Glad you really don't mind a little imperfection along with your music..
I am fortunate to be able to purchase and download quite a wide range of the music I like in higher resolution formats such as 2496, 24192 and DSD. I figure it makes sense to play them back in their native format. Suffice it to say I am sold on 24 bits and sample rates of 48kHz or greater.
I hope that you weren't serious that it is up to Qusp to prove anything to you. The papers exist, you can find them, read them and get from them what you will.
I was serious. I am an electronic engineer, and a VLSI designer, and a DSP engineer, and a programmer - with more than 30 years of experience.
Before you/we go any further - have you heard of 'sinc' function: https://en.wikipedia.org/wiki/Sinc_function and do you know how it is involved in ADC -> DAC chain and in resampling if any ?
If not, all the conversations are about (I learned the terms in microphone DIY list) unobtanioum, hair of Danish virgins, snake oil and golden pinnae.
Before I give you any links (which can be easily found), let's talk about the ADC -> DAC conversion in the ideal world of math.
Math (DSP theory) states and proves that
1) if you have a band limited signal, i.e. a signal which has no components with higher than Nyquist frequency;
2) if the signal is sampled discreetly with Fs = 2 * Fn, where Fn is the Nyquist frequency;
3) if the samples are infinitely precise,
then the the original signal can be restored perfectly. I.e. in all time points (not only at sample points) the value of output signal will be equal to the value of input signal.
Again, please pay attention to band limited.
The above is the fundamentals of all digital recording and reproduction.
Physical implementations have limitations and imperfections, and engineers look for working around the limitations and imperfections.
...
Real signals are not band limited, so we have some aliases and they can not be restored perfectly.
44.1KHz sampling rate is horrible because in order to avoid aliasing it requires a very steep anti-aliasing filter which grossly distorts phase causing, among other thing, pre-echos. ... What else is new ?
I am not into religious arguments. Regarding audio and electronics I am an engineer, and I know stuff is designed by engineers.
The comment was intended to be tongue-in-cheek
Your tongue in a Danish virgin cheek ?
Then surely you understand the problem with your remark about bit perfect, we're not talking about anything other than maintaining a path free of avoidable transmission errors.
You make a good point about sampling theory, however in practice most hardware on the analog side and the converters themselves make achieving the ideal delineated in Nyquist almost impossible to achieve in practice, hence higher sample rates and greater bit depths.
FWIW I suspect a reasonably priced pro-sumer sound card may provide all of the performance you are looking for at a reasonable cost. How many channels do you actually need? M-Audio, EMU and ESI make or have made good high performance sound cards with competent ADC and DAC implementations - perhaps not ideal in a noisy PC environment but technically closer to perfection than most of the cheap bottom feeder sort of hardware we've been discussing. I would be surprised if the hardware in my SMSL dac approaches the measured performance of either my M-Audio 2496 or 24192 PCI neither of which cost appreciably more than that DAC and offer stable, reliable drivers and good enough performance that I trust them for use characterizing my new hardware designs. All of this is predicated on the idea that a PC is involved somewhere in the mix, if not perhaps there is another yet viable solution.
yes the sinc function is involved, in SOME and definitely not all audio playback architectures and delta sigma ADC OSF in recording and production processes. but the waters are very muddy and all dacs handle these filters differently depending on their architecture, standards are not adhered to even on the ADC end. to keep the brick wall filter from having an effect in the audible bandwidth when combined with further analogue filters on the dac side is a mess not easily cleaned up.
the easiest, perhaps laziest way to do this is to push them further out of band.
you know, you need to do some doing, rather than all this reading, we do not live in a theoretical world and standards are not adhered to
and the analogue filter in most of the dacs in your price range is far from perfect let me tell you
I do second the recommendation for an EMU 0404 or similar, well actually it depends on what computer and OS you are running. do you have an objection to used? I dont see any need for new hardware given your
A. purpose and needs
B. World view
I think its a bit humorous you are trying to out math my compatriot up there
also to be clear, i'm just putting info out there, myself I dont upsample my audio and try not to buy it if possible, oversampling I dont have any problem with, particularly if using a synchronously clocked system, but I think its maybe better done on the PC side and then fed to the dac 8x OS, but not upsampled. given the previous conversation, the above reads as if google featured heavily in your day today
Last edited:
Sure not. If you reread what I wrote about band limited signal and perfect restoration, you'll probably guess that I care about perfect restoration and not about bit-perfect restoration.
Suppose I have the same band-limited signal which I sample with two mathematically perfects ADCs. The only difference between the two ADCs is that one samples half (for example) sample period later than the other.
Both sequences of samples will be perfectly restored to the original signal by perfect DACs, yet the vast majority of samples in the two sequences will differ simply because they were performed at different times - remember - we have half sample period shift.
Now look at this from a different standpoint: since mathematically the original signal can be perfectly restored, it can also be resampled (with the same frequency) at different (i.e. shifted) times. And because it is resampled at different (i.e. shifted) times, most of the samples will differ from the original.
I'm trying to say that I don't care what the DSP algorithm is doing - as long as it produces sample sequence which can be reconstituted into the same input signal. In fact, I am using a lot the thing called "DSP fractional delay".
And from yet another standpoint - bit-perfectness is, of course, a sufficient condition for perfect signal restoration, but pursuing it often deprives the pursuer of other benefits.
...
On a practical note. Pursuit of bit-perfectness is ridiculous in the world of recorded (not synthesized) CD audio music. This is because professional equipment typically works with 48/96/192KHz sample rates, so final downsampling to 44.1KHz is non bit-perfect by construction.
Rather than pursuing bit-perfectness one should simple be able to choose the right resampler when needed. In real world due to unavailability of infinite time one has to truncate the 'sinc' function, but this can be made with any desired accuracy. When the errors introduced by the resampler are (significantly) less than 1 bit quantization error, the resampler in practical (not mathematical) terms is bit-perfect.
...
And I haven't yet touched acoustical SNR in real living rooms/concert halls ...
Alas, I am originally an analog guy. First of all, I am a radio-physicist by diploma (i.e. electronics, RF, quantum physics, lasers, V/UHF). Secondly, I did design a successive approximation ADC in the eighties.
It was rather interesting taking into account available to us at the time analog components.
For example, I used R-2R ladder, and I used bipolar transistors as switches and, interestingly enough, by playing with base current I could achieve zero voltage on the switches in ON condition.
Another interesting feature was the comparator. It consisted of an uncompensated OpAmp, but I had to put a DC preamplifier with 3 .. 5 gain in front of it in order to achieve the needed speed of switching. Luckily, the package contained two OpAmps
....
If anybody claims that at home by simple means he/she achieved better than 80db SNR/dynamic range, I am in doubt. For example, with my simplest amplifiers being powered OFF I see about 10mVpp at output. And the frequency is about 90MHz. "Surprisingly" there is a nearby radio station called "90FM".
When the amplifiers are powered ON, I still see the 10mVpp. And it's becoming really interesting when I need to get rid of parasitic generation .
I mean, in my EMI jungle reality I have to be really careful about possible IMD between the 90MHz and sound.
Very interesting and your comments on analog design issues is dead on IMO. In a past life I designed semiconductor ATE test hardware and getting anything much beyond 100dBSFDR was quite a task in that environment.
In audio design I have the impression that it is not just the noise, but the nature of the noise that makes the difference.
In audio design I have the impression that it is not just the noise, but the nature of the noise that makes the difference.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.