[Decals-svnlog] SF.net SVN: decals: [1] mp4-als_format_notes.txt
Status: Inactive
Brought to you by:
jbr79
Menu
▾
▴
[Decals-svnlog] SF.net SVN: decals: [1] mp4-als_format_notes.txt
|
From: <jb...@us...> - 2007-08-20 00:07:56
|
Revision: 1
http://decals.svn.sourceforge.net/decals/?rev=1&view=rev
Author: jbr79
Date: 2007-08-19 17:07:52 -0700 (Sun, 19 Aug 2007)
Log Message:
-----------
initial SVN import
Added Paths:
-----------
mp4-als_format_notes.txt
Added: mp4-als_format_notes.txt
===================================================================
--- mp4-als_format_notes.txt (rev 0)
+++ mp4-als_format_notes.txt 2007-08-20 00:07:52 UTC (rev 1)
@@ -0,0 +1,635 @@
+
+MPEG-4 Audio Lossless Coding (ALS) Documentation
+------------------------------------------------
+
+
+MPEG-4 Audio Lossless Coding[1][2], also called MP4-ALS, is one of the lossless
+audio coding formats which is part of the MPEG-4 audio specification. It is
+based on the LPAC audio codec[3] by Tilman Liebchen. This document covers the
+raw ALS format, which consists of the ALS header and a sequence of frames.
+
+
+
+KEY CONCEPTS
+============
+
+Original file preservation
+--------------------------
+The ALS header can provide everything that is needed to reconstruct the
+original WAVE, AIFF, or raw audio file. This includes the file type,
+sample rate, number of channels, bits-per-sample, sample type, and byte order.
+The original header and/or footer are also preserved exactly in the ALS
+header. The ALS format only supports from 8-bit to 32-bit audio in 8-bit
+increments. Either integer or IEEE-floating-point sample types are supported.
+
+
+Frame size and block switching
+------------------------------
+The ALS frame size (number of audio samples) is fixed for a given stream or
+file. All audio samples for a single channel in a frame is referred to as a
+block. In order to adapt to transient signals, each block can be divided
+into recursive subblocks. Each subblock can be split into 2 smaller subblocks,
+down to 1/32 of the original frame size.
+
+
+Random access
+-------------
+Seeking is an important feature for any audio format. It allows the decoder to
+jump forward or backward to the start of a frame without having to decode all
+frames in between. In the ALS format, the encoder can optionally include
+so called "random access" frames to achieve this goal. The ALS header
+specifies a random access interval from 0 to 255. A value of 0 indicates no
+random access frames, 1 means every frame is a random access frame, 2 means
+every other frame, etc... Each set of frames between random access frames is
+call a random access unit. The encoder indicates the size of each random
+access unit by either writing them all in the header or by writing it at the
+start of each random access frame. The ra_location field in the ALS header
+tells which of the 2 locations are used.
+
+I'm going to go ahead and gripe here about this design. The problem I see is
+that using random access frames is optional, and that there is no unique
+identifier for frame boundaries. This prevents a decoder from finding random
+frames without both a reference point and either random access information or
+by decoding every frame between the reference and the destination.
+
+
+Channels
+--------
+ALS has 2 different ways of reducing redundancy between multiple channels. The
+first method is joint stereo coding. The encoder calculates the difference
+between 2 channels and selects 2 of left+right, left+diff, or right+diff. For
+files with more than 2 channels, joint stereo can still be used by grouping the
+channels into sets of pairs and individual channels. For example, encoding of
+standard 5.1 audio could be done as L+R, C, LFE, LS+RS. The format allows for
+reordering of the channels from their original positions in order to pair them
+together if necessary.
+
+The 2nd method is called multi-channel correlation, or MCC. This method uses
+adaptive subtraction to compare the prediction residual of reference channels
+to the remaining channels. This is a more complex procedure which I will
+descibe later once I fully understand the math behind it.
+
+
+Forward Prediction
+------------------
+The primary prediction method in ALS is forward linear predictive coding, or
+LPC[4][5]. There are several ways to select good prediction coefficients. The
+recommended method for ALS is Levinson-Durbin Recursion[6][7]. This is a good
+method for 2 reasons. First of all, it sequentially produces higher order
+coefficients, which is ideal for adapting the prediction order for each block.
+Secondly, it produces parcor (reflection) coefficients as an intermediate
+result, which are the values that are quantized and written to the ALS
+bitstream. The standard defines a bit-exact method for transforming the
+quantized parcor coefficients into the LPC coefficients used for linear
+prediction.
