EMUSIC-L Digest Volume 45, Issue 1 This issue's topics: 8-bit vs. 16-bit (16 messages) Your EMUSIC-L Digest moderator is Joe McMahon . You may subscribe to EMUSIC-L by sending mail to listserv@american.edu with the line "SUB EMUSIC-L your name" as the text. The EMUSIC-L archive is a service of SunSite (sunsite.unc.edu) at the University of North Carolina. ------------------------------------------------------------------------ Date: Sun, 4 Oct 1992 13:46:34 -0400 From: Dana Nibby Subject: 8bit vs 16bit? Can anyone tell me what the difference is in sound quality between 8bit and 16bit? I understand about sampling rate, and that if the rate is too low, aliasing/clipping occurs. I've asked this question before and gotten no respon. People tend to say, "Well, 16bit sounds better." Gee, well, in what respect? Is 8-bit sound tinny, whereas 16-bit sound is warmer? For example, does an Ensoniq ESQ-1 sound tinny/weak because it has an 8-bit processor? How does 8-bit/16 bit sound compare to what you hear on vinyl? I mean, what are the fequency limits of vinyl? And if anything much past 8-bit is useless for vinyl... In other words, is 8-bit sound good enough for vinyl records? Can a comparison even be made here? Can *someone* out there help me understand the difference between 8-bit sound and 16-bit sound? Is 8-bit sound 'fuzzy', etc. etc. Thanks, Dana ------------------------------ Date: Sun, 4 Oct 1992 16:36:28 -0500 From: William Flanagan Subject: Re: 8bit vs 16bit? Dana- go to the local computer store, and listen to the difference. 8-bit is much less clear than 16-bit samples. If you like, liken it to an 8-bit GIF and a 16-bit GIF. What's the difference there? Well, one looks a whole lot better (more colors). The same goes for samplers (more sound). William ------------------------------ Date: Sun, 4 Oct 1992 21:58:09 EDT From: Keith Hall Subject: Re: 8bit vs 16bit? > > Can anyone tell me what the difference is in sound quality between 8bit > and 16bit? I understand about sampling rate, and that if the rate is too low, > aliasing/clipping occurs. I've asked this question before and gotten no respon. First, quality of sound (good, bad) is usually a very personal thing. Sure a synth puts out a noisey hissy sound it could be considered bad, but not everybody hates that (especially if you are interested in the technique involved in creating Techno music). As William Flanagan explains the technical difference is the density of the digitized material. As you know, a sample which has a sampleing rate of say 40k (for numerical conveinience) is simply sampling a signal 40 thousand times a second. This actual sampleing is convertind the amplitude of the wave, which in electerical terms is simply the amount of electricity moveing through the cable hooked into the input of the sampler. Each sample is a number which represents a relative amplitude. If you only have 256 possible values, then man different voltages will be represented as the same value, meaning that the definition of a wave will be limited to the larger amplitude difference and not the smaller. All this means that lower amplitude signals (high frequency is usually at a lower amplitude) will not have as much definition in the wave. This usually means that you will never get the truly bright sound of a crash symbol of a high hat. But hey, I have settled for 12-bit for a year now and with a 32k sampling rate. But its a Roland S-550 which is a truly incredible machine(an earlier version of the 770 which should be on everybodies wish list). If you are really interested in Techno musich I would advise working with a fullbodied sampler an not use the computer, it will make your life a thousand times easier and will amaze you with the sound quality difference. Keith Hall khall@s850.mwc.edu ------------------------------ Date: Sun, 4 Oct 1992 22:50:00 LCL From: DOWRJ@VAX1.COMPUTER-CENTRE.BIRMINGHAM.AC.UK Subject: 8-bit vs. 16-bit It is probably easier to think in terms of graphics when it comes to the problems of 16-bit v's 8-bit sound. When there are only 256 values being represented (8-bit) everything is a bit grainy, the more bits, the less things seem 'digitized'. 16-bits gives you 65536 different values and so changes are less abrupt. If one thinks of a picture represented by 1 bit i.e. there is only black *or* white (no grey scale) there is really very little detail there, one only gets an outline. The more steps there are the more subtly one sees, more detail. In sound one 'hears' as it where more detail. Ok the anaolgy kind of breaks down there, but if you think of a pure sound say a sine wave (or a high flute if you like) This would normally be a continuous curve. Digitizing makes it stepped. The more bits, the smaller the steps. Well in one direction (amplitude). sampling rate steps in the time domain. So... What one hears is actually a kind of distortion in reality. In one bit sampling things would either be on or off, so a sine wave would become square. most distortion is about curves becoming square. In the worst cases 8-bit sounds