. However, in Encoder 1, the code table of No. 2 is most frequently used in source B as well. The code tables Quad (A) and Quad (B) are also frequently used, but for the count1 area in each frame, the code table Quad (A) or Quad (B) is always used alternatively. Because.
[0027]
From FIG. 5, the code table having a larger number of selections compared to other code tables is Table 2, 3, 5, 15, Quad (A), Quad (B), etc., but this is realized when decoding by the direct method is performed. For ease of explanation, only Tables 2, 3, Quad (A), and Quad (B) having the shortest long code length are decoded by the direct method here, and the other code tables are decoded by the binary tree search method. .
[0028]
The amount of calculation required for the Huffman decoding process by the above method and the amount of calculation required for the Huffman decoding process purely by the binary tree search method were calculated and compared. The results are shown in FIG.
[0029]
FIG. 6 shows the calculation amount required for Huffman decoding per frame and arranged on the time axis. For the four types of bit streams used above, (A-1) and (A- 2), (B-1), and (B-2) correspond. In this figure, Bin is the number of processing steps per frame when decoding is performed only by the binary tree search method, and per frame when DTL is decoded by combining the direct method and the binary tree search method according to the method of the present invention. Is the number of processing steps.
[0030]
Note that when Table 2, 3, Quad (A), and Quad (B) are decoded by the direct method as described above, the calculation amount of the music source A is greatly reduced, and not so much in the case of the music source B. This indicates that the selection of the Huffman code table to be decoded by the direct method is more suitable for the music source A, and in order to decode the music source B more efficiently (reduce the amount of calculation). , Tables 2, 5, 15, and 26 may be decoded by a direct method.
[0031]
Thus, in the decoding apparatus, the code table to be decoded by the direct method may be changed according to the genre of the audio source that is often decoded (reproduced).
[0032]
【The invention's effect】
As described above, according to the present invention, since the direct method and the binary tree method are switched according to the Huffman code table, it is possible to perform decoding processing with a small decoding table and to perform decoding at high speed. It becomes possible.
[Brief description of the drawings]
FIG. 1 is a block diagram of a Huffman code decoding apparatus according to an embodiment of the present invention; FIG. 2 is a diagram showing a Huffman code table used in the Huffman code decoding apparatus; FIG. 4 is a diagram showing an example of input data decoded by the Huffman code decoding device. FIG. 5 is a diagram showing each Huffman code when PCM audio data is encoded into MP3 data. FIG. 6 is a diagram showing a distribution of frequency of use of the table. FIG. 6 is a diagram comparing the number of processing steps required for decoding when the MP3 data is decoded by the binary tree search method and the method of the present invention. ]
DESCRIPTION OF SYMBOLS 10 ... RAM, 11 ... Huffman code table number analysis part, 12 ... Decoding method selection part, 13 ... Acquired bit number calculation part, 14 ... Bit extraction / rewinding part, 15 ... Direct method decoding part, 16 ... Binary tree decoding part

Claims (2)

A method of decoding time-series audio data such as MP3 standard encoded by switching and applying a plurality of Huffman code tables as appropriate in units of frames while one song is composed of a plurality of frames,
For each frame of the time-series data, analyze which Huffman code table is applied to which the frame is encoded,
For a frame encoded by applying a partial Huffman code table having a short maximum code length among the plurality of Huffman code tables, data is decoded by a direct method using a decoding table, and the partial Huffman code table is used. A method of decoding a Huffman code in which data is decoded by a binary tree search method for frames encoded by applying most of the Huffman code table. An apparatus for decoding time-series audio data such as the MP3 standard, which is encoded by appropriately switching and applying a plurality of Huffman code tables in units of frames while one song is composed of a plurality of frames,
Direct method decoding means for decoding data by a direct method using a decoding table of a frame encoded by applying a partial Huffman code table having a short maximum code length among the plurality of Huffman code tables;
Binary tree decoding means for decoding data using a binary tree search method for frames encoded by applying most of the Huffman code tables other than the part of the Huffman code tables;
Analysis means for sequentially analyzing from the top which Huffman code table is applied to which the encoded data is encoded;
A selection unit that selects a decoding unit based on the analysis result of the analysis unit and sequentially inputs encoded data to the selected decoding unit;
A Huffman code decoding apparatus.

Images