+
+One of the unique features of the ALS format is that it uses progressively
+increasing prediction orders for the first samples in random access frames,
+which cannot reference previous samples. Other formats, such as FLAC, write the
+first samples directly, then proceed with prediction using the full number of
+coefficients. In ALS, the first sample is written directly, but each subsequent
+sample is predicted using increasingly higher orders until the full prediction
+order is reached.
+
+
+Long-Term Prediction
+--------------------
+Also called pitch coding, long-term prediction as implemented in ALS reduces
+signal redundancy by predicting the value of an LPC residual sample using the
+average value of 5 residual samples at a specified time lag with a specified
+gain value. The optimal lag and gain values are determined for each frame by the
+encoder. This is called pitch coding because the frequency, or pitch, of a
+signal leads to periodic redundancy. If the period is calculated, it can be
+exploited in order to remove that redundancy from the encoded output. This
+algorithm is typically used in speech codecs, such as Speex.
+
+
+RLSLMS (Recursive Least Squares, Least Mean Square) backward prediction
+-----------------------------------------------------------------------
+ALS has an alternative backward prediction mode, which uses the recursive least
+squares algorithm[8].
+
+Entropy Coding
+--------------
+In ALS, both the prediction coefficients and the residual signal are compressed
+using entropy coding. The encoder can choose between 2 different methods,
+Golomb-Rice Coding[9] and Block Gilbert-Moore Coding (BGMC). Rice coding is a
+form of Huffman coding, while BGMC, also called Shannon-Fano-Elias coding, is a
+precursor of arithmetic coding.
+
+
+Floating-Point Data
+-------------------
+The ALS format provides the ability to losslessly encode floating-point[10]
+audio. The floating-point signal is basically broken down into an integer signal
+plus the residual signal. The integer signal is encoded using the standard ALS
+algorithm, while the residual is encoded using the Masked-Lempel-Ziv
+dictionary-based compression scheme[11].
+
+
+References
+----------
+[1] http://www.nue.tu-berlin.de/forschung/projekte/lossless/mp4als.html
+[2] http://en.wikipedia.org/wiki/Audio_Lossless_Coding
+[3] http://www.nue.tu-berlin.de/wer/liebchen/lpac.html
+[4] http://en.wikipedia.org/wiki/Linear_predictive_coding
+[5] http://en.wikipedia.org/wiki/Linear_prediction
+[6] http://en.wikipedia.org/wiki/Levinson_recursion
+[7] http://en.wikipedia.org/wiki/Autocorrelation
+[8] http://en.wikipedia.org/wiki/Recursive_least_squares_filter
+[9] http://en.wikipedia.org/wiki/Golomb_coding
+[10] http://en.wikipedia.org/wiki/IEEE_floating-point_standard
+[11] http://en.wikipedia.org/wiki/Dictionary_coder
+
+
+
+BITSTREAM FORMAT
+================
+
+ALS Header
+----------
+In the raw ALS file format, the ALS header is located at the beginning of the
+bitstream, and is transmitted only once. It contains information about the
+audio and global properties of the encoded ALS bitstream.
+
+als_header() {
+ fixed_portion() 22 bytes
+ if(channel_config) {
+ ch_config 16 bits
+ }
+ if(channel_reorder) {
+ // number of bits used to code each channel position
+ ch_bits = max(1, ceil(log2(channels)))
+ for(i=0; i<channels; i++) {
+ channel_position[i] ch_bits
+ }
+ }
+ header_size 32 bits
+ trailer_size 32 bits
+ header() header_size
+ trailer() trailer_size
+}
+
+The fixed portion of the ALS header is always 22 bytes long. The field layout is
+given in Table 1. If the channel configuration flag is indicated, the
+configuration data is written. The channel configuration data is currently not
+used, and neither its purpose nor its format are given in the draft spec or in
+the reference software. If the channel reorder flag is indicated, the channel
+layout is written. The number of bits used to encode the position of each
+channel is determined as the minimum number of bits needed to encode the
+highest channel index. Next, the header size and trailer size (in bytes) are
+written. These sizes are used by the decoder to read the original header and
+trailer, which make up the last 2 fields of the ALS header.
