Analog or Digital for your tunes

[The Expanding Man]

so realistically a CD is never going to faithfully reproduce sound all the way up to 22KHz

In fact, at 22khz, all redbook can reproduce is a square wave. That's a major loss of resolution (it can't get much worse!!).

[The Expanding Man]

Vinyl sounds awful unless you love distortion

One of these, or a combination of these, is your culprit:

Worn out stylus
Bad Cartridge alignment
Incorrect tonearm balancing / weight
Dust (on the stylus and/or vinyl)
A poor RIIA correction in your phonostage.

Vinyl is quite laborious. It's a lifestyle decision

[mesh control]

[Ex-Cyber]

so realistically a CD is never going to faithfully reproduce sound all the way up to 22KHz

In fact, at 22khz, all redbook can reproduce is a square wave. That's a major loss of resolution (it can't get much worse!!).

A standard feature of digital playback/reproduction systems is a reconstruction filter. Basically, this is a low-pass filter on the output with a cutoff frequency near the Nyquist frequency. It's there precisely to get rid of the higher harmonics of the raw digital signal. In short, even if the raw samples on the disc look like a square wave, that's not what you'll hear (actually you would never hear a true square wave since something would attenuate some of the higher harmonics).

Yeah, it seems like logically that the reduced resolution-per-wavecycle close to nyquist should result in some amount of (potential?) distortion for very high frequencies. It seems though like a lot of that would be the responsibilities of the ADC (when recording) and the DAC (when playing back).

Antialiasing (recording) and reconstruction (playback) filters are generally considered standard parts of the sampling and reproduction processes, and are often integrated into ADC and DAC chips, but they're not strictly part of the digital domain. They still have limitations, though (some theoretical, some practical). I'm not an expert on audio applications, so I don't really know what kinds of things have been done in that field to minimize distortion.

I'm not sure what difference a signal being synthesized in the digital domain would have: Fourier's theorem says that any sound, recorded or otherwise, can be precisely decomposed into sinewaves.

A discrete-time signal can't really represent a frequency component higher than the Nyquist frequency (aka "folding frequency" aka fs/2). When sampling an analog source, frequencies higher than the Nyquist frequency don't just disappear, though, they get "reflected" and show up as frequencies below it. For example, if you sample a 23KHz component with a sampling rate of 44KHz, it will produce a 21KHz component in the digital signal. Likewise e.g. a 43KHz component would produce a 1KHz component. Thus a recording system must have a low-pass filter on the input to attenuate higher frequency components. The difference is that a digitally synthesized signal (assuming it's synthesized at or below the sampling rate at which it will be recorded) isn't being sampled and thus won't have higher frequency components and won't require an antialiasing filter.

[cools]

Square waves make me wet.

[Domino]

That picture reminds me that I need to pick up Blue Train (Coltrane) one of these days.

[thegreathopper]

Don't bin the michell! I'll have it!

My dad used to have a michell hydraulic reference, the same one used in clockwork orange.

Unfortunately it is not a michell hydraulic reference, Mine is a Focus one with red mat. I was just about to say you could have it but maybe I should ebay it instead.
http://www.avforums.com/forums/hi-fi-sy ... eeded.html

[kernow]

nahhh



Is our friendship not worth £300-350

[louisg]

I'm not sure what difference a signal being synthesized in the digital domain would have: Fourier's theorem says that any sound, recorded or otherwise, can be precisely decomposed into sinewaves.

A discrete-time signal can't really represent a frequency component higher than the Nyquist frequency (aka "folding frequency" aka fs/2). When sampling an analog source, frequencies higher than the Nyquist frequency don't just disappear, though, they get "reflected" and show up as frequencies below it. For example, if you sample a 23KHz component with a sampling rate of 44KHz, it will produce a 21KHz component in the digital signal. Likewise e.g. a 43KHz component would produce a 1KHz component. Thus a recording system must have a low-pass filter on the input to attenuate higher frequency components. The difference is that a digitally synthesized signal (assuming it's synthesized at or below the sampling rate at which it will be recorded) isn't being sampled and thus won't have higher frequency components and won't require an antialiasing filter.

Right, but the point is that you're brickwalling out those >22khz frequencies so they don't alias. Normal humans can't hear that high anyway, so even a little bit of rolloff isn't going to hurt much.

