Hacker News new | comments | show | ask | jobs | submit login

> You can record the playback of a vinyl record and transfer it to a digital record without losing any information.

No. Sampling truly analog data always involves approximating.

See http://en.wikipedia.org/wiki/Sampling_%28signal_processing%2...

Notice how they use the term "theoretical ideal sampler".




No. This is wrong. First, sampling need not imply quantization[1]. Second, the signal to noise ratio of the continuous signal source is important. With sufficiently many bits and sufficiently many samples, you can exactly reproduce any band-limited continuous signal above the noise floor[2]. Those last four words are important. Analog data never comes without noise, and a "sufficient number of bits" is nowhere near as high as you think (particularly with noise-shaping techniques like dithering).

Now, it is the case, as user rayiner mentioned, that there can be advantages to using more than the required number of bits and samples (though not quite for the reasons he mentioned). For starters, you do need an anti-aliasing filter between the signal source and your ADC. Increasing the sample rate reduces the complexity of the AA filter, and digital-domain math can make up for it (and increase the effective number of bits, to boot!). But when you go to reproduce the signal, there's no good reason to use more samples and bits than necessary. That was the whole point of the "Niel Young paper" that has been linked to so many times.

Vinyl does not come close to the limits of consumer-grade recording devices (except maaaaaybe in certain bands, but that can be dealt with by using noise-shaping techniques). Does that make those who like the sound of vinyl bad people? No. Just like those who enjoy sitting in front of a wood fireplace are not bad people. Gas fireplaces have a lot of advantages, but some people like the crackling of the wood and the smell it produces.

[1] Tektronix made a very nice set of (entirely analog) sampling oscilloscopes in the 1960's which used sampling techniques to measure high frequency signals (on the order of 1GHz when contemporary continuous-signal CROs could barely reach 100MHz). These oscilloscopes displayed discrete-time, continuous-amplitude signals, and deliberately excluded the anti-aliasing filter I mentioned above (though it wasn't true heterodyning as most RF people would think of it, because samples were taken based a delay from a trigger recognizer, and thus not necessarily equally spaced in time).

[2] Signals below the noise floor are outside the scope of this discussion and usually require some form of synchronous detection (like a lock-in amplifier) or frequency-spreading / -despreading (like GPS).


The level of 'approximation' involved here would be like making a square that was .0001% too long on one side. You wouldn't be able to tell at all.

The page you listed also says: "There has been an industry trend towards sampling rates well beyond the basic requirements; 96 kHz and even 192 kHz are available.[1] This is in contrast with laboratory experiments, which have failed to show that ultrasonic frequencies are audible to human observers"


The criticisms of these sorts of things are often as unscientific as the original claims. E.g. the fact that humans can't hear 96 KHz doesn't mean that sampling at that rate doesn't make it e.g. easier to design the roll-off filter in the DAC. It doesn't mean it doesn't make it easier to do transformations on the audio like simulated surround. There are a lot of steps between the digital signal on a CD and your ears, and just because your ears can't hear 20 KHz doesn't mean it doesn't make the intermediate steps easier to build.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: