To that end, I'm of the opinion that the SB16 was, and is, better suited to Windows than DOS.
I guess that exactly is the nail in the SB16's coffin:
There's only one reason why you'd buy a Sound Blaster: 100% hardware compatibility. The SB16 could not provide that. There were various clones that delivered similar compatibility, at reduced cost and/or better quality.
In Windows, hardware compatibility was no longer an issue. Windows uses a hardware abstraction layer (HAL), which basically means that as long as you have drivers, the exact hardware is not relevant. In Windows, an SB16 is not any more or less compatible than any other OPL3 + 16-bit CODEC card.
And the SB16 had nothing going for it really... You could get cheaper cards with similar quality and features. Cards at the same pricepoint as an SB16 would generally give considerably better quality.
Creative should have nailed the SB16 with perfect SB compatibility and best-in-class 16-bit audio, but they failed on both counts. In practice, many people preferred the SB Pro 2, because it at least got the compatibility right.
This is precisely what older DACs do. Like, exactly. Look at the schematic for the MT-32. It has a one-channel DAC and IIRC 6 channels of audio. The DAC outputs a fixed voltage via an analog switch into a S&H latch, which holds that voltage until the DAC comes back around to service that channel again. It's been a while since I've looked at this circuit so I don't remember the exact DAC model number, but it was super common in the late 80s, early 90s. Roland used it, or similar, in all their synths. Some highish-end CD players used it, etc., and that's just the ones I know about.
I think you're missing my point:
These analog components do not have an infinitely fast transition time, either deliberately, or because of physical limitations.
They tend to 'sag' a bit, smoothing out the waveform.
And the SB16 had nothing going for it really... You could get cheaper cards with similar quality and features.
Agreed, although it can be argued (albeit, mostly by today's standards) that the ASP/CSP functionality and integration within Windows is a decent feature-add.
In late 1992, the Sound Blaster 16 was probably the most feature-laden card of its type.
Cards at the same pricepoint as an SB16 would generally give considerably better quality.
Not as far as the available comparisons have yet shown. That's sort-of the main contention for this entire thread - while not best-in-class, the Sound Blaster 16 isn't nearly as bad a 16-bit performer as it's often made out to be.
Not as far as the available comparisons have yet shown. That's sort-of the main contention for this entire thread - while not best-in-class, the Sound Blaster 16 isn't nearly as bad a 16-bit performer as it's often made out to be.
Well, SNR alone doesn't tell the whole story.
While I do not have an actual SB16 in my collection, I can draw a parallel with the AWE32 that I do have: its frequency response is all over the place.
What I would consider an interesting comparsion would be the following:
Rip a PCM track from a CD to act as a reference recording. Perhaps one of the 'standards' from the audiophile-world, such as Enya or some Toto or such. Or Michael Jackson's Bad album perhaps...
Then take a (semi-)professional audio device to record various 16-bit soundcards playing back that PCM track (would be great if you could record it in 24/192).
Then offer the original ripped track, plus the sound card recordings, so everyone can play the whole set back on their own device, for a comparison.
My experience is that my GUS MAX plays back a PCM track pretty much the same as my CD player does, very neutral. The AWE32 sounds like some kid has been playing with the eq to boost the bass and treble, and scoop the midrange.
Likewise, when I play tracker music on the GUS, it sounds the way I expect it to. Playing it on the AWE32... mutilates it.
I mean, for my personal listening experience, I don't mind a bit of noise in the background, when I get a rich and natural reproduction of the sound. But when the midrange is missing, musical detail gets lost, and that I do mind.
Last edited by Scali on 2019-05-13, 20:21. Edited 1 time in total.
Scali wrote:Perhaps you need to look at the videos again.
The waveform is sampled, which indeed is quantized in time and dynamic range. Howe […] Show full quote
SirNickity wrote:
The waveform is stair-cased. Digital audio is quantized in the X and Y axis, so I don't even understand how anyone can fathom it as anything else. The lower the resolution on either axis, the more stepping occurs.
Perhaps you need to look at the videos again.
The waveform is sampled, which indeed is quantized in time and dynamic range. However, samples are just points, as in 'infinitely short' in terms of time.
The waveform would only get 'stairstepped' if you stretch out each sample in time in the horizontal direction. In practice this is never the case, not even with the cheapest of DACs.
This is precisely what older DACs do. Like, exactly. Look at the schematic for the MT-32. It has a one-channel DAC and IIRC 6 channels of audio. The DAC outputs a fixed voltage via an analog switch into a S&H latch, which holds that voltage until the DAC comes back around to service that channel again. It's been a while since I've looked at this circuit so I don't remember the exact DAC model number, but it was super common in the late 80s, early 90s. Roland used it, or similar, in all their synths. Some highish-end CD players used it, etc., and that's just the ones I know about.
