Integer is King? - final thoughts about the EQ challenge

[JonHodgson]

No I don't, you're also not an electronics engineer obviously.

You are wrong. I have said education, while I'm employed in artistic field, registered with our Ministry of Culture.

My apologies for this error.

But, in certain period of my life I also actively designed custom analog audio circuitry for recording usage, which was deployed and used in some local studios where I live. Not a big deal whatsoever, but I happen to know what I speak of while talking of impendance mismatches, because this were the most painful and sound-degrading issues on an _otherwise_healthy_ signal path, sometimes requiring re-buffering and similar solutions.

Why did you underline "otherwise healthy"? As an electronics engineer impedance is the most basic part of what we learn, I'm going to guess that pretty much the first thing both of us learned was ohm's law. Talking about impedance mismatches in an otherwise healthy signal path is like a doctor talking about ebola in an otherwise healthy patient.

Not that it matters, or I own such equipment, but I was also introduced to _audibly_ different output of Waves TDM versus VST versions (on ppc boxes). Which _other_ engineers noticed first! Maybe this is over-simplified and I don't know the technical details (do I really need to know them ...), but I believe there's float vs. int, too, going on there. In other words: this difference isn't anything new and lots of people know it.

Straw man argument. I have repeatedly stated that it is not surprising if two different implementations of ostensibly the same algorithm sound different. This can be due to the developer not being aware of any difference in the first place, not knowing how to deal with the problem, or having limited processing resources... or as Urs suggests for marketing reasons. None of these support the extrapolation that all integer algorithms share certain characteristics in contrast to their floating point counterparts, nor that there are always any differences at all.

No, you implied the connection, I simply responded sarcastically to it.

Again wrong, repeatedly so. I emphasized a "similarity" in the audible aspect. I said it reminded me on "what happens in case of such impendance mismatch". Then again, I can understand that you want to "prove your right ..." and you are sticking to your guns.

Do you understand the meaning of the word "imply"? You may not have meant to imply it, but you did imply it.

Electronic engineers (that's me) would agree that impedance mismatches can be heard, but from your description I concluded that you don't really understand what is actually going on in there at an electronics level.

I gave absolutely _no_ such description from which the above quoted conclusion could be made as a positive conclusion. This, Sir _you_have_made_up_ in it's entirety. Speaking as an engineer myself, this very moment.

You're right, I made the mistake of "reading between the lines". Sorry, I should have stuck to the specifics in your post.

By the way, if you fed a 10k source into a similar impedance the most notable effect would be the drop in level, frequency response changes would usually be secondary... but hey, you knew that anyway didn't you?

[JonHodgson]

No, you implied the connection, I simply responded sarcastically to it.

Again wrong, repeatedly so. I emphasized a "similarity" in the audible aspect. I said it reminded me on "what happens in case of such impendance mismatch". Then again, I can understand that you want to "prove your right ..." and you are sticking to your guns.

Do you understand the meaning of the word "imply"? You may not have meant to imply it, but you did imply it.

Just to add to this... I do apologize for my sarcasm which was based on the misconception that your implication was intentional.

[mauseoleum]

This is killing the thread already, but there's at least two things happening when impendance mismatching, sort-of "from the book" (not including what happens if capacitors are in signal path):

- if impendances are "matched", there's both current and voltage "driving" the input stage. Sounds "about right" usually ... pleasantly "dry" and "inplace" I should say.
- if lower impendance feeds unproportionally high impendance input, there's unproportionally more voltage component ivolved in "driving" the input stage. I describe the resulting sound as "watery", "overblown", if bass gets turned up, it's quickly "booming" in most cases (maybe this is explanation: "bigger than life sound, but watery").
- opposite happens if the higher impendance source "drives" low-impendance input (in case of guitars, there's "no tone" happening), sound is "dried", "sucked up" in most cases.


Similar _sound_ impressions I got from said integer vs. float comparisons and described as such _already_ (re-read ... ) Absolutelly there isn't any "impendance mismatching" in digital ...

I do recognize the resulting _digital_ sound, though. Integer sounds (to me) flatter, dryer, less "watery", more "inplace", esp. if you turn up the bass. What is _the_cause_ of said resulting sound, I don't know - and, don't really care per se.

But, if somebody explains why is that so, I certainly won't dismiss it - and will surely listen to it or read it.

Please, stop obsessing now.

[JonHodgson]

This is killing the thread already, but there's at least two things happening when impendance mismatching, sort-of "from the book" (not including what happens if capacitors are in signal path):

- if impendances are "matched", there's both current and voltage "driving" the input stage. Sounds "about right" usually ... pleasantly "dry" and "inplace" I should say.
- if lower impendance feeds unproportionally high impendance input, there's unproportionally more voltage component ivolved in "driving" the input stage. I describe the resulting sound as "watery", "overblown", if bass gets turned up, it's quickly "booming" in most cases (maybe this is explanation: "bigger than life sound, but watery").
- opposite happens if the higher impendance source "drives" low-impendance input (in case of guitars, there's "no tone" happening), sound is "dried", "sucked up" in most cases.

Your description is flawed.

The proportion of voltage and current "driving" the circuit is set by the input impedance of the circuit.

What changes is that the output Impedance of the source unit and the input umpedeance of the destinations unit together form an Impedance network, this affects the amount of voltage (and therefore current) driving the circuit, but does nothing to the proportion.

Taking the simplest example of a resistive only Impedance in both cases...

We'll give the destination unit 100k input impedance.

If the source has 10k output impedance, then it is driving a total of 10+100 = 110 Ohms, any voltage appearing across that will be split 100/110 so you get 91% of it appearing at the input.

give the source a 100k output impedance and you get 50%, a 1 meg output impedance and you wind up with only 9% of the signal driving the input.

At the voltage at the input to the load goes down, so does the current, at no time does their proportion change... ohm's law, remember?

The reason the change isn't always just one of level and can also be tonal (but does not have to be) is twofold. Firstly the impedances are likely to be reactive, in other words they are frequency dependent, therefore the network created by the two together will have a different response which depends on both networks. The second is that when driving a low load impedance the driver may not be able to generate enough current to maintain the voltages required, so it will distort.

[aciddose]

about how low impedance output feeding a high impedance input causes problems:
if this were to ever occur in a circuit it would be due to the lack of a proper output buffer. "low impedance output" means that the output can drive a lot of current, it doesnt mean the output requires a pull-down current source or resistor to bias it.

in an ideal world we would have zero impedance outputs driving infinite impedance inputs. the next best thing is biased bipolar buffer outputs driving fet inputs which is what we normally use these days. with such a configuration there are no notable effects on the signal which should usually be the desired outcome.

the only place in audio where impedance matching really matters is either in obsolete standards or interfaces to passive devices. the only interface to a passive device should ideally be a buffer (input or output) which provides either zero or infinite input/output impedance so that we can safely interface with the rest of the world without worrying about impedances.

[JonHodgson]

about how low impedance output feeding a high impedance input causes problems:
if this were to ever occur in a circuit it would be due to the lack of a proper output buffer. "low impedance output" means that the output can drive a lot of current, it doesnt mean the output requires a pull-down current source or resistor to bias it.

Never said it did, you are quite correct though.

As an aside, impedance matching on certain digital connections can be far more important, because you can end up with reflections screwing everything up.

[aciddose]

i was actually referencing maus when he said "the resulting sound as "watery", "overblown", if bass gets turned up, it's quickly "booming" in most cases"

that sounds like.. maybe an output circuit which has a isolation transformer or maybe a dc blocking capacitor and depends upon a certain pull-down resistance on the output to keep the frequency spectrum flat.

using that kind of circuit is just silly.. that would've been the usual configuration for balanced outputs in the "olden days", but who still uses transformers these days? it just creates impedance issues of course and with the cost of semiconductors so low, going the balanced-buffer route is actually cheaper than the passive route.

i dont know what else he might've been referring to.. actually i should read where this discussion came from in the first place to get an idea what he might be even bothering to mention this for. ...and how did this conversation begin in a thread called "integer is king?"

[JonHodgson]

i was actually referencing maus when he said "the resulting sound as "watery", "overblown", if bass gets turned up, it's quickly "booming" in most cases"

that sounds like.. maybe an output circuit which has a isolation transformer or maybe a dc blocking capacitor and depends upon a certain pull-down resistance on the output to keep the frequency spectrum flat.

using that kind of circuit is just silly.. that would've been the usual configuration for balanced outputs in the "olden days", but who still uses transformers these days? it just creates impedance issues of course and with the cost of semiconductors so low, going the balanced-buffer route is actually cheaper than the passive route.

i dont know what else he might've been referring to.. actually i should read where this discussion came from in the first place to get an idea what he might be even bothering to mention this for. ...and how did this conversation begin in a thread called "integer is king?"

It started because he claimed he heard in audible similarity between the difference between integer and float, and the effects of impedance mismatches in analogue audio.

I'd be amused if it wasn't for the fact that I can imagine it getting round on the net and hundreds of hard working engineers and developers having to try to repeatedly explain to moaning customers and potential customers that their use of floating point (or integer, naturally there will be some who believe the floating point "sound" is better) is not detrimental to quality.

[JCJR]

in an ideal world we would have zero impedance outputs driving infinite impedance inputs. the next best thing is biased bipolar buffer outputs driving fet inputs which is what we normally use these days. with such a configuration there are no notable effects on the signal which should usually be the desired outcome.

Plug a phantom-powered condenser mic into a 30 foot balanced cable. Then plug the balanced cable into a balanced 10 megohm preamp.

For the final feat of magic, walk across the stage holding the mic, dragging the cable along behind you, and dig on all that LOUD mechanical noise that the high-z input picks up offa dat cable and squirts out the PA speakers!!

That's one practical example where infinite impedance inputs ain't always a good idea. Many more examples can be easily thunk up.

[aciddose]

"That's one practical example where infinite impedance inputs ain't always a good idea"

actually...

"the only place in audio where impedance matching really matters is either in obsolete standards or interfaces to passive devices."

if you have problems there, you're either using a microphone which uses a passive isolation transformer for driving the balanced output, or it is connected directly (100% passive) from the transducer itself. if you take advantage of the phantom power and run a buffer you should be able to run 10km of cable without too many negative effects taking place.

the problem would be that the mic output is high impedance, adding an output buffer using phantom power would reduce the output impedance. the transformer also causes impedance dependent frequency/phase effects which can be solved by using an active balanced buffer rather than a passive transformer.

most of the "vintage" mics are passive and without full buffers, obsolete technology. you might find that the modern mics sound "sterile".. that is because they are - you should have no impedance problems with modern gear.

the ideal configuration would be a modern buffered mic with and adc and wireless transmission. you'll get pretty much absolutely lossless and extremely low noise performance there.

oh, also there are different types of balanced cables. twisted pair, shielded, braided shield, etc. you should be using straight balanced cable with a single (non-braided) shield for the purpose of a mic, otherwise you'll have the cable acting as an inductor and the shield may pick up additional interference rather than "shielding" from it.

[JonHodgson]

Except that what's being described does not seem to be the microphone's inability to drive the load, but changes caused to the impedance of the connection through physical movement.
It's ages since I looked at this stuff, but your cable is going to be a resistive and capacitive element in the network, and its characteristics will change somewhat when it is physically manipulated (the capacitance at least), now what will be the overall effect of that on the impedance of the combined network (and thus the signal entering the load) when one end is connected to a very high impedance versus a more moderate one?

[bmanic]

_audibly_ different output of Waves TDM versus VST versions

This is probably a bug. Or, that company deliberately crippled the VST version to justify the extra $$$ that they charged for TDM versions, while the VST market already had reasonable prices.

You can listen to some TDM vs RTAS files here:

Waves C4 RTAS
Waves C4 TDM
Waves L2 RTAS
Waves L2 TDM
Waves RenComp RTAS
Waves RenComp TDM

Cheers!
bManic

[aciddose]

if you've got an infinite impedance input driven by a signal passing through a cable, the effect that the cable has on the signal will depend upon the capability of the driver to supply current. in order to drive a capacitive/inductive load, you need way more current than you normally would. using a sufficiently powerful driver, the effects of the cable can be pushed to levels which no longer matter.

with a cable connected to an infinite impedance input, the driver basically sees a capacitive and inductive network between the wires in the cable. so, you can model this by placing a capacitor + inductor circuit on the output of your driver.

when the signal at the driver's input goes high, it needs to push a lot of current into the cable in order to both overcome the inductance and charge up the capacitance. when the signal goes low, it needs to pull very hard likewise. so long as the driver has the ability to push and pull sufficiently hard (can require thousands of times the voltage desired at the input at the other end of the cable) these effects dont matter.

when the driver isnt capable of pushing or pulling at the currents/voltages required, you get distortion. most of the time, for reasonable lengths of cable you can run a buffer/driver which is capable of avoiding any problems. if you do run into problems something is probably unreasonable about the configuration.

[JonHodgson]

if you've got an infinite impedance input driven by a signal passing through a cable, the effect that the cable has on the signal will depend upon the capability of the driver to supply current. in order to drive a capacitive/inductive load, you need way more current than you normally would. using a sufficiently powerful driver, the effects of the cable can be pushed to levels which no longer matter.

with a cable connected to an infinite impedance input, the driver basically sees a capacitive and inductive network between the wires in the cable. so, you can model this by placing a capacitor + inductor circuit on the output of your driver.

when the signal at the driver's input goes high, it needs to push a lot of current into the cable in order to both overcome the inductance and charge up the capacitance. when the signal goes low, it needs to pull very hard likewise. so long as the driver has the ability to push and pull sufficiently hard (can require thousands of times the voltage desired at the input at the other end of the cable) these effects dont matter.

when the driver isnt capable of pushing or pulling at the currents/voltages required, you get distortion. most of the time, for reasonable lengths of cable you can run a buffer/driver which is capable of avoiding any problems. if you do run into problems something is probably unreasonable about the configuration.

You're describing a static system, but the cable's characteristics are affected by physical manipulation, therefore the overall response of the system will change when you move the cable about (which is what JCJR described), it's the effect of those dynamic changes which I was trying to ascertain.

[aciddose]

actually i'm not sure about that side of it, you're right.

i think that in that situation you're introducing currents at some point down the cable while the driver can only compensate for the properties of the cable at it's own end. it sees the cable as a l/c/r network. when you introduce the currents at some mid point in the cable the driver isnt able to compensate correctly since it cant "see" those currents until it is already too late. the result would be that the output of the driver at the time when the currents were introduced was correct for a static cable, but will now be over or under compensating. you'll also get currents bouncing back and forth through the cable creating an oscillation as the driver attempts to stabilize it's output.

that is normal for a l/c/r network. if you change the properties of the network or introduce currents at any point within it, you'll get some oscillation before the circuit will stabilize again. i'm not absolutely sure, but i think this is indeed an unavoidable problem since the driver can not "predict" the changes in the cable, so what you need to do is adjust the properties of the network in order to reduce the level of oscillation. sucking current at the far-end of the cable would produce that effect, so that is indeed the purpose of using a low impedance input in these situations. what you're really doing though is not using a low impedance input, you're rather attaching a pull-down resistance before the high impedance input.

i've never experimented with very long lengths, i've really only done 40ft, so i dont really know what it takes to create major issues. thinking about it though it seems like you'd need to do a ton of un/coiling and moving about to create substantial currents in the cable. emi i'd think would be the real issue and that should be possible to solve with sufficient shielding.

anyway if somebody with more experience wants to chime in and talk about some specific situations and problems/solutions with long cables i'd love to be educated in this area. this is far more interesting than the float vs. int junk.