[Head Injury]

Quote:

I think the lines will still be straight, if the D/A doesn't alter them.

No. Analog signals are sine waves. They don't have straight lines. The D/A process takes the points of a digital wave form and basically draws sine waves between them. It doesn't matter how smooth the digital waveform is. Twenty points in a period won't draw a more accurate sine wave than two points in the period. All you need is an approximation.
 
Think about it like this. You have two devices that can draw a circle. One is a regular compass, with two contact points (the center and the pencil). The other is a weird extra complex compass with a center and 19 pencils arrayed in a circle around it. One won't draw a better circle than the other.
 
Once again, a higher sampling rate (which is pretty much what your graph represents) only increases the maximum frequency. It doesn't magically make the other frequencies smoother, just because it gives the D/A process more points to play with.

 

[kiteki]

Quote:

That's a faulty representation, for one thing, the low pass filter at the end of the D/A conversion doesn't allow those straight lines, the the analog electrical signal at the output will actually be nice curves and not some jagged mess.
 

I'm not sure I follow how a low-pass/high-cut filter affects the straight lines.  If an amplifier intended for a sub-woofer has a low-pass filter, a 60Hz square-wave is still 'square' with 'sharp edges' ...
 
 
Quote:

You assert that analog copy is perfect, that's quite far from the truth:
- the signal is tortured through the phono stage when the huge RIAA equalization is applied
- to make the vinyl master, you actually need to carve a disk with a diamond cutting head, a far from lossless/perfect process, softer details are lost in the cutting process, from an engineering pov, SNR is much lower for LPs than for CDs.
- left and right are never totally separated, crosstalk is abysmal on LPs
- not to mention all the playback issues related to the differences between the center and the exterior of the disk, the stylus erasing details at each playback
- mono bass instead of stereo bass...

 
I think you mean SNR is lower on CD's.
 
24/192 vinyl or cassette tape rips are pretty quizzical...
 
 

 

[kiteki]

Quote:

They don't have straight lines. The D/A process takes the points of a digital wave form and basically draws sine waves between them.

Nah, it draws straight lines between them.
 
That's why for example nintendo music and electronic music sounds so synthetic and artificial, it's all straight lines, triangles and polygon's.
 
 

 

[stv014]

Quote:

 
Nah, it draws straight lines between them.

It does not, unless it is a really poor DAC (worse than newer generation onboard audio). The correct continuous time representation of a single sample of value 1 is sin(x*PI)/(x*PI), where x is the relative time in samples. So it ideally looks something like this:

With some DACs, the ringing is only on the right side, which means they use a minimum phase filter, rather than the mathematically "correct" linear phase as shown above. But it is not just simple straight lines.
 
For comparison, here are some 44.1 kHz samples recorded at 192 kHz from actual sound card DACs. These show both the linear and minimum phase type of lowpass filter. The former is from Realtek onboard audio, and even that is not that far from the "ideal" representation. I also included the spectrum of a 100 Hz pulse train, which shows better how the DAC filter rolls off (note that anything above 22050 Hz is garbage, and should not be there).
 
   
 
   
 
And finally here is how bad the simple linear interpolation would be (you want the response to be flat up to 20 kHz, and have no signal above 22050 Hz):
 

 

 

[khaos974]

Quote:

Quote:

That's a faulty representation, for one thing, the low pass filter at the end of the D/A conversion doesn't allow those straight lines, the the analog electrical signal at the output will actually be nice curves and not some jagged mess.
 

I'm not sure I follow how a low-pass/high-cut filter affects the straight lines.  If an amplifier intended for a sub-woofer has a low-pass filter, a 60Hz square-wave is still 'square' with 'sharp edges' ...
 
Quote:

You assert that analog copy is perfect, that's quite far from the truth:
- the signal is tortured through the phono stage when the huge RIAA equalization is applied
- to make the vinyl master, you actually need to carve a disk with a diamond cutting head, a far from lossless/perfect process, softer details are lost in the cutting process, from an engineering pov, SNR is much lower for LPs than for CDs.
- left and right are never totally separated, crosstalk is abysmal on LPs
- not to mention all the playback issues related to the differences between the center and the exterior of the disk, the stylus erasing details at each playback
- mono bass instead of stereo bass...

 
I think you mean SNR is lower on CD's.
 
24/192 vinyl or cassette tape rips are pretty quizzical...

No, that's exactly what I mean, in the best cases, with absolute perfect playback, SNR for LPs is just above 70 dB, in most cases, it's between 5070 dB while the SNR for a CD player is easily above 90 dB.
Straight lines and sharp edges is high frequency content, to reproduce those, you need higher order harmonics (ie high frequency content), otherwise the edges are rounded, so the low pass filter does really eliminate those jaggies.
 
See how you need higher frequencies to reproduce a square wave.
 

 

 

[khaos974]

Here's an article dedicated to the limits of vinyl and why analog in the form of vinyl is quite far from the 1:1 soundwave/record some supporter pretend it is, you'll find that what's on the LP is quite changed from what the signal originally was.
 
Quote:

[size=x-small]PRODUCING GREAT SOUNDING PHONOGRAPH RECORDS[/size]
[size=x-small](or Why Records Don’t Always Sound Like the Master Tape)[/size]

[size=x-small]BY: KEVIN GRAY 5/3/97[/size]

 

[size=x-small]The phonograph record is a marvelous medium for storing and reproducing sound. With frequency response from 7 Hz to 25kHz and over 75 dB dynamic range possible, it is capable of startling realism. Its ability to convey a sense of space, that is width and depth of sound stage, with a degree of openness and airiness, is unrivaled by anything but the most esoteric digital systems. [/size]

 

[size=x-small]That having been said, it is important to understand the limitations of this medium in order to make great sounding records. The first limitation is recording time and level (volume). The amount of time possible on a record side is entirely dependent on the cutting level (volume) and the amount of low frequency information (bass). Bass uses more space than treble.[/size]

 

[size=x-small]The record groove is an analog of a sound wave. Try to picture looking down on a narrow river or stream. The left bank is the left channel and the right bank is the right channel. Your turntable’s stylus is a wide round raft that is going to travel that river. For simplicity, imagine that the banks stay parallel, (left and right the same) which means the sound is monaural. The louder the sound and or the heavier the bass, the wider the whole river (and your boat) wiggles side to side. The higher the pitch (frequency), the closer together the wiggles get. In other words the sharper the twists and turns, the higher the pitch. Obviously, everything from bass to treble is happening at once, so the gently sweeping wide curves (bass guitar and bass drum) have smaller, more jagged wiggles (vocals, guitars, keyboards, cymbals, percussion etc.), superimposed on them. It should be mentioned here that if the bass information is too loud, your raft gets thrown over the embankment (skips). So now you should be able to see that the louder the music is cut, the wider the groove wiggles, and the less time can fit on the side. Or looking at it the other way around, the longer the side, the less room for wiggles (volume and bass). [/size]

 

[size=x-small]Next limitation: treble. You can put as much treble on a DAT or CD as you want. Unfortunately this is not true on a record (or analog tape for that matter). Although 25kHz response is possible, excessive transients are a problem. There are several reasons for this. It was decided with the advent of the first electrical transcription phonograph record, to reduce bass and boost treble in the cutting of the master record. This reduces bass wiggles and makes treble louder. And we aren’t talking about a little bit of cut and boost here, we’re talking about a 40 dB change from bottom to top! Without the bass cut, you’d only have about 5 minutes on your LP side. Without the treble boost, you would hear mostly surface noise. You don’t have to worry about this drastic cut and boost sounding funny, because the phono preamplifier in your amplifier or receiver has an inverse curve which boosts the bass and reduces the treble by the same amounts used in cutting, so the whole process comes out linear. This was standardized worldwide in 1953 and is called the RIAA record and reproduce curves.[/size]

 

[size=x-small]I said you don’t have to worry about the RIAA curve, but the cutting engineer sure does! Power amplifiers (100 to 400 plus watts) are used to drive the tiny coils (one for each channel) in the cutting head. They’re like miniature speakers which instead of just moving air, push the stylus that etches the groove in your record. With 20 dB of treble boost, you can only imagine the beating that the cutting head takes from cymbal crashes and the like. The coils are helium cooled but still can reach 200 degrees Centigrade. A circuit breaker is used to prevent catastrophic destruction. This doesn’t all add up to the limitation it seems, because it is still possible to cut levels higher than can be played back. [/size]

 

[size=x-small]Let’s take a look at cymbals and vocal sibilance (those loud ‘S’ sounds). "Why", do you ask, "Do they sound OK on the tape but sometimes so awful on the record?" The answer is twofold. First, the problem is aggravated by the high frequency boost we just discussed. Further excessive boost in your mix makes it that much worse. Unlike a cymbal crash in which the impulse is short (the actual hit of the stick on the cymbal), the duration of an ‘S’ is considerably longer, so it is even more pronounced. And second, the worst part: Remember the river? Suppose the river’s twists and turns are actually tighter than your raft? Ever watch a raft attempting rapids? Well, that is exactly what your stylus is doing when it hits a loud cymbal crash or a loud ‘S’ in the record groove. At the instant that the curvature of the groove is tighter than the tip radius of your stylus (raft), it goes over instead of through ‘the rapids’, and you have 100 percent distortion. The higher the frequency and or level, the greater the curvature and distortion. [/size]

 

[size=x-small]The cutting engineer can usually tell if treble peaks are going to ‘break up’ on playback, by the amount of current drawn by the cutting amplifier. This is measured by current meters on the amplifiers. If the current is excessive, the only way to prevent this is to use a very fast-attack treble limiter to reduce the intensity, and therefore, the groove curvature. [/size]

 

[size=x-small]While we’re on the curvature subject, it is necessary to explain one more thing. Ever wonder why outside diameter cuts on a record sound clearer and cleaner than inside ones? Unfortunately it’s a fact. Why? The answer is geometry, curvature again. One turntable revolution at 33 1/3 rpm on an LP takes 1.8 seconds. That 1.8 seconds is spread over a circumference of 36 inches on the outside of the record. At the minimum allowable inside diameter that same 1.8 second revolution would only cover 14.9 inches. You can see from this, that a gentle wiggle spread over 36 inches would get quite ‘scrunched’ over 14.9 inches. A jagged groove at 36 inches would get really scrunched at 14.9 inches (remember the rapids). Excessive treble can even cause the cutting stylus to accelerate so fast that its back edge wipes out what the front edge just cut! It’s unfortunate, but treble rolls off, and distortion goes up as you approach the center of the record. It is quite gradual, but if you compare the source recording to the disc, this actually starts to become noticeable after the second cut or so. Any attempt to compensate for this by boosting the treble, only makes the problem worse (greater curvature remember).[/size]

 

[size=x-small]I’ll discuss stereo very briefly. If the sides of the river don’t stay parallel, it’s stereo. In other words, any difference between the two channels causes the stylus to move up and down in addition to sideways. As the stylus digs deeper, it is using more precious disc space. The moral for engineers is: If you are looking for hot levels or long sides, don’t pan instruments like drums and percussion hard left and right. Keep the bass and bass drum in the center, and keep everything in phase. An out of phase snare or bass drum can wreak havoc. Use an oscilloscope if possible![/size]

 

[size=x-small]All else being equal (bass, volume and depth of cut), by allowing the end of the record to finish farther out from the label, instead of spreading the grooves farther apart to fill all the space, will actually make the record sound better. However, I understand the concept of making the record look ‘full’.[/size]

 

[size=x-small]So much for the primer on record cutting. Now let me give you some additional tips on making your record sound great. First, keep it as short as possible. I know this isn’t always possible, but particularly if hot levels are important, keep it short! How short? As a general rule an LP should be under 20 minutes and 24 minutes maximum. 16 to 18 minutes is ideal. Also, try to balance the side times, preferably within one minute. If one side has to be longer, put more of the quiet material on that side. This will insure even levels. If the sides are long, remember that the more bass, the lower the cutting level (volume). It is possible to squeeze 30 minutes on a side but the level will be so low you’ll have to crank it just to hear it, and you will hear the surface noise! [/size]

 

[size=x-small]A hot club record should be under 12 minutes, 8 to 10 minutes is ideal. Some of the top club DJs tell me they won’t even play records that are over 12 minutes long because they know the levels will be low and don’t want to adjust gain.[/size]

[size=x-small]Watch excessive treble boost in the 8 to 16 kHz range in mixing, you won’t get it back on your record. You can’t break the laws of physics, sorry. A good idea is to check your mix against a record you like with lots of cymbals. If you hear a lot more sizzle on your tape, chances are it won’t make it to the record. Particularly watch those ‘S’s. Use a de’esser on vocals. I don’t do endorsements, but dbx makes a great one. This will give you more overall treble because in cutting your record, the treble limiter won’t be chomping on your cymbals too. [/size]

 

[size=x-small]Put your hottest, brightest most dynamic mixes on the beginning of the disc and they’ll stay that way. If possible keep the quieter material on the inside tracks.[/size]

 
The entire article can be found here: http://www.recordtech.com/prodsounds.htm

 

[kiteki]

Quote:

Quote:

 
Nah, it draws straight lines between them.

It does not, unless it is a really poor DAC (worse than newer generation onboard audio). The correct continuous time representation of a single sample of value 1 is sin(x*PI)/(x*PI), where x is the relative time in samples. So it ideally looks something like this:

With some DACs, the ringing is only on the right side, which means they use a minimum phase filter, rather than the mathematically "correct" linear phase as shown above. But it is not just simple straight lines.
 

Right
 

 
Also this article is relevant - http://www.cirrus.com/en/pubs/whitePaper/DS668WP1.pdf
 

---

 
Is "sin(x*PI)/(x*PI)" the sinc interpolation function / anti-aliasing?
 
If 16/44.1 data requires low-pass filters, anti-aliasing, oversampling/upsampling and a minimum-phase filter in the D/A process to sound natural, it doesn't seem like 16/44.1 is an ideal format.
 
 
 

 

[stv014]

Quote:

Originally Posted by kiteki /img/forum/go_quote.gif
 
Is "sin(x*PI)/(x*PI)" the sinc interpolation function / anti-aliasing?

 
Yes.
 
Quote:

Originally Posted by kiteki /img/forum/go_quote.gif
 
If 16/44.1 data requires low-pass filters, anti-aliasing, oversampling/upsampling and a minimum-phase filter in the D/A process to sound natural, it doesn't seem like 16/44.1 is an ideal format.

 
Well, these are not that hard to implement with currently available hardware. The second DAC chip I have shown on the graphs is available for about $3 when ordered in large quantities.
 

 

[astroid]

Anyone who thinks 44100 samples per second isnt enough to accuratley represent an analog waveform is wrong.

 

[kiteki]

Quote:

 
Well, these are not that hard to implement with currently available hardware. The second DAC chip I have shown on the graphs is available for about $3 when ordered in large quantities.
 

Do you have any minimum-phase DAC's to recommend?  Thanks for your posts they are helpful.

Quote:

Anyone who thinks 44100 samples per second isnt enough to accuratley represent an analog waveform is wrong.

It's not enough, if you need low-pass filters, anti-aliasing, oversampling/upsampling etc. to achieve a satisfactory result.
 
It's like this,
 
 -->  -->

 

 

[astroid]

Its not like that

 

[MaciekN]

Nyquist- Shannon theorem, does it ring a bell? For reference: http://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem
Higher sampling rates allow for less ringy filters, and that's all as far as I know.E
 
EDIT: kiteki, low pass filtering, anti-aliasing, over/upsampling are all recommended for hi rez formats just as well as for 44.1/16, although, they are not obligatory. Some poeple like filterless DACs even though they measure worse.

 

[miceblue]

This is an interesting thread; I've been wanting to know if there is a difference between vinyl and CD's as well.

I don't know if I can contribute much to this thread, but I remember learning about the Nyquist theorem for signal sampling (at least 2X the maximum frequency; so I think a standard CD's 44.1 kHz would be adequate enough) and using the Fourier series to approximate a signal as a big 'ol continuous sinusoidal function....that was a very complicated and mind-boggling class that I took.

 

[mikeaj]

Quote:

This is an interesting thread; I've been wanting to know if there is a difference between vinyl and CD's as well.
I don't know if I can contribute much to this thread, but I remember learning about the Nyquist theorem for signal sampling (at least 2X the maximum frequency; so I think a standard CD's 44.1 kHz would be adequate enough) and using the Fourier series to approximate a signal as a big 'ol continuous sinusoidal function....that was a very complicated and mind-boggling class that I took.

 
For a number of practical considerations, there are some advantages to a higher sampling rate.  For 20 kHz reproduction, you definitely don't want the sampling rate to be the Nyquist 40 kHz.  Maybe CD audio should have just been 48 kHz too, to make things easier and because some people hear a bit above 20 kHz, but 44.1 kHz is not a problem.
 
Also, Fourier series does even better than just approximate--in most situations of interest, including for the real-world kind of signals we're looking at, there is convergence (in some sense).
 
I was going to briefly touch on some of those practical considerations and the conditions for convergence for Fourier series, but they're of course both of wikipedia:
http://en.wikipedia.org/wiki/Nyquist-Shannon#Practical_considerations
http://en.wikipedia.org/wiki/Convergence_of_Fourier_series

 

[kiteki]

 
So I've been reading some papers and the Meyer & Moran one costs $20 what's up with that - https://secure.aes.org/forum/pubs/journal/?ID=2
 
Anyway... in the introduction it says "The noise of the CD-quality loop was audible only at very elevated levels."
 
So, they are saying, that even with very sophisticated A/D/A equipment (HHB CDR850 Professional), the duplicate CD had audible noise to the participants?
 
Note, they made a duplicate CD for testing purposes, this is not the same as A/B/X'ing whether there is an A/D/A in the listening chain or not, as far as the theoretical total transparency of a DAC is concerned, that should be self-evident.