like a cassette recorder being driven too hard! let's face it - 8-bit is crap ------------------------------ Date: Mon, 5 Oct 1992 07:30:54 EDT From: "(Derek Burney)" Subject: Re: 8 bit vs. 16 bit Just a quick addition to the question of sampling quality. Although the number of bits used to represent the data matters a great deal, the rate matters as much, if not more. You always need to sample at a rate twice that of the highest frequency you want to hear. That's because the sampling algorithms require two sample points before they can determine the frequency, so if you want to get a note that is 20k for instance, you need to be sampling at least at 40k (again for numerical convenience) for the algorithm to find the frequency. One more thing... One reader mentioned that "sound quality" is subjective since noise to one person may be preferable to someone else. I beg to differ. While an individual may prefer to have noise in a particular sample, it's still a sample of lesser sound quality than a cleaner one. All other parameters being equal, 16 bits will always return a better sample than 8. Cheers, Derek Burney ------------------------------ Date: Mon, 5 Oct 1992 09:49:00 EDT From: Mario Vergona Subject: 8/16 bit hi, I do not care particularly about sine waves of the co-sine of the tangent of the pi root definitions of the sample rate: The bottom line is this I bought a synth which sounded very good to me and was 8 bit sampled. After a short time I bagan to notice things, the piano, amoung other sounds, had a static/break up in certain registers, the recording of cerain things didn't sound as good as I wanted them to be. I eventually traded that synth in (a K1) because as time went on the sound bothered me more and more. I got the K4, which is 16 bit, and there is a real great improvement in the sound quality. It is true that you can cover up these 8 bit sample break ups with reverb, as Kawai must have realized and now includes reverb on the K1-II, but 16 bit sound makes a difference in dry applications. m ------------------------------ Date: Mon, 5 Oct 1992 11:12:00 EDT From: wbf@ALUX2.ATT.COM Subject: 8 vs. 16 bits For those who are interested, the dynamic range of a digital system may be expressed by: n DR in dB = 20 log ( 2 - 1 ) where n is the number of bits. Therefore, a 16 bit system has a theoretical limit to its dynamic range of approximately 96.3 dB and an 8 bit system's limit will be about 48.1 dB. As you can see, a violin playing a dynamic of mf or p will have some overtones being more than 48 dB below the digital system's maximum output volume. In the 16 bit system, these overtones are captured where the 8 bit system looses it altogether. Even a "bad" analog system would reproduce the overtones better than the 8 bit digital system. The sound may be in the noise floor, but, like listening to an old 78, the ear has an amazing ability to hear through the noise at least a little bit. (No pun intended, but I'll take what I can get!) Bill ------------------------------ Date: Mon, 5 Oct 1992 12:37:23 +0100 From: Martin Roth Subject: >Ok the anaolgy kind of breaks down there, but if you think of a pure sound >say a sine wave (or a high flute if you like) This would normally be a >continuous curve. Digitizing makes it stepped. The more bits, the smaller >the steps. Well in one direction (amplitude). sampling rate steps in the >time domain. So... >What one hears is actually a kind of distortion in reality. In one bit >sampling things would either be on or off, so a sine wave would become >square. most distortion is about curves becoming square. In the worst >cases 8-bit sounds like a cassette recorder being driven too hard! >let's face it - 8-bit is crap Well, the question about 8 vs. 16 bit interested me, but so far I have not seen any good aswers! The sine-wave is a good exaple, by digitizing it you get no longer one sine wave but a step-like wave with the rough apperance of a sine. Sure 16 bits steps are a better approximation to the original, but WHY? Sampling Therem states that disortions caused by the sampling in discrete time steps causes only disortions higher in frequency than half the sampling frequence. So if you add a low-pass after the DAC (what is done allways), you have NO distortion. If I think about it, I believe that Sampling Theorem is true for discrete-step sampled continuous signals, but assumes continous values at each step. Is there a sort of extension to reduce the single sample steps to a finite number of levels? How does that affect frequency response? To illustrate my questions, imagine a 2-level digitised sound at an extremely high sampling rate. I'm sure it can not reproduce anything like music, but why? How can the Sampling Theorem be extended to take into account the discrete signal levels? -Martin _______________________________________________________________________ _ Martin Roth Martin Roth ETHZ, ips, RZ F16 |\ /|_) Mail: roth@ips.id.ethz.ch Sandacker 14 g 01/256 55 68 | \/ | \ CH-8154 Oberglatt p 01/850 32 75 (Eng. Comp. Sci. ETH) Switzerland (F-)emails welcome! ----------------------------------------------------------------------- ------------------------------ Date: Mon, 5 Oct 1992 12:39:41 +0100 From: Martin Roth Subject: Re: 8/16 bit >Ok the anaolgy kind of breaks down there, but if you think of a pure sound >say a sine wave (or a high flute if you like) This would normally be a >continuous curve. Digitizing makes it stepped. The more bits, the smaller >the steps. Well in one direction (amplitude). sampling rate steps in the >time domain. So... >What one hears is actually a kind of distortion in reality. In one bit >sampling things would either be on or off, so a sine wave would become >square. most distortion is about curves becoming square. In the worst >cases 8-bit sounds like a cassette recorder being driven too hard! >let's face it - 8-bit is crap Well, the question about 8 vs. 16 bit interested me, but so far I have not seen any good aswers! The sine-wave is a good exaple, by digitizing it you get no longer one sine wave but a step-like wave with the rough apperance of a sine. Sure 16 bits steps are a better approximation to the original, but WHY? Sampling Therem states that disortions caused by the sampling in discrete time steps causes only disortions higher in frequency than half the sampling frequence. So if you add a low-pass after the DAC (what is done allways), you have NO distortion. If I think about it, I believe that Sampling Theorem is true for discrete-step sampled continuous signals, but assumes continous values at each step. Is there a sort of extension to reduce the single sample steps to a finite number of levels? How does that affect frequency response? To illustrate my questions, imagine a 2-level digitised sound at an extremely high sampling rate. I'm sure it can not reproduce anything like music, but why? How can the Sampling Theorem be extended to take into account the discrete signal levels? - -Martin _______________________________________________________________________ _ Martin Roth Martin Roth ETHZ, ips, RZ F16 |\ /|_) Mail: roth@ips.id.ethz.ch Sandacker 14 g 01/256 55 68 | \/ | \ CH-8154 Oberglatt p 01/850 32 75 (Eng. Comp. Sci. ETH) Switzerland (F-)emails welcome! - ----------------------------------------------------------------------- ------------------------------ Date: Mon, 5 Oct 1992 13:33:48 +0100 From: "(Adam MIROWSKI)" Subject: > I believe that Sampling Theorem is true for discrete-step > sampled continuous signals, but assumes continous values at each step. > Is there a sort of extension to reduce the single sample steps to a finite > number of levels? How does that affect frequency response? These are linear systems, so the initial-quantization error propagates simply as an additional signal. I don't think it affects the frequency response. > To illustrate my questions, imagine a 2-level digitised sound at an extremely > high sampling rate. Easy. It is right in front of me. > I'm sure it can not reproduce anything like music, but why? So this George Duke CD is not music. OK. Your opinion. :-) > How can the Sampling Theorem be extended to take into account the > discrete signal levels? By lowering the cutoff frequency of the reconstruction filter. ------------------------------ Date: Mon, 5 Oct 1992 14:12:14 GMT From: Martin Rootes Subject: Re: 8/16 bit Martin Roth writes:- > ............................................................. Sure 16 bits > steps are a better approximation to the original, but WHY? Sampling Therem > states that disortions caused by the sampling in discrete time steps causes > only disortions higher in frequency than half the sampling frequence. So if > you add a low-pass after the DAC (what is done allways), you have NO distortion. > > If I think about it, I believe that Sampling Theorem is true for discrete-step > sampled continuous signals, but assumes continous values at each step. > Is there a sort of extension to reduce the single sample steps to a finite > number of levels? How does that affect frequency response? The sampling theorem you refer to only concerns itself with the highest frequency that can be represented and the phenomona of quantisation noise. A low pass filter is an absolute necessity on the output of a DAC, otherwise quantisation noise would be present (which is higher than half the sampling frequency). The number of bits in a sample does not affect the frequency response but the Signal to Noise ratio. The S/N ratio is affected because if a system has less values to respresent the original analog system errors will creep in, which will be perceived as noise. If you have an 8 bit system the level of a signal at any point can only be represented by 1 of 256 values, any signal level which falls anywhere between two of these values (a very common event) will therefore be incorrectly represented. This discrepency is effectively noise, and is 1/256 of the maximum signal level for an 8bit system. Offhand I can't remember what this is in term of db, but it is very bad and gets even worse for complex waves. 