TW315561B - A multi-channel predictive subband audio coder using psychoacoustic adaptive bit allocation in frequency, time and over the multiple channels - Google Patents

Abstract

A subband audio coder 12 employs perfect/non-perfect reconstruction filters 34, predictive/non-predictive subband encoding 72, transient analysis 106, and psycho-acousti c/minimum mean-square-error (mmse) bit allocation 30 over time, frequency and the multiple audio channels to encode/decode a data stream to generate high fidelity reconstructed audio. The audio coder windows 64 the multi-channel audio signal such that the frame size, i.e. number of bytes, is constrained to lie in a desired range, and formats the encoded data so that the individual subframes can be played back as they are received thereby reducing latency. Furthermore, the audio coder processes the baseband portion (0-24kHz) of the audio bandwidth for sampling frequencies of 48kHz and higher with the same encoding/decoding algorithm so that audio coder architecture is future compatible.

Description

A7 B7 315561 V. Description of the invention (1) (Please read the notes on the back before filling in this page) Background of the invention The field of the invention. The invention is related to the high-quality encoding and decoding of multi-channel audio signals, especially related to time , Frequency and multi-frequency channels using complete / incomplete luck filters, predictive / non-predictive subband coding, transient analysis, and mental hearing / minimum mean square error (mmse) bit allocation subbands Encoders are used to generate a data stream with limited decoding computational load. Related Art Description It is known that high-quality audio and music encoders can be divided into two broad systems. The first is a sub-band / transform encoder suitable for high-frequency resolution in the sub-band or coefficient selection in the quantitative analysis window based on a mental auditory masking calculation. The second is the use of ADPCM to process the sub-band selection, so as to form its low-resolution sub-band encoder with poor frequency resolution. The Ministry of Economic Affairs Central Standards Bureau Employee Consumer Cooperative printed the first type of encoder by allowing the bit Distribute and use the large-scale short-term spectrum variation of general music signals to adapt to the spectrum energy of the signal. The high resolution of these encoders allows frequency-converted signals to be directly applied to the mental auditory model based on the critical band theory of hearing The AC-3 audio encoder (Todd et al., February 1994 Convention of the Audio Engineering Society > " AC 3 Flexible Perceptual Coding for Audi. Transmission and St. rage " usually calculates l〇24- on individual PCM signals ffts and apply a mental hearing model to the 1024 frequency coefficients in each channel to determine the bit transmission rate of each coefficient. The Dolby system uses a transient analysis to reduce the window size to 256 samples to isolate each transient 0 The AC-3 encoder uses a patented reverse adaptation script. The paper size is applicable to China National Standards (CNS) A4 specification (210 X 297 ) 315561 a? ____B7 ____ 5. Description of the invention (2) The algorithm decodes the bit allocation. This reduces the amount of bit allocation information sent by the marginally encoded audio data. The result can be used for the audio frequency bandwidth in the forward adaptation system. The increase leads to an improvement in sound quality. In the second type of encoder, the quantization of each differential subband signal is fixed or adapted to minimize the quantization noise power in all or part of the subband and minimize There is no obvious reference to the theory of mental hearing masking. Generally, the recipient is a direct mental hearing loss threshold. Predictive / differential subband signals cannot be applied, because it is difficult to estimate the performance of the predictor before the bit allocation process. The problems are further mixed with the quantization noise in the prediction process. The work of these encoders is that the critical audio signal due to perception is roughly periodic in the long term. This periodicity is caused by the predictive difference quantization Use. Divide the signal into a small number of sub-bands, reduce the audible effect of noise modulation, and allow the use of long-term spectrum changes of the audio signal If the number of sub-bands is increased, the predicted gain in each sub-band is reduced, and at some time the predicted gain will approach zero. Printed by the Consumer Labor Cooperative of the Central Bureau of Standards of the Ministry of Economic Affairs ( (Please read the notes on the back before filling this page)

Digital Theater Systems, LP (DTS) uses an audio encoder, where each PCM audio channel is filtered into four sub-bands, and each sub-band is encoded using a reverse ADP CM encoder to adapt the predictor coefficients In the subband data, the bit allocation is fixed, the same for each channel, the lower. The frequency subband is assigned more bits, which is more than the higher frequency subband. The bit allocation provides a Fixed compression ratio, such as 4: 1 〇DTS encoder described in 1991 Proceedings of the 10th International AESC ο nference, 4 1-5 6 pages, Mike S myth and S tephen S myth This paper scale is applicable to Chinese national standards ( CNS) Λ4 specification (210X297 mm) 315561 Printed by the Central Sample Bureau of the Ministry of Economic Affairs Employee Consumer Cooperative V. Invention Description (3) " APT-X100: A LOW-DELAY, LOW BIT-RATE, SUB-BAND ADPCM AUDIO CODERFOR BROADCASTING " ○ Class 2 audio encoders have other common limitations. First, known audio encoders use a fixed frame size for encoding / decoding, that is, the sample represented by a frame The target or time period is fixed. The result is that when the encoding transmission rate increases relative to the sampling rate, the amount of data (bytes) in the frame also increases. Thus, the decoder buffer size must be designed to accommodate the worst case System to prevent data overflow. This increases the amount of RAM, which is the main cost component of the decoder. Second, the known audio encoder is not easy to expand to a sampling frequency greater than 48 kHz (000 Hz). Doing so will make Existing decoders are not compatible with the format required by the new encoder. This lack of future compatibility is a serious limitation. Furthermore, the known format used to encode each PCM data needs to be adjusted before playback is started. The frame is read in by the decoder. This requires the buffer size to be limited to about ms blocks of data, so that the delay or latency is not troublesome. Annoying. In addition, although these encoders have encoding capabilities up to 24 'kHz, Higher sub-bands are often abandoned. This reduces the high frequency or environment of the reconstructed signal. Known encoders generally use one of two error detection systems. The most commonly used is the Read Solo mo η encoding method. The middle encoder adds error detection bits to the marginal information in the data stream. This helps in the detection and correction of any errors in the marginal information. However, the errors in the audio data are not detected. For example, 0 of the invalid code state check of the audio header. For example, a specific 3-bit parameter may only have 3 valid states. If one of the other 5 states is identified, an error must occur. This only provides detection capabilities. The paper size is applicable to China National Standard (CNS) Α4 specification (210Χ297mm) — 6 — (please read the precautions on the back before filling this page) --- ^ 装 —

Sufi ml nn ffml I order Λ: ----

5. Description of the invention (4) without detecting errors in the audio data The rate has better quality than CD, and the improved sensory quality at low bit transmission rate, with reduced playback latency, simplified error detection, improved previous echo loss, and future expansion to higher sampling rates 0

Printed by the Consumer Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs (please read the precautions on the back before filling in this page) This is a sub-band encoder, which frames each audio channel window into a sequence of audio frames and filters each frame Form a fundamental band and a high frequency range, and decompose each fundamental band signal into multiple sub-bands. When the bit transmission rate is low, the subband encoder normally selects an incomplete filter to decompose the fundamental band signal, but when the bit transmission rate is sufficiently high, a complete filter is selected. The high frequency coding stage is independent Encode the high-frequency signal of the fundamental band signal. The fundamental band encoding stage includes a VQ and an ADPCM encoder, which encode the higher and lower frequency sub-bands respectively. Each time the band frame includes at least one frame , Each of them is further subdivided into multiple sub-frames. Each sub-frame is analyzed to estimate the prediction gain of the ADPCM encoder. When the prediction gain is low, the prediction ability is disabled, and each transient is detected to adjust the transient and The latter SF 0-wide area bit management (GBM) system assigns each bit to each subframe, which utilizes the differences between multiple audio channels, multiple subbands, and subframes within the current frame. The GBM system first calculates its predicted gain adjusted SMR to allocate bits to each subframe to satisfy a mental hearing model 0. Then the GBM system allocates any remaining bits according to a MSE processing method, or immediately switches to -MMSE allocation , Reduce all noise base values, or gradually transform into a MMSE sub-sheet. The paper standard is applicable to the Chinese National Standard (Shenyang) 84 specifications (2) 0 297 mm) 1 Ί-3ί556ί at Β7 5. Description of the invention (5 ) Match. -The multiplexer generates each output frame, which includes a sync (sync) word, a frame header, an audio header, and at least one sub-frame, which are multiplexed into a data stream at a transmission rate. The frame header includes the size of the window and the size of the current output frame. The audio table header indicates a compression configuration and one of the audio frame encoding formats. Each audio sub-frame includes marginal information used to decode the audio sub-frame without reference to any other sub-frames, high-frequency YQ codes, multiple fundamental-band audio sub-frames, where each channel uses lower-frequency sub-band audio frequencies Data is compressed and multiplexed with other channels, a high-frequency audio section, in which high-frequency range audio data for each channel is compressed and multiplexed with other channels, so that multi-channel audio signals can be multiple Decode at the decoded sample rate, and one to verify the decompression synchronization at the end of the subframe. The window size is selected as a function of the ratio of transmission rate to encoder sampling rate, so that the size of the output frame is limited to a desired range. When the amount of compression is quite low, the window size is reduced so that the frame size does not exceed a higher maximum value. The result is that a decoder can use an input buffer with a fixed and relatively small amount of RAM. When the amount of compression is quite high, the window size is increased. The result is that the GBM system can distribute each bit to a larger time window, thereby improving the performance of the encoder. These and other features and advantages of the present invention are detailed from the following preferred specific forms together with the drawings and tables. It will be obvious to the skilled personnel, among them: Brief description-Figure 1 is a block diagram of a 5-channel audio encoder according to the present invention; This paper scale is applicable to China National Standard (CNS) Α4 regulation (210X297mm) ~ ^ ^ -8- Λ (please read the precautions on the back before filling in this page)-installed

