Perhaps I can address this question.
Jean-Luc Ohl, a well-known French pro, has pointed out
on a French forum an old open access paper from James A. Moorer, Sonic Solution, whom I already mentioned in my post
#105.
This paper is still online and accessible for free, contrary to
its AES version that is behind paywall or subscription, but it's the same content :
Breaking the Sound Barrier : Mastering at 96 kHz and beyond.
The publication date of the AES paper is 6th November 1996.
Yet, at this early stage, Mr Moorer already described ways (and challenges) to implement digital signal processing on DSD streams to adjust volume or realize equalization. But he stated that the required computational power of available DSP chips of the times (he mentionned the Motorola 56000 family) were insufficient to obtain the required precision for pro quality standard and that special hardware had to be built (hence the development of a Sony DSP processor, obviously the CXD2926). At several instances, he wrote about the impractibility of signal processing on 1 bit stream because of the insufficient available processing power at the time of writing his article and the necessity to develop special hardware tools.
I suspect that the myth about the infeasiblity of editing DSD comes from this state of affair at the time, ie that no off-the-shelf processor were able to do that and
impractibility in Moorer's text has became
impossibility in the mind of some. That myth has been perpetuated throughout the following years among the general public even when more and more powerful hardware became available.
That being said, putting aside most of the mathematics used by Mr Moorer in his article, which are mostly way over my head, there are also some descriptions of practical processing on one bit streams that are easy to understand. Mr Moorer explained in relatively simple terms how to change the gain on a 1 bit stream in the digital (discrete) domain in page 10 of his article. Suffice to say that it must be understood that a 1 bit stream consisting of 1s and 0s must be looked as symbols whose meaning are actually values of 1 and -1. Hence, it is easy to scale those values by an arbitrary factor, for instance 0.5 to bring the gain down by 6 dB. The very same process is described in Ayataka Nishio's presentation of his chip. It is a straightforward process. The difficulties bare on the shoulder of the following delta-sigma modulator for reasons explained by Mr Moorer in his article if one have to remodulated back to a 1 bit stream (to produce a release format for SA-CD distribution, for example), but the overal process is really easy to understand*. It is also possible to not remodulate back to 1 bit and convert the results of the gain adustment computation directly in analogue with a multibit converter. That's what is done in some Cirruc Logic digital to analogue audio converter chips such as the
CS4365, CS4385, CS4398... with additionnal tricks, because
Cirrus Logic combines the function of both gain adjustement and low-pass filtering of the DSD shaped quantization noise to comply with the recommended 50 kHz filter of the Scarlett book, but to my mind the principle of the gain adjustement function remains fundamentally the same.
In a David Walstra's article in Studio Sound (
DSD: Where are We Now?, October 1998, page 106), it is written that a working DSD editing system designed jointly by Sony and Sonic Solution has been demonstrated at the 104th AES convention in Amsterdam in 1998. Obviously, DSD were already editable before SA-CD discs and players were put on the consumer market.
*Note : to my understanding, the whole process in fundamentally identical to the technique Sharp used to use to adjust the volume on DSD streams in his memorable "1 bit" amplifiers, such as the SM-SX1, SM-SX100, SM-SX200... In this integrated amplifiers, the DSD stream, once converted back from discrete values to analogue in the form of square signals is simply pass through... a potentiometer wired as an attenuator ! The square signals whose voltage amplitude has just been reduced by the potentiometer is modulated to another 1 bit stream by a delta-sigma modulator (7th order) to drive the switching transistors of the class D output stage. This process to adjust the gain of the bitstream in the analogue (continuous time) domain is really comparable to the process in the discrete (digital) domain : the potentiometer is just replaced by the computation with the chosen scaling factor applied on the discrete 1 bit data. One can follow closely the signal path from the output of the Sony SA-CD decoder in the DX-SX1 transport associated to the SM-X1 amplifier up to the delta-sigma modulator chip in the service manuals of each device to figure out the technique :
Sharp DX-SX1 (from page 25 from output pins 47 and 48 of the Sony CXD2751 decoder) and
Sharp SM-SX1 (from page 20, "SOC101" 1 bit terminal).