Not entirely true.
Oversampling alone can reduce noise levels.
If for example you have a converter with 4x oversampling, then the noise power goes all the way up to fs/2, assuming the noise spectrum is white. If you then reduce the bandwidth back to the bandwidth of the incoming signal, the noise power will also be reduced by the same factor of the oversampling. In this case the noise is reduced by 6dB. To get useful noise reduction you need lots of oversampling and therefore this method is never used.
Oversampling in combination with noise shaping is much more effective. This works in exactly the same way, but now the noise shaping shifts the spectrum of the noise to the high frequencies. If you then reduce the bandwidth back to the bandwidth of the incoming signal, the noise power will be reduced more than the oversampling factor.
You are of course right. I was trying to keep things simple without introducing too many caveats.
Hi Julf,
I continued the transducer discussion here, please check the link......
http://www.diyaudio.com/forums/full...ferences-paper-alum-titanium-poly-etc-15.html
I continued the transducer discussion here, please check the link......
http://www.diyaudio.com/forums/full...ferences-paper-alum-titanium-poly-etc-15.html
Personally I believe op-amp's with approximately 20 microsecond settling time sound vastly different than op-amp's with 100 nanosecond settling time.
For instance OPA2111KP versus AD8620.
The transducer must be able to respond to the difference or I wouldn't hear it.
But now I'm speaking in vagueness without evidence.
Though I recommend you or anyone else try it for yourself!
Julf said:
Alright.
Thanks, but my interest is in digital audio these days, not headphones. If I need another decent set of headphones, I go and buy them, rather than try to build them myself.
But do you think one with a 10 nanosecond settling time sounds different from one with a 100 nanosecond settling time?
What matters is that the "speed", slew rate and settling time is fast enough not to affect audible frequencies even at full amplitude. If that requirement is satisfied, going beyond that doesn't buy you anything.
Anyway, this doesn't really have much to do with resolution or quantisation noise, does it?
Indeed.
My summary after this thread is
- Linear is ideal when there is zero pre-/post-echo
- Minimum is ideal when there is pre-/post-echo
- Filterless is ideal for all low latency applications like instruments or audio/visual
Last edited:
AD797, AD8620, AD828, LT1363 all sound the same in "speed" however the last two seem to sound a little sharper and stricter.
Julf said:
It does not.
Something more on filters and how they reconstruct the original signal.
The value of the sample is changed into a short pulse wave witch is then fed into a filter. In a perfect world these pulses should be infinitely short and the impulse response of the filter should be a perfect (sin(x))/x function. (aks sinc function). This function is of infinite size.
In the real world the pulses have a certain pulse width and the sinc function is truncated, witch means it is chopped to have a finite size.
Here's the useful part of the sinc function:
It is important to notice that the sampling is taking place at the whole numbers of the x axis. So at ....-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6....
Different sample values produce different size sinc functions, only the y-axis amplitude of the sinc function changes. And all these different sinc functions are added together to reproduce the original signal.
This picture shows how that's done:
On latency:
Latency is not an issue for musicians if its less than about 3-4msec. A normal converter has no problems with this at all. The drivers of the soundcard can be much more of a problem here. In these day's its not difficult to get latencies of 2-3 msec for computer based instruments. Most soundcards have analogue monitoring so latency doesn't matter at all. Basically its a non existing issue these day's.
The value of the sample is changed into a short pulse wave witch is then fed into a filter. In a perfect world these pulses should be infinitely short and the impulse response of the filter should be a perfect (sin(x))/x function. (aks sinc function). This function is of infinite size.
In the real world the pulses have a certain pulse width and the sinc function is truncated, witch means it is chopped to have a finite size.
Here's the useful part of the sinc function:
It is important to notice that the sampling is taking place at the whole numbers of the x axis. So at ....-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6....
Different sample values produce different size sinc functions, only the y-axis amplitude of the sinc function changes. And all these different sinc functions are added together to reproduce the original signal.
