ATRAC3plus - Revision history

Maxpol: More links.

2018-11-29T15:31:23Z

Maxpol: /* Coding techniques */

2014-06-22T20:34:35Z

Coding techniques

← Older revision		Revision as of 13:34, 22 June 2014
Line 41:		Line 41:
	=== Coding techniques ===		=== Coding techniques ===

	ATRAC3plus is a hybrid subband/MDCT codec like [[MP3]]. The signal is split into 16 subbands using [http://en.wikipedia.org/wiki/Polyphase_quadrature_filter Polyphase Quadrature Filter] before MDCT and bit allocation. The sample-frame size is 2048 samples per channel.		ATRAC3plus is a hybrid subband/MDCT codec like [[MP3]]. The signal is split into 16 subbands using [http://en.wikipedia.org/wiki/Polyphase_quadrature_filter Polyphase Quadrature Filter] (further PQF) before MDCT and bit allocation. The sample-frame size is 2048 samples per channel.

	After the subband splitting ATRAC3plus tries to extract sine waves from each subband using Generalized Harmonic Analysis (further GHA). GHA encodes parameters of extracted sine waves such as frequency, amplitude and phase into final bitstream.		After the subband splitting ATRAC3plus tries to extract sine waves from each subband using Generalized Harmonic Analysis (further GHA). GHA encodes parameters of extracted sine waves such as frequency, amplitude and phase into final bitstream.
Line 70:		Line 70:
	"Bitstream decoder" decodes various sound parameters from supplied frame data. First the residual signal will be decoded by applying inverse quantization, power compensation, inverse MDCT and gain compensation. Then the sine waves will be synthesized according with their parameters such as frequency, amplitude and phase. Then the residual and the synthesized sine waves will be added together. Optionally, some white noise can be added if specified in the bitstream.		"Bitstream decoder" decodes various sound parameters from supplied frame data. First the residual signal will be decoded by applying inverse quantization, power compensation, inverse MDCT and gain compensation. Then the sine waves will be synthesized according with their parameters such as frequency, amplitude and phase. Then the residual and the synthesized sine waves will be added together. Optionally, some white noise can be added if specified in the bitstream.

	This processing will be repeated for each of 16 subbands. Finally the ~~QMF~~ synthesis filter will be applied in order to sum all subbands together and reconstruct the encoded audio signal.		This processing will be repeated for each of 16 subbands. Finally the PQF synthesis filter will be applied in order to sum all subbands together and reconstruct the encoded audio signal.

	Various algorithms are used to improve compression results:		Various algorithms are used to improve compression results:

Maxpol: /* Coding techniques */

2014-06-22T20:32:49Z

Coding techniques

← Older revision		Revision as of 13:32, 22 June 2014
Line 41:		Line 41:
	=== Coding techniques ===		=== Coding techniques ===

	ATRAC3plus is a hybrid subband/MDCT codec like [[MP3]]. The signal is split into 16 subbands using [http://en.wikipedia.org/wiki/~~Quadrature_mirror_filter~~ Quadrature ~~Mirror~~ Filter] before MDCT and bit allocation. The sample-frame size is 2048 samples per channel.		ATRAC3plus is a hybrid subband/MDCT codec like [[MP3]]. The signal is split into 16 subbands using [http://en.wikipedia.org/wiki/Polyphase_quadrature_filter Polyphase Quadrature Filter] before MDCT and bit allocation. The sample-frame size is 2048 samples per channel.

	After the subband splitting ATRAC3plus tries to extract sine waves from each subband using Generalized Harmonic Analysis (further GHA). GHA encodes parameters of extracted sine waves such as frequency, amplitude and phase into final bitstream.		After the subband splitting ATRAC3plus tries to extract sine waves from each subband using Generalized Harmonic Analysis (further GHA). GHA encodes parameters of extracted sine waves such as frequency, amplitude and phase into final bitstream.

