Audio Engineering Society Convention Paper - Jeroen …

1 Audio Engineering Society Convention Paper Presented at the 116th Convention 2004 May 8 11 Berlin, Germany This Convention Paper has been reproduced from the author's advance manuscript, without editing, corrections, or consideration by the Review Board. The AES takes no responsibility for the contents. Additional papers may be obtained by sending request and remittance to Audio Engineering Society , 60 East 42nd Street, New York, New York 10165-2520, USA; also see All rights reserved. Reproduction of this Paper , or any portion thereof, is not permitted without direct permission from the Journal of the Audio Engineering Society . Low complexity parametric stereo coding Erik Schuijers1 , Jeroen Breebaart2 , Heiko Purnhagen3 , Jonas Engdeg ard3. 1 Philips Digital Systems Laboratories, Glaslaan 2, 5616 LW, Eindhoven, The Netherlands 2 Philips Research Laboratories, Prof.

2 Holstlaan 4, 5656 AA, Eindhoven, The Netherlands 3 Coding Technologies, D obelnsgatan 64, 11352 Stockholm, Sweden Correspondence should be addressed to Erik Schuijers ABSTRACT. Parametric stereo coding is a technique to efficiently code a stereo Audio signal as a monaural signal plus a small amount of stereo parameters. The monaural signal can be encoded using any Audio coder. The stereo parameters can be embedded in the ancillary part of the mono bit stream creating backwards mono compatibility. In the decoder, first the monaural signal is decoded after which the stereo signal is reconstructed from the stereo parameters. In this Paper , a low complexity decoder solution is described based on complex-modulated filter banks. Combinations of the parametric stereo decoder with both a parametric coding scheme and with aacPlus will be elucidated. 1 INTRODUCTION Spectral Band Replication (SBR) [3].

3 Current develop- ments in parametric Audio coding focus on Parametric Parametric coding techniques gained major momentum Stereo (PS) techniques [2], [4]. Using PS, a stereo sig- in the field of Audio coding. These techniques have been nal is represented as a mono signal plus a small amount applied to create complete full bandwidth codecs of parameters describing the stereo image. Recent re- [1], [2] as well as bandwidth extension algorithms like SCHUIJERS ET AL. LOW COMPLEXITY PARAMETRIC STEREO CODING. search on PS coding has led to a high quality stereo re- PS employs three types of parameters to describe the construction at bit rates below 10 kbit/s for the stereo stereo image (see [5]): parameters [5]. However, as will be elucidated below, this system comes at high computational cost, especially 1. Inter-channel Intensity Differences (IID); describ- in terms of memory usage.

4 In this Paper , a low com- ing the intensity differences between the channels, plexity implementation is presented. Furthermore, com- binations of this low complexity PS coding tool with a 2. Inter-channel Phase Differences (IPD); describing full bandwidth parametric Audio coding scheme [2] and the phase differences between the channels and with aacPlus, the bandwidth-extended High Efficiency Advanced Audio Codec (HE-AAC) [6], are examined. 3. Inter-channel Coherence (IC); describing the coher- The low complexity implementation of the PS system, as ence between the channels. The coherence is mea- described in this Paper , was proposed to MPEG-4 [7], [8] sured as the maximum of the cross-correlation as a where the technical specification was finalized [9] and is function of time or phase. awaiting formal approval. In principle, these three parameters allow for a high qual- The structure of this Paper is as follows.

5 Section 2 in- ity reconstruction of the stereo image. However, the IPD. troduces the parametric stereo coding model. Section 3. parameters only specify the relative phase differences be- presents a low complexity implementation of the para- tween the channels of the stereo input signal. They do not metric stereo decoding process. Combinations of this prescribe the distribution of these phase differences over process with a parametric Audio codec and with aacPlus the left and right channels. Hence, a fourth type of pa- are described in Sections 4 and 5, respectively. Finally, rameter is introduced, describing an overall phase offset conclusions are drawn in Section 6. or Overall Phase Difference (OPD). 2 PARAMETRIC STEREO CODING In order to reconstruct the stereo image, in the PS de- coder a number of operations are performed, consist- Parametric Stereo coding aims at describing a stereo sig- ing of scaling (IID), phase rotations (IPD/OPD) and de- nal as a mono signal plus a set of parameters charac- correlation (IC).