+
+Table 1 : 22-byte fixed portion of the ALS header
++------------------------------------------------------------------------------+
+|bits |parameter |description |
+|==========+==================+================================================|
+|32 |magic number |"ALS\0" or 0x414C5300. Indicates that this is |
+| | |an ALS file |
+|----------+------------------+------------------------------------------------|
+|32 |sample rate |Audio sampling frequency, e.g. 44100 |
+|----------+------------------+------------------------------------------------|
+|32 |samples |Total number of samples in the file |
+|----------+------------------+------------------------------------------------|
+|16 |channels |Number of channels - 1 |
+|----------+------------------+------------------------------------------------|
+|3 |file type |Type of the original audio file |
+| | |0=raw/unknown 1=WAVE 2=AIFF |
+|----------+------------------+------------------------------------------------|
+|3 |resolution |Bits-per-sample/8 - 1 |
+|----------+------------------+------------------------------------------------|
+|1 |sample type |0=integer samples 1=floating-point samples |
+|----------+------------------+------------------------------------------------|
+|1 |source order |Source sample endianness 0=little 1=big |
+|----------+------------------+------------------------------------------------|
+|16 |frame size |Source frame size - 1 (measured in samples) |
+|----------+------------------+------------------------------------------------|
+|8 |random access |Random access frame interval 0=no RA frames |
+|----------+------------------+------------------------------------------------|
+|2 |ra location |Random access data location |
+| | |0=none 1=frames 2=header |
+|----------+------------------+------------------------------------------------|
+|1 |adapt |Indicates use of adaptive LPC order |
+|----------+------------------+------------------------------------------------|
+|2 |coef table |Selects 1 of the 3 tables of Rice code |
+| | |parameters used to encode parcor coefficients. |
+| | |Selection is generally based on the sample rate.|
+| | |3=use 7-bit quantizer instead of Rice coding. |
+| | |See Tables 2, 3, and 4 |
+|----------+------------------+------------------------------------------------|
+|1 |long-term pred |Indicates use of long-term prediction |
+|----------+------------------+------------------------------------------------|
+|10 |max order |Maximum LPC order |
+|----------+------------------+------------------------------------------------|
+|2 |block switching |Indicates use of block switching and maximum |
+| | |level of block division |
+|----------+------------------+------------------------------------------------|
+|1 |entropy coder |Type of entropy coder used 0=Rice 1=BGMC |
+|----------+------------------+------------------------------------------------|
+|1 |subblock part |Indicates partitioned entropy coding |
+|----------+------------------+------------------------------------------------|
+|1 |joint stereo |Indicates use of joint stereo coding |
+|----------+------------------+------------------------------------------------|
+|1 |mcc |Indicates use of multi-channel correlation |
+|----------+------------------+------------------------------------------------|
+|1 |channel config |Indicates presence of channel configuration |
+|----------+------------------+------------------------------------------------|
+|1 |channel reorder |Indicates presence of channel reordering |
+|----------+------------------+------------------------------------------------|
+|1 |has crc |Indicates presence of reference CRC checksum |
+|----------+------------------+------------------------------------------------|
+|1 |rlslms |Indicates use of RLSLMS backward prediction |
+|----------+------------------+------------------------------------------------|
+|5 |reserved |Not used |
+|----------+------------------+------------------------------------------------|
+|1 |aux enabled |Indicates presence of auxilliary data |
++------------------------------------------------------------------------------+
+
+Table 2 : Coef table 0 : Rice code params for encoding of parcor coefficients
+ Recommended for 44kHz and 48kHz
++--------------------------------+
+|coef number |offset |rice param |