Then there's the Covox and DSS. Heck even the SB with its microcontroller-based DSPs did this. 😀 I know those aren't representative of mass consumer audio, but my point is, it's way more common than you would think. Not anymore, since successive approximation and similar techniques are cheap and easy now.
Thank you for your clear and informative explanation. 😀
AFAIK, this and similar issues are also the main reason software samplers cannot approximate the output of vintage hardware samplers such as Akai's cult machines, as they lack the old ADCs and DACs present in those instruments.
On top of that, the samples found in many older romplers (and most wavetable sound cards!) are not merely digitized waveforms, but resynthesized using specific algorithms to be able to fit them into as little memory as possible, which often introduces additional artifacting. Many of those waveforms are therefore not 'clean' or 'smooth', and some will show rather interesting phenomenons when visualised using a scope due to the tricks employed by the manufacturer to keep them sounding good (see Roland JV-1080).
Right. So you don't want to back up your points with numbers, yet complain when I do?
Well, first of all, as someone else already pointed out, your own prices prove that the SBs are not exactly the most affordable options.
Secondly, I'm not sure why you are throwing in random Yamaha MIDI devices from 1996, with no relation to PCs, gaming, or even the Sound Canvas (or the era we're discussing).
Lastly, you are actually going to stick to the point that you don't know what an Amiga 500/600 cost in the early 90s?
Really, look it up yourself. Even if I were to give you a link, you'd continue arguing.
It becomes a semantic argument "whether they exist at all, then" I suppose
not really, because:
SirNickity wrote:
But you have to understand, this is because the stepping has been filtered out
so it doesnt exist. Every time you see someone advertising/mentioning stair steps it is a straight up SCAM, using a neat sound bite with a small grain of truth in theory, same with audiomorons blabbing about jitter in CD players and trying to sell you $300 "crystal module". Creative was quite a scammy company.
SirNickity wrote:
Then there's the Covox and DSS.
both shipped with proper low pass filters and do not produce stairstep signal
They absolutely do. There's wiggle room in this argument either way depending on whether you consider the reconstruction filter as an integral and immutable part of the system. On modern DACs, sure. There's no point in the accessible signal where jaggies exist. In very low end, or just older tech DACs, they do.
And no, there is no technical limitation that prevents the jump between one audio sample and the next from existing at a discrete level. Electronics do have slew rate limitations, but for a transistor, 1/44100th of a second is an eternity, and 1/65536th of 5V is 0.07mV, which is not a lot but probably still above the resolution limitations of a decent transistor. (Noise from external systems notwithstanding.) The only thing that is going to affect the ability of a DAC to hold a sample at a discrete level, like an ideal voltage source, is a low pass filter.
For a fixed-rate device like a CD player, you may as well assume the reconstruction filter is part of the DAC. Fine. On particular DACs, jaggies still exist in the system, in the analog domain, but not at the output of the system. Semantics. I'll let that lie.
But on devices - like early SBs with fixed filters and continuously variable sample rates - it is a thing. To what degree depends on where you measure, and I'll even go so far as to say it's more or less negligible. You will either have aliasing or crush legitimate highs, or both with a shallow filter with a low -3. But a blanket statement that usually accompanies the same single video held up as proof and evidence is not a very strong argument. It's a technical fallacy that, admittedly, doesn't mean much. I just hate to see it used as a "gotcha!" argument against anyone who mentions stair-steps. It's just not as true as some people think it is. It's like the "distortion kills speakers" fallacy. It raises the RMS level of high frequencies, which can lead to thermal overload, but it isn't a direct consequence.
But on devices - like early SBs with fixed filters and continuously variable sample rates - it is a thing. To what degree depends on where you measure, and I'll even go so far as to say it's more or less negligible.
You will either have aliasing or crush legitimate highs, or both with a shallow filter with a low -3.
But a blanket statement that usually accompanies the same single video held up as proof and evidence is not a very strong argument. It's a technical fallacy that, admittedly, doesn't mean much.
You are everything thats wrong with audiomoron movement 😉 Finding some minor tidbit that is _not_ the IRL general case, but still kinda sorta true on a very narrow theoretical level, and then going around convincing people sky is not blue, nothing worse than nanosecond jitter in audio signal, and only quality speaker cables will give you good sound! This is what Creative is guilty of here while lying about source of shit sound out of their noisy cards:
SB16CQA.TXT
14. GENERAL 2 3 4.1 ELIMINATING UNWANTED NOISES 4 Q2. I just upgraded from SBPro to the SB16 and now some of my games 5 sound hissy. Do I need to return the SB16 for replacement? 6 7 A2. The high quality 16-bit CODEC plays back every detail of the 8 8-bit sample including the coarse resolution of the staircase 9 waveform. Hearing the "hissing sound" is like seeing all the 10 sharp edges of a 320x240 picture on a very sharp 1280x1024 11 monitor. SB16 has a Treble control to suit your needs on 12 different sound quality. To eliminate the unwanted "hissing 13 sound", you can set the Treble level to zero.
This answer, while technically correct, insinuates the problem is with the source material, and not analog audio section of your card sounding like this: https://www.youtube.com/watch?v=GdcVedXM88s
Well, I think the main point is that some people try to argue about what 'DAC output' is, or where exactly one should measure.
Clearly we are talking about sound cards, and the only reasonable point where you would measure is at the output, because that's the only point in the signal chain that end-users would actually care about. So you would measure the entire signal chain. The DAC does not operate in a vacuum.
Reminds me a bit of guitar amps. If you were to judge a guitar amp as a regular hi-fi amp, it would be considered ridiculously poor, because its frequency response and dynamic range are laughable.
However, combine it with an electric guitar, and things start to make sense: the electric guitar has very crude magnetic pickups, and the resulting sound is not exactly pretty or useful, if you would send it directly to a hi-fi amp.
A guitar amp however, has a very specific impedance, which matches well with the passive circuit that is the magnetic pickup. And its 'poor dynamic range' is actually a very nice form of natural compression, which means that the volume of your guitar is not 'all over the place', but you can actually play with all sorts of nuances, and not worry about your sound not being heard. The nuances are preserved, and the volume is 'flattened out'. Likewise, you can push this compression too far, and get distortion, which actually sounds very ncie and musically useful.
Also, the 'weird' frequency response works like a charm to make your guitar sound balanced over the entire range of the instrument.
So it's the entire signal chain that makes the sound, not just the guitar, not just the pre-amp, not just the power-amp, and not just the speaker.
In fact, in the studio, these amps are actually recorded with microphones (yes, plural). So in that sense, even the microphones (and their position relative to the speaker) are part of the signal chain. Even the natural ambience of the room is.
That's how I think sound cards should be discussed: As a whole.
Scali wrote:What I would consider an interesting comparsion would be the following:
Rip a PCM track from a CD to act as a reference recordin […] Show full quote
What I would consider an interesting comparsion would be the following:
Rip a PCM track from a CD to act as a reference recording. Perhaps one of the 'standards' from the audiophile-world, such as Enya or some Toto or such. Or Michael Jackson's Bad album perhaps...
Then take a (semi-)professional audio device to record various 16-bit soundcards playing back that PCM track (would be great if you could record it in 24/192).
Then offer the original ripped track, plus the sound card recordings, so everyone can play the whole set back on their own device, for a comparison.
I'm planning to use the RightMark Audio Analyzer to capture a set of measurements that will hopefully complement the available magazine findings. With that information, I'm also hoping to produce a "best" configuration for a given card. I like your idea of providing a comparison of recorded music playback as well, where ideal measurements may not equate to ideal listenability, and vice-versa. The main obstacle here is just a lack of time...
The AWE32 sounds like some kid has been playing with the eq to boost the bass and treble, and scoop the midrange.
Yep. This is largely correctable though. Consider the following quick measurement comparison from my CT3980:
The difference in frequency response is the result of default settings of 160 for both the Bass and Treble (white), versus modified values of 128 for both (green), as well as a slight volume boost to the Master and Voice mixer channels (224 -> 236). Do note the other stats too, which (concerning the CT3980_2 results) are a consequence of the following overall mixer settings, and use of the line output:
Enabling additional mixer inputs/channels has a negative impact that isn't reflected in the above measurements, of course, but that shouldn't be surprising.
It absolutely doesn't have to sound like that though; the mixer configuration in that video is pretty far from optimal. Noise related to the mixer stage isn't just a Creative or SB16 problem either.
Do you have any practical experience with the SB16 or its contemporaries? I'm curious to know what is driving the negative comparison and commentary, because it's coming across as little more than an anti-Creative bent.
Yeah, and that's fine. I don't question that. Look -- I know the discussion here is mostly relevant only to the SB16. I know your reply... this one:
rasz_pl wrote:
used car salesman snakeoil. Neo, there are no stair steps.
... is in reference to the ... uh .. dubious FAQ entry posted earlier. I haven't scoped an SB16 playing an 8-bit waveform. I don't know whether there's any truth to that statement at all. On a traditional DAC, that is clocked to some multiple of either/or 44.1kHz or 48kHz, and has a sliding reconstruction filter (assuming it's not locked to one frequency), it would be mostly nonsense. "Mostly" only because you aren't going to get more than 8-bit resolution out of a 16-bit DAC playing an 8-bit wave, so there will be granularity issues with the resulting signal. But... not really any more or less than an 8-bit DAC, so the argument still fails as a reason why the SB16 would sound worse than an SB Pro with the same input. The only variable is quantization noise, which I feel is probably outside the scope of their answer, and probably not the cause of discrepancy anyway.
