AC-3编码算法研究与DSP上的实现
|
摘要
AC-3是杜比实验室开发的数字音频编码技术,该技术可以传输和存储多达5个全频带声道和一个低频效果声道,而所占用的存储空间比CD上一路线性PCM编码所占用的空间还要少。由于AC-3系统编码灵活,在消费电子领域,目前大多数的电影制作都使用了该技术,美国等国家的数字电视系统也采用该技术作为音频编码标准。
TMS320DM642数字多媒体处理器是德州仪器开发的一款高性能定点DSP。此款DSP的核处理器拥有64个通用32位寄存器和8个带有VelociTI.2扩展功能的独立功能单元:2个乘法单元和6个算术逻辑单元(ALUs)。DM642还采用了两级cache结构并拥有多种强大的外设。
本文主要研究了AC-3音频编码器算法的开发,以及基于DM642硬件平台的移植和优化。论文详细研究了AC-3编码器核心算法模块:输入滤波、暂稳态判决、MDCT、指数编码、比特分配。设计了输入滤波器;改进了暂稳态判决方法;推导并实现了用N/4点FFT实现N点MDCT的快速算法,并将短窗MDCT转换为与长窗MDCT一致的计算方式;设计了指数编码策略选择,指数平滑算法;在分析比特分配与噪声之间关系的基础上设计了一种高效的比特分配算法。接下来,该编码器被移植到DM642平台上,针对平台自身特点改善算法提高效率。在定点化和充分利用开发工具优化的基础上,解决了IIR滤波器精度要求高的问题。目前,该编码器在DM642平台上在190MHz频率下可实现实时编码。
最后,该编码器通过了21个测试序列的客观模拟主观测试。在128kbps的码率下PESQ分值为4.036,在320kbps的码率下PESQ分值为4.220。
AC-3 is a digital audio coding technology developed by Dolby Lab. The technology can transfer or store as many as 5 full frequency band channels and a low frequency effect channel, the memory required is less than one channel linear PCM data in CD. Due to the flexibility of AC-3 system, in the custom electronic domain, most of movies are using this technology now, and the digital TV system of several countries, like America, also make use of this technology as the standard of audio coding.
The TMS320DM642 Digital Media Processor is a high-performance fixed-point DSP of Texas Instruments. The DSP core processor has 64 general purpose 32-bit registers and 8 independent functional units with VelociTI.2 extensions: 2 multipliers and 6 arithmetic logic units (ALUs). The DM642 uses a two-level cache-based architecture and has a powerful and diverse set of peripherals.
The objective of this thesis is to develop AC-3 audio encoder algorithm, transplant and optimize it based on DM642 hardware platform. At first, AC-3 encoder core arithmetic modules are presented in detail, including input filter, transient detection, MDCT, exponents encoding, and bit allocation. Here, an input filter is designed, the method of transient detection is improved, the N point MDCT fast algorithm by N/4 point FFT is deduced and realized, short windowed MDCT is conversed to the form similar to long windowed MDCT, the arithmetic on the exponent coding strategy and smoothing is carried out, and an efficient bit allocation algorithm is designed based on the relation of bit allocated and noise introduced. Then, the encoder is transplanted to DM642 platform, the encoder’s efficiency is improved by making use of the characteristic of platform. By fixing point and making use of the development tools sufficiently, the problem of the high resolution requirement of IIR filter is resolved. For the moment, this encoder can run real-time on DM642 platform at 190 MHz.
At last, this encoder is passed the objective test, which simulate subjective test by consideration of human hearing characteristic. Here 21 test sequences are used, the average PESQ value is 4.036 at 128kbps bitrate, and 4.220 at 320kbps bitrate.