+|============+=======+===========|
+|0 |-52 |4 |
+|------------+-------+-----------|
+|1 |-29 |5 |
+|------------+-------+-----------|
+|2 |-31 |4 |
+|------------+-------+-----------|
+|3 |19 |4 |
+|------------+-------+-----------|
+|4 |-16 |4 |
+|------------+-------+-----------|
+|5 |12 |3 |
+|------------+-------+-----------|
+|6 |-7 |3 |
+|------------+-------+-----------|
+|7 |9 |3 |
+|------------+-------+-----------|
+|8 |-5 |3 |
+|------------+-------+-----------|
+|9 |6 |3 |
+|------------+-------+-----------|
+|10 |-4 |3 |
+|------------+-------+-----------|
+|11 |3 |3 |
+|------------+-------+-----------|
+|12 |-3 |2 |
+|------------+-------+-----------|
+|13 |3 |2 |
+|------------+-------+-----------|
+|14 |-2 |2 |
+|------------+-------+-----------|
+|15 |3 |2 |
+|------------+-------+-----------|
+|16 |-1 |2 |
+|------------+-------+-----------|
+|17 |2 |2 |
+|------------+-------+-----------|
+|18 |-1 |2 |
+|------------+-------+-----------|
+|19 |2 |2 |
+|------------+-------+-----------|
+|2k, k>=10 |0 |2 |
+|------------+-------+-----------|
+|2k+1, k>=10 |1 |2 |
+|------------+-------+-----------|
+|k>=127 |0 |1 |
++--------------------------------+
+
+Table 3 : Coef table 1 : Rice code params for encoding of parcor coefficients
+ Recommended for 96kHz
++--------------------------------+
+|coef number |offset |rice param |
+|============+=======+===========|
+|0 |-58 |3 |
+|------------+-------+-----------|
+|1 |-42 |4 |
+|------------+-------+-----------|
+|2 |-46 |4 |
+|------------+-------+-----------|
+|3 |37 |5 |
+|------------+-------+-----------|
+|4 |-36 |4 |
+|------------+-------+-----------|
+|5 |29 |4 |
+|------------+-------+-----------|
+|6 |-29 |4 |
+|------------+-------+-----------|
+|7 |25 |4 |
+|------------+-------+-----------|
+|8 |-23 |4 |
+|------------+-------+-----------|
+|9 |20 |4 |
+|------------+-------+-----------|
+|10 |-17 |4 |
+|------------+-------+-----------|
+|11 |16 |4 |
+|------------+-------+-----------|
+|12 |-12 |4 |
+|------------+-------+-----------|
+|13 |12 |3 |
+|------------+-------+-----------|
+|14 |-10 |3 |
+|------------+-------+-----------|
+|15 |7 |3 |
+|------------+-------+-----------|
+|16 |-4 |3 |
+|------------+-------+-----------|
+|17 |3 |3 |
+|------------+-------+-----------|
+|18 |-1 |3 |
+|------------+-------+-----------|
+|19 |1 |3 |
+|------------+-------+-----------|
+|2k, k>=10 |0 |2 |
+|------------+-------+-----------|
+|2k+1, k>=10 |1 |2 |
+|------------+-------+-----------|
+|k>=127 |0 |1 |
++--------------------------------+
+
+Table 4 : Coef table 2 : Rice code params for encoding of parcor coefficients
+ Recommended for 192kHz
++--------------------------------+
+|coef number |offset |rice param |
+|============+=======+===========|
+|0 |-59 |3 |
+|------------+-------+-----------|
+|1 |-45 |5 |
+|------------+-------+-----------|
+|2 |-50 |4 |
+|------------+-------+-----------|
+|3 |38 |4 |
+|------------+-------+-----------|
+|4 |-39 |4 |
+|------------+-------+-----------|
+|5 |32 |4 |
+|------------+-------+-----------|
+|6 |-30 |4 |
+|------------+-------+-----------|
+|7 |25 |3 |
+|------------+-------+-----------|
+|8 |-23 |3 |
+|------------+-------+-----------|
+|9 |20 |3 |
+|------------+-------+-----------|
+|10 |-20 |3 |
+|------------+-------+-----------|
+|11 |16 |3 |
+|------------+-------+-----------|
+|12 |-13 |3 |
+|------------+-------+-----------|
+|13 |10 |3 |
+|------------+-------+-----------|
+|14 |-7 |3 |
+|------------+-------+-----------|
+|15 |3 |3 |
+|------------+-------+-----------|
+|16 |0 |3 |
+|------------+-------+-----------|
+|17 |-1 |3 |
+|------------+-------+-----------|
+|18 |2 |3 |
+|------------+-------+-----------|
+|19 |-1 |3 |
+|------------+-------+-----------|
+|2k, k>=10 |0 |2 |
+|------------+-------+-----------|
+|2k+1, k>=10 |1 |2 |
+|------------+-------+-----------|
+|k>=127 |0 |1 |
++--------------------------------+
+
+
+ALS Frame
+---------
+if(RA is enabled, RA location is in frames, and current frame is an RA frame) {
+ * random access unit size, 32 bits
+}
+if(MCC is enabled) {
+ if(joint stereo is not enabled) {
+ * use MCC for this frame
+ } else {
+ * read 8 bits to determine if this frame uses MCC
+ seems only 1 bit should be needed. not sure what the other 7 bits are
+ used for...
