Analog computing / computers

[serpretetsky]

http://en.wikipedia.org/wiki/Analog_computer

Most analog computing is pretty old. If you guys have any examples of modern analog computing i would love to hear about it.

You guys think analog computing will ever make a comeback? I would imagine probably integrated with digital computing.

Wikipedia has a good example:

For example, voltage can simulate water pressure and electric current can simulate rate of flow in terms of cubic metres per second (in fact, given the proper scale factors, all that is required would be a stable resistor, in that case). Given flow rate and accumulated volume of liquid, a simple integrator provides the latter; both variables are voltages. In practice, current was rarely used in electronic analog computers, because voltage is much easier to work with.

In other words: why do difficult calculations to find answers to difficult questions when the solutions often present themselves in the natural world using natural phenomena.

You could simulate water on a computer, but the water simulation of the real world is running in real time with pretty good precision! (using water to simulate water is probably a bad example, but hopefully it gets the idea across)


But with the speed of modern digital computers, I wonder if such circuits would even be any faster? Plus such circuits would probably be designed for very specific applications.

Still, makes me wonder:
1) Whats faster, integrating/differentiating a difficult equation using digital computers running at high frequency but requiring many clock cycles or integrating a difficult equation using an analog component using a single 'step' but at the cost of waiting for the circuit to complete.

2)Would that analog circuit be able to handle various generic problems?

 

[SecurityTheatre]

Analog calculations have very limited use. You are limited to problems which closely mirror those of analog electronics. Your example of the flow of water and electricity is one of those, but there are very few others, although many types of simple integration and differential equations can be simulated to some marginal degree of accuracy.

In addition, this type of circuit is sensitive to variables like temperature. Increase the temperature and you change the resistance of an analog circuit, affecting the results. Controlling for these variables would be challenging to get any degree of precision.

And what if you want to modify your calculation to account for a higher temperature of water? Or for some sort of turbulence? What if you want to model the flow of water accounting for other factors such as gravity, or various specific gravity of fluids?

There are discreet equations for this that can be calculated on a general purpose computer, but changing the logic on an analog computational circuit requires substantial re-engineering of the circuit.

There used to be some value in integrator circuits for certain types of calculations that were expensive to perform on digital hardware, but now, within a few billion CPU cycles and a few megabytes of RAM, most any equation simple enough to conceive in an analog sense can also be modeled in a digital one. Now that a few billion CPU cycles and a megabyte of RAM are trivial numbers, it seems unlikely to me that you would find substantial practical use for analog computing.

I once read a paper on using certain limited analog counters for obtaining a type of fuzzy logic in AI circuitry. This research was relying on one of the fundamental problems of analog circuitry - that they are somewhat variable dependent on environmental conditions, etc, and using this as a source of entropy to provide a semistable state for storing analog "feelings" and randomizing actions and behaviors.

Beyond that, I'm not sure the use... but I'm willing to be proven wrong.

 

[Peter McNair]

Interesting posts.

An analog computer can run 1000 faster just by making the feedback capacitor 1000 times smaller. BUT Nature has this wonderful way of not allowing something for nothing and the price paid is increased (electrical) noise - which reduces the accuracy of the solution. A more accurate solution takes more time. BUT this is true in the digital world too. And I suspect - at some deeper level - all computers are in some sense equivalent - by which I mean there is no 'best' (in general) method. (Dare I say it, but a corollary of this would be that there's no 'best' programming language...) Of course, in real life, some computing methods are more suited to particular tasks than others.

It's a matter (I think) of being in an appropriate reference frame for the problem at hand: it is much easier to model a cup of tea being stirred by a spoon - by getting a cup of tea and stirring it with a spoon. And it's a lot easier to find the area of a circle (using calculus) in a circular coordinate frame rather than in a rectangular coordinate frame...

I suspect analog computing will make a comeback - integrated into digital computing - maybe clunky at first, but at some point seamless in some sense.

In the meantime I thought it would be fun to build my own proper (i.e. analog) computer - it's early days but you can read about it here:

http://analog-ontology.blogspot.co.uk/

 

[sm625]

It is amazing what they used to do with op amps. But it is hell on earth working with those circuits. It is so much easier just to use a $2 microcontroller.

 

[SecurityTheatre]

It is amazing what they used to do with op amps. But it is hell on earth working with those circuits. It is so much easier just to use a $2 microcontroller.

Agreed. I could have said the same. My analog electronics classes at University felt like wrestling with a greased pig.

 

[aigomorla]

oldest analog computer...

i suppose a sundail could also satisify the analog computer aspect as well.

 

[Harvey]

In electronic circuits, the term, "analog computing" can be interpreted in a couple of ways: It can refer to any number of circuits using op-amps and/or comparators to perform static calculated results from a set of inputs. It is more commonly to refer to a family of analog calculating circuits known as analog multipliers. This group is further catagorized as linear or exponential circuits.

Typical linear multiplier can perform calculations in the form (X x Y) / Z, where X, Y and Z are any linear values within the range the circuit can accomodate. These values may be voltage or current, depending on the architecture of the circuit, and the resulting output is not temperature dependent.

Typical exponential circuits usually depend on the exponential characteristic of the voltage across one or more transistor base-emitter junctions and/or diodes. They are capable of wider ranging calculations, but the gain of these ciruits is temperature dependent due to the temperature characteristics of the diode junction voltage. Typical means of temperature compensating these circuits involve temperature dependent resistors (thermistors) with an inverse temperature coefficient that approximates the inverse of the temerature characteristic of the multiplier circuit.

One example is a circuit know as a log-anti-log multiplier, commonly used as an analog volume control. The exponential control characteristic allows using a linear voltage or current to control the volume of an audio signal in decibles (dB), which correlates to the way human beings perceive relative loudness.

I hold two patents for analog multiplier circuit topologies having extremely low distortion (> 0.003%) and wide dynamic range (>119 dB) , and I have developed control circuits that, unlike thermistors, calculate and eliminate the temperature coefficient from calculations using exponential multipliers in real time.

One major advantage of analog control is that, aside from propogation through the circuit, the resulting output is an analog value produced in absolute real time with infinite resolution for both the signal and the controlling value.

Digital audio is like digital image resolution. In audio, the number of bits defines relative steps of loudness. In graphics, it defines relative steps of a grey scale. In audio, the sampling frequency is the equivalent of dot pitch in graphics. There are two major problems with linear PCM (pulse code modulation):

1. Distortion is lowest at full amplitude (maximum loudness), and it increases with lower amplitude signals.

2. The inherent distortion products are non-harmonic in nature (frequency products unrelated to musical content)

In both cases, the human ear is more sensitive to such distortion products.

I hated early CD's. 44 kHz may satisfy the Nyquist requirement for sampling a single sine wave up to 20 kHz, but the math falls apart when dealing with more than one high frequency component at a time or with signals that vary in amplitude during the sampling period. Although John Cage produced a well known recording of 4 minutes and 33 seconds of silence, to date, I don't know of anyone who has produced a hit recording of a steady state monotone.

I hate MP3's and other sub-Nyquist lossy sampling systems even more. Today, advanced processing and algorithms can reduce the offensive nature of many of the artifacts to produce less irritating content, but it still doesn't produce the accuracy of the output signal relative to the input.

When people asked why I disliked digital audio, my answer was, "Not enough bits." When they asked how many I would want, I would answer, "All of them. How many do you have? I've got more."

At 24 bits x 192 kHz, digital audio approaches the equivalent of "photo-realistic" graphic images because the artifacts caused by the residual errors are suppressed below the threshold of human perception. It's the same reason you need greater resolution on larger monitors, TV screens and theater projections. However, it still takes a finite amount of time to compute and output the desired result, as opposed to the real time product of analog multiplication.

 

[serpretetsky]

Interesting posts.

An analog computer can run 1000 faster just by making the feedback capacitor 1000 times smaller. BUT Nature has this wonderful way of not allowing something for nothing and the price paid is increased (electrical) noise - which reduces the accuracy of the solution. A more accurate solution takes more time. BUT this is true in the digital world too. And I suspect - at some deeper level - all computers are in some sense equivalent - by which I mean there is no 'best' (in general) method. (Dare I say it, but a corollary of this would be that there's no 'best' programming language...) Of course, in real life, some computing methods are more suited to particular tasks than others.

It's a matter (I think) of being in an appropriate reference frame for the problem at hand: it is much easier to model a cup of tea being stirred by a spoon - by getting a cup of tea and stirring it with a spoon. And it's a lot easier to find the area of a circle (using calculus) in a circular coordinate frame rather than in a rectangular coordinate frame...

I suspect analog computing will make a comeback - integrated into digital computing - maybe clunky at first, but at some point seamless in some sense.

In the meantime I thought it would be fun to build my own proper (i.e. analog) computer - it's early days but you can read about it here:

http://analog-ontology.blogspot.co.uk/

Welcome to the forums and AWESOME website! I salute you sir! Is it difficult to create arbitrary function generators for your differential and integral computers? I am not very familiar with this subject.

 

[serpretetsky]

oldest analog computer...

i suppose a sundail could also satisify the analog computer aspect as well.

I would actually argue the abacus is a digital computer, not analog. It's usually used in discrete ways. In other words, the beads can either be on one side, or the other, there is no continuous middle ground.

 

[aigomorla]

I would actually argue the abacus is a digital computer, not analog. It's usually used in discrete ways. In other words, the beads can either be on one side, or the other, there is no continuous middle ground.

The beads can also be moved in a negative.
0 would be neutral ground on a abacuss, and u would either + or - in the value needed.

I always considered the abacuss the analog calculator, which a calculator could fill in the definition for computer.

 

[serpretetsky]

The beads can also be moved in a negative.
0 would be neutral ground on a abacuss, and u would either + or - in the value needed.

I always considered the abacuss the analog calculator, which a calculator could fill in the definition for computer.

but you see what im saying right?
http://en.wikipedia.org/wiki/Digital

abacus uses discrete values (there are a finite amount of "states" those beads can take). Therefore you can consider it as being digital.

As oppose to using two metered cups of water and pouring them into a larger metered cup to obtain a sum of the two. That's a true analog computer, it's continuous, there are no discrete values (i guess it's hard to classify that as a full blown computer, but you get the idea)

Now you can argue that if you zoom far enough down onto the water it becomes quantuamized and discrete, but water is generally thought of as being continuous. Similar to most natural aspects of our world from human perspective.

edit: for example, you can store digital or analog data on a magnetic tape, it depends how you treat it. You can choose to store discrete values on the tape, where you don't care about any other values except those discrete values, or you could store a continous signal on that tape, in which case you are using it in an analog fashion.

 

[Peter McNair]

Welcome to the forums and AWESOME website! I salute you sir! Is it difficult to create arbitrary function generators for your differential and integral computers? I am not very familiar with this subject.

In a word: yes! I will need bits of electronics which convert x volts into (for example) sin(x) and so on. But it's something of a can of worms - it looks like there isn't a single unified approach to any of this: an approach which works for sin(x) might be useless for another function. But this comment is also true of the digital world - probably something of an understatement, but special functions can be quite problematic to compute on a digital computer.

I'd be very interested to know if there's a reason why special functions are 'difficult' to compute either in the analog or digital doman...

 

[videogames101]

You can solve differential equations with opamps. One of my older professors tells a story of a room sized analog computer with racks of modular opamp circuits for solving systems of differential equations. That's just the way you had to do it back in the day. The university had a full time guy just tuning the whole thing, temperature and other variables made it hard to maintain.

Nobody wants to deal with non-ideal secondary effects, build tolerances, temperature, etc. Getting accuracy and speed is really hard.

Analog computing is quite capable, but as anyone who's worked with analog circuits can tell you: if you can do it in digital, do it in digital. (and do it synchronously)

 

[aigomorla]

edit: for example, you can store digital or analog data on a magnetic tape, it depends how you treat it. You can choose to store discrete values on the tape, where you don't care about any other values except those discrete values, or you could store a continous signal on that tape, in which case you are using it in an analog fashion.

rofl... then its irony how the first PC was a digital one vs an analog as the abacuss dates during ancient Arabia and china even.

 

[videogames101]

In electronic circuits, the term, "analog computing" can be interpreted in a couple of ways: It can refer to any number of circuits using op-amps and/or comparators to perform static calculated results from a set of inputs. It is more commonly to refer to a family of analog calculating circuits known as analog multipliers. This group is further catagorized as linear or exponential circuits.

Typical linear multiplier can perform calculations in the form (X x Y) / Z, where X, Y and Z are any linear values within the range the circuit can accomodate. These values may be voltage or current, depending on the architecture of the circuit, and the resulting output is not temperature dependent.

Typical exponential circuits usually depend on the exponential characteristic of the voltage across one or more transistor base-emitter junctions and/or diodes. They are capable of wider ranging calculations, but the gain of these ciruits is temperature dependent due to the temperature characteristics of the diode junction voltage. Typical means of temperature compensating these circuits involve temperature dependent resistors (thermistors) with an inverse temperature coefficient that approximates the inverse of the temerature characteristic of the multiplier circuit.

One example is a circuit know as a log-anti-log multiplier, commonly used as an analog volume control. The exponential control characteristic allows using a linear voltage or current to control the volume of an audio signal in decibles (dB), which correlates to the way human beings perceive relative loudness.

I hold two patents for analog multiplier circuit topologies having extremely low distortion (> 0.003%) and wide dynamic range (>119 dB) , and I have developed control circuits that, unlike thermistors, calculate and eliminate the temperature coefficient from calculations using exponential multipliers in real time.

One major advantage of analog control is that, aside from propogation through the circuit, the resulting output is an analog value produced in absolute real time with infinite resolution for both the signal and the controlling value.

Digital audio is like digital image resolution. In audio, the number of bits defines relative steps of loudness. In graphics, it defines relative steps of a grey scale. In audio, the sampling frequency is the equivalent of dot pitch in graphics. There are two major problems with linear PCM (pulse code modulation):

1. Distortion is lowest at full amplitude (maximum loudness), and it increases with lower amplitude signals.

2. The inherent distortion products are non-harmonic in nature (frequency products unrelated to musical content)

In both cases, the human ear is more sensitive to such distortion products.

I hated early CD's. 44 kHz may satisfy the Nyquist requirement for sampling a single sine wave up to 20 kHz, but the math falls apart when dealing with more than one high frequency component at a time or with signals that vary in amplitude during the sampling period. Although John Cage produced a well known recording of 4 minutes and 33 seconds of silence, to date, I don't know of anyone who has produced a hit recording of a steady state monotone.

I hate MP3's and other sub-Nyquist lossy sampling systems even more. Today, advanced processing and algorithms can reduce the offensive nature of many of the artifacts to produce less irritating content, but it still doesn't produce the accuracy of the output signal relative to the input.

When people asked why I disliked digital audio, my answer was, "Not enough bits." When they asked how many I would want, I would answer, "All of them. How many do you have? I've got more."

At 24 bits x 192 kHz, digital audio approaches the equivalent of "photo-realistic" graphic images because the artifacts caused by the residual errors are suppressed below the threshold of human perception. It's the same reason you need greater resolution on larger monitors, TV screens and theater projections. However, it still takes a finite amount of time to compute and output the desired result, as opposed to the real time product of analog multiplication.

But what is the alternative? Obviously we'd love perfect analog recording and playback of audio signals, but in reality analog audio deviates from this ideal. Anytime you operate on signals in the analog domain you're looking at thermal noise from every component. If you have any sort of non-linear device, you have shot noise as well. Not to mention tuning analog circuits for variations in component values, temperature, etc. The sooner and more accurately you can run an audio signal into an ADC, the better, at least in terms of signal integrity. Obviously you lose infinite resolution, but you cut your losses in terms of noise.

Digital storage is really the only reasonable option for audio.

History has shown people are more than happy with lossy compression of 16-bit 44.1k PCM. Even moderately trained ears can't successfully ABX between 24/192k and 16/44.1k.

As a footnote, I buy both vinyl and 16/44.1k versions of most music. I do love some analog audio from time to time. It certainly has a different flavor.

 

[SecurityTheatre]

But what is the alternative? Obviously we'd love perfect analog recording and playback of audio signals, but in reality analog audio deviates from this ideal. Anytime you operate on signals in the analog domain you're looking at thermal noise from every component. If you have any sort of non-linear device, you have shot noise as well. Not to mention tuning analog circuits for variations in component values, temperature, etc. The sooner and more accurately you can run an audio signal into an ADC, the better, at least in terms of signal integrity. Obviously you lose infinite resolution, but you cut your losses in terms of noise.

Digital storage is really the only reasonable option for audio.

History has shown people are more than happy with lossy compression of 16-bit 44.1k PCM. Even moderately trained ears can't successfully ABX between 24/192k and 16/44.1k.

As a footnote, I buy both vinyl and 16/44.1k versions of most music. I do love some analog audio from time to time. It certainly has a different flavor.

Has a different flavor, but if you encode the hissing and popping, you can convince even "skilled" audiophiles that they are listening to a vinyl, even when it's a digital recording of a vinyl output.

 

[serpretetsky]

In electronic circuits, the term, "analog computing" can be interpreted in a couple of ways: It can refer to any number of circuits using op-amps and/or comparators to perform static calculated results from a set of inputs. It is more commonly to refer to a family of analog calculating circuits known as analog multipliers. This group is further catagorized as linear or exponential circuits.
.

Interesting

The analog vs. digital tradeoff in multiplication

In most cases the functions performed by an analog multiplier may be performed better and at lower cost using Digital Signal Processing techniques. At low frequencies a digital solution is cheaper and more effective, and allows the circuit function to be modified in firmware. As frequencies rise, the cost of implementing digital solutions increases much more steeply than for analog solutions. As digital technology advances, the use of analog multipliers tends to be ever more marginalised towards higher-frequency circuits or very specialized applications.

-http://en.wikipedia.org/wiki/Analog_multiplier


How many instruction cycles does multiplication take on modern processors from intel and amd? Just one? or more?

 

[Harvey]

But what is the alternative? Obviously we'd love perfect analog recording and playback of audio signals, but in reality analog audio deviates from this ideal.

I didn't say analog was the end all and be all of audio. I could post a lot of reasons why 16 bits x 44 kHz is inadequate for really fine, accurate audio reproduction of complex material (i.e. music), and early CD's had more major problems that generate extremely harsh, anti-musical products.

I also said that, at 192 kHz x 24 bits, we're approaching the visual equivalent of a photo-realistic printer because those unpleasant artifacts are suppressed below the threshold of human perception.

Think of it this way -- A photo of an attractive model in a conventional newspaper format (low dot pitch, low grey scale) wouldn't be that exciting, but with a photo of the same model published as a Playboy centerfold, as viewed from five feet, a young boy could get off single handed.

In other words, good enough really is good enough, but if really fine audio accuracy is required, 44 kHZ X 16 bits isn't.

Anytime you operate on signals in the analog domain you're looking at thermal noise from every component.

True, but if the noise of the circuitry is well below the noise of the source material, the noise from the source will dominate the signal. If the noise of the circuit is above the noise from the source, but it is low enough that it still leaves excellent dynamic range, it may not matter.

As an example, the thermal noise of an ideal 150 ohm resistor is -131.5 dBm (20 - 20 kHz) at room temperature. That defines the noise from an otherwise ideal microphone having a source impedence of 150 ohms before any amplification. If the signal from the microphone requires 40 dB of gain to bring it to line level, its self noise will be amplified to -91.5 dBm (-131.5 dBm + 40 dB).

If the inherent noise of the analog circuit is -110 dB, that noise is so far below -91.5 dBm that it won't contribute significantly to the noise at the output of the circuit.

If you have any sort of non-linear device, you have shot noise as well.

That could be one reason they granted my patent applications for class A analog multipliers (fancy name for volume controls) with a dynamic range 119 dB (clipping to noise floor). My co-inventor and I were the first to accomplish this performance, and our worst case distortion was around 0.003%

Not to mention tuning analog circuits for variations in component values, temperature, etc.

As I also posted, linear multipliers have no temperature coefficient, and I have developed a precision technique that absolutely (not approximately) removes the temperature coefficient of the control path of exponential multiplier circuits. The audio signal path has no tempco.

The sooner and more accurately you can run an audio signal into an ADC, the better, at least in terms of signal integrity. Obviously you lose infinite resolution, but you cut your losses in terms of noise.

True, AS LONG AS the ADC has sufficiently enough bits at a sufficiently high sampling rate, and there's still the matter of the accuracy (distortion) of the AD process.

Digital storage is really the only reasonable option for audio.

For storing a large number of independent tracks in a conveniently sized medium, yes. For a single end product such as an album, no.

Side note -- My late co-inventor/business partner built a prototype analog FM recording system that produced simply stunning recordings.

History has shown people are more than happy with lossy compression of 16-bit 44.1k PCM. Even moderately trained ears can't successfully ABX between 24/192k and 16/44.1k.

Yeah, and the kids are happy with Spaghettios... until they grow up, and their tastes mature enough to understand the subtleties of better pasta dishes. Distortion is ANY unintended change in the program material. The altered end result (i.e. MP3) may (or may not) be pleasant in its own right, but it is not the same as the original material. Accuracy is far more imparative for professional production and archival storage of finished originial artistic productions.

I fully acknowledge the production benefits and the non-volatility of digital audio recording, and I am quite happy with the audio quality of 24 bit x 192 kHz recordings, and 24 bit x 96 kHz is good for most program material.

End products are their own separate issue. We produce products to sell to the masses in the popular formats because that what people buy. I have no problem with that. The artists and producers have to pay their bills, just like the rest of us.

I can, and I've been able to hear the difference between MP3's and high quality original source material many times in some of the best audio facilities in the world. I participated in another lab demonstration with skilled, world class recording engineers where we started with a high quality digital music signal, processed it through a lossy compression system and subtracted the results from the original digital source. That allowed us to hear just the material that was removed by the system, the material the computer "thought" no one could hear.

The result was a shock to those engineers who, to a person, stated that they would not want those subtle details they had worked so hard to include removed from any of their creative recording efforts.

 

[Cerb]

How many instruction cycles does multiplication take on modern processors from intel and amd? Just one? or more?

5-6, it seems (AT's Haswell article, Agner Fog's Bulldozer tables). It's been done in as little as 3 on more than a few reasonable CPUs (IE, not DSP-type weird things), and less for low-bit-depth imuls (IE, 8-bit x 8-bit could surely be done in 1 cycle, but does anyone really care, when most modern data is 32 bits or wider?).

 

[videogames101]

Interesting
-http://en.wikipedia.org/wiki/Analog_multiplier


How many instruction cycles does multiplication take on modern processors from intel and amd? Just one? or more?

depends on what you are multiplying

for IMUL, you're looking at 6-8, more for extra wide data

14.4 INSTRUCTION LATENCY

in

http://www.intel.com/content/dam/ww...4-ia-32-architectures-optimization-manual.pdf

 

[Cerb]

depends on what you are multiplying

for IMUL, you're looking at 6-8, more for extra wide data

14.4 INSTRUCTION LATENCY

in

http://www.intel.com/content/dam/ww...4-ia-32-architectures-optimization-manual.pdf

Also depends on what it's running on. Note that those are Atom CPU latencies.

 

[oynaz]

I didn't say analog was the end all and be all of audio. I could post a lot of reasons why 16 bits x 44 kHz is inadequate for really fine, accurate audio reproduction of complex material (i.e. music), and early CD's had more major problems that generate extremely harsh, anti-musical products.

I also said that, at 192 kHz x 24 bits, we're approaching the visual equivalent of a photo-realistic printer because those unpleasant artifacts are suppressed below the threshold of human perception.

Think of it this way -- A photo of an attractive model in a conventional newspaper format (low dot pitch, low grey scale) wouldn't be that exciting, but with a photo of the same model published as a Playboy centerfold, as viewed from five feet, a young boy could get off single handed.

In other words, good enough really is good enough, but if really fine audio accuracy is required, 44 kHZ X 16 bits isn't.



True, but if the noise of the circuitry is well below the noise of the source material, the noise from the source will dominate the signal. If the noise of the circuit is above the noise from the source, but it is low enough that it still leaves excellent dynamic range, it may not matter.

As an example, the thermal noise of an ideal 150 ohm resistor is -131.5 dBm (20 - 20 kHz) at room temperature. That defines the noise from an otherwise ideal microphone having a source impedence of 150 ohms before any amplification. If the signal from the microphone requires 40 dB of gain to bring it to line level, its self noise will be amplified to -91.5 dBm (-131.5 dBm + 40 dB).

If the inherent noise of the analog circuit is -110 dB, that noise is so far below -91.5 dBm that it won't contribute significantly to the noise at the output of the circuit.



That could be one reason they granted my patent applications for class A analog multipliers (fancy name for volume controls) with a dynamic range 119 dB (clipping to noise floor). My co-inventor and I were the first to accomplish this performance, and our worst case distortion was around 0.003%



As I also posted, linear multipliers have no temperature coefficient, and I have developed a precision technique that absolutely (not approximately) removes the temperature coefficient of the control path of exponential multiplier circuits. The audio signal path has no tempco.



True, AS LONG AS the ADC has sufficiently enough bits at a sufficiently high sampling rate, and there's still the matter of the accuracy (distortion) of the AD process.



For storing a large number of independent tracks in a conveniently sized medium, yes. For a single end product such as an album, no.

Side note -- My late co-inventor/business partner built a prototype analog FM recording system that produced simply stunning recordings.



Yeah, and the kids are happy with Spaghettios... until they grow up, and their tastes mature enough to understand the subtleties of better pasta dishes. Distortion is ANY unintended change in the program material. The altered end result (i.e. MP3) may (or may not) be pleasant in its own right, but it is not the same as the original material. Accuracy is far more imparative for professional production and archival storage of finished originial artistic productions.

I fully acknowledge the production benefits and the non-volatility of digital audio recording, and I am quite happy with the audio quality of 24 bit x 192 kHz recordings, and 24 bit x 96 kHz is good for most program material.

End products are their own separate issue. We produce products to sell to the masses in the popular formats because that what people buy. I have no problem with that. The artists and producers have to pay their bills, just like the rest of us.

I can, and I've been able to hear the difference between MP3's and high quality original source material many times in some of the best audio facilities in the world. I participated in another lab demonstration with skilled, world class recording engineers where we started with a high quality digital music signal, processed it through a lossy compression system and subtracted the results from the original digital source. That allowed us to hear just the material that was removed by the system, the material the computer "thought" no one could hear.

The result was a shock to those engineers who, to a person, stated that they would not want those subtle details they had worked so hard to include removed from any of their creative recording efforts.

Agreed. People telling you that they cannot hear the difference between a CD and an MP3 is not listening closely enough, though 192 kbp MP3 or better are OK.
Even the difference between CDs and DVD audio (24bit 48Khz, IIRC) is quite noticeable, given the right source material. I cannot tell the difference between 24bit 48Khv and 24bit 192Khz, though.

 

[Harvey]

Agreed. People telling you that they cannot hear the difference between a CD and an MP3 is not listening closely enough, though 192 kbp MP3 or better are OK.
Even the difference between CDs and DVD audio (24bit 48Khz, IIRC) is quite noticeable, given the right source material. I cannot tell the difference between 24bit 48Khv and 24bit 192Khz, though.

Another issue regarding bit depth and sampling rates in production is that, in summing several signals, and especially in non-linear processing such as equalization, which entails multiplication of values, rather than addition, the system may require extra bits of overhead in order to arrive at the correct value of the output signal of summed or multiplied input when it is truncated down to the maximum of the system.