TMS320DM642数字多媒体处理器是德州仪器开发的一款高性能定点DSP。此款DSP的核处理器拥有64个通用32位寄存器和8个带有VelociTI.2扩展功能的独立功能单元:2个乘法单元和6个算术逻辑单元(ALUs)。DM642还采用了两级cache结构并拥有多种强大的外设。
本文主要研究了AC-3音频编码器算法的开发,以及基于DM642硬件平台的移植和优化。论文详细研究了AC-3编码器核心算法模块:输入滤波、暂稳态判决、MDCT、指数编码、比特分配。设计了输入滤波器;改进了暂稳态判决方法;推导并实现了用N/4点FFT实现N点MDCT的快速算法,并将短窗MDCT转换为与长窗MDCT一致的计算方式;设计了指数编码策略选择,指数平滑算法;在分析比特分配与噪声之间关系的基础上设计了一种高效的比特分配算法。接下来,该编码器被移植到DM642平台上,针对平台自身特点改善算法提高效率。在定点化和充分利用开发工具优化的基础上,解决了IIR滤波器精度要求高的问题。目前,该编码器在DM642平台上在190MHz频率下可实现实时编码。
最后,该编码器通过了21个测试序列的客观模拟主观测试。在128kbps的码率下PESQ分值为4.036,在320kbps的码率下PESQ分值为4.220。
AC-3 is a digital audio coding technology developed by Dolby Lab. The technology can transfer or store as many as 5 full frequency band channels and a low frequency effect channel, the memory required is less than one channel linear PCM data in CD. Due to the flexibility of AC-3 system, in the custom electronic domain, most of movies are using this technology now, and the digital TV system of several countries, like America, also make use of this technology as the standard of audio coding.
The TMS320DM642 Digital Media Processor is a high-performance fixed-point DSP of Texas Instruments. The DSP core processor has 64 general purpose 32-bit registers and 8 independent functional units with VelociTI.2 extensions: 2 multipliers and 6 arithmetic logic units (ALUs). The DM642 uses a two-level cache-based architecture and has a powerful and diverse set of peripherals.
The objective of this thesis is to develop AC-3 audio encoder algorithm, transplant and optimize it based on DM642 hardware platform. At first, AC-3 encoder core arithmetic modules are presented in detail, including input filter, transient detection, MDCT, exponents encoding, and bit allocation. Here, an input filter is designed, the method of transient detection is improved, the N point MDCT fast algorithm by N/4 point FFT is deduced and realized, short windowed MDCT is conversed to the form similar to long windowed MDCT, the arithmetic on the exponent coding strategy and smoothing is carried out, and an efficient bit allocation algorithm is designed based on the relation of bit allocated and noise introduced. Then, the encoder is transplanted to DM642 platform, the encoder’s efficiency is improved by making use of the characteristic of platform. By fixing point and making use of the development tools sufficiently, the problem of the high resolution requirement of IIR filter is resolved. For the moment, this encoder can run real-time on DM642 platform at 190 MHz.
At last, this encoder is passed the objective test, which simulate subjective test by consideration of human hearing characteristic. Here 21 test sequences are used, the average PESQ value is 4.036 at 128kbps bitrate, and 4.220 at 320kbps bitrate.
引文
[1]钱征海,吕宗伟,数字音频技术与杜比AC-3,电子技术应用,2005年,第9期:49-51
[2] Ted Painter,Perceptual Coding of Digital Audio,Proceedings of the IEEE,vol 88,No.4,april 2000
[3] D. Krahe,New source coding method for high quality digital audio signals, NTG Fachtagung Hoerrundfunk,pp. S.371–S.381,1985
[4] E. F. Schroder,W.Voessing, High quality digital audio encoding with 3.0 bits/sample using adaptive transform coding, in Proc. 80th