+ }
+}
+if(MCC is not enabled and RLS-LMS is not enabled) {
+ for each channel or channel pair {
+
+ * enabled coupled block switching if joint stereo is enabled, this is
+ the first channel in a pair, and this is not the last channel
+
+ if(block switching is enabled) {
+ * read block switch flags (8-bit, 16-bit or 32-bit)
+ coupled block switching can be explicitly turned off here
+ } else {
+ * use only 1 subblock at full frame size
+ }
+
+ if(coupled block switching is enabled) {
+ for each subblock {
+ * decode subblock for 1st channel in pair
+ * decode subblock for 2nd channel in pair
+ * restore original signal if one of the pair is difference signal
+ }
+ } else {
+ for each subblock {
+ * decode subblock.
+ * if coupled block switching is explicitly disabled for this,
+ pair, reconstruct the difference signal after reading the 2nd
+ channel in the pair. this is for use as predictor context for
+ the next frame.
+ }
+ }
+ }
+} else if(RLS-LMS is enabled) {
+ TODO: complete this section
+} else if(MCC is enabled) {
+ TODO: complete this section
+}
+
+Block switch flags
+------------------
+The block switch value in the header indicates the number of flag bits that are
+contained in each block. If all 32 bits are not used, the flag bits are left
+aligned in a 32-bit integer. This 32-bit integer making up the block switch
+flags is treated as an array of 1-bit values containing a binary tree. Index 0
+(the high bit) indicates if channel coupling is turned off in order to do
+independent block switching. The root node is located at index 1. The layout
+follows the standard binary-tree-in-array implementation. For a node at index i,
+the parent node is at i/2, the first child is at i*2, and the second child is at
+i*2+1. If a node has a value of 1, the corresponding subblock is split into 2
+smaller equal-sized subblocks. If a node has a value of 0, the subblock is not
+split. The smallest subblock size is 1/32 of the original frame size. For the
+last frame in the stream, which may be smaller than all the other frames, the
+subblocks are divided as if it were a full-size frame, but only the number of
+subblocks which are needed to fit the final number of samples are used, with
+the last subblock being truncated if needed.
+
+
+ALS Subblock
+------------
+* align bitstream to 8-bit boundary
+* 1-bit frame type. 0=zero or constant, 1=normal
+if(frame type is zero or constant) {
+ * 1-bit frame type. 0=zero, 1=constant
+}
+* 1-bit difference flag indicates if this subblock contains a normal
+ signal or a difference signal.
+if(frame type is constant) {
+ * align bitstream to 8-bit boundary
+ * read 1 sample. size based on sample resolution in the ALS header.
+} else if(frame type is normal) {
+ // determine entropy partitioning and read entropy coding parameters
+ if(coding type is Rice) {
+ if(entropy partitioning is enabled) {
+ * 1-bit partition flag. 0=1 partition, 1=4 partitions
+ } else {
+ * only 1 partition
+ }
+ * read 1st Rice parameter (4-bit if depth<=16-bit, 5-bit if >16-bit)
+ if(partitioning is enabled) {
+ * read 3 remaining delta-encoded Rice parameters. deltas are Rice
+ encoded using parameter=0.
+ }
+ } else if(coding type is BGMC) {
+ if(entropy partitioning is enabled) {
+ * 2-bit partition order. 2^x partitions (1, 2, 4, or 8)
+ } else {
+ * 1-bit partition flag. 0=1 partition, 1=4 partitions
+ }
+ * read 1st BGMC parameter (8-bit if depth<=16-bit, 9-bit if >16-bit)
+ * read remaining delta-encoded Rice parameters. deltas are Rice
+ encoded using parameter=2.
+ * separate grouped BGMC parameters:
+ s = p >> 4
+ sx = p & 0x0F
+ }
+ // shift (unused LSB's)
+ * read 1-bit shift flag
+ if(shift flag) {
+ * read 4-bit shift value, add 1. this is the number of unused LSB's.
+ }
+ // decode and reconstruct LPC coefficients
+ if(using LPC prediction [not RLS-LMS]) {
+ // determine LPC order
+ * order = fixed/max prediction order (from ALS header)
+ if(adaptive prediction order is used) {
+ * limit max order to range [1 ... (subblock size / 8) - 1]
+ * read order for this subblock using number of bits required to
+ store max order.
+ }
+ // decode quantized coefficients
+ if(not using a coefficient table [ct=3]) {
+ * read each quantized coefficient as 7-bit value - 64
+ } else {
+ * read quantized coefficients from bitstream using rice params from
+ the coefficient table.
