Tradeoff between complexity and memory size in the 3GPP enhanced aacPlus decoder: Speed-conscious and memory-conscious decoders on a 16-bit fixed-point DSP

This paper investigates tradeoff between complexity and memory size in the 3GPP enhanced aacPlus decoder based on 16-bit fixed-point DSP implementation. In order to investigate this tradeoff, the speed- and the-memory conscious decoders are implemented. The maximum number of operations for the implemented speed-conscious decoder is 29.3 million cycles per second (MCPS) for a 32 kb/s bitstream. The maximum number of operations for the memory-conscious decoder, where 70% of the data are allocated to an external memory area, increases by 5.7 MCPS (19%) for the bitstream. The investigation of this tradeoff provides an actual relationship between the computational complexity and the internal memory size of the 3GPP enhanced aacPlus decoder. The implemented decoders enable music download and streaming services on next-generation mobile terminals.


INTRODUCTION
The 3rd generation partnerships project (3GPP) [1] has al ready standardized many technical specifications for 3G mobile systems. In these specifications, highly efficient speech and audio codecs have been included. 3GPP en hanced aacPlus [2] is a newly standardized audio co dec for mobile applications. It enables high quality stereo audio compression at a low bitrate in the range of 24-32 kb/s thanks to integration of MPEG-4 parametric stereo (PS) [3] [4] into MPEG-4 high efficiency advanced audio coding (HE-AAC) [5].
When 3GPP enhanced aacPlus decoder is implemented on a DSP for music download and streaming services on mobile terminals, there is a tradeoff between low computational complexity and low internal memory size. However, this tradeoff in the 3GPP enhanced aacPlus decoder has not been reported.
This paper investigates tradeoff between complexity and memory size in the 3GPP enhanced aacPlus decoder based on 16-bit fixed-point DSP implementation. In Section 2, the 3GPP enhanced aacPlus algorithm is explained in reference to the HE-AAC algorithm. Section 3 demonstrates computa tional complexity of the implemented decoder with and without an external memory area on the DSP in order to investigate this tradeoff.     :g.
.P In the specification of 3GPP enhanced aacPlusI, two modes are standardized for stereo signals. At a bitrate higher than 36 kb/s, the encoder is the same as the HE-AAC encoder illustrated in Fig. 2 (a). Tn this case, the decoder operates in the LP-SBR mode to reduce computational complexity. On the other hand, the framework illustrated in Fig.3 (a) is used to encode stereo signals at lower than or equal to 36 kb/s for improving sound quality.   incorporating PS and additional tools into the HE-AAC de coder described in [6]. The decoding process has been opti mized by maximizing the parallel use of ALUs and MACs, in order to achieve low computational complexity.

Performance evaluation of speed-conscious decoder
The number of operations was evaluated with two bit streams encoded at 48 kb/s and 32 kb/s. They were gener ated, by the 3GPP floating-point reference encoder [12], from a typical stereo music signal sampled at 48 kHz. Com plexity of these bitstreams was measured by the 3GPP fixed point reference decoder [13].     buffers were moved to the external memory area. Although these data are accessed frequently, the access is limited to a specific period. To reduce the latency in the memory access, the data are copied to the internal memory area just before the RAM access. After the access, they are stored to the ex ternal memory area again. The remaining internal memory area and computing power can be used for other applications.
The computational complexity of the memory-conscious decoder was evaluated in comparison with that of the speed conscious decoder. The conditions are the same as the earlier evaluation. Figure 6 illustrates decoder is suitable for music-only players in order to reduce battery consumption. Both of the implemented decoders en able music download and streaming services on next generation mobile terminals.

Complexity comparison
Computational complexity of the implemented 3GPP en hanced aacPlus decoder was compared with that of conven tional AAC and HE-AAC decoders. The AAC, the HE-AAC and the 3GPP enhanced aacPlus bitstreams were encoded at 96, 48 and 32 kb/s from typical stereo signals sampled at 48 kHz. These bitrates were chosen so that the bandwidth of each decoded stereo signal is about 16 kHz.