[5] P. Hedelin,A tone-oriented voice-excited vocoder,in Proc. IEEE Int. Conf. Acoustics Speech, and Signal Processing (ICASSP-81)
[6] L. Almeida, Nonstationary spectral modeling of voiced speech,IEEE Trans. Acoust., Speech, Signal Processing,vol. ASSP-31,pp.374–390, June 1983
[7] X. Serra and J. O. Smith III,Spectral modeling and synthesis: A sound analysis/synthesis system based on a deterministic plus stochastic decomposition, Comput. Mus. J.,pp. 12–24,Winter 1990
[8] T. Moriya,N. Iwakami,K. Ikeda, and S. Miki, Extension and complexity reduction of TWINVQ audio coder, in Proc. Int. Conf. Acoustics,Speech,and Signal Processing (ICASSP-96),May 1996,pp. 1029–1032
[9] A. Spanias,Speech coding: A tutorial review,Proc. IEEE,vol. 82,pp. 1541–1582,Oct. 1994
[10] L. Fielder , M. Bosi , G. Davidson , AC-2 and AC-3: Low-complexity transform-based audio coding , Collected Papers Digital Audio Bit-Rate Reduction,pp. 54–72,1996
[11] L. Fielder,G. Davidson,AC-2: A family of low complexity transform-based music coders,in Proc.,10th AES Int. Conf.,Sept 1991
[12] J. Princen,A. Bradley,Analysis/synthesis filter bank design based on time domain aliasing cancellation, IEEE Trans. Acoust,Speech,Signal Processing, vol. ASSP-34,pp. 1153–1161,Oct. 1986
[13] M. Davis,The AC-3 multichannel coder,in Proc. 95th Conv. Aud. Eng. Soc., Oct. 1993,preprint 3774
[14] C. Todd,G. Davidson,M. Davis,et ,AC-3: Flexible perceptual coding for audio transmission and storage,in Proc. 96th Conv. Aud. Eng. Soc.,Feb. 1994,reprint 3796
[15] G. Davidson,Digital audio coding: Dolby AC-3, in The Digital Signal Processing Handbook,V. Madisetti and D. Williams,Eds. Boca Raton,FL: CRC Press,1998,pp. 41.1–41.21
[16] ATSC,A/52, Standard:Digital Audio Compression (AC-3),Washington,D.C:Advanced Television Systems Committee ,Dec.20,1995
[17] Keith Wright,Notes on Ogg Vorbis and the MDCT, http://www.free-comp-shop.com/vorbis.pdf,2003-5-23
[18] Wang . Y , Yaroslavsky . L , Vilemo . M .On the Relationship Between MDCT,SDFT and DFT ,the 5th International Conference on Signal Processing, August 21-25,2000,Beijing,China,44-47
[19]窦维蓓,刘若珩.基于DSP的IMDCT快速算法.清华大学学报,自然科学版,2000年,第3期:99-103
[20]徐盛,胡剑凌等.一种新的MDCT快速算法.数据采集与处理,2000年,第4期:431-434
[21] Texas Instrument,TMS320C64x Technical Overview,Jan., 2001
[22] Texas Instrument, TMS320DM642 Digital Media Processor Product Preview 2002.7
[23]汪春梅,张崇明,李蕊等,TM S320DM 642多媒体处理系统中高性能音频功能的实现,电声技术,2006-06:53-55
[24]万福,高俊,DSP芯片编程中数的定点运算,通讯与广播电视,2004年第1期
[25]张卫宁,赵子婴,定点DSPs的定标及其运算方法,计算机工程,2002年3月,28卷第3期:223-225
[26]周日贵,胡景春,叶水生,定点DSP的开发实践,南昌航空工业学院学报(自然科学版),2003年3月,第17卷第1期:59-62
[27]胡广书,数字信号处理——理论、算法与实现,北京:清华大学出版社,2003.8,581-601
[28] Proakis J G,Manolakis D G,Introduction to Digital Signal Processing,New York:Macmillan Publishing Company,1998
[29] Tretter S A,Introduction to Discrete-Time Signal Processing,New York:John Wiley and Sons,1976
[30]李方慧,王飞,何佩琨等,TMS320C6000系列DSPs原理与应用,北京:电子工业出版社,2003,220-224
[2] Ted Painter,Perceptual Coding of Digital Audio,Proceedings of the IEEE,vol 88,No.4,april 2000