Maxpol: /* Huffman tables for delta coding */

2010-12-30T22:34:20Z

Huffman tables for delta coding

← Older revision		Revision as of 15:34, 30 December 2010
Line 699:		Line 699:
	==== Huffman tables for delta coding ====		==== Huffman tables for delta coding ====

	PLEASE NOTE: delta values indicated in the tables below will be added using modular arithmetic as described [[#Delta coding ~~based on modular arithmetic~~\|here]], so in the case of "wrap around" the value of "7" will be treated as "-1", the value of "6" = "-2" and so on.		PLEASE NOTE: delta values indicated in the tables below will be added using modular arithmetic as described [[#Delta coding\|here]], so in the case of "wrap around" the value of "7" will be treated as "-1", the value of "6" = "-2" and so on.

	* '''vlc_tab_index = 0''', delta range -1...1		* '''vlc_tab_index = 0''', delta range -1...1

Maxpol: /* Mode 1 (master) */

2010-12-30T22:25:00Z

Mode 1 (master)

← Older revision		Revision as of 15:25, 30 December 2010
Line 633:		Line 633:
	if (weigths_tab_indx) {		if (weigths_tab_indx) {
	for (pos = 0; pos < [[#num_quant_units\|num_quant_units]]; pos++)		for (pos = 0; pos < [[#num_quant_units\|num_quant_units]]; pos++)
	wl_coeffs[pos] = [[#Tables of weights\|wl_weights]][channel_num][weights_tab_indx - 1][pos];		wl_coeffs[pos] += [[#Tables of weights\|wl_weights]][channel_num][weights_tab_indx - 1][pos];
	}		}

Maxpol: /* Method D: sequence of numbers in ascending order */

2010-10-03T17:03:28Z

Method D: sequence of numbers in ascending order

← Older revision		Revision as of 10:03, 3 October 2010
Line 269:		Line 269:
	Let us return to our sequence. The 2nd value will be coded directly as well using 5 bits because ilog2(31 - 5) = 5. Similar for the 3rd one. No delta coding is applied in that case. The 4th value will be delta-coded using 4 bits:		Let us return to our sequence. The 2nd value will be coded directly as well using 5 bits because ilog2(31 - 5) = 5. Similar for the 3rd one. No delta coding is applied in that case. The 4th value will be delta-coded using 4 bits:

	num_delta_bits = ~~ilog~~(31 - 15) = 4 bits;		num_delta_bits = ilog2(31 - 15) = 4 bits;
	delta = 18 - 15 - 1 = 2		delta = 18 - 15 - 1 = 2

Line 277:		Line 277:

	Therefore the resulting sequence will be 27 bits long. Compared to the unpacked version (8 x 5 bits = 40 bits) this packing method will yield a bit-reduction of 13 bits (32% smaller).		Therefore the resulting sequence will be 27 bits long. Compared to the unpacked version (8 x 5 bits = 40 bits) this packing method will yield a bit-reduction of 13 bits (32% smaller).


	===== Vector quantization with residual encoding =====		===== Vector quantization with residual encoding =====

Maxpol: /* Value grouping with "group coded" flag */

2010-10-01T11:50:21Z

Value grouping with "group coded" flag

← Older revision		Revision as of 04:50, 1 October 2010
Line 351:		Line 351:
	(0, 0); flag = 0 (group not coded)		(0, 0); flag = 0 (group not coded)

	Thereafter, each "not coded" group requires only one bit to be transmitted indicating that all values in that group are zero. On the other hand, each "coded" group requires one extra bit to be transmitted indicating that at least one value in that group is non-zero. In the case above that overhead is worthwhile because the half of th signal contains zeroes.		Thereafter, each "not coded" group requires only one bit to be transmitted indicating that all values in that group are zero. On the other hand, each "coded" group requires one extra bit to be transmitted indicating that at least one value in that group is non-zero. In the case above that overhead is worthwhile because the half of the signal contains zeroes.

	The encoded signal is 4 x 1 + 4 x 7 = 32 bits long while the unpacked one is 16 x 3 = 48 bits long. The bit-reduction is thereafter 16 bits (33% smaller).		The encoded signal is 4 x 1 + 4 x 7 = 32 bits long while the unpacked one is 16 x 3 = 48 bits long. The bit-reduction is thereafter 16 bits (33% smaller).

Maxpol: /* Mode 1 (slave) */

2010-08-22T21:31:25Z

Mode 1 (slave)

← Older revision		Revision as of 14:31, 22 August 2010
Line 643:		Line 643:
	Data stored in the bitstream:		Data stored in the bitstream:

	* 2/7/9 or more bits (depending on mode): info for the [[#Reconstruction of trimmed word-length coefficients\|reconstruction of trimmed ~~coeffcicients~~]]		* 2/7/9 or more bits (depending on mode): info for the [[#Reconstruction of trimmed word-length coefficients\|reconstruction of trimmed coefficients]]
	* 2 bits: indicates which huffman table from [[#Huffman tables for delta coding\|this set]] should be used for decoding		* 2 bits: indicates which huffman table from [[#Huffman tables for delta coding\|this set]] should be used for decoding
	* for each [[#num_coded_vals\|num_coded_vals]]		* for each [[#num_coded_vals\|num_coded_vals]]
	** huffman-coded delta value to be added to the corresponding value of the master channel		** huffman-coded delta value to be added to the corresponding value of the master channel

	/* parse mode/num_coded_vals/split_point_delta parameters for [[#Reconstruction of trimmed word-length coefficients\|trimmed ~~coeffcicients~~]] */		/* parse mode/num_coded_vals/split_point_delta parameters for [[#Reconstruction of trimmed word-length coefficients\|trimmed coefficients]] */

	vlc_sel = get_bits(2); /* selects a huffman table from [[#Huffman tables for delta coding\|this set]] */		vlc_sel = get_bits(2); /* selects a huffman table from [[#Huffman tables for delta coding\|this set]] */

Maxpol: /* Mode 1 (master) */

2010-08-22T21:30:41Z

Mode 1 (master)

← Older revision		Revision as of 14:30, 22 August 2010
Line 597:		Line 597:

	* 2 bits: index of the table of weigths, "0" - indicates "no table used"		* 2 bits: index of the table of weigths, "0" - indicates "no table used"
	* 2/7/9 or more bits (depending on mode): info for the [[#Reconstruction of trimmed word-length coefficients\|reconstruction of trimmed ~~coeffcicients~~]]		* 2/7/9 or more bits (depending on mode): info for the [[#Reconstruction of trimmed word-length coefficients\|reconstruction of trimmed coefficients]]
	* 5 bits: number of directly coded coefficients ('''num_direct_coeffs'''). This value must be < [[#num_coded_vals\|num_coded_vals]]		* 5 bits: number of directly coded coefficients ('''num_direct_coeffs'''). This value must be < [[#num_coded_vals\|num_coded_vals]]
	* 2 bits: size of deltas in bits ('''delta_bits''')		* 2 bits: size of deltas in bits ('''delta_bits''')
Line 610:		Line 610:
	weigths_tab_indx = get_bits(2); /* get index of weights table to be added after decoding */		weigths_tab_indx = get_bits(2); /* get index of weights table to be added after decoding */

	/* parse mode/num_coded_vals/split_point_delta parameters for [[#Reconstruction of trimmed word-length coefficients\|trimmed ~~coeffcicients~~]] */		/* parse mode/num_coded_vals/split_point_delta parameters for [[#Reconstruction of trimmed word-length coefficients\|trimmed coefficients]] */

	num_direct_coeffs = get_bits(5);		num_direct_coeffs = get_bits(5);
Line 629:		Line 629:
	}		}

	/* reconstruct trimmed ~~coeffcicients~~ as described [[#Reconstruction of trimmed word-length coefficients\|here]] */		/* reconstruct trimmed coefficients as described [[#Reconstruction of trimmed word-length coefficients\|here]] */

	/* add weighting coefficients if requested */		/* add weighting coefficients if requested */