So what am I carrying on about? The headstrong notion that "stair-steps are a myth." They're just not. At least not always. That's not an opinion, it's fact, and I have a screen-shot from my scope (... somwhere) to prove it. 😀 I re-watched the Xiph video just now because I remember it being designed to prove a point -- I just didn't remember what. The point he was trying to make is that your end-product of digital audio, even at a "measly" 16-bit 44kHz, will not be plagued with jaggies. So you don't need to be so concerned with HD resolution downloads for music, nor terribly concerned about the DAC you use. And that is absolutely true.
(Although the supporting analog circuitry is probably worth some scrutiny. I took issue with his statement that passing a signal through an aliasing filter more than once won't band-limit the signal more. It definitely will. The DC-blocking cap is not infinitely large, and the aliasing filter is not infinitely high, so there will be a cumulative effect. Enough to care about? Depends on the -3dB points of the band-limiting. For the purposes of his argument, "no" will suffice most of the time.)
However, even he states -- counter to his argument, in a way -- that simple DACs that use a zero-order hold topology will output those "point in time samples" as a continuous voltage. This is exactly what I was talking about with the Burr-Brown DACs from the 80s and 90s that used a sample-and-hold IC to allow a single expensive DAC to multiplex several channels of audio. These old 90s sound cards worked the same way. It wasn't until DSP got cheap and easy enough to surpass ladder-DAC topologies that that was no longer the norm. Further, I highly doubt even the SB16 uses dithering to reduce the quantization noise on the Y axis.
This may not convince anyone of any different ideas than they had two pages ago. And, well, so be it. I can't imagine I have much to add that I haven't said yet. It does get to a point of semantics, like Scali said, of where you measure. Xiph guy is absolutely 100% correct in all his assertions. It's just that people with only a vague understanding of digital audio take his point and shove it down people's throats without understanding how and why it is both true and false. Not that I'm an expert -- he could walk circles around me in DSP theory, I'm sure. But I do understand enough to know his results require qualified assumptions before they can be taken as fact. He knows that as well, and even said at the end of the video that he took liberties.
That is all. 😎
EDIT: Oh!, actually I found the scope screen-shot I took while measuring the MT-32.
This is directly after the S&H stage, before the reconstruction filter. Kinda looks like stair-steps to me. So they don't exist AT ALL, hm? (Remember, Xiph guy said at one point "not even the digital signal". Which is actually true -- the digital samples are infinitely small points in time. It's the "not even" part that leads to confusion, I think.)
I'd be interested if those voltage deltas translate through a power amplification stage. High frequency elements like that could really cook out your tweeter drivers.
Some of that is lost in cable capacitance (at line stages) and various parts in a typical amplifier. Most amplifers intentionally limit the bandwidth to some extent and gain at the final stage is much lower for ultrasonics than audible range so that more bandwidth is left for where it matters to work against non linearities present. Luckily energy of the ultrasonic stuff is fairly low and at normal listening levels (with typical speakers it isn't even 1W range) there shouldn't be any danger. I'm not sure how damaging it can be to the high freq elements at louder volumes when the levels reach tens of W or even beyond (I'm in no hurry to test hahahaha).
I would anticipate filtration by the input stage helping a lot, but that sort of output would have me checking. Usually tweeter death is something you see with clipping.
First, the output on that scope IS NOT indicative of what you see on the line outputs. The output chain is: (1) DAC -> (2) Analog Switch -> (3) Sample-and-Hold latch -> (4) Lowpass filter -> (5) output. The scope sample was taken between (3) and (4). The reason it looks like this is because the DAC outputs the voltage for Main Left out through the analog switch. S&H #1 then latches that voltage for the 1/32000th period (IIRC, the MT-32 is 32kHz.) In the meantime, the DAC then outputs the voltage for Main Right out and the analog switch directs that to S&H #2. Then so on for S&H #3, #4, etc.
(It's been a while, but I think there was a reverb return that is mixed in analog, or something like that -- hence the multiple channels. I want to say there were six channels, but I think I could be mixing that up with the D-110 or something else.)
The S&H then goes through a lowpass filter (reconstruction filter) which removes those high-frequency components. The MT-32 uses a 4th-order filter, I think. Then you go to the output.. maybe through another buffer, or the volume control or something like that.
OK, and second... yes, amplifiers will limit the bandwidth. For one, amplifiers have limited slew-rate, so they can't produce a true square wave anyway. If you try, the output of the amp will be delayed slightly from the input, and the feedback loop will try to compensate for this until the phase errors cause the amp to start oscillating uncontrollably. So, a band-limiting filter is pretty much mandatory. (Some ICs have a band filter integrated though, so you can use a negative feedback with no added impedance.)
