ESS claim DAC pushes the boundaries of audio science
Reply #6 – 2013-01-23 19:10:20
pdq, it wasn't all about hearing below -96 dB - he mentioned why that is, and the guy seemed to know his stuff and explain a lot of it pretty well to a lay audience, partly by analogy. He also had a decent enough analogy of noise shaping (error-diffusion type of idea) to get the idea across. It seemed to be about some type of variation in the quantization noise spectrum and amplitude spikes within it as DC level offset slowly adjusts or as you switch from say a 1kHz tone to a 20kHz tone, which the sigma-delta modulators tended to slowly converge upon with an unstable type of feedback loop. There's something plausible there, and it could well be that they've made it inaudible if it was audible before. It's not clear which components they're comparing and the exact nature of the blind tests, but they claimed that different features in their FPGA could be turned on or off, allowing them to select 5 features to control 5 states (modelling it as a state machine), the last of which apparently made it a match for a linear DAC in the blind testing. There was a 3D plot of their device pictured as a state machine (they couldn't plot all 5 dimensions they controlled for) which seemed to show it tighly constrained, while the device without their magic had a more unconstrained chaotic pattern in the mysterious parameters, and there were spectral plots showing spikes in the noise floor presumably of their modulator without their FPGA's features enabled (or maybe an unnamed competitor's SDM) and a clean spectrum with it enabled. These were at a plausible level for a 16-bit signal with small amounts of inadequate dither to become potentially audibly degraded by quantization noise in the presence of a -30dB test tone, though they ought to be rather well masked in real music, examples of which included an old Frank Sinatra track, an un-named piece and a modern recording where a shaker is used, which somehow it supposed to reveal the difference. The guy claimed that their measurements would get them in the door at the consumer companies, this 3D plot would really pique interest, and the companies would only decide when their 'golden-ear' audiophile gave it the thumbs up. It all has a mixture of an air of plausibility and reasonableness in the presenter, and a mix of test & measurement plots and computer modelling plots, combined with the secrecy and lack of detail and transparency that leads one to have enough suspicion that it could all just be a blind-them-with-science marketing approach with some superficially plausible theory behind it. The best approach to truly demonstrate the truth of their claims to skeptical listeners at a conference like that would probably be to provide a digital sound clip and a digital zero source generated on the same clock, two or more DACs with analogue outputs level-matched in a low-noise summing amplifier sent to a single headphone amplifier and a set of high-grade headphones, and send the non-zero digital signal to one DAC or the other DAC one at a time, while the other DAC receives zero, under computer control to provide A/B/X comparisons (potentially with software very much like the software ABX comparators we use here, allowing the choice of a specific revealing part of the clip, such as this shaker). There might need to be time-alignment of the two channels also if switching is allowed during playback.