Digitally synthesized sounds also have to deal with aliasing: you can try this if you generate a sine wave above the nyquist rate. That's one thing that makes even simple synthesis so processor-intensive: A lot of applications will oversample like crazy or use non-naive non-table-driven bandlimited synthesis to avoid those nasties. So, even if everything is generated digitally, you will have a little bit of rolloff as you get near nyquist.

I'd even argue that recordings have an advantage in that you can use expensive non-realtime algorithms to downsample from a high rate (where nyquist is twice human hearing's limit) instead of having to settle for a cheaper on-the-fly method.

BTW, to The Expanding Man, if CD players generated a squarewave near nyquist, you'd know pretty fast! I think phase shifts manifest as a loss of amplitude around that pitch.

[subcons]

Digital certainly dominates my listening time for the sake of convenience. But I do love my modest vinyl collection. Nothing really compares to vinyl as far as the type of sound you get from it compared to an MP3.

[Ed Oscuro]

Vinyl is often perceived as warm and punchy. Warm is generally a lack of high frequencies, and punchy is often the consequence of non-linear distortion.

Also, LULZ @ people who say "warm" and "cool" in terms of audio.

I actually do enjoy listening to stuff on tape now and then, but good digital masters are simply cleaner and more of the frequencies come through. (I never have listened to vinyl, but that typical cassette noise hiss has its charm sometimes.)

http://i908.photobucket.com/albums/ac28 ... 010701.jpg

Ahh, lovely aliasing on the cover; vinyl strikes again!

[Alec]

Uncompressed Digital

[Ex-Cyber]

Digitally synthesized sounds also have to deal with aliasing: you can try this if you generate a sine wave above the nyquist rate. That's one thing that makes even simple synthesis so processor-intensive: A lot of applications will oversample like crazy or use non-naive non-table-driven bandlimited synthesis to avoid those nasties.

Simple synthesis is not especially intensive compared to its sample rate. Early sound chips mostly just had counters and/or sine ROMs hooked up to a DAC, and there are a bunch of other applications where you pretty much just use a phase accumulator indexing a sine table or other sampled waveform (IIRC this is the basis for Yamaha's "FM" chips as well as the SID, but it's been a while since I've read up on them so I could be getting mixed up). Anyway, my point was that if you're doing direct digital synthesis at the target sample rate, you can't really get "aliasing" in the conventional sense, because those higher frequencies just don't exist. As for oversampling, that's typically done for the sake of making it easier to digitally filter the signal (because analog filters tend to be pretty finicky and unreliable compared to digital ones). For example, I've heard that CD players commonly interpolate to something like 176KHz (I don't recall the exact number) for digital reconstruction filtering.

[louisg]

Simple synthesis is not especially intensive compared to its sample rate. Early sound chips mostly just had counters and/or sine ROMs hooked up to a DAC, and there are a bunch of other applications where you pretty much just use a phase accumulator indexing a sine table or other sampled waveform (IIRC this is the basis for Yamaha's "FM" chips as well as the SID, but it's been a while since I've read up on them so I could be getting mixed up). Anyway, my point was that if you're doing direct digital synthesis at the target sample rate, you can't really get "aliasing" in the conventional sense, because those higher frequencies just don't exist.

That's just not true; here are a couple audio examples:
http://www.gorenfeld.net/lou/alias1.wav
This is a simple sinewave steadily climbing, going above nyquist (22khz), until it settles at 43800hz. But, what you hear is not a sinewave at 43800hz but a sinewave at 300hz since it's aliased. This happens because the rate at which the audio "snapshots" are taken of the sinewave are not fast enough to capture the rate-- it's kind of how if you shine a strobelight on a wheel, and spin the wheel fast enough, the wheel will appear to be going backwards even though it's actually traveling forwards.

http://www.gorenfeld.net/lou/alias1.wav
This is the exact same thing but with a simple/naive sawwave (ramp up). In this, you hear aliasing *well before nyquist* because the frequency components which make up the signal have already folded over the top! You can probably hear this in many cheap digital synthesizers (sid, gameboy, whatever) unless they are running at a very high rate (which they might be.. I haven't played with them too much).

Yamaha FM does alias quite a bit since it's not oversampled like modern digital synthesizers are. To hear this, make a really busy patch with a high modulation index and play high notes. It's especially audible if you use the pitch wheel. A lot of other digital synths have this problem too, like the MT-32 for instance.

