He says nothing that controverts sinc envelope distortion, and offers only 1982-level opinion otherwise IAC. I've already shown for which purposes it can be compensated for and for which it can't. The question is: when are its audible effects tolerable? Obviously not so much for professional audio.
Last edited:
If the distortion goes up at -60dB compared to 0dB, it should go up even more at -90dB. It doesn't. You're positing a non-existent problem.
Nothing special:
A loop back recording of a CD ABXed against the rip.
Make sure the start and end points are sample accurate, levels are within 0.1dB. Provide the files used with a description on how you got a positive result, so others can verify your results.
DSP is a different animal to playback.
Maybe you missed my edit, he specifically mentioned it in his later paper at the same conference, no matter you obviously have your own ideas. BTW did you ever try putting an oscillator sweep into a sound card or better a white noise track from a test CD into a wide band spectrum analyzer?
I actually am a DSP engineer. The last big DSP project I did was an ultrasound system for locating discharges under oil in high voltage electrical plant. We actually used envelope detection as part of the algorithm, and agonised for ages over whether it was OK to use sigma-delta converters and so on.
If you want an accurate record of the peak value of a waveform, then you do indeed have to sample pretty fast, and a sigma-delta converter is no good as it will replace a sharp pulse with its own impulse response, and the answer you get depends on where the peak of the original waveform fell between samples, as much as the magnitude of it.
The real question is whether this is relevant to audio, are our ears accurate peak detectors? I don't think so, as there is no proof that a brickwall filter is audible. A waveform can look different to the original and yet sound the same.
Another experiment you can try is to feed a tone near the Nyquist frequency through a sigma delta ADC and DAC. As you get close to Nyquist, you will see a horrible beating in the envelope that gets bigger and bigger, until it goes from zero to full amplitude. But this is not audible either. The tone has passed above the limit of hearing before the beating becomes significant.
If you want an accurate record of the peak value of a waveform, then you do indeed have to sample pretty fast, and a sigma-delta converter is no good as it will replace a sharp pulse with its own impulse response, and the answer you get depends on where the peak of the original waveform fell between samples, as much as the magnitude of it.
The real question is whether this is relevant to audio, are our ears accurate peak detectors? I don't think so, as there is no proof that a brickwall filter is audible. A waveform can look different to the original and yet sound the same.
Another experiment you can try is to feed a tone near the Nyquist frequency through a sigma delta ADC and DAC. As you get close to Nyquist, you will see a horrible beating in the envelope that gets bigger and bigger, until it goes from zero to full amplitude. But this is not audible either. The tone has passed above the limit of hearing before the beating becomes significant.
Last edited:
Sorry, you're downright wrong! My own experiments demonstrate that "crap imaging and trashy hi end" is all about insufficient engineering of the analogue side - poor implementation is poor implementation, just sort that out and then everything falls into place as it should ...
It's actually quite an interesting point. Sound travels at about 1000ft/sec in air, so 1 foot per millisecond, 44 Red Book samples to the foot. Each sample is about 1/4 inch of sound wave, and the impulse response of a sigma-delta audio converter could span a good few inches in its first few major ripples.
If we had superb sonar devices for ears, then we might see (hear? auralise?) that the impulse response of our audio converters would blur a point source in the stereo soundstage into something like an Airy disk - Wikipedia, the free encyclopedia . This was what we observed to happen in our ultrasonic location experiments with sigma-delta converters.
I saw one source claiming that the ear/brain can resolve time of arrival differences of 10 microseconds (about half a Red Book sample) between ears, so maybe we do have superb sonar devices for ears.
If we had superb sonar devices for ears, then we might see (hear? auralise?) that the impulse response of our audio converters would blur a point source in the stereo soundstage into something like an Airy disk - Wikipedia, the free encyclopedia . This was what we observed to happen in our ultrasonic location experiments with sigma-delta converters.
I saw one source claiming that the ear/brain can resolve time of arrival differences of 10 microseconds (about half a Red Book sample) between ears, so maybe we do have superb sonar devices for ears.
Last edited:
Our *differential* spatial perceptions have evolved to be able to locate the angular location of a transient such as a snapping twig in a natural ambient environment to within a couple of degrees which gives an idea of the extent that a stereophonic image can be perceived. Of course, nothing anywhere close to such accuracy is within the capability of what is detected by only one ear.
Last edited:
Is this something from the recent work of Bob Stuart & Peter Craven about time smearing from anti-aliasing/anti-imaging filters?
Google Meridian MQA for more info. This is the Stereophile report:
Meridian Presents MQA
I know this as the aperture effect, its very well known. And there are lots of way's to deal with this, as others have pointed out to you. Its not an existing problem for most converters.
This is a very interesting paper.
They used very steep reconstruction filters with a 500Hz transition band (normally this is 4x bigger in redbook).
And they used extremely high sound pressure levels of 120dB.
Its the only paper in existence that demonstrates an (barely) audible difference between "normal" and hires formats. There's speculation on what is the cause of this audible difference, but certainly no consensus. IM could be a strong candidate and the very steep reconstruction filter an other.
Imo this doesn't seem like a very strong case in favour of hires formats.
Concerning timing resolution: Sampling theory clearly states that you can PERFECTLY (this includes the timing resolution) reconstruct a signal with a sampling frequency of more than twice the upper frequency limits of that signal.
To be clear (since this is a common bit of nonsense thrown around by people with little understanding of digital method, and you KNOW that's coming next), timing resolution is not equal to sample spacing (i.e., 1/sample rate). It's actually several orders of magnitude better than the 10us interaural sensitivity of the sharpest ears.
I prefer an alternative statement of this: In order to sample and reconstruct a signal perfectly, you must remove all content above the Nyquist frequency before sampling. The reconstruction can be perfect, but the anti-alias filtering before sampling always loses something.
The reason for this is that continuous time signals always have infinite bandwidth in theory, there is no frequency for which you can say for sure that the signal contains no energy. (In practice, the energy gets so small above some frequency that it can be neglected.)
The design of the anti-alias filter is subject to time-frequency duality. You can have a nice sharp transition band that suppresses aliasing without attenuating wanted high frequencies, or you can have a nice impulse response that maintains timing resolution. You can't have more of both unless you crank up the sampling rate.
Moved to the lounge by moderation.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.