Maxpol: /* Mode 2 (master) */

2010-08-22T21:30:04Z

Mode 2 (master)

← Older revision		Revision as of 14:30, 22 August 2010
Line 663:		Line 663:
	Data stored in the bitstream:		Data stored in the bitstream:

	* 2/7/9 or more bits (depending on mode): info for the [[#Reconstruction of trimmed word-length coefficients\|reconstruction of trimmed ~~coeffcicients~~]].		* 2/7/9 or more bits (depending on mode): info for the [[#Reconstruction of trimmed word-length coefficients\|reconstruction of trimmed coefficients]].
	* 1 bit: '''enable_grouping''' flag. "1" indicates that residual values were coded pairwise (in groups of two).		* 1 bit: '''enable_grouping''' flag. "1" indicates that residual values were coded pairwise (in groups of two).
	* 1 bit: selects one of the first two huffman tables from [[#Huffman tables for delta coding\|this set]].		* 1 bit: selects one of the first two huffman tables from [[#Huffman tables for delta coding\|this set]].

← Older revision		Revision as of 08:31, 29 November 2018
Line 7:		Line 7:
	= ATRAC3plus introduction =		= ATRAC3plus introduction =

	ATRAC3plus is a proprietary audio compression algorithm developed by [[Sony]]. As in the case of ATRAC3 ATRAC3plus represents the next generation of the ATRAC codec introduced in 1992 with the MiniDisc. Common use of that codec is in nowel Minidisc players and Portable Playstations made by [[Sony]].		ATRAC3plus is a proprietary audio compression algorithm developed by [[Sony]]. As in the case of its predecessor [[ATRAC3]], ATRAC3plus represents the next generation of the [[ATRAC]] codec introduced in 1992 with the MiniDisc. Common use of that codec is in nowel Minidisc players and Portable Playstations made by [[Sony]].

	Streams coded with ATRAC3plus are usually stored either in the [[Microsoft_Wave\|WAV]] container (those files have the ".at3" extension though) or in the Sony's proprietary [[Oma\|Oma/Omg]] container. In the case of the [[Microsoft_Wave\|WAV]] container the undocumented [[GUID]]:		Streams coded with ATRAC3plus are usually stored either in the [[Microsoft_Wave\|WAV]] container (those files have the ".at3" extension though) or in the Sony's proprietary [[Oma\|Oma/Omg]] container. In the case of the [[Microsoft_Wave\|WAV]] container the undocumented [[GUID]]:

← Older revision		Revision as of 14:31, 22 August 2010
Line 643:		Line 643:
	Data stored in the bitstream:		Data stored in the bitstream:

	* 2/7/9 or more bits (depending on mode): info for the [[#Reconstruction of trimmed word-length coefficients\|reconstruction of trimmed ~~coeffcicients~~]]		* 2/7/9 or more bits (depending on mode): info for the [[#Reconstruction of trimmed word-length coefficients\|reconstruction of trimmed coefficients]]
	* 2 bits: indicates which huffman table from [[#Huffman tables for delta coding\|this set]] should be used for decoding		* 2 bits: indicates which huffman table from [[#Huffman tables for delta coding\|this set]] should be used for decoding
	* for each [[#num_coded_vals\|num_coded_vals]]		* for each [[#num_coded_vals\|num_coded_vals]]
	** huffman-coded delta value to be added to the corresponding value of the master channel		** huffman-coded delta value to be added to the corresponding value of the master channel

	/* parse mode/num_coded_vals/split_point_delta parameters for [[#Reconstruction of trimmed word-length coefficients\|trimmed ~~coeffcicients~~]] */		/* parse mode/num_coded_vals/split_point_delta parameters for [[#Reconstruction of trimmed word-length coefficients\|trimmed coefficients]] */

	vlc_sel = get_bits(2); /* selects a huffman table from [[#Huffman tables for delta coding\|this set]] */		vlc_sel = get_bits(2); /* selects a huffman table from [[#Huffman tables for delta coding\|this set]] */