This picture shows how that's done:
On latency:
Latency is not an issue for musicians if its less than about 3-4msec. A normal converter has no problems with this at all. The drivers of the soundcard can be much more of a problem here. In these day's its not difficult to get latencies of 2-3 msec for computer based instruments. Most soundcards have analogue monitoring so latency doesn't matter at all. Basically its a non existing issue these day's.
Still, it's typical that legend, folktale, anecdote is closer to the truth than the available data.
Today I saw "new data" on TV that ancient Greeks arrived in America thousands of years prior to Christopher Columbus, that they excavated thousands of tons of Copper and took it back to Europe.
The discovery of "new data" in ancient Egypt, such as Nicotine, is further data which "illustrates" ancient trans-Atlantic trade.
This indicates that millions of history books worldwide, are telling unequivocal lies.
Looking at the documentary, this ancient Greek trade looks like it will soon become "truth".
If we rewrite them, according to the 2014 data, later they'll very likely just be rewritten further.
You can say audio is very different, which it is, since it's only electricity and air so to speak, but the laws of data versus truth still apply right.
According to the Cirrus Logic paper linked earlier we can perceive down to 10 microseconds.
It won't surprise me if it's somewhat evidenced later, that we can perceive differences in let's say 75 nanosecond versus 750 nanosecond settling time performance.
Which will result in accepted "higher" or "different" sound quality.
It's just the data which is not available right now =)
Just my take on the state of affairs.
Analog monitoring does not interact with any software or digital realms at all, it's just a direct signal from the microphone to the musician, according to what I've read.
Are you saying we can achieve two to three milliseconds inclusive of the analog monitoring or two to three milliseconds in the digital realm?
Is this using ASIO or something else?
Thx
You mean for phase perfect reconstruction we need phase perfect FIR or IIR filters right?, which indicates linear-phase FIR / IIR filters.
With "heavy" pre-/post-echo in normal media files you don't think minimum-phase is the most ideal?
The human ear is much more perceptive to pre-echo, rather than post, according to the Wolfson paper.
The ensuing slight phase difference is a trade in "removing" the pre-echo.
I can't quite see the point myself to talk of 'perfect reconstruction' as perfection requires perfect bandlimiting prior to sampling. It follows that no sample will ever get to be a non-zero value because a perfect brickwall linear phase filter has infinite latency.
<afterthought> What kinds of errors get introduced if the AAF and AIF are both minimum phase, rather than linear phase? Does anyone have an analysis that covers that case?
Last edited:
Not "infinite".
A tenth of a nanosecond sampling rate is sufficient to convey the speed of atomic-level movement in the air we live in.
Molecular Dynamics Cinema
Surpassing that will eventually lead to less efficiency, which I tried to illustrate earlier with "turning water into light".
No. It is typical that once strong evidence is found for a theory, some of the thousands of anecdotes turn out to have been closer to the truth than others - purely by accident.
If it was on TV, it must be true. I have seen "new data" on TV proving very convincingly that the pyramids were built by aliens from space.
I suggest you read the wikipedia entry for Pre-Columbian trans-oceanic contact. Among the many crackpot theories, your italian professor with his pineapple gets almost half a paragraph.
They are, but for completely different reasons - mainly because history gets written by the victors.
Another good read for you: The scientific method
You realize that that corresponds to sound travelling less than 5 mm?
In 75 nanoseconds, sound travels about the width of a human hair.
It might also surprise you that a 16 bit signal sampled at 44.1 kHz can resolve time differences smaller than a nanosecond.
Which qualifies it as "wild speculation".
This has been asked before - What is the point of this thread? Just give some exposure to your wild amateur armchair speculation?
2-3 msec latency is normal these day's. ASIO is one driver that can do this.
Look at the second picture I posted. You see that the peaks and valley’s of the impulse responses of the filter cancel each other out. This means that the pre echo is not audible.
Pre echo can be an issue where the fir phase linear filter has frequency cure deviations. For instance with loudspeaker filters.
1-Because this makes explanation easier.
2-Not only must the filter have infinite length, if we want mathematical perfect reproduction, the sample pulsewidth must be zero. But I explained this in my post.
<afterthought> I'm very curious about this to, I could find nothing about it.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.