Thanks for the info. The Marantz SA113S uses the TI/Burr-Brown DSD1792 delta-sigma converter, with specs here:
http://www.ti.com/lit/ds/symlink/dsd1792a.pdf
Going by the spec sheet, the Burr-Brown DSD1792 does not accept 128fs DSD, and is restricted to SACD and 64fs DSD-download playback. The distortion is 0.0004% at 44.1kHz, 0.0008% at 96kHz, and 0.0016% at 192kHz. The Resonessence Invicta, based on the ESS 9018, is slightly lower at less than 0.00032% (with no mention of frequency).
http://resonessencelabs.com/tech-specifications/
http://resonessencelabs.com/wp-content/uploads/2012/05/InvictaMeasNotes.pdf
Page 20 of the PDF has an interesting section mentioning how noise level in delta-sigma converters are related to the DC level of the signal. DC-dependent noise can increase as much as 20 db (!) with some delta-sigma converters; the relationship between DC level and noise depends on the selection of noise-shaping algorithm by the chip vendor.
The steady-state distortion specs don't tell us much about the sound. The Burr-Brown 1792 and ESS 9018 are both delta-sigma architectures, but with different noise-shaping algorithms, so we can expect them to sound different.
The noise-shaping algorithm (which is designed by the chip vendor) is responsible for a 40 to 60 dB improvement in converter performance. 40 to 60 dB of multiple-loop digital feedback around a high-speed 5 or 6-bit converter is a big deal; without noise shaping, a 5-to-6 bit converter would not be suitable for high-quality audio at all.
This is a purely subjective opinion, but I feel that the noise-shaping algorithm dominates the sound of delta-sigma converters, although low-speed analog electronics can mask the differences between converters and overlay another whole layer of coloration. In particular, low-speed electronics will react quite differently to the ultrasonic comb spectra of PCM and the ultrasonic spread spectra of DSD.
Here's a link to a seemingly unrelated topic:
http://arstechnica.com/science/2013...tric-grid-by-keeping-generators-out-of-synch/
This is what happens when you have multiple feedback loops that interact with each other; small, difficult-to-predict transient conditions can make the whole system go unstable. Noise-shaping, as used in delta-sigma and DSD converters, are high-order feedback systems wrapped around an array of switches, and as mentioned in the ESS and Resonessence literature, can behave unpredictably under dynamic conditions. Steady-state sinewave testing will not reveal the instabilities; the sliding-DC test exposes some of the instabilities, but not all of them.
Part of what I like about ladder/R2R converters is they are flash converters; no noise-shaping, no feedback, no stability issues, just a switch array. Yes, there are issues with monotonicity (due to physical limits of resistor-matching), but Philips dynamic-matching and Burr-Brown's CoLinear are effective ways around this. The monotonicity errors are also steady-state; that is, they are always there, instead of a dynamic distortion that comes and goes with the signal. Loudspeakers, after all, have pretty substantial amounts of distortion compared to amplifiers and digital systems, but the distortion is steady-state.
At the risk of going all meta on you guys, I'm attracted to non-feedback audio technology, with the signal only traveling in the forward direction, and backward shunt paths through the power supplies kept to a minimum.
http://www.ti.com/lit/ds/symlink/dsd1792a.pdf
Going by the spec sheet, the Burr-Brown DSD1792 does not accept 128fs DSD, and is restricted to SACD and 64fs DSD-download playback. The distortion is 0.0004% at 44.1kHz, 0.0008% at 96kHz, and 0.0016% at 192kHz. The Resonessence Invicta, based on the ESS 9018, is slightly lower at less than 0.00032% (with no mention of frequency).
http://resonessencelabs.com/tech-specifications/
http://resonessencelabs.com/wp-content/uploads/2012/05/InvictaMeasNotes.pdf
Page 20 of the PDF has an interesting section mentioning how noise level in delta-sigma converters are related to the DC level of the signal. DC-dependent noise can increase as much as 20 db (!) with some delta-sigma converters; the relationship between DC level and noise depends on the selection of noise-shaping algorithm by the chip vendor.
The steady-state distortion specs don't tell us much about the sound. The Burr-Brown 1792 and ESS 9018 are both delta-sigma architectures, but with different noise-shaping algorithms, so we can expect them to sound different.
The noise-shaping algorithm (which is designed by the chip vendor) is responsible for a 40 to 60 dB improvement in converter performance. 40 to 60 dB of multiple-loop digital feedback around a high-speed 5 or 6-bit converter is a big deal; without noise shaping, a 5-to-6 bit converter would not be suitable for high-quality audio at all.
This is a purely subjective opinion, but I feel that the noise-shaping algorithm dominates the sound of delta-sigma converters, although low-speed analog electronics can mask the differences between converters and overlay another whole layer of coloration. In particular, low-speed electronics will react quite differently to the ultrasonic comb spectra of PCM and the ultrasonic spread spectra of DSD.
Here's a link to a seemingly unrelated topic:
http://arstechnica.com/science/2013...tric-grid-by-keeping-generators-out-of-synch/
This is what happens when you have multiple feedback loops that interact with each other; small, difficult-to-predict transient conditions can make the whole system go unstable. Noise-shaping, as used in delta-sigma and DSD converters, are high-order feedback systems wrapped around an array of switches, and as mentioned in the ESS and Resonessence literature, can behave unpredictably under dynamic conditions. Steady-state sinewave testing will not reveal the instabilities; the sliding-DC test exposes some of the instabilities, but not all of them.
Part of what I like about ladder/R2R converters is they are flash converters; no noise-shaping, no feedback, no stability issues, just a switch array. Yes, there are issues with monotonicity (due to physical limits of resistor-matching), but Philips dynamic-matching and Burr-Brown's CoLinear are effective ways around this. The monotonicity errors are also steady-state; that is, they are always there, instead of a dynamic distortion that comes and goes with the signal. Loudspeakers, after all, have pretty substantial amounts of distortion compared to amplifiers and digital systems, but the distortion is steady-state.
At the risk of going all meta on you guys, I'm attracted to non-feedback audio technology, with the signal only traveling in the forward direction, and backward shunt paths through the power supplies kept to a minimum.
Last edited: