[Eduardo AAVM]

Appeared in LA Audiophile

The Future is Analog

Los Angeles, CA - The director of the non-profit Institute for the Future sees analog electronics dwarfing digital in 20-30 years. As heard recently on KPFK’s Digital Village, IF director Paul Saffo said the future is analog. “Digital does not disappear, but there is a whole new industry that is more focused on analog electronics, growing off of a very small base now, that in 20-30 years will probably dwarf digital electronics.” Saffo went on to explain why: “Today, computers don’t really do anything important for us. They just manipulate information and doodle with symbols and the like. What we are on the verge of is giving our computers primitive sensory organs - eyes, eyes, and other devices, if you will - to become aware of the physical environment around us. And the physical environment is not digital; it is analog. So, what you’re going to see is a world in which our machines will be sucking up more-and-more analog information. Initially, they’ll process it in digital form, but as time goes on, more-and-more of the processing will be done with new kinds of analog circuitry.” Digital Village, a program about now computers and the Internet affect our lives, can be heard in southern California every Saturday at 10:00 a.m. on KPFK-FM 90.7 (http://www.kpfk.org/programs/dv/). The Web site for the Institute for the Future is http://www.iftf.org .

[JohnR]

Initially, they’ll process it in digital form, but as time goes on, more-and-more of the processing will be done with new kinds of analog circuitry

Err... is he serious? Any word on what kind of analog circuitry this is going to be?

I reckon this is one of those "I'm a visionary" bets. How this works is, you get yourself a title that sounds like it might mean you're important, then you go about making a bunch of outrageous predictions. Then, if any of them ever come true you can go about saying how you predicted it years ago and aren't you brilliant and by the way your consulting/speaking fee is incredibly high now.

Generate enough long-shot predictions, and you have a reasonable probability of one of them coming true.

JohnR

[Jay S]

I don't think information can be processed on a loss-less basis unless it is done digitally.  The problem is perhaps in accurately capturing analog information and converting it to digital (and the reverse of converting digital back to analog, as we all know).  But a computer will need to process in the digital medium.

I think the future is probably more like a refinement of digital technology so that it is able to capture nature as it is, rather than taking an imperfect sampling.  Take music for example: if your sampling rate approaches infinity then the resulting waveform is essentially analog.

[DeanSheen]

Arent digitial cameras now being surpassed & augmented w/analog devices to capture a richer picture beyond what more pixels gives you?

[8thnerve]

I don't think information can be processed on a loss-less basis unless it is done digitally.

I would argue that digital is only lossless once it's in the digital form.  The conversion to digital is certainly a lossy process.  The trick is to find out how much resolution we need to not notice that the process is imperfect.

And analog does not necessarily mean what we may assume it means.  Analog is defined as data that is represented by continuously variable quantities, while Digital is generally defined as on and off, binary.  New advances in computer technology may allow analog processors that simply have the ability to recognize more than two levels of fundamental states.  This would allow the digital storage of analog information.  Maybe.

[JoshK]

As a mathematician by training I am fascinated with this topic and have thought about this a lot.  Most of my training has been in numerical analysis which is precisely the issue here, i.e. using discrete methods of solving for approximate solutions to continous but unsolvable (in closed form) equations.

Not everything that can be done with mathematics can be done with electronics unfortunately.  For instance if you could develop a one-to-one transformation between digital and analog then you could 100% accurately get back the analog signal. I think the problem is with how to electronically implement the inverse transformation.  I am not sure it can be done.

I will admit that I don't know exactly how the sampling rate is taken but it seems that it is done via constant sampling.  This we know in mathematics isn't a very efficient way of sampling.  In the time domain this might be sufficient but it seems like in the frequency domain there would be a better way.  It seems to me that higher sampling at higher frequencies relative to lower frequencies would provide a better sense of realism.  Maybe this has been studied before.  Then again, in the time domain, it seems like if you sampled once through constantly and then sampled additionally in places of large dynamics things would get better still (kinda like a quasi-cbr/vbr sampling).

8thnerve,

Not sure I understand a few of your points. You say digital is only lossless when it is in digital form.  Not sure how you argue that digital is every not a lossless process.  Second is with analog processors.  If you have more than two levels--remember Russian's first computers? there were trinary processors, but this caused many problems with stability--you still don't have analog.  Theorectically you need infinite levels to represent analog.  However there are efficient ways of doing this without infinite levels.  Remember calculus?  You can reasonably approximate a function in a localized area with its level, slope and curvature (i.e. taylor's theorem).

[Eduardo AAVM]

Hi guys, JoshK what I think I understood from 8th Nerve is that digital when acquiring data is not loseless.

On the other isseu maybe 8th Nerve refered to the experimental computers with CPU's based on quantum physics laws andnot just plain binary.

[JohnR]

There are so many misconceptions in the post by 8thnerve that it's hard to even know where to start...

I'd like to address this oft-repeated statement that digital "has only two levels, on and off." This is true only on the most fundamental and/or naive level - it's a bit like saying that the financial systems of the world are based on a system that has only ten numbers. Digital information is stored and manipulated in binary numbers only because it happens to be the best way to make digital circuitry (i.e. the underlying hardware). As JoshK pointed out, ternary logic circuits exist(ed); I'm sure it would be quite possible to make ten-level logic circuits, it just wouldn't be very practical.

So, a binary number is just a number. It can be manipulated like any other number, it just has a different base. If you need bigger numbers, use more bits (note: "bit" is a contraction of "binary digit"); if you need smaller differences between numbers, use more bits again. This is just like saying that if your bank account has a million dollars in it, you need more digits to represent it than if it has only a thousand dollars in it. And, if you need to write down the number of cents as well as the number of dollars, again you need more digits!

Digital is just numbers -- the fact that it's binary or "has only ones and zeroes" is COMPLETELY IRRELEVANT.

Now, the notion that digital information can be manipulated losslessly is FALSE. Presumably you have all done a tax return, and on your calculator produced a number like, say, $1528.476248243. What do you write on line 463 of your tax return? You write $1528.47! (Or $1528.48, depending on how much of a pedant you are...) You have just LOST INFORMATION. The exact same thing occurs when processing audio represented digitally.

For example, suppose you have two 16-bit numbers (as you do with CDs). If you multiply them together, you get a 32-bit number. Multiply two 32-bit numbers together and you get a 64-bit number. And so on and so on. This is not a made-up example -- multiplication is one of the two fundamental operations in ALL digital filtering -- ie eq. (The other is addition.) But sooner or later you are going to have get rid of some of those extra bits you have accumulated! Voila, LOSS OF INFORMATION. This is EXACTLY like removing those extra digits when you have to write the number into the box on your tax return.

Now, the IRS might not care about a few cents either way, but in audio processing the loss of these bits causes NOISE. In some types of digital processing and if done incorrectly, it can even cause instability.

JoshK raises some much more interesting points, to which I'd like to respond a bit later

JohnR

[8thnerve]

That's right Eduardo, the conversion to digital is all based on sampling.  As you said Josh, analog would have an infinite amount of levels and time samples.  With standard redbook we sample 44,000 times per second, which in mathematical theory allows us to capture sound perfectly since this sampling rate allows us to record a 22KHz sound wave which is above the limit of our hearing.  We have seen over the past 20 years though that this is not all that there is to it.

Bit depth is also a problem as our ears are certainly more sensitive than 16.7 million levels of volume that the 16 bit standard allows for.  I agree with your suggestion that the level capturing process should be more dynamic so that we use all the available resolution based on the dynamic range of a particular segment of the recording, simliar to DBX.  Another interesting possibility would be to store the level information logarithmically as opposed to linearly.  Our ears are much more sensitive to the low level sounds and most harmonic information which we use for spatial clues, character, etc.  So why not store information in the lower amplitude levels.  Of course with compression these days, the quality of 99% of the recordings out there would probably decrease dramatically.

Suffice to say though that the conversion to digital is certainly a lossy process, and I mean just the theory based process, not the implementation errors that occur in transmission, stamping of the CD, and the incredibly high number of errors in playback that are "corrected" for with the built in error correction stored on every 16th (I think) bit of the CD.

And as far as analog computers, I was alluding to research in biological processors, which is certainly an analog medium.  You are right though Josh, simply using more base states is still digital.  Similar to the difference between RISC and CISC processor architectures.  While many UNIX based systems use a smaller (Reduced Instruction Set Computer, RISC) set of instruction that are more reliable and run cooler, Wintel based systems use a complex (Complex Instruction Set Computer, CISC) set of base instructions that are less reliable and slower, but can accomplish more at a time.

And John, I'm not sure what misconceptions you are referring to from my post.  I did say "generally defined as on and off", as most peoples understanding of digital ends there.  And I absolutely DISAGREE that it is irrelevant.  It is completely relevant as it is the basis for everything above it.  It is precisely because of it's binary nature that digital systems have to "LOSE INFORMATION" as you said, because at its simplest form, it must be represented by a defined amount of ones and zeros.  So I fail to see how it is "COMPLETELY IRRELEVANT".

Cheers,

8th Nerve

[Dan Banquer]

Just to clear up any misconceptions that anyone might have: Analog isn't exactly lossless either. For Digital to be lossless in a pure theoretical domain, especially when you are doing a conversion, you would need an infinite number of bits.
Considering all of the above, there are limits to what we can reproduce, mostly due to "analog" issues.