[3] D. Krahe,New source coding method for high quality digital audio signals, NTG Fachtagung Hoerrundfunk,pp. S.371–S.381,1985
[4] E. F. Schroder,W.Voessing, High quality digital audio encoding with 3.0 bits/sample using adaptive transform coding, in Proc. 80th
[5] P. Hedelin,A tone-oriented voice-excited vocoder,in Proc. IEEE Int. Conf. Acoustics Speech, and Signal Processing (ICASSP-81)
[6] L. Almeida, Nonstationary spectral modeling of voiced speech,IEEE Trans. Acoust., Speech, Signal Processing,vol. ASSP-31,pp.374–390, June 1983
[7] X. Serra and J. O. Smith III,Spectral modeling and synthesis: A sound analysis/synthesis system based on a deterministic plus stochastic decomposition, Comput. Mus. J.,pp. 12–24,Winter 1990
[8] T. Moriya,N. Iwakami,K. Ikeda, and S. Miki, Extension and complexity reduction of TWINVQ audio coder, in Proc. Int. Conf. Acoustics,Speech,and Signal Processing (ICASSP-96),May 1996,pp. 1029–1032
[9] A. Spanias,Speech coding: A tutorial review,Proc. IEEE,vol. 82,pp. 1541–1582,Oct. 1994
[10] L. Fielder , M. Bosi , G. Davidson , AC-2 and AC-3: Low-complexity transform-based audio coding , Collected Papers Digital Audio Bit-Rate Reduction,pp. 54–72,1996
[11] L. Fielder,G. Davidson,AC-2: A family of low complexity transform-based music coders,in Proc.,10th AES Int. Conf.,Sept 1991
[12] J. Princen,A. Bradley,Analysis/synthesis filter bank design based on time domain aliasing cancellation, IEEE Trans. Acoust,Speech,Signal Processing, vol. ASSP-34,pp. 1153–1161,Oct. 1986
[13] M. Davis,The AC-3 multichannel coder,in Proc. 95th Conv. Aud. Eng. Soc., Oct. 1993,preprint 3774
[14] C. Todd,G. Davidson,M. Davis,et ,AC-3: Flexible perceptual coding for audio transmission and storage,in Proc. 96th Conv. Aud. Eng. Soc.,Feb. 1994,reprint 3796
[15] G. Davidson,Digital audio coding: Dolby AC-3, in The Digital Signal Processing Handbook,V. Madisetti and D. Williams,Eds. Boca Raton,FL: CRC Press,1998,pp. 41.1–41.21
[16] ATSC,A/52, Standard:Digital Audio Compression (AC-3),Washington,D.C:Advanced Television Systems Committee ,Dec.20,1995
[17] Keith Wright,Notes on Ogg Vorbis and the MDCT, http://www.free-comp-shop.com/vorbis.pdf,2003-5-23
[18] Wang . Y , Yaroslavsky . L , Vilemo . M .On the Relationship Between MDCT,SDFT and DFT ,the 5th International Conference on Signal Processing, August 21-25,2000,Beijing,China,44-47
[19]窦维蓓,刘若珩.基于DSP的IMDCT快速算法.清华大学学报,自然科学版,2000年,第3期:99-103
[20]徐盛,胡剑凌等.一种新的MDCT快速算法.数据采集与处理,2000年,第4期:431-434
[21] Texas Instrument,TMS320C64x Technical Overview,Jan., 2001
[22] Texas Instrument, TMS320DM642 Digital Media Processor Product Preview 2002.7
[23]汪春梅,张崇明,李蕊等,TM S320DM 642多媒体处理系统中高性能音频功能的实现,电声技术,2006-06:53-55
[24]万福,高俊,DSP芯片编程中数的定点运算,通讯与广播电视,2004年第1期
[25]张卫宁,赵子婴,定点DSPs的定标及其运算方法,计算机工程,2002年3月,28卷第3期:223-225
[26]周日贵,胡景春,叶水生,定点DSP的开发实践,南昌航空工业学院学报(自然科学版),2003年3月,第17卷第1期:59-62
[27]胡广书,数字信号处理——理论、算法与实现,北京:清华大学出版社,2003.8,581-601
[28] Proakis J G,Manolakis D G,Introduction to Digital Signal Processing,New York:Macmillan Publishing Company,1998
[29] Tretter S A,Introduction to Discrete-Time Signal Processing,New York:John Wiley and Sons,1976
[30]李方慧,王飞,何佩琨等,TMS320C6000系列DSPs原理与应用,北京:电子工业出版社,2003,220-224