（八）用JAVA编写MP3解码器——解码增益因子

解码一帧Layer3第2步：解码增益因子  -- getScaleFactors_1()和getScaleFactors_2()方法

 

      增益因子（Scale Factor）数据存储在帧内的主信息中。主信息包含增益因子和用哈夫曼编码的主数据（main_data）。增益因子简单讲就是逆量化公式的指数中的一个因子。手册中解码主数据的伪码算法如下，其中MPEG 1.0由ISO/IEC 11172-3给出，MPEG 2.0由ISO/IEC 13818-3给出。官方并没有MPEG2.5这个版本，这个版本是民间版本。

 

      解码MPEG 1.0的增益因子时用到的输入值：

• slen0[]、slen1[]：从上图中看出增益因子的值为位流中读取0..5位的值，当前究竟应该取几位，由这两个常量表给出。
• 粒度组信息gri：用于判断块的类型的成员变量block_type、mixed_block_flag值已经在上一步中初始化。
• scfsi：其值已在上一步初始化，若scfsi[0..3]=1表示增益因子共用。

 

解码MPEG 2.0的增益因子时用到的输入值：

• i_slen2[]：强度立体声(intensity stereo)的增益因子(scalefactor)比特数，变量名中的2表示是MPEG 2.0版。从伪码算法看出解码时需要从位流中读入0..4位，这个常量表的值给出当前应该读多少位。
• n_slen2[]：正常(normal)编码的增益因子(scalefactor，简称sf)比特数，这个常量表的值给出当前应该读多少位。
• 常量表i_slen2[]和n_slen2[]的取值见ISO/IEC 13818-3 subclause 2.4.3.2 slenx x=1..4，这两个常量表已经在Layer3的构造方法内被初始化。
• nr_of_sfb[3][6][4]：MPEG 2.0的增益因子被分割为4部分，故nr_of_sfb的第3维长度为4；根据块的类型和scalefac_compress值增益因子频带(简称sfb)每部分的个数(number)分为6种情况，6种情况下其个数各不相同，故nr_of_sfb的第2维长度为6；根据帧边信息中block_type和mixed_block_flag这两个成员的值，共分为3种不同的块类型，故nr_of_sfb的第1维长度为3。我把nr_of_sfb设计为这样的结构，是考虑到和逆量化取得增益因子的值顺序一致。

 

解码增益因子得到的值：

 

• 增益因子用去的比特数保存在part2_bits中，供后续的哈夫曼解码时计算主数据的比特数。
• 解码得到的增益因子的值，长块的放入scfL[2][23]；
• 得到的增益因子的值，短块(包括纯短块和混合块中的短块)的值保存在scfS[2][3][13]中，得到的这些值供后续的逆量化使用，这些值在数组中的存放顺序要和逆量化时读取的顺序一致，这样效率更高一些。

 

 

　　解码MPEG 1.0和MPEG 2.0/2.5增益因子的方法差别很大，这里定义了两个版本的方法。这里涉及到的“纯短块”、“长块”、“混合块”、“窗”概念，在逆量化中再作说明。根据上面对常量及变量值的描述中提及的“共用”、“块的类型”、“窗”就很容易看懂getScaleFactors_x()方法中的if语句和for语句的作用；短块内每个子带分3窗，所以有2重循环。

 

      class Layer3内定义的getScaleFactors_x()方法源码：

	//2.
	//>>>>SCALE FACTORS========================================================
	private static int[][] scfL;		// [2][23];
	private static int[][][] scfS;		// [2][3][13];
	private static int[] i_slen2;		// MPEG 2.0 slen for intensity stereo
	private static int[] n_slen2;		// MPEG 2.0 slen for 'normal' mode
										// slen: 增益因子(scalefactor)比特数
	private static byte[][][] nr_of_sfb;//[3][6][4]

	/*
	 * MPEG 2.0/2.5
	 */
	private void getScaleFactors_2(final int ch, final int gr) {
		byte[] nr;
		int i, bandIdx, win, slen, num, n = 0, scf = 0;
		boolean i_stereo = objHeader.isIStereo();
		GRInfo gri = objSI.ch[ch].gr[gr];
		int[] l = scfL[ch];
		int[][] s = scfS[ch];

		rzero_bandL = 0;
		if ((ch > 0) && i_stereo)
			slen = i_slen2[gri.scalefac_compress >> 1];
		else
			slen = n_slen2[gri.scalefac_compress];

		gri.preflag = (slen >> 15) & 0x1;
		gri.part2_bits = 0;
		if (gri.block_type == 2) {
			n++;
			if ((gri.mixed_block_flag) != 0)
				n++;
			nr = nr_of_sfb[n][(slen >> 12) & 0x7];

			for (i = 0; i < 4; i++) {
				num = slen & 0x7;
				slen >>= 3;
				if (num != 0) {
					for (bandIdx = 0; bandIdx < nr[i]; bandIdx += 3) {
						for (win = 0; win < 3; win++)
							s[win][scf] = bsMainInfo.getBits17(num);
						scf++;
					}
					gri.part2_bits += nr[i] * num;
				} else {
					for (bandIdx = 0; bandIdx < nr[i]; bandIdx += 3) {
						for (win = 0; win < 3; win++)
							s[win][scf] = 0;
						scf++;
					}
				}
			}