[8thnerve]

Dan makes a good point.  It is important to realize that neither analog or digital is inherently lossless at all.  We only have potential for loss when we try and manipulate something.  In the case of this argument, I am speaking of the lossy nature of the conversion from analog sound storage to digital sound storage.  Dan is correct that the recording of live sound to an analog medium is lossy due to imperfections in the equipment used, and even due to environmental effects to the magnetic based storage medium that the sound is stored on (i.e DAT or other tape...).

[JohnR]

It is precisely because of it's binary nature that digital systems have to "LOSE INFORMATION"

Wrong. It's because of its DIGITAL nature.

For Digital to be lossless in a pure theoretical domain, especially when you are doing a conversion, you would need an infinite number of bits.

Yes, but while converting it, you don't need an infinite number of bits to produce 144dB dynamic range, which should be enough for anybody. You only need 24  

However, if you are doing any math on it, you need more bits, or else you lose some of that dynamic range. If you want it to be "lossless" then you need an infinite number of bits while processing it

[Eduardo AAVM]

About quantum computers:

http://www.uni-konstanz.de/quantum-optics/quantech/

[Dan Banquer]

O.K. John, you have just given us an excellent example here of conversion loss. The best in Digital to Analog Converters can now take those 24 bits in Digital domain and get only 18 bits out in analog domain, and that's only for the best of the converters. The spec is called dynamic range and it is one the most crucial spec's for a converter. The reason for this loss is noise, and most of it is from the analog domain. In all probability, this will not improve in our lifetime, unless there is a paradigm shift in semiconductor physics.
In any case it makes me wonder why the industry went to 24 bits. Marketing maybe?

[8thnerve]

It is precisely because of it's binary nature that digital systems have to "LOSE INFORMATION"

Wrong. It's because of its DIGITAL nature.

John,

Would you be kind enough to explain the difference here?

[JohnR]

Sure, but first, did my tax return example make sense?

Dan, 18 bits in the D/A is the absolute best there is?   I know there are tricks with using multiple DACs to increase resolution, but still... Could you explain a bit more what determines the noise floor?

[JoshK]

...  For instance if you could develop a one-to-one transformation between digital and analog then you could 100%  ...

OK, that was a stupid statement since there exists no such transformation....duh!

[Dan Banquer]

I understand about the taxes example, their appears to be two ways of dealing with the "excess". One is to use dither,. which is noise, to mask it, and I think the other is DSP. Don't quote me on the DSP, I'm pretty hazy on some of those techniques.
Let's see if I can explain the noise thing. If not, let me know and I will attempt to send you in the right direction.
If we have a 20 bit system, and audio assigns each bit at 6 db of range then we potentially have a system with 120 db of dynamic range. This 120 db of dynamic range has to "fit" in a 2 volts rms signal with all of it's bits being used, so in analog we need to have a -120db down form 2 volts rms signal to noise ratio. That represents a maximum of 2 microvolts of noise.
Does the above help you to understand? I think you have been around enough to understand just how little 2 microvolts of noise is.

[JohnR]

Hi Dan, the tax return question was for 8thnerve.

I see your point, put in those terms 24-bit convertors do look like hocus-pocus... still, at the 120 dB level, I believe the IRD preamp has a spec of 120dB S/N, so why can't a DAC do it? Anyway, I was more wondering what the fundamental limitation was, as you said solving it required a paradigm shift in semiconductor physics, implying that it's something to do with the silicon rather than thermal noise - ?

J

[Dan Banquer]

120 db s/n from where? Is that from 10 volts or higher? Pre amps used to be rated at db down from one volt, because power amp sensitivity is typically anywhere from 0.5 volts rms to 1.5 volts rms.
Let me try to explain this. I will start with the DAC. Odb full scale voltage is typically 2 volts rms to 2.2 volts rms. Well designed line stage pre amps have a typical signal to noise ratio of -100 db down from one volt, and well designed power amps have a signal to noise ratio at one watt into 8 ohms (2.83 volts) of -90 to -95 db down from 2.83 volts rms. One of the things that confuses people here is that for some reason everyone decided to rate their S/N ratios at full scale. This makes the numbers look good but it is not realistic to our listening levels which average 1 to 2 watts. Now if you add those noise figures up and add the inherent ground loops between pieces of equipment, and an ambient noise level of around 40 db spl, plus deduct for the lack of low level linearity in most loudspeakers, you start to come to the conclusion that we have yet to fully reproduce 16 bits.
Shocking: isn't it.
The major thing you  need to remember is that the db is relative, and relative to where is the key.