Printed by the Huanggong Consumer Cooperatives, Central Bureau of Standards, Ministry of Economic Affairs. I A7 —_B_L— '__ 5. Description of the invention (6) Figure 2 is a block diagram of a multi-channel encoder; Figure 3 is a block diagram of a baseband encoder and decoder Figure; Figure 4 a and 41) is divided into a block diagram of a high sampling rate encoder and decoder, Figure 5 is a block diagram of a single channel encoder; Figure 6 is a variable transmission rate per frame bit group Figure of the frame size; Figure 7 is a diagram of the amplitude response to the NPK and PR reconstruction filters; Figure 8 is a diagram of the subband alias setting for a reconstruction filter; Figure 9 is a diagram of the NPR and PR filters Figure 10 is a schematic diagram of a single primary band encoder; Figure 11a and lib are the transient detection of the primary frame and the calculation of the scale factor; Figure 12 illustrates the entropy encoding method of quantized TM0DES Procedure; Fig. 13 illustrates the scale factor quantization procedure; Fig. 14 illustrates a signal masking: the convolution, which has a signal frequency response to generate SMR, "Figure 15 is a diagram of the human auditory response;" Figure 16 is the subwave Diagram of SMR with band; Figure 17 is a diagram of the error signal of mental hearing and mmse bit allocation; Figure 18a lSb is the sub-band energy level diagram and inversion diagram respectively, exemplifying the mms e "water-filling" bit configuration procedure; Figure 19 is a block diagram of a single frame in the data stream; Figure 20 is a schematic diagram of the decoder; Figure 2 1 is a block diagram of the encoder hardware. One of the examples; the paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210 father 197 ^ _) '™ ^ ~ "' -9 — --- ------ 、 装 — (Please read the precautions on the back before filling in this page) -9 Printed by the Employees ’Consumer Cooperative of the Central Bureau of Standards of the Ministry of Economic Affairs ^ 15561 A7 __ B7 5. Description of the invention (7) Figure 22 shows the decoding Block diagram of one of the hardware embodiments of the device. Brief description of the table. Table 1 is the maximum: t box size vs. sampling rate and transmission rate; Table 2 is the maximum 4 allowed frame size (bytes) vs. sampling rate and transmission rate Table 3 shows the relationship between the τκ ABIT index value, the number of quantization levels and the SNR of the formed sub-band. The present invention details the multi-channel audio coding system. As shown in FIG. 1, the present invention has combined The characteristics of the known coding system plus both the additional characteristics of a single multi-channel audio encoder 1 Q. The coding algorithm It is designed to be performed at the studio quality level, that is, better than CD " quality, and provides a wide range of applications that change the compression level, sampling rate, word length, number of channels, and perceived quality. The encoder 12 uses PCM audio Multiple channels of data H (usually sampled at 48 kHz and word length between 16 and 24 bits) are encoded into a data stream 16 at a known transmission rate (preferably within the range of 32-40 9 6 kbps). Known audio encoders, the architecture of this case can be expanded to a higher sampling rate (48-92 kHz) without being incompatible with existing decoders designed for the fundamental band sampling rate or any intermediate sampling rate. In addition, the PCM data 14 is windowed and coded in a one-frame, and each frame is preferably divided into 14 subframes. The size of the audio window (that is, the number of I ”CM samples) is based on the relative value of the sampling rate and the transmission rate, so that the size of an output frame read by each frame of the decoder 18 (that is, the number of bytes) is restricted, which is appropriate For the introduction. Between 5.3 and 8 kilobytes. This paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210X297mm) -10- (please read the precautions on the back before filling this page), τ printed by 3l556i at r ~ — ___ _ B7 V. Description of the invention (8) As a result, the amount of RAM needed to enter the data stream in the decoder is kept relatively low, which reduces the cost of the decoder. At low rates, the larger window size can be used to frame PCM data, which improves coding performance. At higher bit rates, a smaller window size must be used to meet the data limit. This will inevitably reduce the encoding performance, but this will not be significant at higher rates. Also, the way the PCM data is framed allows the decoder 18 to start playback before the entire output frame is read into the buffer. This reduces the delay or latency of the audio encoder. The encoder 12 uses a high-resolution filter bank, which preferably reconstructs the filter in terms of incompleteness (NPR) and completeness (PR) according to the bit transmission rate Switching between them, each audio channel 14 is decomposed into many subband signals. The predictive and vector quantization (VQ) encoder is used to encode the lower and higher frequency subbands, respectively. The initial YQ sub-band can be fixed or can be dynamically determined as a function of the nature of the current signal. The joint frequency coding method can be used at a low bit rate to simultaneously encode multiple channels in the higher frequency subband. The predictive encoder preferably switches between APCM and ADPCM modes based on the subband predictive gain. A transient analyzer separates each subband subframe into pre-echo and post-signal (sub-sub-frame), and calculates the respective scale factor for the sub-frames of pre- and post-echo to reduce the pre-echo miss. The encoder adapts the available bit transmission rate to the current frame of all PCM channels and sub-bands according to their respective needs (mental hearing or mse), so as to optimize the coding efficiency. By combining predictive coding with a mental hearing model, the coding efficiency of the low bit transmission rate is improved, thereby reducing the bit transmission rate that achieves subjective transparency. A programmable controller 19, such as a computer or keyboard, is connected to the encoder 12 to relay audio mode information, including, for example, the desired bit transmission. The paper standard is applicable to the Chinese National Standard (CNS) A4 specification (2i〇X297 Mm) ^ ~ -11- (please read the precautions on the back before filling in this page) Packing_ Order A7 A7 Printed Bag B7 of the Employees ’Consumer Cooperative of the Central Standards Bureau of the Ministry of Economy V. Invention Description (9) Rate, Number of Channels, PR Or parameters such as NPR reconstruction, sampling rate and transmission rate. Encoded signal and sideband information. It is compressed and multiplexed to the data stream 16, so that the decoding calculation load is constrained to be within the desired range. Data stream 16 is encoded or broadcast on a transmission medium 2: 0 such as CD, digital video disc (DVD), or direct broadcast satellite. Decoder 18 decodes individual sub-band signals and performs inverse filtering operations To generate a multi-channel audio signal 22, which is subjectively equivalent to the original multi-channel audio signal H ύ-audio system 24, such as a home theater system or a multimedia computer, to reproduce the audio signal for the user. The multi-channel encoder is shown in Figure 2. The encoder 12 includes a plurality of (preferably 5) individual channel encoders 26 (front left, center, front right, rear left, and rear right), which generate encoding subbands for each group Signal 28 (suitably 32 subband signals per channel). The encoder 12 uses a wide area bit management (GBM) system 30, which consists of a common bit pool in each channel, between subbands within a channel, and within a known subband Each element is dynamically allocated in an individual box. The encoder 12 may also use joint frequency coding techniques to take advantage of inter-channel correlation in the higher frequency sub-band. In addition, the encoder 12 can use vector quantization for higher frequency subbands that are particularly imperceptible, and provide a basic high frequency or environment at an extremely low bit transmission rate. In this method, the encoder Utilize the disparate signal requirements of multiple channels, such as the rms value of the sub-band and the level of mental hearing masking, and the uneven distribution of the signal energy in the frequency of each channel and the time in a fixed frame. Bit allocation overview GBM system 30 first determines which channel's secondary wave band will be used as the joint frequency. This paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210X2? 7mm): _ -12- (please read the back side first (Notes and fill in this page again) m- -1-—---. I ......... mm 1 I-·--ϊ I----_ «νι ^ ϋ —HI— vm mn In i 1 ^ 1 n ^ i Printed Α7 Β7 by the employee consumer cooperative of the Central Bureau of Standards of the Ministry of Economy V. Description of the invention (lϋ) Encoding and average the data, and then determine which sub-band will be encoded with the quantification of the child, And the available bit transmission rate minus these bits. The determination of which sub-band row vector quantization can be made first, because all sub-bands above a threshold frequency are vector quantized, or according to each frame The mental auditory masking effect of the individual sub-bands is created. Thereafter, the GBM system SO uses mental auditory masking to allocate bits (AB I T) on the remaining sub-bands to optimize the subjective quality of the decoded audio signal. If additional bits are available, the encoder can be converted into a pure mmse system, ie '^ Filled water', and all bits are redistributed according to the relative rms value of each wave band to reduce the rms value of the error signal to The smallest. This applies to extremely high bit transfer rates. The preferred method is to retain the mental auditory bit allocation, while according to the m m s e system, only additional bits are allocated. This maintains the shape of the noise signal generated by mental hearing and hearing shielding, but moves the noise base evenly downward. Alternatively, the preferred method can be modified to allocate additional bits based on the difference between r m s and the psychoacoustic level. As a result, psychoacoustic distribution changes to mm se distribution when the bit transmission rate increases, thereby providing a smooth transition between the two technologies. The technology above 0 is particularly suitable for fixed bit transmission rate systems. Alternatively, the encoder 12 can Set a missing level (subjective or mse) and allow the total bit transmission rate to change to maintain the missing level 0-the multiplexer 32 multiplexes the sub-band signals and marginal information into a specified data format Data stream 16 0. The details of this data format are discussed in Figure 20 below. For a band coding for sampling rates in the range of 8-4 8 k H z, the channel encoder 26, as shown in Figure 3, uses a uniform 512 tap 32 filter band analysis filter bank This paper size is applicable China _ Home Rubbing (CNS) Α4 specification (2ΙΟΧ297 order pen) -13 — Γ * • --------------- ^ Pack ------ ^, order- -: --- Λ;. (Please read the note Ϋ on the back before filling this page) A7 ____ B7 _ ^ __ printed by the Consumer Cooperative of the Central Escalation Bureau of the Ministry of Economic Affairs V. Invention description (11) 34, at 48 kHz The sampling rate operation divides the sound spectrum (0-24 kHz) of each channel into 32 sub-bands, and each sub-band has a bandwidth of 750 Hz. Encoding stage 3 6 Encode each wave band signal and pre-multiplex 3 8 into a compressed data stream 1 6 ◦Decoder +18 receives the compressed data stream and uses a decompressor 4 〇 to encode each wave band Separate the data, decode each band signal 4 2 and use a 5 1 2 tap 3 2 band uniform interpolation filter group 4 4 to reconstruct the PCM digital audio signal for each channel (Fsamp = 48 kHz). In this architecture, all encoding strategies, for example, 96 or 192kHz sampling rate, use 32-band encoding / decoding procedures at the lowest (fundamental band) audio frequency, such as 〇-2 4 kHz. Therefore, decoders designed and constructed today based on the 48 kHz sampling rate will be compatible with future encoders designed to utilize higher frequency components. The current decoder reads the fundamental band signal (〇-24 kHz) and omits the coded data at higher frequencies. Sampling rate coding For a sampling rate in the range of 48-96 kHz, the channel encoder 26 preferably divides the audio spectrum into For both, use a uniform 32-band analysis filter bank for the lower half, and an 8-band analysis filter bank for the upper half. As shown in Figure 4 and 4b, the audio frequency spectrum (0-48 kHz) First, use a 2 56 tap 2 wave band ten to take a type of front filter bank 4 6 to be divided to obtain an audio bandwidth of 24 kHz per band. The bottom band (0-24 kHz) is as described in FIG. 3 It is divided and coded into 32 uniform bands. However, the upper band (24-48 kHz) is divided and coded into 8 uniform bands. If ten of the 8 bands take one / internal filter bank 48 If the delay is not equal to the 32-band filter bank, a delay compensation stage 50 must be used somewhere in the 24-48kHz signal path to ensure that the paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) — 14 — (Please read the precautions on the back before filling out this page) Binding—Binding Central Standards of the Ministry of Economic Affairs Printed by Aiko Consumer Cooperative A7 ____ _B7_ V. Description of the invention (I2). The time waveforms of the two are aligned at the decoder before the 2-band composite filter bank. In the 96 kHz sample coding system, 24-48 kHz audio The band is delayed by up to 384 samples, and then divided into 8 uniform bands using a 128-tap interpolation filter bank. Each 3kHZ sub-band is encoded 52 and compressed together with the encoded data from the 0-24 kHz band 54, to form a compressed data stream 16 〇 When reaching the decoder 18, the compressed data stream 16 is decompressed 56, and 32-band decoder (ϋ-24 kHz region) and 8-band decoder (24-48kHz) both The codes used are separated and fed to their respective decoding stages 42 and 58. The 8 and S2 decoded sub-bands are uniform inner pan filter banks 60 and 4 using 128 taps and 512 taps respectively It is reconstructed. The decoded sub-band is subsequently compounded using a 2 5 6 tap 2 band homogenous filter box 6 2 to produce a single PCM digital audio signal with a sampling rate of 96 kHz. The operation of half the compressed data stream sampling rate, which can be discarded by Band coded data portion (2 "4 8kHz) and only the sub-band decoder 32 within the region of 0-24 kHz audio and easy to implement. Channel Encoder In all the encoding strategies described above, the 32-band encoding / decoding procedure is implemented for the fundamental band portion of the audio bandwidth between G-24 kHz. As shown in FIG. 5, a frame grabber 64 windowizes the PCM audio channel n to divide it into consecutive data frame 66 segments. The PCM audio window defines the number of consecutive input samples, for which the encoding process generates an output frame in the data stream. The window size is set based on the amount of compression (that is, the ratio of the transmission rate to the sampling rate) to limit the amount of encoded data in each frame. Each continuous data frame 6 6 is divided into a private paper standard by a 32-wave FIK ten-in-one filter bank 34 with 512 taps. It is applicable to the Chinese National Standard (CNS) A4 specification (210X297 public sauce) -15- (please first Read the notes on the back ^ item and fill out this page) Install ------ I order 丨 · ----- C, ----- Α7 3ί556Ι ____ Β7 5. Description of the invention (1 3) ™ 3 2 Uniform frequency band 68 0 The sample selected by each band output is buffered and applied to the 3 2 band coding stage 3 6 〇 (please read the precautions on the back before filling this page)-analysis stage 7ϋ (details (In Figures 10-1 9) Generate the best predictor coefficients, differential quantizer bit allocation and the best quantizer scale factor for each buffered subband sample selection. The analysis stage 70 can also determine which subband will be combined with Vector quantization (VQ) 'and which will be combined frequency coding, if these decisions are not fixed. This data, or marginal information, is fed forward to the selected ADPCM stage 72, VQ stage 73 or Joint Frequency Coding (JFC) stage 74, and to the data multiplexer 32 (compressor). Then, the sub-band selection is encoded via the ADPCM or VQ program, and the quantization code is input to the multiplexer. JFC stage 7 4 does not actually encode the sub-band selection, but generates a code indicating which channel's sub-band is combined and where it is placed in the data stream q the quantization code and margin of each sub-band The information is compressed into the data stream 16 and transmitted to the decoder. Printed by the Consumer Cooperative of the Ministry of Economic Affairs of the Ministry of Economic Affairs and arrived at the decoder 1S, the data stream is multiplexed 40 (or decompressed) and returned to the individual sub-bands. Scale factor and bit of each sub-band The distribution together with the predictor coefficients is first established in each inverse quantizer 75. Then, the differential codes of the respective designated sub-bands are directly reconstructed using the ADPCM program 76 or the inverse vector quantization program 77 or the inverse JFC program 78. Finally, the 32-band interpolation filter bank 4 4 is used to mix each sub-band back to a single p C M audio signal 22. The PCM signal is framed as shown in FIG. 6. When the transmission rate changes a fixed sampling rate, the frame grabber 6 shown in FIG. 5 changes the size of the window 7 9 so that each output frame has 80 bits. The number of groups is constrained and the paper size of, for example, 5.3k bytes and 8k bytes is in accordance with the Chinese National Standard (CNS) A4 specification (210X297 ^ t) ~ 1 A7 B7 Central Bureau of Standards, Ministry of Economic Affairs-industrial consumption Printed by the cooperative. 5. Description of Invention (H) Group. Tables 1 and 2 are the design tables for the designer to select the optimal window size and decoder buffer size (frame size) for a fixed sampling rate and transmission rate. At low transmission rates, the frame size can be larger. This allows the encoder to use the audio signal versus time; non-flat variance distribution and improve the performance of the audio encoder. At high transmission rates, the frame size is reduced so that the total number of bytes does not overflow the decoder buffer. As a result, the designer can provide the decoder with 8K bytes of KAM to meet all transmission rates. This reduces the cost of the decoder. In general, the size of the audio window is obtained as follows: Audio window = (frame size) * Fsamp * (8 / Trate) where the frame size is the size of the decoder buffer, F samp is the sampling rate, and Tr ate is the transmission rate . The size of the audio window has nothing to do with the number of audio channels. However, as the number of channels increases, the amount of compression must also increase to maintain the desired transmission rate. ^ _1_ F s a nip (kHz)

Trate 8-12 16-24 32-48 64-96 128-192 (please read the notes on the back before filling this page)

· —C installed I S 5.1 2 kbps $ 1024 kbps ^ 2048 kbps ^ 4096 kbps 1024 2 048 1024 40 96 2048 1024 ice 氺 氺 2 04 8 1024 * wood * 2 0 48

Trate 8-12 < 512 kbps 8-5, 3 k < 1024 kbps * < 2048 kbps * < 4096 kbps · * Table_2 Fsamp (kHz) 16-24 32-48 64-96 128-192 8 -5.3k 8-5. 3k 8-5. 3k 8-5. 3 k * 8-5. 3k 氺 氺

8-5. 3k * 8-5 · 3k 8-5, 3k, this paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210X297 g) A7 B7 Printed and printed by the Consumer Cooperative of the Ministry of Economic Affairs Zhongsui Standard Bureau Description (l5) Sub-band filter 3 2 band 5 1 2 Tap-to-even filter bank 3 4 Select from two polyphase filter banks to divide the data frame 6 6 into Figure 5 The 3 2 uniform sub-bands 68 are shown. The two filter banks have different reconstruction properties. The latter exchanges reconstruction accuracy with sub-band coding gain. There is a filter called a complete remapping (PR) filter. When the complete reconstruction decimation (encoding) filter and its interpolation (decoding) filter are placed back-to-back, the reconstructed signal is' " completeness ", where completeness is defined as 24 bits The resolution is within 0.51sb. Another kind of filter is called incomplete reconstruction (NPR) filter, because the reconstructed signal has a non-zero noise base value, which is related to the incompleteness of the filter program. The alias setting function is related to the nature of elimination. The individual transfer functions 82 and 84 of the NPH and PR filters for a single primary band are shown in FIG. 7. Since the NP1 {the wave filter is not constrained to provide complete reconstruction, its performance is much larger than that of the PR filter and is close to stop Band Rejection (NSBR) ratio, that is, the ratio of the passband to the first side lobe (110 dB to 85dB). As shown in Figure 8, the side lobe of the filter results in a natural location in the third band The signal 8 6 enters the adjacent sub-band with an alias. The sub-band gain measures the repulsive effect of signals in each adjacent sub-band, and thus indicates the Hb force of the filter to decorrelate the audio signal. Since ΝΡί (the wave filter has an NSBR ratio that is much larger than the PR wave filter, so its: # will also have a far larger subband gain. As a result, the NPR filter provides better coding efficiency 0 As shown in Figure 9, for both Pfi and NPβ filters, as the total bit transmission rate increases, the total amount of loss in the compressed data stream decreases. However, at low transmission rates, the two filter types The difference between the second band gain performance

It ml i «(Please read the note ^ h on the back before filling in this page) Pack ----, order ... _____