			n = (n << 1) + 1;
			for (i = 0; i < n; i += 3) {
				for (win = 0; win < 3; win++)
					s[win][scf] = 0;
				scf++;
			}
		} else {
			nr = nr_of_sfb[n][(slen >> 12) & 0x7];
			for (i = 0; i < 4; i++) {
				num = slen & 0x7;
				slen >>= 3;
				if (num != 0) {
					for (bandIdx = 0; bandIdx < nr[i]; bandIdx++)
						l[scf++] = bsMainInfo.getBits17(num);
					gri.part2_bits += nr[i] * num;
				} else {
					for (bandIdx = 0; bandIdx < nr[i]; bandIdx++)
						l[scf++] = 0;
				}
			}

			n = (n << 1) + 1;
			for (i = 0; i < n; i++)
				l[scf++] = 0;
		}
	}

	/*
	 * MPEG 1.0
	 */
	private static final int slen0[] = { 0, 0, 0, 0, 3, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4 };
	private static final int slen1[] = { 0, 1, 2, 3, 0, 1, 2, 3, 1, 2, 3, 1, 2, 3, 2, 3 };

	private void getScaleFactors_1(final int ch, final int gr) {
		GRInfo gri = objSI.ch[ch].gr[gr];
		int scale_comp = gri.scalefac_compress;
		int length0 = slen0[scale_comp];
		int length1 = slen1[scale_comp];
		int sfb, win;
		int[] l = scfL[ch];
		int[][] s = scfS[ch];

		gri.part2_bits = 0;

		if (gri.window_switching_flag != 0 && gri.block_type == 2) {
			if (gri.mixed_block_flag != 0) {
				// MIXED block
				gri.part2_bits = 17 * length0 + 18 * length1;
				for (sfb = 0; sfb < 8; sfb++)
					l[sfb] = bsMainInfo.getBits9(length0);

				for (sfb = 3; sfb < 6; sfb++)
					for (win = 0; win < 3; win++)
						s[win][sfb] = bsMainInfo.getBits9(length0);

				for (sfb = 6; sfb < 12; sfb++)
					for (win = 0; win < 3; win++)
						s[win][sfb] = bsMainInfo.getBits9(length1);
			} else {
				// pure SHORT block
				gri.part2_bits = 18 * (length0 + length1);
				for (sfb = 0; sfb < 6; sfb++)
					for (win = 0; win < 3; win++)
						s[win][sfb] = bsMainInfo.getBits9(length0);
				for (sfb = 6; sfb < 12; sfb++)
					for (win = 0; win < 3; win++)
						s[win][sfb] = bsMainInfo.getBits9(length1);
			}
		} else {
			// LONG types 0,1,3
			int[] scfsi = objSI.ch[ch].scfsi;
			if (gr == 0) {
				gri.part2_bits = 10 * (length0 + length1) + length0;
				for (sfb = 0; sfb < 11; sfb++)
					l[sfb] = bsMainInfo.getBits9(length0);
				for (sfb = 11; sfb < 21; sfb++)
					l[sfb] = bsMainInfo.getBits9(length1);
			} else {
				gri.part2_bits = 0;
				if (scfsi[0] == 0) {
					for (sfb = 0; sfb < 6; sfb++)
						l[sfb] = bsMainInfo.getBits9(length0);
					gri.part2_bits += 6 * length0;
				}
				if (scfsi[1] == 0) {
					for (sfb = 6; sfb < 11; sfb++)
						l[sfb] = bsMainInfo.getBits9(length0);
					gri.part2_bits += 5 * length0;
				}
				if (scfsi[2] == 0) {
					for (sfb = 11; sfb < 16; sfb++)
						l[sfb] = bsMainInfo.getBits9(length1);
					gri.part2_bits += 5 * length1;
				}
				if (scfsi[3] == 0) {
					for (sfb = 16; sfb < 21; sfb++)
						l[sfb] = bsMainInfo.getBits9(length1);
					gri.part2_bits += 5 * length1;
				}
			}
		}
	}
	//<<<<SCALE FACTORS========================================================

 

 

上一篇：（七）用JAVA编写MP3解码器——解码帧边信息

下一篇：（九）用JAVA编写MP3解码器——哈夫曼解码

 

【本程序下载地址】http://jmp123.sourceforge.net/

评论

2 楼 lfp001 2013-12-06  

逆量化混合块时：前8个频带是长块，用长块公式逆量化；后9个频带是短块。

你从ISO/IEC 11172-3引用过来的说明是指逆量化时如何使用增益因子。解码增益因子时从位流读入增益因子的值，存放到数组内（长块和短块分别存放在各自的数组内），存放的顺序刚好是解码时依次读取的顺序

1 楼 clearstarrysky 2013-12-03  

if block_type is 2 and switch_point is 1:
slen1: length of scalefactors for the scalefactor bands 0 to 7 (long window scalefactor band) and 4 to 5 (short window scalefactor band) Note: Scalefactor bands 0-7 are from the "long window scalefactor band" table, and scalefactor bands 4-11 from the "short window scalefactor band" table. This combination of partitions is contiguous and spans the entire frequency spectrum.
slen2: length of scalefactors for the scalefactor bands 6 to 11
文档里说混合块的短块slen1从4-5,代码里却是从3开始,为什么?
for (sfb = 3; sfb < 6; sfb++) 
for (win = 0; win < 3; win++) 
  s[win][sfb] = bsMainInfo.getBits9(length0);