Here is a popular method people use to avoid this problem:
http://www.experimentalscene.com/articles/minbleps.php

So, I won't argue that a not-so-good synth will go to this trouble, but on the topic of audio quality it's a huge issue in digital synthesis.

As for oversampling, that's typically done for the sake of making it easier to digitally filter the signal (because analog filters tend to be pretty finicky and unreliable compared to digital ones). For example, I've heard that CD players commonly interpolate to something like 176KHz (I don't recall the exact number) for digital reconstruction filtering.

Yeah, you get fewer problems with filters too IIRC at high rates. If you oversample before output, you can go with a filter with a gentler slope.. I'm not 100% clear on the details there.

[greg]

Minidisc! I'm the only person who has a car MD player, it seems. Like my dear friend Specineff, I still love MDs. I prefer to own music on CD because of the tangibility issue, but I do have a subscription to emusic.com and I download a lot of stuff from there each month. For making compilations and such, MD is the best format. A month or so ago a friend of mine gave me a SANSA Clip 2gb, so now I finally own an official mp3 player that isn't an MD player. It's great to stick something onto it that may or may not be worthy of having its own MD, but I still get more use out of my MDs.

[Ex-Cyber]

That's just not true; here are a couple audio examples:
http://www.gorenfeld.net/lou/alias1.wav
This is a simple sinewave steadily climbing, going above nyquist (22khz), until it settles at 43800hz. But, what you hear is not a sinewave at 43800hz but a sinewave at 300hz since it's aliased. This happens because the rate at which the audio "snapshots" are taken of the sinewave are not fast enough to capture the rate-- it's kind of how if you shine a strobelight on a wheel, and spin the wheel fast enough, the wheel will appear to be going backwards even though it's actually traveling forwards.

That's the kind of scenario I intend to exclude when I say "synthesis at the target sampling rate". Naturally, you can generate aliasing by doing what amounts to decimation in your synthesis process. Maybe I'm mangling the terminology or something, but the crucial issue is that you are calculating the samples, not sampling a physical wave that actually exists independently of your system and thus must be filtered.

[Warp_Rattler]

I do have a subscription to emusic.com

I love eMusic so very much; it's the supplier for probably 90% of my digital listening right now. I just hate the terrible realization that there's a lot of stuff out there, and just the albums I've got in my saved list will take me about a year and a half to acquire on my current plan.

SANSA Clip 2gb

Clip or Clip+? Either way, they're terrific little players. My wife's got the 2GB Clip and I've got a 4GB Clip+ with 4GB microSD card, which makes for more portable listening options than I can exhaust in one go. Gotta love that something so tiny has FLAC and OGG support, something a good amount of higher-profile portables lack.

[louisg]

That's just not true; here are a couple audio examples:
http://www.gorenfeld.net/lou/alias1.wav
This is a simple sinewave steadily climbing....

That's the kind of scenario I intend to exclude when I say "synthesis at the target sampling rate". Naturally, you can generate aliasing by doing what amounts to decimation in your synthesis process. Maybe I'm mangling the terminology or something, but the crucial issue is that you are calculating the samples, not sampling a physical wave that actually exists independently of your system and thus must be filtered.

I'd argue that it's equivalent: the sinewave generated digitally above nyquist is a discrete representation of a continuous function just as a digital recording of audio is a discrete representation of a continuous waveform. As such, the aliasing can be dealt with in either case in a similar way: If you generate that sinewave at a higher sampling rate and just filter out the frequencies you don't want, they will no longer appear in the 44khz wave when decimated/downsampled. This is similar to an ADC oversampling and slicing out frequencies before the signal is passed onto whatever processes/plays it. In both cases, the digital signal is attempting to represent frequencies above its limit and, consequently, they are aliasing.

[drunkninja24]

Pffft, Vinyl has nothing over the ultimate analog recording format......WATER! [/dethklok]

But seriously, for albums I really enjoy, I try to track down the Vinyl for simply because I really feel like I am purchasing something tangible. But I have a huge digital library for putting on my portable players or making big playlists when I just want to chill for a while. I'll whip out the Vinyl if I want to listen to a specific album for a while though.

[GaijinPunch]

http://www.wired.co.uk/news/archive/201 ... and-vinyly

Vinyl +1

EDIT: Totally unrelated but years ago I used to bump elbows w/ one of the members of Subhead, but apparently he's not involved in this project.

EDIT 2: He's apparently passed away now, a few clicks away from the original link has taught me.