I —...... w ----- I—! I The paper standards apply the Chinese National Standards (CNS) Μ specifications (2) 〇 > < 297mm-18- The Central Bureau of Economic Development of the Ministry of Economic Affairs A7 B7 printed by the Employee Consumer Cooperative V. Description of the invention (The noise base related to the NPR filter is large. Therefore, the relevant missing curve 90 of the NPR filter is below the missing curve 92 related to the Pft filter. Therefore, When the transmission rate is low, the audio encoder selects NPR dropper group 0 at a certain point 94. The quantization error of the encoder falls below the noise base value of the NPK filter, thus adding an extra bit to the Αΰ PC Μ encoder The element does not provide additional benefits. At this point, the audio encoder switches to p R filter bank. 0 ADPCM encoding. ADPCM encoder 7 2 generates a predictive sample selection from the linear combination of H previous reconstruction samples. η). Then, this prediction sample is subtracted from the input X (η) to obtain a difference sample d (n). The difference sample is divided by the RMS (or PEAK) scale factor to pre-calibrate to make the difference The amplitude of the selected sample is matched with the characteristic Q of the quantizer. The sampled difference ud U of the calibration is applied to a step size SZ with L levels ( Determined by the number of ABITs allocated to the current sample selection) quantizer characteristics. The quantizer generates a hierarchical code QL (n) for each calibration difference sampling continent (η). These hierarchical codes are finally transmitted To the ADPCM stage of the decoder. To update the history of the predictor, the quantizer hierarchical code QL (η) uses an inverse quantizer 1 / Q with the same characteristics as Q for local decoding to generate a quantized calibration difference sample d ( n). The selected sample (η) is multiplied by the RMS (or PEAK) scale factor and re-calibrated to produce ί (η). The original sample input sample x (n)-quantization type〗 U) is added by The initial prediction sample p (n) to the quantization difference sample〗 (n) is reconstructed. Then this sample is used to update the predictor history. Vector quantized predictor coefficients and high frequency subband sampling system Use vector quantization (VQ) to encode 0 predictor VQ has a vector dimension of 4 sample selections, based on the Zhongguanjia standard (CNS) A4 specification (2 丨 GX297mm) '---- ~~ '--------- 装 ------ J- 訂 I .---.--- (. (Please read the note on the back side before filling in this page) ff Central Ministry of Economy Quasi-bureau employee spending She printed bag A7 ____ B7 _ _ V. Description of the invention (17) and the sample transmission rate of 3 bits per sample. Therefore, the final code book is composed of 4096 code vectors of dimension 4. Matching vector The search is constructed as a two-level dendritic circuit, and each node in the dendritic circuit has 64 branches. The upper level stores Ψ4 node code vectors, which are only needed by the encoder to assist the search process. The bottom layer contacts 4096 final code vectors, which are required by the encoder and decoder. For each search, I28 MSE calculations with a dimension of 4 are required. The codebooks and node vectors at the upper level are trained using the LBG method, and there are more than 5 million prediction coefficient training vectors. The training vectors are accumulated over all sub-bands, which presents a positive predictive gain, and at the same time encodes a wide range of audio material. ◦Average average SNR is obtained for each test vector in a training group. High frequency VQ has a number of 32 The vector dimension of the sample selection (the length of the primary frame) and the bit transmission rate of each sample selection ϋ. 3 1 2 5 bit. Therefore, the final codebook consists of 1 0 2 4 code vectors of dimension 32. The matching vector search system is constructed as a two-level branch circuit, and each node in the branch circuit has 32 branches. The upper hierarchy stores 32 node code vectors, which are only needed by the encoder. The bottom layer contains 1 ϋ 2 4 final code vectors, the latter being required by the encoder and decoder. For each search, MSE calculations with 64 dimensions of 32 are required. The codebook and node vectors at the upper level are trained using the LBG method, and there are more than 7 million sub-band training vectors of still frequency. For a wide range of audio materials, the sampling rate of 4 8 k H z consists of The selection of vectors is accumulated from the outputs of subbands 16 to 32. At a sampling rate of 4 8 kHz, each training sample represents an audio frequency in the range of 12 to 24 kHz. For each test vector in the training group, an average SNK of about 3 dB is expected. Although this paper is again applicable to the Chinese National Standard ('.CNS) A4 · present grid (2 x 297 mm) — 5 Ω — ........:--. 1--.... .. I-II I ..... I 一 -I ^ n I—. Ini, (please read the precautions on the back before filling out this page) A7 B7 3l556i 5. Description of the invention (1 8) 3 dB is one A small SNK, however, is sufficient to provide high frequency or environment at these high frequencies. It feels far better than the known technology of simply abandoning high frequency subbands. Joint Frequency Coding: In very low bit rate applications, the overall reconstruction can be improved by only coding the sum of high frequency subband signals from two or more audio channels instead of coding them independently . Since high-frequency sub-bands often have similar energy distributions, and because the main strength of the human auditory system and the v-intensity of high-frequency components are sensitive rather than their fine structure, joint frequency coding is possible. Therefore, since more bits can be used to encode sensoryly important low frequencies at any bit transmission rate, the reconstructed average signal provides a good overall frequency. The joint frequency coding index (JO i Μ) system Directly transmitted to the decoder to indicate which channels and subbands have been combined and where the encoded signal is placed in the data stream. The decoder reconstructs the signal in the designated channel and then copies it to every other channel. Then, each channel is calibrated according to its special RMS scale factor. Since the joint frequency coding work averages the time signal according to the similarity of its equal energy distribution, the reconstruction frequency is reduced by 0. Therefore, its application is usually limited to low bit transmission rate applications, while the 2020 kHz signal is mainly at the high bit. In the medium of the meta-transmission rate application, the joint frequency coding usually fails. The 0th-order band encoder encodes the sideband coding procedure for single-use ADPCM / APCM procedures, especially the analysis stage 70 shown in FIG. 5 and the ADPCM encoder 72 and This paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210X297mm) -21- I --------- (installed-I (please read the precautions on the back before filling this page) Printed by the Central Bureau of Standards' Staff Consumer Cooperative Printed by the Ministry of Economic Affairs of the Ministry of Economic Affairs, Central Bureau of Standards and Staff's Consumer Cooperative Printed A7 Β7 V. Interpretation (IS) 2 shows the interaction of the wide area bit management system 30, which is illustrated in Figure 2 〇 Figure 11-19 details the component program shown in Figure 1S. The filter bank 34 divides the PCM audio signal 14 into 32 sub-band signals x (n), which are written into their respective sub-band sample selection buffers 9 6 in 0 assuming an audio window size For 4 0 9 6 sample selections, each time the wave with sample buffer 9 6 stores a frame of 1 2 8 sample selections, the latter is divided into 4 3 2 sample selection sub-frames 0 1 〇2 The window size of the 4 sample selections will generate a single 32 sample selection subframe. Each sample selection is guided to the analysis stage 70 to find the prediction coefficient, predictor mode (PM ODE), and transient mode of each subframe ( TM0DE) and scale factor (SF). Each sample selection x (n) is also provided to the GBM system 30. The latter determines the bit allocation (AB1T) of each subframe of each band of each audio channel. After that, each The sample selection χ (η) is sent to the ADPCM encoder 72 in a frame-by-frame manner. The estimation of the best prediction coefficient uses the standard autocorrelation method optimized on the sub-band sampling Χ (η) of a segment 98 is the Weiner-Hopf or Yule-Walker equations to generate H prediction coefficients (preferably to the power of 4) for each subframe. Optimal prediction system is quantized. The predictor coefficients of a group of four are preferably used. The element dendritic circuit searches for a 12-bit vector codebook (3 bits per coefficient) to be quantized. A 12-bit vector codebook contains 4 0 9 6 uses The standard clustering algorithm optimizes the coefficient vector for a desired probability distribution. A vector quantization (VQ) search 100 selects the coefficient vector that exhibits the smallest mean squared deviation between itself and the optimal coefficient. Then, each frame The best coefficients are replaced with these quantized " vectors. An inverted VQ LUT 1 Q 1 is provided to the ADPCM encoder 7 2 to provide a quantized prediction. This paper size is applicable to the Chinese National Standard (CNS) Α4 specification (210X297 mm) -22 — --------- (install — (please read the precautions on the back before filling out this page) Order Λ mmmmmKmm: · ί A7 B7 printed by the Consumer Cooperative of the Central Standards Bureau of the Ministry of Economy i. Description of invention (^) The detector coefficient q predicts the estimation of the difference signal d (η). One of the major dilemmas of Jt ADPCM is that the difference sample sequence d (n) cannot be easily predicted before the actual recursion procedure 72. The basic requirement of the forward adaptive subband ADP CM is that the difference signal energy is known before A1JPCM encoding, in order to calculate a quantizer that will produce a known quantizer error or noise level in the reconstruction selection Appropriate bit allocation. Knowledge of the difference energy signal is also needed to allow the determination of an optimal difference scale factor before encoding 0. Unfortunately, the difference signal energy is not only related to the characteristics of the input signal, but also to the performance of the predictor. In addition to known limitations such as In addition to the optimality of the predictor order and predictor coefficients, the performance of the predictor is also affected by the quantization error or noise level induced by the reconstruction sampling Q. Because the quantization noise is subject to the final bit allocation AB1T and difference The degree factor KMS (or PEAK) is dominated by itself, and the difference signal energy estimation must be achieved before repeating 1. Step 1: Assuming that the first difference signal estimation of the error is made by sampling the buffer subband X (η) through an ADPCM procedure, the latter does not quantize the difference signal. This is the quantization and KMS in the ADPCM encoding loop The calibration is invalid. By estimating the difference signal d (ro) in this way, the effects of the scale factor and bit allocation value are eliminated from the calculation. However, the procedure takes into account the effect of quantization errors on the predictor coefficients by using vector quantized prediction coefficients. A reverse VQ LUT 104 is used to provide quantized prediction coefficients. To further improve the accuracy of the estimated predictor, the actual ADP CM accumulated at the end of the previous section before the calculation is selected In the predictor. This ensures that the predictor applies the Chinese national standard (CNS) A4 specification (2IOX297 mm) from the paper size previously entered. — -23 — (please read the precautions on the back and fill in this page). : --- The Ministry of Economic Affairs, Central Bureau of Standards, Employee and Consumer Cooperative Printed A7 * B7 V. Description of the invention (21) Where the actual WPCM predictor stops when the balancer ends. This estimate ed U) and the actual program d (η) The main difference between them is that the effects of quantization noise on the reconstruction sample X (η) and the reduced prediction accuracy are ignored. For a quantizer with a large number of levels, the noise level will generally be smaller (assuming proper scaling), so the actual difference signal energy will closely match the one calculated in the estimate. However, when the number of quantizer hierarchies is small, as in the case of general low bit rate audio encoders, the actual predicted signal, and thus the difference signal energy, will be significantly different from the estimator. This results in a coding noise base value that is different from what was predicted earlier in the adaptive bit allocation process. Nonetheless, the change in prediction performance may not be significant for the application or bit transmission rate. Therefore, the estimated value can be directly used to calculate the bit allocation and scale factors without repeating. An additional improvement is to compensate for the performance loss by deliberately overestimating the energy of the difference signal, if a quantizer with a small number of levels may be allocated to the sub-band. This overestimation can also be graded based on changes in the number of quantizer levels for improved accuracy.

If Step 2: Use the estimated bit allocation and the recalculation of the scale factor Once the bit allocation U13 Π ') and the scale factor (SF) have been generated using the first estimated difference signal, their optimality can be obtained by Using the estimated AB1T and RMS (or PEAK) values in the ADPCM loop 72, a further ADPCM estimation procedure is tested. As in the case of the first estimation value, the history of the estimated predictor is copied from the actual ADPCM predictor before starting the calculation, to ensure that both predictors start from the same point. Once the buffer input sample selection has passed through this second estimation loop, the The noise base value obtained in each band is compared with the assumed noise base value in the adaptive bit allocation procedure. Any significant differences can be applied to the Chinese National Standard (CNS) A4 specification (210X297 mm) by this paper scale -24- (please read the note ^ h on the back and then fill in this page) Packing _ Ordered by the Central Sample Bureau of the Ministry of Economic Affairs Employee Consumer Cooperative Printed Α7 Β7 V. Description of invention (22) Corrected bit allocation and / or scale Factor compensation 0 Every time the latest difference signal estimation value is used to calculate the next set of bit allocation and scaling factors, step 2 can be repeated to properly improve the noise base value distributed on each wave band. Generally speaking, If the scale factor will change by more than about 2-3 dB, it will be recalculated. Otherwise, the bit allocation will run the risk of violating the mental auditory masking procedure or the mms e procedure to the signal-to-masking ratio. Usually, a single repetition is sufficient. ~ The calculation of the sub-band prediction mode (PMODJj) is to improve the coding efficiency. A controller 106 can set a PM0DE flag when the prediction gain in the current sub-frame falls below a threshold value. The prediction program is closed. When the _predicted gain (the ratio of the input signal energy to the estimated difference signal energy) measured during the estimation phase for the input selection of a section exceeds a positive threshold value, the PM0DE flag is set to the opposite. If the prediction gain is measured to be less than the positive threshold value, the ADPCM predictor coefficients are set to zero for the subband at both the encoder and decoder, and the respective _PM0DE is set to zero. The prediction gain threshold value is set to be equal to the missing rate of the coefficient vector of the predictor transmitted above. This method attempts to ensure that the coding gain of the ADPCM program is always greater than or equal to the forward adaptive PCM when PM0DE = 1. (APCM) coding program. Otherwise, by setting PM0DE to zero and resetting the predictor coefficients, the ADPCM procedure reverts to APCM. If the ADPCM coding gain change is not important to the application, the P Μ 0 DE in any or all sub-bands can be Set to high value. On the contrary, _P Μ 0 DE can be set; set to low value, if, for example, some sub-bands will not be encoded at all, the bit transmission rate of the application is high enough so that there is no need to predict the gain to maintain the audiobook paper The standard is applicable to the Chinese National Standard (CNS) Α4 specification (210Χ29B / when worn) (please read the $ item on the back and fill in this page)

A7 B7 Employee consumption cooperation of the Central Prototype Bureau of the Ministry of Economic Affairs. Printed by Du. V. Invention description (Subjective quality of Μ, the signal is highly transient content, or the ADPCM coded audio editing feature is not required at all, as is the case for audio editing applications. For each waveband, the individual prediction mode (PMODE) is transmitted at an update rate equal to the linear predictor in the encoder and decoder ADPCM procedures. The purpose of the PMODE parameter is to indicate to the decoder whether the particular subband will be With any prediction coefficient vector address associated with its encoded audio data section. When PM〇DE =] in any primary band, then-the predictor coefficient vector address will always be included in the data stream. When in any primary band When PM0DE = 0, the vector address of a predictor coefficient will never be included in the data stream, and the predictor coefficient is set to zero in the ADPCM stage of the encoder and decoder. The calculation of PMODE begins with the relative The corresponding buffer estimate difference signal energy analysis obtained in the first-stage estimation (that is, assuming no quantization error) analyzes the input signal energy of the buffer subband. Input sample selection X (η) and estimated difference selection The two samples, e d (η), buffer each sub-band. The buffer size is equal to the number of samples included in the update period of each predictor, such as the size of the sub-frame. Then calculate the predicted gain as follows:

Pgain (dB) = 20, 0 * L〇gl〇 (RMSX (n, / RMSed (η)) where RM S x (a, = buffer input selection χ (n) root mean square value, and rμ S ed (n) = root mean square value of the buffer estimate difference sample ed U). For positive prediction gain, the difference signal is on average less than the input signal. Therefore, for the same bit transmission rate, the ADPCM program can be obtained on APCM A reduced reconstruction noise base value. For negative prediction gain, the ADPCM encoder makes the difference signal larger than the input signal on average, which results in a higher noise base value than APCM for the same bit transmission rate. PM〇D] The standard of pre-private paper opened by E is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) -26 «rt ------ τΐτI .--- ^ --- (please read the back page first (Notes to fill out this page) 31556J Α7 Β7 printed by the Employee Consumer Cooperative of the Central Bureau of Samples of the Ministry of Economic Affairs V. Description of the invention (24) The measured gain threshold will be positive, and will have a coefficient vector bit due to the transmission predictor The additional channel capacity consumed by the site is taken into account. The calculation controller of the square wave band transient mode (TMODE) 1 ϋ 6 pairs of each Each subframe in the band calculates each transient mode (TMODE.). The TMOUE indicates that it is in the estimated difference signal ed (n) buffer when PM0DE = 1, or the round subband signal x when PM013E = 〇 (n) The equivalent scaling factors and the number of samples in the buffer. TMODE is updated at the same rate as the prediction coefficient vector address and transmitted to the decoder. The purpose of the transient mode is to Reduce the audible coding when the signal transient exists ~ pre-echo_ " Artificial phenomenon. The transient is defined as a rapid transition between a low amplitude signal and a high amplitude signal. Because the scale factor is selected in the first band difference The averaging is performed on the segments, so if a rapid change in signal amplitude occurs in a segment, that is, when a transient occurs, the calculated scaling factor tends to be much larger than the best one for low amplitude sampling before the transient. Therefore, the quantization error in the sample selection before the transient may be very high. This noise is perceived as a pre-echo loss. In practice, the transient mode is used to correct the average band length of the subband scale factor , To limit the effect of transients on the calibration of the difference sample selection immediately before it. The motive of the practice is the pre-shadowing phenomenon inherent in the human auditory system, which implies that in the presence of transients, the noise can be masked before the transients, but the delay is kept short. Depending on the value of PMOD £, the sub-band selection The contents of the buffer x (n) or the estimated difference buffer edU) (that is, the subframe) are copied into a transient analysis buffer. Here, the content of the buffer is evenly divided into 2, 3 or 4 sub-frames. The paper size is applicable to the Chinese national standard (〇yang) 84 Siluo (210 \ 297 this dream) --------- C Install ------ (please read the precautions on the back before filling in this page),? Τ_ Printed and printed on the A7 ____ B7 by the Central Preparatory Bureau of the Ministry of Economic Affairs A5 ____ B7 V. Invention description (25), depending on the analysis buffer The sample selection depends on the size. For example, if the analysis buffer contains 3 2 subband samples (2 1.3 ms @ 1 5 0 〇H z), the buffer and is divided into 4 each 8 The sub-frame of sample selection, and the time resolution of 5.3 Hz for the sub-band sampling rate of 1500 Hz. Alternatively, if the analysis window is composed of 16 subband selections, the buffer only needs to be divided into two subframes to obtain the same time resolution. · Analyze the signals in each subframe and determine each transient state except the first one. If any sub-frames are declared as transient, two separate scale factors are generated for analysis buffer, that is, the current sub-frame. The first scale is calculated from the sample selection in the sub-frame before the transient sub-frame Factor. Calculate the second scale factor from the transient sub-frame and all the samples of the intermediate sub-frame. The transient state of the first sub-frame is not calculated, because the quantization noise is due to the start of the analysis window itself. And automatically restricted. If more than one subframe is declared as transient, only the first to appear is considered. If the transient sub-buffer is not detected at all, then all samples in the analysis buffer are used and only a single scale factor is calculated. In this method, the scale factor including transient sample selection is not used to calibrate the earlier sample selections in the past that exceed one sub-frame period in the past time. Therefore, the quantization noise before transient is limited to one sub-sample During the subframe. The transient officer informs the sub-frame if its energy ratio in the previous sub-buffer exceeds a transient threshold (T Ί '), and the energy in the previous sub-frame is lower than a pre-transient threshold (PTT) ), Declared as transient. The values of TT and PTT will depend on the bit rate and the degree of suppression of the required pre-echo. They are usually subject to change until the echoes are missed before being detected and match the position of other coding artifacts (if any). The standard applies to the Chinese National Standard # 4 (CNS) A4 (210X297 mm) ... " " ~ — (Please read the precautions on the back before filling in this page) Binding · Order 31556i A7 B7 5. Invention Description (26) (Please read the note ^^ on the back before filling in this page). Increasing ΙΤ and / or decreasing PTT value will reduce the possibility of the subframe being declared as transient, and thus will reduce the bit transmission rate related to the scale factor transmission. On the contrary, decreasing TT and / or increasing The PTT value in the middle will increase the subframe declaration: the possibility of being transient, and therefore will increase the bit transmission rate related to the scale factor transmission. Since TTT and PTT are set individually for each band, the encoder The sensitivity of the transient detection can be set arbitrarily for any primary band. For example, if the perceptibility of the front echo in the high-frequency sub-band is found to be lower in the lower-frequency sub-band, then each can be set. In order to reduce the possibility of being declared transient in the higher order wave band. However, because TMODE is embedded in the compressed data stream, the decoder does not need to know the transient detection algorithm used at the encoder in order to correctly decode the information of the Μ 0 D β quadruple buffer mallet

As shown in FIG. 11a, if the first sub-frame 108 in the sub-band analysis buffer 109 is transient, or if no transient sub-frame is detected, the TM0DE Employee Consumer Cooperative System = 〇〇 If the second sub-frame is transient, but the first is not, TM0DE = 1 〇 If the third sub-frame is transient, but the first or second is not, then TM 0DE = 2 〇If only the fourth subframe is transient, then TM0DE = 3 〇The scale factor is shown in lib. When TM0DE = 0, each scale factor 110 is calculated for all subframes. When TM0DE = 1, the first scale factor is calculated for the first subframe, and the second scale factor is calculated for all previous subframes. When TM0DE = 2, the first scale factor is calculated based on the first and second subframes, and the second scale factor is applied to the Chinese National Standard Specification (210X297 mm) for all previous substandard papers ) &Quot; ~ -29- A7 _____________B7_ printed by the Employee Consumer Cooperative of the Central Bureau of Economic Development of the Ministry of Economic Affairs V. Invention description (27) Subframe for calculation. When TMUDE = 3, the first scale factor is calculated for the first, second, and third subframes, and the second scale factor is calculated for the fourth subframe. Use τ Μ 〇 ADPC Μ encoding and decoding of DE When TMQDE = 0, use a single scaling factor to calibrate the entire analysis buffer (that is, the primary frame). The delayed subband difference sample is selected and transmitted to the decoder to assist in the reverse calibration work . When TM0DE> 0, then use two scaling factors to calibrate the subband difference sample selection and transmit the two to the decoder. For any TMODE, use each scale factor to calibrate the sample used to select the difference. The calculation of the sub-band temperature W (RMS or PEAK) depends on the PMODE value of the sub-band. The estimated difference sample ed (η) or the input sub-band sample x (n) is used to calculate the appropriate Scale factor 〇 In this calculation, use 1'MODE to find the number of scale factors and confirm the corresponding sub-frame in the buffer. 0 RMS scale factor calculation for the jth sub-band The rms scaling factor is as follows: When TM0DE = G, the single rms value is: L 2 0 5 RMS corpse (£ ed (n) / L) _ n = 1 where L is the number of samples in the sub-frame. When TM0DE > At 0, the two rms values are: it RMSlj = (Σ ed (n) 2 / L) ° '5 n = 1 k + 1 2 05 RMS2. = (Σ ed (n) / L)' 'n = 1 This paper standard is applicable-No-National Standards (CNS) A4 specifications (2IOX29 »f_) ^ '~~! (Please read the note ^^ on the back side and fill in the main text) The A7 B7 is printed by the Consumer Cooperative of the Bureau of the P. V. Explanation of the invention (28) where k = (TMODE * L / NSB), and the NSB is the number of uniform subframes. If PMODE = 0, then edjU) sample selection replacement to input sample selection Xj (n) 〇PEAK scale due to failure meter m • Calculate the pe ak scale factor for the j-th wave band as follows: When T Μ 0 DE = 0 , The single peak value is: PEAK] = MAX (ABS (edj (n))), for n = l, L When TM 0 DE > 0, then the two peak values are: PEiiKlj = MAX (ABSUdj ( n))), for n = (TMODE * L / NSB) PEAK2j = MAX (A! 3S (edj (n) :)), for n = (TMODE * L / NSB), L if PM0DE = 0, then edjU ) Sample selection replacement to input the sample selection Xj (n) 〇_PMODE, and the scale of the PMQDE. The prediction mode flag has only two values, on or off, and with a 1-bit code Transfer directly to the decoder. The catastrophe of TMODE. Transient Pennant flags have up to 4 values · 0, 1, 2, and 3, and use 2-bit unsigned integers. Digital words or random transmission directly through a 4-level entropy table to the decoder To try to reduce the average word length of TMODE to less than 2 bits. Typically, this random entropy coding is used to save bits in low bit rate applications. The entropy is detailed in Figure 12. The coding procedure 11 2 is as follows: the transient mode codes TMODE (j) · of j sub-bands are mapped to (p) number of 4 layers of rising-line variable-length 'codebooks, where each codebook is one to one difference Enter the statistical characteristics for optimization. The TMODE value is mapped to the 4-level table 114 and calculated 116. For each table paper size, the Chinese National Standard (CNS) A4 specification (210X297 public product) is applied. ^ Install I 1 ^ Order 4 ^ ^ ^ (Please read the note $ item on the back and then fill in this page)-A7 B7 3155ei V. Description of the invention (29) (NBp) The total number of bits used. Use the TflUM index selection 118 to provide the table with the lowest bit used in the entire mapping process. 0 The mapping code VTMODE (j) is extracted from this table, compressed, and transmitted with the THUFF index word. Delivered to the decoder, 〇 holds the same group of 4 levels The decoder of the reversal table uses the THUFF index to guide each incoming variable-length code YTM〇DE (j) to the appropriate table for decoding and returning to the TMODE index. Quantization of sub-band scale factor To transmit the scale factor to the decoder, it must be quantized into a known code format. In this system, it is quantified with a uniform 64-level logarithmic characteristic, a uniform 128-level logarithmic characteristic, or a variable rate-coded uniform 64-level logarithmic characteristic 12ϋ 〇ϋ4-level quantizer presents 2.25 dB in two cases The step size is 128 levels and the step size is 1.25dB. The 64 level quantization is used for low-to-moderate bit transmission rates. The extra variable rate coding is used for low-bit transmission rate applications, while the 128 level is generally Used for high bit rate. An example of the quantization procedure Γ20 is shown in FIG. 13. The scale factor (RMS or PEAK) is read from a buffer 1 2 1, transformed into a log domain 1 2 2 and then applied as determined by the encoder mode control 128 64-level or 128-level uniform quantizer 1〗 4, U6. Then, the quantized scaling factor is written into a buffer 13ϋ. The range of the 128 and ti4 hierarchical quantizers is sufficient to cover the scale factors with a dynamic range of approximately 16 ϋ dB and I44 dB respectively. The upper limit of the 128 hierarchical level is set to cover the dynamic range of the 24-bit input PCM digital audio signal. The upper limit of 64 levels is set to cover the dynamic range of 20-bit input PCM digital audio signals. Map the logarithmic scale factor to the quantizer, and use the nearest quantizer order for the financial reward (CNS) A specification (2 丨 GX 297mm) ''--32- ------- --f 装 ------ r 定 ----- (Please read the precautions on the back before filling out this page) I printed by the Central Bureau of Economic Development of the Ministry of Economic Affairs

Printed bags of employees ’consumer cooperatives of the Central Bureau of Economics of the Ministry of Economic Affairs. 5. Description of invention (w) The layer code KMSql (or PEAK ") replaces the scale factor. In the case of a 64-level quantizer, these codes are 6 bits long And the range is between 0-6 3. In the case of a 128-level quantizer, each code is 7 bits long, and the range is between 0-127. The realization of the inverse conversion 1 3 1 is simply to map each hierarchical code back to its inverse quantization characteristic. In order to obtain the RMSq (or P £ AKq) value. For ADPCM (or APCM when PM0DJl '= 0) differential sampling calibration, the quantization scale factor is used for both encoder and decoder, so ensure that the calibration and reverse calibration procedures are the same. The 64-level quantizer code When the bit transmission rate needs to be reduced, additional i, quotient, or nj variable-length coding is performed. (34 hierarchical codes are in the jth order. The band is on the second waveband (j = 2) to the highest effect subband. First order differential code 132. This program can also be used to encode the PE.AK scale factor. The marked differential code DRMSQ "j) (or DP £ 'AKy £ (j)) has a maximum range of +/- 63' and is stored in A buffer 1 3 4. In order to reduce the bit transmission rate of the original 6-bit code, the differential code is mapped to the (P) loss of the 127-level rising variable-length code book. Each code book is one Different input statistical characteristics are optimized. 'The procedure for entropy encoding the marked differential codes is the same as the entropy encoding procedure for the transient mode illustrated in Fig. 12. Only P I2? Hierarchy variable length code tables 0 are used. The table that provides the lowest bit used in the mapping program is selected using the SHUFF index. The mapping codes VDK, MSv L (j) are extracted from this table, Combined with the SHUFF index word and transmitted to the decoder. Decoders holding the same set of (p) 127-level inversion tables use the SHUFF index to guide the incoming variable-length codes to the appropriate table for decoding to return to the differential quantizer code Hierarchy 〇The following paper standards are applicable to the Chinese National Standard (CNS) A4 (210X29 Yi Xi Shou_)-installed-(please read the precautions on the back before filling out this page) to set the central standard of the Ministry of Economic Affairs A7 B7 printed by Juzheng Consumer Cooperatives V. Description of invention (31) The routine is converted back to absolute value: EMSq L (1) = DRMSq l (1)

RMSql (j) = 13RMS (i: L (j) + RMSQL (j-l) for j = 2,.. .K and P E A K differential code level is converted back to absolute value using the following routine: PEAKql ⑴ = DPEAKq “1)

PEAKql (j) = DPEAKQL (j) + PEAKQL (1) For j = 2,... K where in both cases, K = the number of sub-bands of action. Wide-area bit allocation The wide-area bit management system so shown in FIG. 10 allocates U β i τ for multi-channel audio encoder management bits, determines the number of acting subbands (s UBS) and the joint frequency strategy (cJ Q i MX) and VQ strategies to provide subjectively transparent encoding with a reduced bit rate. This increase can encode and store the number of audio channels and / or playback time on a fixed medium while maintaining or improving the audio frequency. Generally speaking, the GBM system 30 firstly allocates each bit to each sub-band according to the psychoacoustic analysis corrected by the encoder prediction gain, and then allocates the remaining bits according to the mms e system to reduce the total noise base value. In order to optimize the coding efficiency, the GBM system simultaneously allocates bits to all audio channels, all sub-bands, and the entire frame. In addition, a joint frequency coding strategy may be used. In this way, the system uses the non-uniform distribution of the signal energy between the audio channels in frequency and time. Psychoacoustic analysis The psychoacoustic measurement system is used to determine information that is not related to sensations in the audio signal. Definition of information that is not related to sensations Because the audio signal cannot be heard by human listeners, it can be applied to the China National Standard Rate (CNS) A4 specification (210X297mm) in the time domain, frequency domain, or in some other benchmark paper standards. J Order I,: ^. (Please read the precautions on the back before filling out this page) Printed 315561 A7 B7 ______— " " " 'The invention description (32) On the measurer. J’D’Johnston: " Transform Coding of Audio