16bit systems have 65536 discrete levels, and a far better S/N ratio (over 100db). For a thorough explanation of this and other related matters see Hal Chamberlin's Musical Applications of Microprocessors. > To illustrate my questions, imagine a 2-level digitised sound at an extremely > high sampling rate. I'm sure it can not reproduce anything like music, but > why? How can the Sampling Theorem be extended to take into account the > discrete signal levels? See above. - Martin. Martin Rootes - Senior Systems Programmer/Analyst, Sheffield Hallam University Email : M.Rootes@scp.ac.uk Disclaimer: Sheffield Hallam University has no opinions, the ones above are mine. ------------------------------------------------------------------------------ ------------------------------ Date: Tue, 6 Oct 1992 20:06:00 +0000 From: Nick Rothwell Subject: 8-bit Ratners >It is probably easier to think in terms of graphics when it comes to the >problems of 16-bit v's 8-bit sound. When there are only 256 values being >represented (8-bit) everything is a bit grainy, the more bits, the less >things seem 'digitized' This is not necessarily a bad thing. 8-bit Grunge is good in places. Listen to the PPG. >let's face it - 8-bit is crap Well, it works for me. Nick. ------------------------------ Date: Sat, 10 Oct 1992 17:16:20 MDT From: "(Adam Schabtach)" Subject: Re: 8-bit Ratners >This is not necessarily a bad thing. 8-bit Grunge is good in places. Listen >to the PPG. Or the Fairlight Series II. Or the Emax. >>let's face it - 8-bit is crap > >Well, it works for me. And Tangerine Dream, Peter Gabriel, Kate Bush, Thomas Dolby, etc. etc... I.e. just about anyone who used the early Fairlights, samplers, and digital/analog hybrid synthesizers. --Adam ------------------------------ Date: Wed, 7 Oct 1992 23:03:36 -0500 From: William Flanagan Subject: Re: 8-bit Ratners Yes... it does. But, 16-bit can simulate the grunge of 8-bit with the addition of white noise. 8-bit can't begin to compete with the crystal-clear sound of 8-bit. But what are you guys arguing about? We all know this. 8-bit and are different price range tidbits of the same game... So why argue???? :) William ------------------------------ Date: Thu, 8 Oct 1992 09:34:21 EDT From: ronin Subject: Re: 8-bit Ratners actually, adding white noise to the 16 bit signal only gives you 16 bit noise. think of it as really big dither. the amplitude will still vary with 16 bit resolution. to make 8 bit sound with 16 bit works, you gotta clip it, either directly (overdrive) or indirectly (algorithmic bit reduction). i haven't actually seen much (any?) 16 bit gear that makes this easy. thank you to whoever it was that got here first with the decsription of bitwidth as a determinant of dynamic range. people really do need to be clear about which parameters of sound are affected by which factors, and there does seem to be some confusion around here. to flog the nag just a little, let me point out (reiterate?) two things: 1) yes, exactly... the sampling theorem is a statement of frequency issues. a 1 bit system, sampled at sufficiently high rate, gives acceptable speech intelligibility. 2) but frequency response, noise, and dynamic range are really all different things, although of course related. the frequency content of a signal does not fully specify its tone. if there is no dynamic range, then in the worst case there is no harmonic envelope. 3) and noise isn't noise the way it is in analog. the correct term is signal to error ratio (i think this was mentioned here, but i'm just being pedantic). the fact that qunatization, if unfiltered, results in upper order harmonic distortion is secondary, and is not part of the noise specification. noise, in the strictest digital sense, is error... it is missing information. the fact that this error is highly correlated to signal amplitude is why analog noise is added to the signal (as dither) improve the ratio. among other things, de-correlating the error converts it into noise (in the psychoacoustic domain), and makes it more amenable to normal masking... it becomes a normal 'noise floor'. 4) on this subject, allow me to recommend the books on digital audio by Ken Pohlmann and John Watkinson. -----------< Cognitive Dissonance is a 20th Century Art Form >----------- Eric Harnden (Ronin) or The American University Physics Dept. 4400 Mass. Ave. NW, Washington, DC, 20016-8058 (202) 885-2748 ---------------------< Join the Cognitive Dissidents >------------------- ------------------------------ Date: Mon, 5 Oct 1992 13:16:40 EDT From: The Radio Gnome Subject: Re: 8/16 bit For a good continuation of this thread, check out the past discussions on AUDIO-L@VMTECMEX. We've already churned over LP vs CD and 16/18/24/32 bit sound and data compression and other various other topics of interest. A general thought is that no matter how good the digital media may look in theory, the circuitry itself remains essentially analog. Conversion is always a weak link. Andrew Wing ------------------------------ End of the EMUSIC-L Digest ******************************