Direct Stream Digital

Abstract: In this paper, an overview of Direct Stream Digital (DSD) signal processing is given. It is shown that 1-bit DSD signals can be dithered properly, so the resulting dithered DSD stream does not contain audible artifacts in a band from 0-100 kHz. It is also shown that signal processing can be done best in a high rate, multi-bit domain. Arguments are given that the minimal frequency span needed to comply with the human auditory system is roughly 0-300 kHz. Following the signal processing, nal conversion to DSD is made. It is demonstrated that Super Audio CD (SACD) is a very eÆcient consumer format: it is the format which, while maintaining all necessary psycho-acoustical characteristics such has high band width, ltering with wide transition bands etc., uses the least bits from the disk; hence o ering the longest playing time. INTRODUCTION In the past few years, there has been an evolving trend in the audio world to move from the standard CD-format (i.e., 16 bit resolution, and a sampling frequency of 44.1 kHz) to other formats, which offer a higher resolution. A nice description of this otherwise rather vague definition is given by the AES high-resolution audio technical committee: every carrier offering more than 2 channels, each 44.1 or 48 kHz and resolution corresponding to 16 bit, is coined as high-resolution. Hence, high resolution embraces both multichannel recordings and audio formats which allow for a higher definition of the audio data per sé, such as SACD, which stores 1-bit words at a sampling frequency of 64 REEFMAN AND NUIJTEN WHY DIRECT STREAM DIGITAL... times 44.1 kHz, and DVD-A, which covers a wide variety of sampling rates (44.1/48 to 176/192 kHz) and word lengths (16-24 bit). Often, the SACD data format is called bitstream or Direct Stream Digital (DSD) to contrast it with the Pulse Code modulation (PCM) used in the CD and DVD-A format. Though sigma delta modulation (or its predecessor, Delta modulation), the basic principle behind DSD, has been known for a long time [11], many of its intrinsic properties have not been well-covered. Though many models exist, they are flawed due to linearization the quantizer, or, if a linearization is not made, from the fact that only a first or secondorder modulator is studied. This contrasts to the situation in the realm of PCM, where many of its properties have been studied in depth both experimentally and theoretically, for example, [18, 10, 6]. The purpose of this paper is to give an outline of the technical properties of the DSD format, and to present an overview of the signal processing required for preparing an SACD loaded with 8 channels of high resolution audio data. PROPERTIES OF THE BITSTREAM FORMAT In this section, some properties of single-bit sigma delta modulation will be displayed. Though most of the results mentioned in this section are not new, we think they deserve mentioning because they are not well-known within the audio community as illustrated, e.g., at the AES 2000 in Los Angeles. Introduction to Sigma Delta modulation

Abstract: A stereo audio delta-sigma A/D converter is implemented to support both the standard pulse-code modulation audio and the direct stream digital (DSD) output format. It provides all the standard audio rates up to 192 kHz. A sixth-order, single-bit modulator is employed to achieve the noise performance as well as the bitstream output required by the DSD format. A novel density-modulated dithering scheme is utilized to dramatically reduce the tone level in the signal band without compromising the stability of the high-order modulator. This analog-to-digital converter achieves a dynamic range of 113 dB and a total harmonic distortion +N of 105 dB. It is fabricated in a 0.35-/spl mu/m CMOS process with a die size of 10.5 mm/sup 2/.

Abstract: An audio coding scheme allowing PCM signal to lossless DSD signal expansion for next generation optical disc formats. The method of encoding an input DSD signal includes up-sampling a corresponding PCM signal to the DSD sample rate. Then a set of loop filter parameters for a noise-shaping loop of a sigma-delta modulator are generated, either using a random starting condition of the sigma-delta modulator or including synchronization parameters. This will allow a decoder to regenerate an almost perfect signal, but still it needs a correction signal to be able to bit identically regenerate the DSD input signal. Therefore, a correction signal is generated based on a difference between a sigma- delta modulated version of the up-sampled PCM signal and the input DSD signal, wherein the sigma-delta modulated version of the up-sampled PCM signal is obtained using the set of loop filter parameters. The correction signal may be adapted to be applied to the low bit PCM signal, to the up-sampled PCM signal or to the output bit stream. Finally, an expansion bit stream is generated where an encoded version of the set of loop filter parameters and an encoded version of the correction signal are included. The decoder can reproduce the original DSD signal based on the already available PCM signal and the described expansion bit stream. Thus, the coding scheme enables top quality audio with minimal storage overhead since the already available PCM signal is used in combination with an expansion bit stream. Since only an additional data stream is required to be stored on a disc, e.g. as part of an MPEG stream, DSD functionality is added to existing systems without causing compatibility problems.