Signals Using Perceptual Noise Criteria’1 February 1988 IEEE Journal Selected Areas in Communications »v. 1 J S A.O ϋ, 2 i 3 1 4-3 2 Page 3 explains the general principles of psychoacoustic coding. . Two main factors affect the measurement of mental hearing. One is the absolute threshold of hearing for frequency-related humans. The other is the shadowing effect of sounds. Human food g ^ ji: The second sound can be played at the same time or afterwards. In other words, the first sound prevents us from hearing the second sound, it is called to cover it. 0 $ > In the sub-band encoder, the final result of the mental hearing calculation is to indicate the noise of each sub-band at that instant The inaudible level—the number of groups 0 is calculated as well-known, and is included in the I’9 942 M PEG 1 compression standard IS 0 / IEC DiS " inforniati〇n technology-Coding of moving pictures and associated audio for digital storage media up to about 1.5 Mbits / s''o These numbers g change dynamically with the audio signal. The encoder attempts to adjust the quantization noise base value of each sub-band through the bit allocation process, so that the quantization noise in these sub-bands is less than the audible level. An accurate mental hearing calculation usually requires a high frequency resolution in time to frequency conversion. This display time has a large analysis window for frequency conversion. The standard analysis window size is corresponding to the subframe of the compressed audio data. 〇24 sample selections. The frequency resolution of a length l〇24. Ff t approximately matches the time resolution of the human ear 0 The size of this paper is suitable for financial purposes H g $ 县 丨 eNS 丨 A4__ Γ2Κ) Χ 297mm) ~~~ ^ —-- 35- * (Please read the precautions on the back before filling in this page) Binding V. Invention description (3 A7 B7 The output of the mental hearing model is a signal to each of the 32 sub-bands to the employees of the Central Standards Bureau of the Ministry of Economic Affairs The cooperative prints the masking (SMR) ratio. The amount of SMR, and therefore also indicates quantification. In other words, a large SMR (> > 1) indicates that fewer bits are required below the threshold, and no bits are required as shown in Figure 14. Calculate a fft frequency coefficient 1 4 2 on the frequency selection sample, 2) Average the gyration coefficient 144 for the God's auditory occlusion to generate the SMii sensory response 146 and reduce the sensitivity of SMR with the human ear downward when the sensitivity of the human ear decreases. Because the signal must be higher than the SMR frequency of a 4 k Hz signal frequency, the outer frequency depends on the transmission to the listener. It should be compressed. The other volume will become the sub-level for the SMK level to generate the bits for the 32 sub-bands. Element allocation routine G β Μ system 3 ϋ First, the VQ and ADPC Μ calculations will be used to indicate the tolerance of a specific sub-band Quantize the number of bits required for sample selection in the sub-band of noise. It is clearly indicated that more bits are required, and small SMR (> 0). If SM1U0, the audio signal is quantized in the noise masking element. The SMK of a continuous performance frame is usually borrowed 1) in the PCM sound (length and frequency of each wave, the positive frequency is expected to rely on this, for a more intense rate signal. Therefore, the optimized normalized S Μ β is It is preferably 1 0 24) to generate tone coefficients with frequency dependence, 3) for each sub-band and 4) according to the normalization shown in Figure 15 to produce 0 · Near 4kHz is the maximum, and the frequency is at the same level Perceive that 0 is therefore important in general. However, the average power of the curve is the best when a volume is increased. As a result, if one is not selected, the normalization work 1 4 8 is shown in Figure 16. A series of human hearing rates increased by the noise signal or an accurate form of 20 kHz close to 4 kHz, the audio frequency The nominal power of the system is suppressed. Choose the appropriate coding strategy, that is, which sub-band coding and JFC will work. After that, each paper standard applies the Chinese National Standard (CNS) Μ specifications (210X297j ^ | _)

Please read the notes on the back. Item Dan A Order Λ Printed by the Employee Consumer Cooperative of the Central Bureau of Economic Affairs of the Ministry of Economic Affairs A7 B7 _ ^ __ 5. Description of Invention (34) The GBM system selects a mental hearing or an MMSE bit allocation processing method. For example, at high bit rate, the system will invalidate the establishment of the mental hearing model, and use a solid mmse distribution system. This reduces the complexity of the calculation without any change in the reconstructed audio signal. On the contrary, the low The system can start the above-mentioned joint frequency coding system to improve the reconstruction of lower frequencies. The GBM system can switch between normal mental auditory distribution and m ni se configuration based on the signal transient content based on each frame. . In the case of high transient content, the assumption used to calculate the rest state of SMK is no longer true, so the mm se system provides better performance. For mental hearing allocation, the system first allocates each available bit to satisfy the mental hearing effect, Then allocate the remaining bits to reduce the overall noise base. The first step is to determine the SMR of each waveband for the current frame as described above. The next step is to adjust the SMR for the predicted gain (pgai η) of the respective sub-bands to produce a masked noise ratio (MN). The principle is that the AD PCM encoder will provide a portion of the required SMR results. , The level of inaudible mental auditory noise can be reduced to less bits. The Μ of the j th sub-band is as follows, assuming PM0DE = 1: Μ NR (j) = SMR (j)-P gai η (j ) ^ PEF (ABIT) where PJiF (AMT) is the predicted efficiency factor of the quantizer. When calculating 〇R (j), the designer must have an estimate of the one-bit allocation (ABIT), which can be based on SMR (j) alone The assigned bit may be generated assuming PliF (ABlT) = 1. When the transmission rate is medium to high, the effective prediction gain is approximately equal to the calculated prediction gain. However, when the transmission rate is low, the effective prediction gain decreases. Use For example, the 5-level quantizer, the effective prediction gain achieved is about the estimated prediction. The paper size is applicable to China's' CNS) M specification (210 > < 297mm) " ~ (Please read the notes on the back before filling in (This page)

V ^ 15561 A7 B7 V. Description of the invention (35 times of ϋ.7 times, and the -6-level quantizer allows the effective prediction gain to be approximately equal to the estimated prediction gain, PEF = 1, ϋ 〇 within the limit, when When the bit transmission rate is zero, the predictive coding is essentially suppressed, so the effective prediction gain is zero. In the first step of each time, the GBM system generates a bit allocation system that meets the MNK of each wave band. It is completed for use One bit equals an approximation of 6 dB signal loss. To ensure that the code loss is less than the audible threshold for mental hearing, the specified bit transmission rate is the maximum integer obtained by dividing the MNR by 6 dB, as shown below: .MNRG ) Printed by the Beigong Consumer Cooperative of the Central Bureau of Economic Development of the Ministry of Economic Affairs, by which it tends to follow the frequency and the frequency of the noise, the level of energy to be related will be more _ when the bit transmission rate is less than the target individual sub-band 〇 For example, the channel average can be localized when the target bit is one way. The level of the noise level in Figure 17 is good, and the unique audio bit is transmitted. This is the bit transmitted by the bit SMK. Or bit divided transmission rate ΑΓΓΓ ① Letter of allocation + Quite high, often greater than Zhi will be small when not dispensing the respective transmission rates on the low-speed channel. Or sub-optimal but more evenly divided into KMS ratios (including time, 6dB in wide area, 1 5 7 in the reconstructed signal itself) It will be inaudible because it will remain indispensable. Sub-waves with mmse noise level transmission rate The distribution of the band, the GBM routines, can be simple for each sound, especially distributed in the general audio frequency distribution. VQ code bit and side bit management routine noise level 156 So, the signal is very strong The signal is quite weak. This mental auditory mode is 1 5 8, but the audible sum of 0 bits is greater than the repeated reduction or increase of the frequency channel. The calculation target hardware is implemented during the implementation of the road, or it can be combined with various information). Super will gradually reduce the locality ---------- f outfit ----- Ίbook T! I --- (please read the notes on the back and then fill out this page) Standard (CNS) A4 specification (210 X 297 mm) -3 8-

V. Description of the invention (36> Bit allocation in sub-bands. Many specific techniques can be used to reduce the average bit transmission rate. First, the bit transmission rate carried by the largest integer function can be discarded. Second, from the one with the smallest MWK The secondary band can be removed by one bit. In addition, the higher frequency secondary band can be turned off, or joint frequency coding can take effect. All bit transmission rate reduction strategies follow the general rule of gradually reducing the coding resolution in a smooth manner. Principle, and the least aggressive strategy is introduced first, and the most aggressive strategy is used last. When the target bit transmission rate is greater than the sum of the local bit allocation (including VQ code bits and marginal information) , The wide-area bit management routine will gradually and repeatedly increase the bit allocation of the local sub-band to reduce the total noise base value of the reconstructed signal. This will result in the sub-band of the previously allocated zero bit. Encoding. If P Μ 0 DE comes into effect, in this way, the total number of bits in the 〃 subband may need to reflect the cost of transmitting any predictor coefficients. Printed by the company -------- installed — (please read the notes on the back and then fill in this page) GBM routine can choose one of three different systems to allocate the remaining bits 0 The other one is to use wms e means, redistribution of all bits to make the resulting noise base value approximately flat. This is equivalent to the initial suppression of the mental auditory model establishment '0 is the resulting mm se noise base value, as shown in Figure 18 a sub-band RM Figure 1 6 ϋ is reversed as shown in Figure 1 8 b, and '" filled with water " until all the bits are exhausted. This known technique is called water filling, which is due to the allocation of bits As the number of cells increases, the missed level drops evenly. In the example shown, the first bit is assigned to subbands 1. The second and third bits are assigned to subbands 1 and 2, the fourth To the seventh bit is allocated to sub-bands 1, 2, 4, and 7, etc. Alternately, one bit can be allocated to each band to ensure that each band will be encoded, and then the remaining bits Yuan is filled with water. This paper scale is applicable to China ’s National Averaging Rate (CNS) A4 specification (210X297 mm) _ 3 9 — Α7 Β7 V. Description of invention (u) One second and more A good choice is to allocate the remaining bits according to the mms e method and the above RMS mapping. The effect of this method is to uniformly reduce the noise base value 1 5 7 shown in Figure 17 while maintaining :; holding the shape related to mental hearing masking 〇This provides a good compromise between mental hearing and mse loss. The third method is to use mms e to allocate the remaining bits, just as it is applicable to the mapping of the difference between the K.MS and MNR of each wave band. When the bit transmission rate increases, the effect of this method is to smoothly transform the noise base shape from the best mental hearing shape 157 to the best (flat) mmse shape 158. In any of these systems, if any wave The coding error of the band falls below 0,5 LS.B relative to the source PCM, then no more bits are allocated to this subband. The arbitrary fixed value of the bit allocation of the subband is used. The maximum value can be used to limit the specific subband. The maximum number of allocated bits. Printed by the Central Standards Bureau of the Ministry of Economic Affairs ΪΚ 工 Consumer Cooperatives-(Please read the note $ item on the back before filling this page] In the above coding system, I assume the average bit transmission rate for each sample selected Fixed, and bit allocation has been generated to make it heavy The frequency of the constitutive audio signal is the largest. Alternatively, the missed level (mse or sensory) can be fixed and allow the bit transmission rate to change to meet the missed level. In the mms e method, the RMS drawing is simply filled with water until the missing level is satisfied. The required bit transmission rate will change according to the ⑽S level of the secondary wave band. In psychoacoustic means, each element is allocated to satisfy individual κ. As a result, the bit transmission rate will vary according to the individual SMK and predicted gain. This type of allocation is not currently useful, because today's decoders operate at a fixed rate. However, alternative transmission systems such as ATM or random access storage media will make variable rate coding practical in the nearest future. Quantification of bit allocation (AB 丨 l) This paper scale is applicable to China National Standard (CNS) Μ specification (210X297 / inclusive) A7 ------- B7___ V. Invention description (38) Bit allocation Index UB 1 T) is generated for each sub-band and each audio channel by the adaptive bit allocation routine in the wide area bit management process. The purpose of the encoder index is to indicate the number of layers 1 62 shown in FIG. 10, which is necessary to quantize the difference signal to obtain a subjectively optimal reconstructed noise base for the decoder audio. It is necessary to play at the decoder. , Etc. indicates the number of layers required for inverse quantization. The index is generated for each analysis buffer, and its value range can be from 0 to 270. The index value, the number of quantizer layers, and the approximate differential subband SNq The relationship between R is shown in Table 3. 〇Because the difference signal is normalized, the step size 1 64 is set to be equal to one -------- ^ ^ — (Please read the notes on the back before filling this page ) Ministry of Economic Affairs, Central Bureau of Precinct Employee Consumer Cooperatives, Indian Emperor ABT1 Index, Q-level digital length (.bit) SNQR (dB) 0 0 0 0 1 3 Variable 8 2 5 Variable 12 3 7 (or 8) Variable (or 3 ) 16 4 9 Variable 19 5 13 Variable 21 6 17 (or 16) (or 4) 24 7 2 5 Variable 27 8 33 (or 32) Variable (or 5) 30 9 65 (or 64) Variable (or 6) 36 10 129 (or 128) variable (or Ό 42 11 256 8 48 12 5 12 9 54 This paper size is applicable to China National Standards (CNS) A4 specification (2 0Χ297 mm) 4 i a

、 1T 31556i A7 B7 Printed by the Employees ’Consumer Cooperative of the Central Bureau of Standards of the Ministry of Economy V. Invention Instructions (39) 13 1024 10 60 14 2 0 4 8 11 66 15 4 0 9 6 12 72 16 8192 13 78 17 1 63 84 14 84 18 3 27 68 15 90 19 6 55 3 6 16 96 20 131072 17 102 21 262144 18 108 22 524268 19 114 23 1048576 20 120 24 2097152 21 126 25 4194304 22 132 26 8388608 23 138 27 16777216 24. 144 bit allocation index (AB IT) is to directly use 4-bit unsigned integer digital words +, 5-bit unsigned integer digital words, or use a 12-level entropy table to transmit to the decoder. Typically, entropy coding will be used for low bit rate applications to save bits. The method of encoding Afi IT is set by the mode control of the encoder and transmitted to the decoder. The entropy encoder uses the procedure shown in FIG. 12 with a 12-level ABIT table to map 166 the ABIT index to a specific codebook, which is identified by the BHUFF index and a specific code VABIT in the codebook. Wide area bit transmission rate Control '. Since both the marginal information and the differential subband selection can be used, the entropy can be changed. The paper scale is applicable to the Chinese National Standard (CNS) 84 specifications (210X297mm) -42 一 -------- -^ 装 ------ ^-定 ---.-- (please read the notes on the back before filling out this page) A7 __B7 _ ^ _ printed by the Staff Consumer Cooperative of the Central Bureau of Standards of the Ministry of Economy V. Description of invention ( 4〇) The length codebook is randomly coded, so when the compressed bit stream will be transmitted at a fixed rate, some mechanism must be used to adjust the resulting encoder bit transmission rate. Once the marginal information is calculated, the normal situation is that it does not need to be modified. Therefore, the best way to achieve bit adjustment is to change the difference in the ADPCM encoder by repeatedly changing the quantization process of the fractional band selection until the constraint on the rate is met. In the system, the wide area rate control (GRC) system 178 of FIG. 10 adjusts the bit transmission rate, which is caused by the process of mapping the quantizer hierarchical codes to the entropy table by changing the statistical distribution of the hierarchical code values. Each entropy table assumes a similar trend of higher code lengths for higher-level code values. In this case, as the probability of lower-level code levels increases, the average bit transmission rate decreases, and vice versa. In the ADPCM (or APCM) quantization process, the size of the scale factor determines the distribution or use of hierarchical code values. For example, when the scale factor increases, the differential sampling will tend to be quantized by the lower layers, so the code value will gradually become smaller. This in turn will result in a smaller codeword length and a lower bit transmission rate. 'The disadvantage of this method is that the reconstructed noise of the subband selection is also increased to the same degree due to the increased scale factor. However, in practice, the adjustment of the scale factor is normally no more than 1 dB to 3 dB. If a larger adjustment is required, it is better to return to the bit allocation and reduce the total bit allocation, rather than risking an audible child. The risk of the possibility of noise in the sub-band that will use the increased scale factor. To adjust the allocation of the ADPCM bits of the entropy coding, the prediction of each band is repeated. The sampling process of the device is repeated during the ADPCM coding cycle. Time is stored in a temporary buffer. Secondly, use the prediction coefficient Aa derived from the sub-band .LPC analysis together with this & wave scale to apply the Chinese National Standard (CNS) Λ4 specification (210X 297mm) ^ ^ ^ -43- (please read the note on the back ^ ^ Item and then fill out this page) Pack-Μ A7 _______B7_ V. Description of invention (41) Scale factor KMS (or PEAK), quantizer bit allocation ABIT, transient mode TMODE, and prediction mode derived from the estimated difference signal pM 〇D] E Encode all subband sampling buffers with the entire ADPCM program. The resulting quantizer hierarchical codes are buffered and mapped to the entropy variable-length codebook. The latter uses the bit allocation index again to determine the codebook size. It is shown that the lowest bit uses 0. Then, the GKC system uses the same bit allocation index covering all indexes to analyze the number of bits used in each wave band. For example, when ABIT = 1, the calculation of bit allocation within wide-area bit management can assume that each band selection is 1.4 cattle average rate (that is, assuming the best hierarchical code amplitude distribution Average rate of tilt code book) 〇If AB IT = 1, the total bit of all sub-bands is greater than 1.4 / (the total number of sub-band selections), then the scale factor can be in all such sub-bands The internal increase is used to complete the bit transmission rate reduction. The adjustment of the scaling factor of each band is preferably reserved until all ABT index rates have been accessed. As a result, an index where the bit transmission rate is lower than that assumed in the bit allocation process will compensate for the bit transmission rate higher than the level 0. This evaluation can also be extended to cover all audio channels where appropriate. Printed by the Employees ’Consumer Cooperative of the Central Sample Agency of the Ministry of Economic Affairs (please read the precautions on the back before filling out this page) The recommended procedure for reducing the total bit transmission rate starts with the minimum ABT index bit transmission rate that exceeds the threshold And increase the proportionality factor in each sub-band with this bit allocation. The actual bit usage is therefore reduced by the number of bits in the equal-order band originally higher than the nominal rate of the allocation. 0 If the modified bit usage still exceeds the maximum allowable value, the next highest ABIT index If the use exceeds the nominal value, the sub-band scale factor increases by 0. This process continues until the corrected bit usage is below the maximum value.