6 A block diagram of the PS decoder is terizing the stereo image. A block diagram of a PS shown in Figure 2. encoder is shown in Figure 1. From the stereo input signal (l[n], r[n]), the time-variant stereo parameters are Mono Audio m[n]. Bit stream demultiplexer decoder Stereo reconstruction estimated on a non-uniform frequency grid, closely re- mono bit l[n]. sembling the Equivalent Rectangular Bandwidth (ERB) stream bit stream De- d[n]. grid [10]. These parameters describe the perceptually correlation relevant spatial cues. Furthermore, a mono downmix r[n]. m[n] is generated. This mono downmix can then be en- Bit stream coded by any mono Audio encoder. The stereo parame- PS bit decoder IID, IPD, stream OPD, IC. ters are quantized and coded into the ancillary part of the mono bit stream yielding a backwards (mono) compati- ble system. Figure 2: Generalized block diagram of PS decoder.

7 Mono bit l[n] Parameter m[n] stream In the FFT-based PS decoder (see [2], [5]) first a de- Bit stream multiplexer Mono Audio estimation &. r[n]. mono downmix encoder correlated signal d[n] is calculated by means of con- bit stream IID, IPD, volving the monaural signal m[n] with a pre-defined se- OPD, IC. PS bit quence. In the stereo reconstruction process, consecu- Parameter Bit stream stream tive windowed segments of both signals m[n] and d[n]. quantization formatter are processed by a time-to-frequency (t/f) transform, per- formed by windowing followed by an FFT, resulting in the complex-valued frequency domain representations Figure 1: Generalized block diagram of PS encoder. M[k] and D[k] respectively. The two frequency domain AES 116TH Convention , BERLIN, GERMANY, 2004 MAY 8 11. 2. SCHUIJERS ET AL. LOW COMPLEXITY PARAMETRIC STEREO CODING. representations of the left and right channels, L[k] and underlying conventional decoder operates at either a re- R[k] respectively, are obtained as linear combinations of duced sampling frequency in the case of SBR, or in mono the signals M[k] and D[k].

8 The mixing parameters are in the case of PS, the additional complexity is at least time and frequency dependent; for each frequency com- partially compensated for. ponent k the mixing process can be described by: The SBR algorithm makes use of complex-exponential modulated (Pseudo) Quadrature Mirror Filter (QMF). L[k] h11 [k] h12 [k] M[k] banks as t/f and f/t transforms, enabling flexible signal = , (1). R[k] h21 [k] h22 [k] D[k] modification at high efficiency [11]. Therefore, they seem like a suitable alterative to the FFT employed in where h11 [k], h12 [k], h21 [k] and h22 [k] are defined by the decoder as presented in Section 2. Furthermore, the the stereo parameters. The signals L[k] and R[k] are fi- potentially very powerful combination of SBR with PS. nally transformed back to the time domain by means of should not result in a decoder much exceeding the com- a frequency-to-time (f/t) transform.

9 In the FFT-based PS plexity of either SBR or PS. Hence, reuse of the t/f and decoder, the f/t transform consists of an inverse FFT fol- f/t transforms of the SBR decoder is desirable. lowed by windowed overlap-add. For a more detailed overview of the actual parameter processing, we refer Quadrature Mirror Filter Bank to [5]. In the analysis QMF bank, the complex-valued sub-band 3 LOW COMPLEXITY IMPLEMENTATION domain signals sk [n] are obtained as: For typical DSP-based applications like mobile devices, L 1. x[n l]p[l]e j K (k+ 2 )(l+ ) , 1. the computational complexity and memory usage of the sk [n] = (2). decoder should be minimized in order to achieve l=0. maximum battery operation time. In the original FFT- based PS decoder [2], the complexity, both computation- where x[n] represents the input signal, p[n] represents the ally as well as in terms of memory, is dominated by the low-pass prototype filter impulse response of order L 1, time-to-frequency (t/f) and frequency-to-time (f/t) trans- represents a phase parameter, K represents the number forms that are applied [7].

10 This is primarily due to the of bands and k the sub-band index with k = 0, 1, .., K 1. length of the windows and FFT as they directly influ- The magnitude responses of the first few lower frequency ence the length of input, output and intermediate storage bands of the 64 bands analysis filter bank are illustrated buffers. in Figure 3. Recently, the SBR [3] tool for bandwidth extension of The sub-band domain signals sk [n] are downsampled by Audio coding has been introduced. Similar to the PS a factor of K resulting in the downsampled complex sub- coding paradigm, also SBR is a parametric Audio coding band domain signals k [n]: enhancement tool that operates as post-processing in the decoder. Moreover, the structure of the SBR decoder and k [n] = sk [Kn]. (3). the PS decoder are quite similar. Both apply a t/f trans- form to obtain a frequency domain representation.