+ * apply the offsets from the coefficient table
+ }
+ // reconstruct parcor coefficients
+ * reconstruct 1st coefficient from the quantized coefficient using the
+ scale parcor coefficient formula below:
+ where qc has a range of -64 to 63,
+ coeff = 32 + ((qc+64)*(qc+65)*128) - (1 << 20)
+ * reconstruct the 2nd coefficient the same way, then reverse the sign.
+ * reconstruct remaining coefficients using
+ coeff = (qc << (Q-6)) + (1 << (Q-7)), where Q=20
+ = (qc << 14) + (1 << 13)
+ = (qc * 16384) + 8192
+ } else if(using RLS-LMS) {
+ * order = 10
+ }
+ if(using LTP) {
+ // decode 5 LTP coefficients and lag
+ c0 = signed rice code (rp=1)
+ c1 = signed rice code (rp=2)
+ c2 = unsigned rice code (rp=2)
+ c3 = signed rice code (rp=2)
+ c4 = signed rice code (rp=1)
+ lag = x-bit unsigned value, where x = 8 + (samplerate / 96000)
+ }
+ // decode residual for this subblock
+ if(no random access for this frame) {
+ for(each entropy coding partition) {
+ if(coding type is Rice)
+ * decode residual for partition using Rice parameters
+ if(coding type is BGMC)
+ * decode residual for partition using BGMC parameters
+ }
+ } else {
+ // in RA frames, first residual values are larger because they are
+ // progressively predicted, so a different entropy coding is used in
+ // order to encode more efficiently
+ * decode up to 3 residuals separately, up to prediction order.
+ r0 = signed rice code (rp=bitdepth-4 [e.g. 16-4=12])
+ r1 = signed rice code (rp=1st rice parameter + 3 [max=31])
+ r2 = signed rice code (rp=1st rice parameter + 1 [max=31])
+ this will make the 1st partition shorter
+ for(each entropy coding partition) {
+ if(coding type is Rice)
+ * decode residual for partition using Rice parameters
+ if(coding type is BGMC)
+ * decode residual for partition using BGMC parameters
+ }
+ }
+}
+if(using RLS-LMS prediction) {
+ // TODO: RLS-LMS parameter decoding
+}
+if(using MCC) {
+ // TODO: decode MCC reference channels and weighting factors
+}
+
+
+Reconstructing the signal (LPC mode, no MCC)
+--------------------------------------------
+When LPC is used, each subblock contains its own set of LPC coefficients.
+After the residual has been decoded, the signal can be reconstructed using
+inverse linear prediction. When decoding the 1st subblock of a random access
+frame, progressive linear prediction is used because of the lack of access to
+previous "warm-up" samples.
+
+if(frame type is zero) {
+ * set all samples for the subblock to zero
+} else if(frame type is constant) {
+ * set all samples for the subblock to previously decoded constant value
+} else if(frame type is normal) {
+ * temporarily apply shift to warm-up samples from previous block
+ if(using LTP) {
+ * reconstruct final residual by reversing LTP
+ }
+ if(not a random access frame) {
+ * convert parcor coefficients to LPC coefficients using pseudo-code
+ below. keep in mind the need to avoid integer data type overflow.
+ for(m=0; m<pred_order; m++) {
+ for(i=0; i<m/2; i++) {
+ temp = lpc[ i] + (((par[m] * lpc[m-i]) + (1<<19)) >> 20);
+ temp2 = lpc[m-i] + (((par[m] * lpc[ i]) + (1<<19)) >> 20);
+ lpc[m-i] = temp2;
+ lpc[ i] = temp;
+ }
+ lpc[m] = par[m];
+ }
+ * inverse linear prediction using code below
+ note that samples below index 0 are from the previous subblock
+ for(each sample, at index n, in the subblock) {
+ sum = 0;
+ for(i=0; i<pred_order; i++)
+ sum += lpc[i] * signal[n-i-1];
+ signal[n] = residual[n] - ((sum + (1 << 19)) >> 20);
+ }
+ } else if(random access frame) {
+ // TODO: RA inverse LPC (progressive prediction)
+ }
+ * undo shift from warm-up samples and block samples
+}
+
+
+TODO: Reconstructing the signal (RLS-LMS mode)
+
+TODO: Reconstructing the signal (MCC mode)
This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
|