Once this is achieved, the old history data is loaded into the predictor, and the paper standard of ADPCM is applicable to the Chinese National Standard (CNS) A4 specification (210X297 ^)

Sl556l A7 B7 V. Description of the invention (42) The coding program 72 repeats for the secondary band whose scale factor has been corrected. Following this, the hierarchical codes are mapped to the best entropy codebook again, and the bit usage is recalculated. If any bit usage :: still exceeds the nominal rate, the scale factor is further increased and the cycle is repeated. The correction of the scale factor can be done in two ways. The first is to transmit an adjustment factor to the decoder for each ABIT index. For example, a 2-bit word can signal an adjustment range, such as 〇, 1, 2, and 3 dB. Because the same adjustment factor is used for all subbands that use ABIT index, and only indexes 1-10 can be used. For coding, the maximum number of adjustment factors that need to be transmitted is 100 for all sub-bands. The scaling factor can be changed in each sub-band by selecting a hierarchical quantizer hierarchy. However, the scale factor adjustment is limited to these steps because the quantizer has steps of 1.2 5 and 2.5 dB, respectively. In addition, when using this technique, if the entropy coding works, the differential coding of the scale factor and the use of the resulting bits may need to be recalculated. Printed by the Central Bureau of Economic Affairs of the Ministry of Economic Affairs. Please read the precautions on the back before filling in this page.) Generally speaking, the corresponding procedure can also be used to increase the bit transmission rate ', that is, when the bit transmission rate is less than the required bit transmission rate. In this case, the scale factor will be reduced to force the differential sampling to make greater use of the external quantizer hierarchy, so the longer codewords in the entropy table are used. If the bit allocation index cannot be used Reduced to a reasonable number of iterations, or the number of adjustment steps has reached the limit under the condition of transmission scale factor adjustment factors, there may be two remedies. First, the secondary wave band within the nominal rate range The scale factor can be increased, thereby reducing the total bit rate. Alternatively, you can abandon the entire ADPCM encoding process and recalculate the sub-bourne paper size to apply the Chinese National Standard (CNS) A4 specification (210X 297 mm) _45_ Printed by the Consumer Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs A7 _____ B7 V. Description of invention (43) With overall suitable bit allocation, this time using fewer bits. Data stream format Multiplexer 32 shown in Figure 10 2 compresses the data of each channel, and then multiplexes the compressed data of each channel into an output frame to form a data stream. 6 〇Compression and multiplex transmission data method , That is, the frame format 186 shown in FIG. 19 is designed to make the audio encoder available for a wide range of applications and can be expanded to a higher sampling frequency, the amount of data in each frame is restricted, and the playback can be independent Start on each sub-frame to reduce latency and reduce decoding error. As shown, a single frame 1 8 6 (409 6 PCM sample / channel) boundary locates the meta-stream boundary, where sufficient information resides to The audio of a sector is properly decoded and consists of 4 sub-frames 188 (1024 PCM sampling / channel), which in turn are each composed of 4 sub-frames 190 (2 5 6 PCM sampling / channel). The frame sync word 192 is placed at the beginning of each audio frame. Frame header information 194 mainly provides information about the structure of the frame 1 86. Encoder configuration, which generates operational data streams and operational features of various options, such as embedded dynamic range control and time code. Option header information 1 96 Inform the decoder if it needs to be down-mixed, whether the dynamic range compensation is completed, and whether the auxiliary data bit group is included in the data stream. The audio coding table header 1 9 8 refers to; ^ the compression configuration and coding format used at the encoder to combine coding, 'marginal information', that is, bit allocation, scale factor, PM ODES, TM ODES, codebook and many more. The rest of the frame is composed of audio subframes 188 in the SUBFS sequence. Each frame starts with the marginal information 200 for audio coding, which relays the information from the key coding system used for most compressed audio to the decoder. The size of this paper is applicable to the Chinese Standard Homepage (CNS) A4 specification (210 X 297 words) (please read the note ^ h on the back and then fill in this page). Installation --- II —1 J1 H-mu * § ^ ιϋ nn vm — ^ n " nfn «1 ^^ — n ^ — _ Α7 Β7 315561 5. Description of the invention (4 4) (please read the precautions on the back before filling out this page) etc. including transient detection , Predictive coding, adaptive bit allocation, high frequency vector quantization, intensity coding and adaptive calibration. Most of this type of data uses the above audio coding header information to decompress from the data stream. 0 High-frequency VQ code array 2 02 is composed of 10 bit indexes per high-frequency subband, indicated by the VQSUB index. The low frequency effect array 2ϋ4 is an option 'and represents very low frequency data that can be used to drive subwoofers, for example. The audio array 206 is decoded with a Huffman / fixed inverse quantizer and divided into multiple subframes (SSC), and each decodes up to 256 PCM samples per audio channel. The audio array 2 0 8 for super sample selection only exists when the sample selection frequency is greater than 48 kHz. To maintain compatibility, decoders that cannot operate at a sampling rate higher than 48 kHz should skip this audio data. Array DSYNC 210 is used to verify the end of the subframe position within the audio frame. If the position is not After verification, the audio decoded in the sub-frame is declared unreliable. As a result, if the frame is not suppressed, the previous frame is repeated. Subband decoder Demonstration of the Consumer Cooperative of the Central Bureau of Standards of the Ministry of Economic Affairs Figure 20 is a block diagram of the subband sampling decoder 18 respectively. The decoder Hunan is very simple compared to the encoder, and does not involve calculations that are fundamental to the quality of the reconstructed audio, such as bit allocation. After synchronization, the decompressor W decompresses the compressed audio data stream 16, detects and corrects transmission-induced errors if necessary, and demultiplexes the data to individual audio channels. The subband differential signal is re-quantized into a PCM signal, and each audio channel is inverted and filtered to convert the signal back into the time domain. The received audio frame and the decompressed header encoded data stream are compressed (or framed) at the encoder ), And in addition to the actual paper size applicable to the national standard (CNS) A4 specification (.2ί〇Χ297 mm) -47 A7 B7 Printed by the Beigong Consumer Cooperative of the Central Standards Bureau of the Ministry of Economy V. Description of invention (45) In addition to the audio code itself, each frame includes additional data for decoder synchronization, error detection and correction, audio encoding status flags, and marginal information for encoding. The decompressor detects the SYNC word and extracts the frame size FSIZE. Encoding The bit stream consists of sequential audio frames, and each starts with a 32-bit (0x7f fe 8 0 0 1) sync word (SYNC). The physical size (FSIZE) of the audio frame is taken from the bytes following the SYNC word. This allows the programmer to set a timer at the end of the frame to reduce the total amount of software. Secondly, NB | ks are extracted, allowing the decoder to calculate the audio window size (32 (Nb Iks + 1)). This tells the decoder which marginal information will be extracted and how many reconstruction samples will be generated. When the header bytes of the frame are received (sync, ftype, surp, nblks, fsize, a mode, sfreq, rate, mixt, dynf, dynct, time, auxcnt, Iff, hflag), the first i2 bytes The validity can be checked using Re ed S ο 1 〇mo u check byte HCiiC. These will correct 1 error byte out of 14 bytes, or flag 2 error bytes. After the error check is completed, the header information is used to update the decoder flag. The headers (fiits, vernum, chist, pcmr, unspec) of the option information following HCRC can be extracted and used to update the decoder flag. Since this information will not change from frame to frame, a majority voting system can be used to compensate for the error of each member. Option header data (times, mcoeff, dcoeff, auxd, ocrt ’) are based on dynf, time, and auxcnt headers. Extract Z ○ The option data can be used at will. Solomon; check byte 0CRC verification. Audio coding frame table header (subfs, s, ubs, chs, vqsub, j ό in X, thuff, shufi, bhuff, sel5, sel7, sel9, sell3, sell7, this paper scale is applicable to China National Standard (CNS) A4 specification ( 21 Ox297mm) " '" -48- (please read the precautions on the back before filling in this page) -install _ order-h A7 3 ^ 556ι ___ Β7 5. Description of the invention (46) sel25, sel33, sel65 , Sell29, ahcrc) transmitted once in each frame 0, etc. You can use the audio Keed Solomon check byte AHCRC verification. Most of the headers are repeated for each audio channel as defined by CHS. The sub-frame coding is used. The audio compression decompression divides the audio coding frame into a majority frame (SUBFS). The necessary marginal information (pmode, pvq, t m o d e, scales, abits, hfreq) are all included to properly decode each frame of audio without referring to any other subframes. Each successive subframe is decoded by first decompressing its marginal information. For each active sub-band and through all audio channels, a 1-bit prediction mode (PM0M) flag is transmitted. The PMQDE flag is valid for the current subframe. PM0DE = 0 implies that the predictor coefficients are not included in the audio box of this subband. In this case, the predictor coefficients of this waveband are reset to zero during the period of the subframe. It shows that the marginal information contains the predictor coefficients of this wave band. In this case, the predictor coefficients are extracted and set up in its predictor during the period of the subframe. For each PM0DE in the pmode array: = 1, a corresponding prediction coefficient .YQ address index is placed in the array PVQ. These indexes are fixed unsigned 12-bit integers, and 4 prediction coefficients are extracted from the check table by mapping 12-bit integers to the vector table 2 6 6. The bit number "Index (AB IT) indicates the number of layers in the inverse quantizer that convert the subband audio code back to an absolute value. The decompression format is different for ΑβίΤ in each audio channel, depending on the BHUFF index and a specific VABIT code depends on 256. Transient mode marginal information (TM 01) E) 2 3 8 is used to indicate that the transient state is applicable to the Chinese national standard (CNS) A4 specification (— -------- if -installation --- --- 7 Order I. --- ^ --- (Please read the notes on the back before filling in this page) A7 B7 printed by the Employee Consumer Cooperative of the Central Bureau of Standards of the Ministry of Economy V. Description of invention (47). Subband The position of the sub-frame relative to the sub-frame. Each sub-frame is divided into; [to 4 sub-frames. In terms of sub-band selection, each sub-frame is composed of 8 samples. The maximum sub-frame size is 32 Sample selection for each subband. If a transient appears in the first subframe, then tm 〇de = 0 〇 When tm 〇de = 1, a transient state is indicated in the second subframe, and so on. In order to control the transient loss (such as pre-echo), when TMODE is greater than 0, the second scale factor is transmitted to each sub-frame band. The ΪΗϋFK index extracted from the audio meter header determines the method required to decode tmODE. When THUFF = 3, TM0DE is decompressed into an unmarked 2-bit integer. Transmit the index of each scale factor to allow proper calibration of the subband audio code in each subframe. If TMODE, etc. At zero, a scaling factor is pre-transmitted. If TM0DE is greater than zero for any primary band, the two scaling factors are transmitted together. The SHUFF index 240 extracted from the audio meter header determines the individual audio channels The method required for SCALES decoding. The vDRMSq l index determines the value of the RMS scale factor. In some modes, the SCALES index is decompressed using five 129-level marked Huffman inverse quantizers. However, the result is the opposite The quantized index is pre-differentially encoded and converted to absolute values as follows: Printed by the Ministry of Economic Affairs Central Standards Staff Consumer Cooperative --------- f Attack II (Please read the note on the back ^^ first Fill in this page)

A ABS_SCALE (n + l) = SCALES (n) -SCALES (n + l) where η is the nth differential scaling factor 0 from the first band in the audio channel. In the low-bit Fu rate audio coding mode At that time, the audio encoder uses vector quantization to efficiently encode high frequency subband audio samples directly. Differential coding is not used in these sub-bands, and all arrays related to normal ADPCM procedures must be kept in a reset state. Use YQ to code the first paper size to apply the Chinese National Standard (CNS) A4 specification (210X297 mm) -50- A7 _ _ B7 _ 5. Description of the invention (4 8) The secondary wave band is indicated by VQSUB, And all the sub-bands that are in SUBS are also coded in this way. The high frequency index (H FREQ) is decompressed 24 8 into fixed 10-bit unsigned integers. The 32 samples required for each band subframe are taken from the Q4 fractional binary LUT by applying appropriate indexes. This is repeated for each channel where the high frequency VQ mode is active. The factor of ten for the effect channel is always XIMO. The number of 8-bit effective samples present in the LFE is obtained as SSC * 2 when PSC = 0, or (SSC + 1) when PSC is not zero. An additional 7-bit scale factor (unsigned integer) is also included at the end of the LFE array, and it is converted into rms 〇g subframe frequency code array with 7-bit LUT to decompress the employee consumption of the Central Standards Bureau of the Ministry of Economic Affairs Printed by the cooperative (please read the precautions on the back before filling in this page) The sub-band audio code extraction process is driven by the AB IT index, and each SEL index is also in the case of ΑΒίll <11. Audio code system Use variable length Huff man codes or fixed linear codes for formatting. Generally speaking, 10 or fewer AB1T indexes will imply Huffman variable length codes. It is represented by the code VQL (n) 2 58 Pre-selected, and ABIT greater than 10 constant means a fixed code. All quantizers have a uniform rising characteristic. The maximum negative level of the fixed code (Y2) quantizer is discarded. Compress the audio code in the sub-frame, each Represents the maximum value of 8 sub-bands for sampling, and these sub-frames are Repeat four times in the frame. If the sampling rate flag (SFREQ) indicates a rate higher than 48 kHz, the ultrasound data array will exist in the audio frame. The first two bytes in this array will indicate the ultrasound frequency. The size of the byte. In addition, the sampling rate of the decoder hardware is based on the Chinese National Standard (CNS) A4 specification (210X297 mm) -51- Printed by the National Bureau of Standards, Ministry of Economic Affairs, Consumer Cooperative A7 B7 V. Invention Note (49) should be set to operate in SFKEQ / 2 or SFKEQ / 4, depending on the high frequency sampling rate. Decompress the sync detection at the end of each sub-frame to detect the data and decompress the sync check word DSYNC = Ο X ffff to allow verification of decompression integrity. In the case of low audio bit transmission rate, the use of variable codewords and audio codes in the marginal information can cause decompression errors, if the header, marginal information, or audio array has been The bit error is corrupted. If the decompression index does not refer to the start of DSYNC, it can be assumed that the previous frame audio is unreliable. Once all the marginal information and audio data are decompressed, the decoder will return to the frame once Multi-channel audio signal reconstruction. 0 Figure 20 examples; ^ Single-band fundamental wave band decoder part in a single channel RMS scale of electrical structure Reconstruct the RMS scale of ADPCM, VQ and JFC algorithms by the open decoder Degree Factor (SCALES). Specifically, the V.'TM〇 [) i ^ THUFF index is inverted to identify the current subframe transient mode (TMODE). Thereafter, the SHUFF index, VD_RMSQL code, and TM0DE are inverted and mapped to reconstruct the differential RMS code. The differential RMS ^ code is inverted by differential encoding 24 2 to select the RMS code, which is then inverted by 2 44 to produce the RMS scaling factor. Inverse quantization of high-frequency optimizing decoders inversely quantizes high-frequency vectors to reconstruct sub-band audio signals. Specifically, the extracted high frequency sample (HFREQ) with 8-bit fractional (Q4) binary digits as identified by the initial VQ subband (VQSUBS) is mapped to an inverted VQ lut 24.8 . The selected table value is reversed and quantified 2 50. The paper size is in accordance with Chinese National Standard (CNS) A4 specification (210X: 297mm). --I Order -----: --- Ci (please read the precautions on the back before filling in this page) Employee Consumer Cooperative of the Central Bureau of Standards, Ministry of Economic Affairs, A7 _____B7 V. Description of Invention (5〇) and marked with RMS The degree factor is calibrated 2 5 2 〇Invert the audio code before _optimization '. Before entering the ADPCM loop, each audio code is inverse quantized and calibrated to produce a reconstructed subband difference sample. The inverse quantization becomes First invert the mapping VABIT and BHUFF indexes to specify the ABIT 'that determines the step size and the number of quantization levels, and invert the mapping SEL index and the VQL (n) audio code that generates the quantizer level code QL (n). After that, each codeword QL (n) is mapped to the check table 260 of the inverse quantizer specified by AB] [and the SEL index. Although each code is sorted by ABIT, each individual audio channel will have a Separate SEL specifier ◦ The review process generates a number of marked quantizer levels that can be converted to unit rms by multiplying the quantizer step size. Then, the unit rms value is converted into a total difference sample by multiplying the assigned K.MS scale factor (SCALES) 262; 1 · QLU] = l / Q [c〇deU]], where 1 / Q is the inverse Checklist for quantizer. 2. Y [η] = Q η]; 丨: S t e p S i z e [a b i t s] 3. Rd [li] = Y [n] * s c a 1 e_f ac tor, where Rd = reconstructed difference sample 0.

The inverted ADPCM performs the ADPCM decoding procedure for each band and difference selection as follows: 1. The predicted coefficients are loaded from the inverted V Q 1 u t 2 6 8. 2. Rotate the current predictor coefficients and maintain the first 4 reconstructed sub-band samples in the predictor history array 268 to generate prediction samples. P [n] = sum (Coeff [.i] * R [ni ]), For i = l, 4, where the paper size is in accordance with the Chinese National Standard (CNS) A4 specification (2I0X297 mm) -53- _, 11: —I --- ^ 装 ----—- ^ 訂--- 1 --- ^ 'ν (Please read the notes on the back before filling in this page) A7 printed by the consumer cooperation of the Central Bureau of Economic Development of the Ministry of Economic Affairs ___B7_ V. Invention description (51) n = Current sampling cycle . 3 · Add the prediction sample to the reconstruction difference sample to produce a reconstructed subband sample 27ϋ. R [η] = K d [n] + 'P [η] 4 · Update the history of the predictor, that is, copy the current reconstruction sub-band sample selection to the top of the history list. K [ni] = R [ii &quot; &quot; i + 1], for i = 4, 1 In the case of PM0 = 0, each predictor coefficient will be zero, the prediction sample is zero, and the reconstructed subband Sample selection is equal to differential subband sampling. Although the calculation of the forecast is not necessary in this case, it is necessary to keep the forecaster history updated, in case PMODE will become effective in the next subframe. In addition, if HfJLAG is active in the current audio frame, the predictor history should be cleared before the first subframe in the frame is decoded. The history should be updated as usual from this point. 0 In the case of the high frequency VQ subband * or when the subband is not selected (that is, above the SUBS limit), the predictor history should remain clear until the subwave When the band predictor becomes active, the selection control of ADPCM, VQ, and JFC decoding-the first, switch &quot; controls the selection of ADPCM or VQ output. The VQSUBS index identifies the starting secondary band of the VQ encoding. Therefore, if the current sub-band is lower than VQSUBS, the switch selects ADPCM output. Otherwise, the VQ output is selected. The second switch 278 controls the choice of direct channel output or JFC encoding output. The J 0 i NX index identifies which channel is combined and in which channel the reconstructed signal is generated. Reconstruct the JFC signal to form the JFC input paper in other channels. The paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) —R 4.— (please read the precautions on the back and fill in this page) k • Install , -0 The employee consumption cooperative of the Central Bureau of Standards, Ministry of Economic Affairs, Yingan 3ι556χ A7 _ 一一 B7-.-V. Description of the invention (52) The source of the intensity of the input. Therefore, if the current sub-band is part of a JFC and is not For the assigned channel, the switch selects the JFC output. Normally, the switch selects the channel output. 0: The torque is randomized. The audio coding mode of the data stream is indicated by AMODE. The decoded audio channel can then be redirected to match the physical output channel configuration on the decoder hardware 280. Dynamic range control data The dynamic range coefficient DCOEFF can be embedded in the audio frame at random during the encoding stage 282. The purpose of this feature is to allow easy compression of the audio dynamic range at the decoder output. Dynamic range compression is particularly important in an audible environment, where high ambient noise levels make it impossible to distinguish low-level signals during loud passes without risking damage to the speakers. This problem is due to dynamics as high as 11 dB The range of 2ϋ-bit PCM audio recording is further mixed with the increased use. Depending on the size of the NBLKS window, for any coding mode (DYNF), one, two or four coefficients are transmitted per audio channel. If a single coefficient is to be transmitted, this is used for the entire frame. When there are two coefficients, the first is used for the first half of the frame, and the second is used for the second half of the frame. Four coefficients are distributed on each frame quadrant. By locally interpolating between transmission values, a higher time resolution is possible. Each coefficient is an 8-bit signed fraction Q2 binary, and represents one of the logarithmic gain values shown in Table (53), and is given in the step of 〇25dB +/- 31.WdB Scope. . Each coefficient is sorted by the number of channels. The size of the dynamic paper is applicable to the Chinese National Standard (CNS) A4 (210x29lf (please read the precautions on the back before filling in this page) h, install _ i --- — ^ Kii ---- B7 B7 Central Ministry of Economic Affairs Printed by the Staff Consumer Cooperative of the Bureau of Standards 5. Description of the invention (53) Range compression is caused by decoding audio samples by multiplying by linear coefficients. The degree of compression can be adjusted by appropriately adjusting the coefficient value at the decoder or ignoring each coefficient. Pre-off completely. The 32-band interpolation filter is fine. 3 The 2-band interpolation filter bank 44 converts the 32nd band of each audio channel into a single PC time domain signal. When FILTS = 0, incomplete Weight coefficient (5 1 2 tap l · 1 U filter). When F〗 LTS = 1, the complete reconstruction coefficient is used. Usually, the cosine modulation coefficient will be pre-calculated and stored in ROM 〇The interpolation program can be expanded to reconstruct a larger data segment and reduce the total amount of loops. However, in the case of the termination frame, the minimum resolution that may be required is 32 PC Μ selection. Interpolation. The algorithm is as follows: generate the cosine modulation coefficient, read 3 2 Select the new subband to the array XI Ν, multiply it by the cosine modulation coefficient and generate the temporary array SUM and DIFF, store the history, multiply by the filter coefficient, generate 32 PCM output samples, update the working array, and output 32 Two new PCM samples 〇Depending on the bit transmission rate and coding system in operation, the bit stream can specify incomplete or reconstructed interpolation filter bank coefficients (FILTS) for completeness. 〇Since the encoder takes The one-type filter bank is calculated with 40-bit floating-point precision, so the ability of the decoder to obtain the maximum theoretical reconstruction accuracy will depend on the source PCM word length used to calculate the rotation and the DSP core precision, and the calibration method for each operation The audio data related to the low frequency effect channel in the low frequency effect PCM has nothing to do with the main audio channel. This channel is used for one of the 20 papers in X128 (120 Hz bandwidth). This paper standard is applicable to Chinese national standards ( CNS) A Shan Jiange (n〇X 297 mm) — 56— ---------, installed— <Please read the notes on the back before filling out this page), ιτ λ Central Bureau of Standards, Ministry of Economic Affairs Printed by employee consumer cooperatives Α7 Β7 5. Description of the invention (54) An 8-bit APCM program operated on a bit-bit PCM input is encoded. The effect audio after the ten-take is aligned with the current subframe audio in the main audio channel in time. Therefore, since the delay on the 32-band interpolation filter bank is 2 5 6 samples (5 1 2 tap ), So care must be taken to ensure that the internal low-frequency effect channel is also aligned with the rest of the audio channel before output. If the effect is within F; [R is also a 5 1 2 tap, no compensation is required. The LFT algorithm uses a 5 1 '2 tap 1 2 8X to interpolate the FIR as follows: map the 7-bit scale factor to rms, multiply it by the step size of the 7-bit quantizer, and generate sub-samples from each normalized value Value, and for example, given for each sample selection. Using a low-pass filter with 1 2 8 interpolation. Hardware implementation Figures 21 and 22 illustrate the basic functions of the six-channel encoder and decoder hardware implementation Constructed to operate at sampling rates of 3, 4, 4 · 1 and 4 8 kHz. Refer to Figure 22, eight ADSP21020 40-bit floating-point digital signal processor (DSP) chips 2 from Analog Devices 9 6 is used to implement a six-channel digital audio encoder. 2 9 S. Six DS Ps are used to encode each channel, and the seventh and eighth are used to implement, and wide-area bit allocation and management, respectively./ And '&quot; corruption stream formatter and error coding &quot; function. Every — ADSP2 1 020 is clocked at 33 MHz, and uses external alpha bit X 32k program ram (PRAM) 300, 4 ϋ bit X 32k data ram (SRAM) 302 performs calculations. In the case of an encoder, an 8-bit X 512k EPROM 304 is also used to store fixed constants such as variable-length codebooks. Data stream formatting DSP uses Reed S ο 1 〇m ο n CKC chip 3 ϋ 6, In order to facilitate error detection and protection at the decoder, the communication between the encoder DS Ρ and wide-area bit distribution and management is based on the use of double-paper standards applicable to the China National Stack Standard (CNS) Α4 specification (210Χ297)-57 · * · I -------- C, installed ----- Μ Order-Γ ---: --- C i _ (Please read the notes on the back before filling this page) Central Ministry of Economic Affairs A7 B7 is printed and printed by the Standard Consumer Cooperative. V. Description of invention (55) Port static RAM 3ϋδ will be implemented. The encoding process flow is as follows. Extract a 2-channel digital audio PCM data stream from the output of each of the three AES / EBU digital audio receivers 3 1 0 ° The first channel of each pair is directed to the CH1, 3 and 5 encoder DSP, and Each second channel leads to CH2, 4 and 6 respectively. By converting the serial PCM words into parallel (s / p), each PCM sample is read into the DSP. Each encoder accumulates a frame of PCM samples and encodes the frame data as described above. Information about the estimated difference signal (ed (η)) and sub-band sampling (X (n)) of each channel is distributed to the wide area bit distribution and management DSP through the dual-port CAM transmission. Then, the encoders The bit allocation strategy is read back in the same way. Once the encoding process is completed, the encoded data and the marginal information of the six channels are transmitted to the data stream formatter DSP through the wide-area bit allocation and management DSP. This stage selectively generates CRC View the bytes and add them to the encoded data for error protection at the decoder. Finally, the entire data packet 16 is combined and output. Figure 22 illustrates an implementation of a six-channel hardware decoder. Use — A single Analog Devices ADSP2 1 0 2 0 4 0 bit floating point digital signal processor (DSP) chip 324 to implement a six-channel digital audio decoder. The ADSP21020 is clocked at 33MHZ and uses an external 48-bit X 32k program ram (PKAM) 32 (3, 40-bit X 32k data ram (SRAM) 328 for decoding. An additional 8-bit X 5 1 2 k EPR 0 Μ 3 3 0 is also used for the storage of fixed constants such as variable length entropy and prediction coefficient vector codebook. The decoding process is as follows. The compressed data stream 16 is serially connected to a parallel converter (s / p) 332 The data input to DSP is decompressed and decoded as shown in the previous example. Each wave is selected for each channel. The sample is selected in a single PCM data stream. 22 The paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 father 297 words #) (Please read the notes on the back and then fill in this page)-* •-·---:-¾. A7 3ι556χ V. Invention description (5 ΰ) structure, and through three parallel to series conversion Device (p / s) 3 3 5 output to three AES / EBU defeated audio transmitter chips 3 3 4 0 Although the present invention has shown and described several exemplary specific forms, but for the skilled person in the industry, there will be many Variations and alternative specific forms 0 For example, as processor speed increases and memory cost decreases The sample selection frequency, transmission rate and buffer size will most likely increase. These changes and alternative specific forms are expected, and can be achieved without departing from the spirit and scope of the invention as defined by the scope of each patent application attached. I Shou. (Please read the note ^ h on the back before filling this page)

—L · — · m tMWUM Ordered by the Ministry of Economic Affairs, Central Standards Bureau, Negative Work Consumer Cooperative Printing This paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210X29p | ^ :)

Claims (1)

6. Patent application scope 1 · A multi-channel audio encoder, including:-a frame grabber (ΰ 4), applying an audio window to each channel of a multi-channel audio signal sampled at a sampling rate to generate its own Audio frame sequence; / Majority filter (3 4), which divides the audio frame of each channel into distinct majority frequency subbands covering a frequency range of a fundamental band, each of which contains a series of subwaves With a frame, each waveband frame has at least a primary frame of the audio data; the majority waveband encoder (26) encodes the audio data in the subbands of each frequency into the coded secondary waveband in a one-time frame mode A signal; a multiplexer (32) that compresses and multiplexes the encoded subband signal into an output frame of each continuous data frame, thereby forming a data stream at a transmission rate; and a controller ( 1 9), according to the sampling rate and transmission rate to set the size of the audio-visual window, so that the size of the output frame is constrained to fall within a desired range ◦ Printed by the Consumer Labor Cooperative of the Central Standardization Bureau of the Ministry of Economic Affairs { Read the precautions on the back first and then fill out this page) 2, such as the multi-channel audio encoder of item χ of the patent application, where the controller sets the audio window size to be less than (frame size PFsampMS / Trate) is two Multiple, where the frame size is the maximum size of the output frame, Fs amp is the sampling rate, and Tr ate is the transmission rate. 3. Multi-channel audio encoder as claimed in item 1 of the patent scope, in which the multi-channel audio signal is encoded at a target bit transmission rate, and the sub-band encoders include most predictive encoders; : This paper uses the Chinese National Standard (CNS) A4 specification (210X297 ^^, printed by the Central Standard Rating Bureau of the Ministry of Economic Affairs, printed and printed by Δδ Β8 CS D8 六 、 Patent application range—Wide Area Bit Manager (GBM) (30), for each frame, calculate a mental auditory signal masking ratio (SMK) and an estimated predictive gain (Pgain), and calculate the masking noise ratio (MNR) by reducing the SMR with their respective predictions: v gain scores ), Allocate each bit to meet each MNR, calculate the transmission rate of the allocated bit on all secondary bands, and adjust the individual allocation so that the actual bit transmission rate is close to the target bit transmission rate _ 4. If applying for a patent The multi-channel audio encoder of the first or third item in the range, wherein the sub-band encoder divides each frame into a plurality of sub-frames, and each band encoder includes one to generate and quantize an error signal for each frame Predictive encoder (7 2), including: an analyzer (98,100,1 0 2,1 04,1 06), which generates an estimated error signal before encoding each frame, and detects the estimated error signal every time Transient within the sub-frame, a transient code is generated to indicate whether a transient exists in any sub-frame other than the first and sub-sub-frames in which the transient occurs, and when a transient is detected These sub-frames before the transient generate a pre-transient scaling factor, and a post-transient calibration factor is generated for the sub-frames that include the transient and after the transient, otherwise generate for the sub-frame A uniform scale factor, the predictive encoder uses the pre-transient, post-transient, and uniform scale factors to calibrate the error signal before encoding to reduce the number of times the scale factor corresponding to the pre-transient Encoding error in the frame 〇5-A multi-channel audio encoder, including:-A frame extractor (64), which applies an audio window to each channel of a multi-channel audio signal sampled at a sampling rate to generate its own The audio frame sequence, the audio frame has a. From DC to approximately The sampling rate is half of this. The paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X297 for L ~ '~~ I ^ n. Order I line (please read the precautions on the back and fill in this page). By the Central Bureau of Industry and Commerce, Beigong Consumer Cooperative Institution A8, B8, C8, D8. 6. Patent application frequency bandwidth;-Front filter (4 6), dividing each audio frame into the fundamental band of the audio bandwidth. A portion of the fundamental wave band frame and a high pomelo sample rate frame representing the rest of the audio bandwidth; a high sampling rate encoder (48, 50, 52), which encodes the high audio channel sampling rate frames into their respective Encode high-sampling rate signal; majority filter (34), divide the fundamental wave band frame of each channel into separate majority frequency sub-bands, the frequency sub-bands each contain a series of sub-band bands, each wave The band frame has at least one primary frame of the audio data; most of the secondary band encoders (26) encode the audio data in the secondary frequency bands of various frequencies in a one-time primary frame mode to generate encoded secondary band signals; and —Multiplexer (32), the coding sub-band The signal and high sampling rate signal are compressed and multiplexed into one output frame of each consecutive data frame, thereby forming a data stream at a transmission rate, so that the fundamental band and high sampling rate part of the multi-channel audio signal It can be decoded independently. For example, the multi-channel audio editing device of the fifth patent application includes: a controller U 9), according to the sampling rate and transmission rate to set the size of the audio window, so that the output box The size is constrained and falls within a desired range. 7. A multi-channel audio encoder, including:-a frame grabber (64) that applies an audio window to each channel of a multi-channel audio signal sampled at a sampling rate to produce its own audio frame paper The scale is applicable to China National Standard (CNS) A4 (210X297 full test I ---------- f .------. Subscribe ------. Ii (please read the back first (Notes for replays are written on this page) r Printed 3i5S6j i __ P8 _ of the Central Consumer ’s Bureau of the Ministry of Economic Affairs Employee's Consumer Cooperatives 六. Patent application sequence; Most filters (3 4) divide the audio frame of each channel into covers A plurality of frequency subbands in the frequency range of a fundamental band, each of which includes a series of subband frames, each of which has at least a primary frame of the audio data; The meta-manager (GBM) (30) calculates a mental auditory signal obscuration ratio (SMR) and an estimated prediction gain (Pga i η) for each frame, and reduces the SMR with their respective scores of the relevant prediction gain Calculate the masked noise ratio (MNR), assign each bit to meet each MNR, and calculate the Distribute the bit transmission rate, and adjust the individual distribution to make the actual bit transmission rate close to the target bit transmission rate. Most band encoders (26), according to the bit allocation, the frequency of each frequency The audio data in the band is coded to produce the coded subband signal; and a multiplexer (32) that compresses and multiplexes the coded subband signal and bit distribution into one of each continuous data frame In the output box, a data stream is formed at a transmission rate _ 〇8. For example, the multi-channel audio encoder of the seventh patent application, where when the transmission rate of the allocated bit is less than the target bit transmission rate, GBM (3 0 ) The remaining bits are allocated according to the minimum mean square error (mms e) system. 9. For example, the multi-channel audio encoder of item 7 of the patent scope, where GBM (30) calculates a mean square (KMS) for each frame Value, and when the assigned bit transmission rate is less than the target bit transmission rate, the GBM redistributes all available bits according to the mmse system applied to the RMS value, until the assigned bit transmission. This paper standard is applicable to the Chinese national standard (CNS_Ta4 Grid (210X297 warehouse father k ^ _ (please read the precautions on the back before filling in this page) f -install _ line Γ 々, the patent application rate is close to the target bit transmission rate 0 1 〇. If the patent application number 7-item multi-channel audio encoder, where GBM (30) calculates a mean square (RMS) value for each frame, and allocates all remaining bits according to the RMS value of the dragon se system until the bit transmission rate is assigned Close to the target bit rate. 11. For example, the multi-channel audio encoder of the seventh patent application, in which GBM (30) is calculated for each frame-the root mean square (RMS) value, and is based on the The mmse system of the difference between the RMS and MNK values allocates all remaining bits until the transmission rate of the allocated bit is close to the target bit transmission rate. 1 2. The multi-channel audio encoder as claimed in item 7 of the patent scope, in which G B Μ (3 ϋ) sets S Μ to a uniform sentence value, so that each bit is allocated according to the minimum mean square error (m m s e) system. 13. A multi-channel fixed loss rate audio encoder, including: a frame grabber (64), applying an audio window to each channel of a multi-channel audio signal sampled at a sampling rate to generate its own Audio frame sequence, the multi-channel audio signal has N-bit resolution; Central Bureau of Standards of the Ministry of Economic Affairs-Industry and Consumer Cooperatives (Please read the precautions on the back before filling this page) Most complete reconstruction filters ( 34), the audio frame of each channel is divided into different majority frequency sub-bands covering a frequency range of the fundamental wave band, the frequency sub-bands each include a series of sub-band sub-frames, each band sub-frame has at least The primary frame of the audio data;-the wide area bit manager (GBM) (30), calculating a mean square UMS for each frame, and assigning bits to each frame according to each KMS value, to Make a code missing + the level is less than half of the least significant bit of the N-bit resolution of the audio signal; this paper ruler applies to the Chinese National Standard (CNS) A4 specification (210X2) _ A8 B8 € 8 D8 Ministry of Economic Affairs, Central Bureau of Prejudice, Unemployment Consumption Printed by Zuosha. 6. Most patented predictive subband encoders (26), according to bit allocation, encode audio data in each frequency band one subframe at a time to generate coded subbands Signal; and—multiplexer (3 2), which compresses the coded subband signal and bit distribution and multiplexes it into one of the output frames of each continuous data frame, thereby forming a data stream at the transmission rate , The data stream can be pre-decoded to make the multi-channel audio signal equal to the N-bit resolution decoded multi-channel audio signal. 1 4. The multi-channel audio encoder as claimed in item i 3 of the patent scope, wherein the frequency range of the fundamental band has a maximum frequency, and the following includes:-front filter (4 calls, within the frequency range of the fundamental band) And the frequency higher than the maximum frequency divides each of the audio frames into a fundamental band signal and a high sampling rate signal, respectively, and the GBM allocates each bit to the high sampling rate signal to satisfy the selected fixed loss; and --High sampling rate encoder (48, 50, 52), which encodes the high sampling rate signals of each audio channel into their respective encoded high sampling rate signals; the multiplexer compresses the encoded high sampling rate signals of each channel Enter the respective output boxes to enable the baseband and the high sampling rate of the multi-channel audio signal to be independently decoded. 1 5. If the multi-channel audio encoder of item 13 of the scope of patent application includes: 1 The controller (19) sets the size of the audio window according to the sampling rate and transmission rate, so that the size of the output frame is constrained to fall within a desired range. Q This paper scale is applicable to China National Standard (CNS) A4 Specifications (2! 〇X2 97 full game 5)-(please read the precautions on the back before filling in this page) -installed- I-line f f ί 6 5 5 ί s ABCD Printed by the Ministry of Economic Affairs Central Bureau of Precision Industry Beigong Consumer Cooperative. Range 16-a multi-channel fixed loss rate audio encoder, including a programmable controller (19) to select a fixed sensory loss and a fixed minimum mean square error (mmse :) loss One of them; a frame grabber (64) that applies an audio window to each channel of a multi-channel audio signal sampled at a sampling rate to produce its own audio frame sequence; most filters (34) Divide the audio frame of each channel into distinct majority frequency subbands covering the frequency range of the one-fundamental band, the frequency subbands each include a series of subband bands. Each frame has at least the audio frequency The primary frame of data; a wide-area bit manager (GBM) (30), which calculates a mean square UM_S for each frame from one time) and assigns each bit to each secondary frame according to the RMS value until the fixed mms e The relevant nrns e system until the missing is satisfied and every time since the box Calculate the mental hearing system's occlusion ratio (SMR) and the estimated hearing gain (P gai η) of the mental hearing system to select and respond to the loss of the selection effect, and reduce the SMR with their respective scores of relevant prediction gains Calculate the masking noise ratio (ΜΜ), and allocate each bit to meet each MNR. The majority frequency band encoder (2 6), according to the bit allocation, the frequency of each frequency band within a single frame Audio data coding to generate coded subband signals; and a multiplexer (32) to compress and multiplex the coded subband signals and bits into an output frame of each continuous data frame , So as to form a data stream at a transmission rate. I7, a multi-channel audio decoder, is used to reconstruct multiple sounds from a data stream (please read the precautions on the back before filling out this page) f -installed., 1T This paper ruler Jbt uses the Chinese National Standard (CNS> M specifications (2 丨 0X297 warehouse replacement 6) Α8 Β8 C8 D8 Printed by the Beigong Consumer Cooperative of the Central Bureau of Economic Affairs of the Ministry of Economic Affairs. 6.Apply for the patent scope frequency channel as an encoder. Sampling rate, each of them The frequency channels are sampled at an encoder sampling rate at least as high as the decoder / decoder sampling rate, and then divided into most frequency subbands, compressed at a transmission rate and multiplexed into the data stream, including:-input buffer (3 2 4), used to read and store the data stream in a frame-by-frame manner, each frame includes a sync word, a frame header, an audio header, and at least one frame, including audio marginal information , Most of them have the subframe of the fundamental frequency band audio code with a frequency range covering a fundamental frequency band, a section of high sampling rate audio code covering a high sampling rate frequency range, and a decompression sync; Tool (4 0), used to a) detect the sync word, b) decompress the frame header to extract-the window size indicating the number of audio samples in the frame and an indication in the frame The frame size of the number of tuples, the window size is set as a function of the ratio of the transmission rate to the encoder sampling rate, so that the frame size is constrained to be smaller than the size of the input buffer, c) decompress the audio header To extract the number of sub-frames in the frame and the encoded audio channel And d) successively decompress each frame to extract audio marginal information, demultiplex the fundamental band audio code in each subframe into multiple audio channels and decompress each audio channel into the second Within the band audio code, demultiplex the high-sampling-rate audio code into the multi-audio channel to the decoder sampling rate and skip the remaining high-sampling-rate audio code to the encoder sampling rate, and detect the decompression sync to verify the end of the sub-frame; a fundamental band decoder (42, 44), using marginal information to decode the sub-band audio code into a reconstructed sub-band signal in a one-time primary frame mode, without C please first Read the precautions on the back and then fill out this page) -Installation.-= The size of the printed paper is applicable to the Chinese National Standard (CNS) A4 specification (210X297 warehouse law 7) _ Printed and printed by the Staff Consumer Cooperative of the Central Bureau of Accreditation of the Ministry of Economic Affairs A8 B8 C8 ___D8 6. For the scope of patent application, please refer to any other sub-frames; a fundamental band reconstruction filter (44), which combines the reconstructed sub-band signals of each channel into a reconstructed fundamental band signal in a one-time primary frame mode; High sampling rate The decoder (58, 60) uses marginal information to decode the high-sampling-rate audio code into a reconstructed high-sampling-rate signal for each audio channel in a frame-by-frame manner; a channel reconstruction filter (62), to The one-time one-frame method combines the reconstructed fundamental wave band and the high sampling rate signal into a reconstructed multi-channel audio signal. 18. The multi-channel audio decoder as claimed in item 1 of the patent scope, in which the fundamental band reconstruction filter (44) includes a non-complete reconstruction (NPE) filter bank and a complete reconstruction (PK) Filter bank, and the frame header includes a filter code that selects one of the NPR and PR filter banks. The multi-channel audio decoder as claimed in item 17 of the patent scope, in which the fundamental band The decoder includes a majority of inverted adaptive differential pulse code modulation (ADPCM) encoders (268,270) for decoding the respective subband audio codes, the marginal information includes the prediction coefficients of the respective ADPCM encoders, and A prediction mode (PM0DE) used to control the application of each prediction coefficient to each ADPCM encoder to selectively make its equal prediction ability function and not function. For example, the multi-channel audio decoding of patent application item 17 Where the marginal information includes: a bit allocation table for each channel ’s sub-band, where the bit transmission rate of each band is fixed on the sub-frame; at least one is used for each Scaling factor for each waveband in the channel; and (Please read the precautions on the back before filling in this page) f. The size of the bound and bound paper is applicable to the Chinese National Standard (CNS) A4 (210X297? H ΛΟ 5 5 nf 8 8 8 8 ABCD Central Bureau of Samples Employee Consumer Cooperative Printing Strategy 6. Applying for patents to teach the surroundings-for the transient mode (TMODE) of each wave band in each channel, identifying the scale factor and the number of related sub-frames, the fundamental wave The tape decoder calibrates the audio code of each wave band with its own scale factor according to its equivalent TMODE to facilitate decoding. This paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210X297 Xi law 9) _ (please read first (Notes on the back then fill this page)