Deep Grooves Mastering - Feedback Transducer

[Bahndahn]

Is this system posted by Bahndahn a similar example to what is being discussed here?

https://cycling74.com/2015/02/04/transforming-room-sound-with-a-max-patch/

The time it takes for a deviation of air particles to travel through air to the obstruction [i.e. sound from snare-drum to panel of glass 5 feet away] is very substantial compared to the time it takes for a feedback signal to be amplified, polarity inverted, and summed etc. therefor the referenced system differs most notably in the time-frame domain, but the two are essentially similar.

There are a number of other differences I can think of between the two, but I am butting myself out in thoughts of this. Depending what way you look at it, it could appear more, or less similar.

Of course, most of the dialog on this form is geared towards making a feedback system that is spectrally transparent, however some crazy things could be done with selective positive & negative feedback, simultaneously. The cutterhead system behaves in a unique way when compared to other 'signal processors'. The 'effects' one could get out of experimental feedback could be very fun -- however that is totally off topic.

[Stevie342000]

What I think I meant is that you could carry out digital equalisation on the signal to simulate the same effect as you would get by using analogue feedback with feedback coils in the cutter head.

This I think is achievable, I am afraid latency in the digital or any other domain is beyond my understanding.

Given the cost of these heads I still think it is a mute point, even with feedback. Just trying to give people options here who may not have deep enough pockets to purchase a feedback head but could achieve it through digital equalisation.

It may of course need a full frequency sweep of each head (they should be fairly close given manufacturing techniques from head sample to head sample)which could be supplied with other specifications(data) which would allow (tell) people who purchased a non-feedback head which frequencies to boost or cut and by how many dBs.

This could help to make a more rounded end product and form part of the package. For those that now how to do the above it may not be necessary as you could do it in the analogue domain by equalisation networks to flatten response. Which you really should be doing any how to ensure that the cutter/replay chain is as flat as possible.

[electronrancher]

What I think I meant is that you could carry out digital equalisation on the signal to simulate the same effect as you would get by using analogue feedback with feedback coils in the cutter head.

This I think is achievable.

This would absolutely work - and is an incredible idea, by the way. Like you said - get the characteristics of the cutter in the type of material you are cutting, and over the whole frequency range.. But then it should be pretty straightforward to make a "correction" EQ which could be applied to future cuts.

Is this a thing? Now that it has slapped me in the forehead, I'm surprised that characterizing and compensating the cutterhead is not a standard part of every setup...

[Bahndahn]

What I think I meant is that you could carry out digital equalisation on the signal to simulate the same effect as you would get by using analogue feedback with feedback coils in the cutter head.

This I think is achievable.

This would absolutely work - and is an incredible idea, by the way. Like you said - get the characteristics of the cutter in the type of material you are cutting, and over the whole frequency range.. But then it should be pretty straightforward to make a "correction" EQ which could be applied to future cuts.

Is this a thing? Now that it has slapped me in the forehead, I'm surprised that characterizing and compensating the cutterhead is not a standard part of every setup...

This is certainly a possibility, but EQ, limiting, and compression will not cut it. To do this 'simulated feedback' idea is more effort than an analog option.

Bryan has nailed this on the head:

I'll chime in on this....
Doing this in the analog world is the only practical way to go. It's real time, with real measurements and adjustments done in real time. However, correction could be done in the digital world but only by developing a model - a stimulus and response model accounting for many (if not infinite) conditions. As we all know by now, it's not just about frequency response under a specific condition like one drive level, or one cutting force. It's a multi point, multi conditional response - and in the digital world it would be an attempt to capture a real life response to generate a simulation of the real world. With enough data under enough conditions it should be theoretically possible.

Antares has used this kind of "holistic" modeling method to develop models of many microphones for use in their Microphone Modelers. I own two of them. The result is you can select the microphone you used as the source, and select the microphone that better fits the situation (or you could never afford). Does the result sound different? Absolutely. Is it an absolute match of the Telefunken U47 I selected? I can't tell you - I have never owned one. But does the response change as level changes ... Yes. Is it correct? Not sure. Does the simulation change when I say the distance is 31 inches compared to 6 feet from the source? Yes. Is the simulation correct? Again, I can't tell you.

I guess a model could be developed for my home-made head - and if I ever make a second one with a hole drilled 0.01mm off, a new model could be developed. I just don't see it as being practical.

If a guy were to make 100 at a time on an automated production line, yes.... I think it's possible.

Bryan

[opcode66]

I still don't think this is viable. So many real-time and instance specific factors determine the head response. Developing a computer model could hypothetically do away with the need for feedback entirely. But, will it be as accurate? That is the thought I keep coming back to. I understand all of this can be done. But, if it isn't as accurate as analog summing, and would require a herculean effort to achieve, why? What would be the advantage?

Here is a short list of factors that I can think of immediately which are either real-time or instance specific that would make a computer model inaccurate.

- Age of the magnets
- Manufacturing inconsistencies of the magnets
- Age of the coils
- Minor inconsistencies in the production of the coils
- The ambient temperature of the room (transducers don't have the same response at all temperatures)
- The exact formulation and therefore cutting resistance of the material you are cutting into as that affects the movement of the transducers which real analog feedback identifies
- Ambient magnetic fields near the cutterhead
- Minor inconsistencies in the production of any moving part or spring. Unless you are willing to pay a king's ransom to have parts machined, they will not be 100% consistent and will therefore need to be accounted for on an individual basis.

The level of engineering required to be able to produce heads in any production run which would account for 99% of what is listed above would make each cutterhead cost something around 10K USD my guess if not more. To insure that every part is perfect and to specification would be unrealistically expensive. If you didn't manufacture to this level of precision, accuracy, and with perfect repetition on each head, then a computer model would inevitably be inaccurate.

Analog processing accounts for all of these variables. Besides, I can't imagine the cost of the test equipment and rig that you would need to build in order to derive a computer model in the first place. Well beyond what any of us will be doing in our spare time with our spare cash. You would have to be able to very accurately measure the movement of the transducers coupled to the torque tube (not on their own). Whatever you used to make measurements would have to be highly engineered so as to not affect that with which you are measuring (otherwise you are affecting the results). You would need to account for the whole frequency range, volume range, as well as in and out of phase data on both channels. That is a lot of permutations.

[markrob]

Hi Todd,

The factors you mention are exactly what any servo feedback system remove. It does not matter if this is done in the digital or analog domain. If the feedback is applied properly, the outcome will be the same. Working in the digital realm has pitfalls that do not exist in the analog world. There is a large body of work that addresses this and with proper application can be accounted for.

You can EQ an non-feedback head to to flatten its response (I have done just that with my design). This works, but suffers from the fact any changes in the system (some of which you noted) will make the head deviate in its performance. Your statements make me think you don't really understand servo control systems.

What is the difference between a summing op-amp with gain (present in the Caruso board) and a piece of code that periodically runs and calculates:

Error_Signal = Servo_Gain * (Actual - Desired)

followed by some IIR or FIR based servo compensation code

If you think that the A/D/A and calculation latencies would be too great (in the ms as you seem to think), look at these parts:

The ADSP-2146X can calcualte a Biquad filter in 4.7ns! They run about $30 in single piece quantities.

http://www.analog.com/en/design-center/landing-pages/001/sharc-benchmarks.html

http://www.digikey.com/product-detail/en/ADSP-21469KBCZ-3/ADSP-21469KBCZ-3-ND/2432859



Mark

[opcode66]

To be clear, there are two digital approaches being discussed. One involves modeling the feedback (and thereby removing the need for it), and the other is to use a realtime digital feedback interpreting system. Also, if everyone wants to continue to discuss this completely off topic subject, I suggest that someone start a separate thread. This thread was created to discuss my achievements with respect to making a feedback transducer using a coil and analog summing. Not a digital system. If you are planning to implement a digital system, please start your own thread. If not, then let's keep to the topic at hand.

In response to you Mark, I'm still not convinced. When you add the interpretation stage in, which you are glazing over, then you have way more than ns latency. Any latency 0.00005s or more will affect the audible range. You can mention servos all you want. That is like comparing apples to oranges. Servo are not used to represent audio. The innaccuracies of doing what you are describing dont matter and aren't readily noticeable for the application of making a servo motor move. Audio representation is not the same. I think we should agree to disagree. Until you or some other Troll actually creates an effective system fully digital, my opinion will likely not change.

[EpicenterBryan]

Also, if everyone wants to continue to discuss this completely off topic subject, I suggest that someone start a separate thread.

I understand Todd's point. The side discussion does make for a tremendously interesting thread though.
So would one of you guys be up for starting a new thread like "Digital Feedback Possible"?

Great stuff guys!

Bryan

[Stevie342000]

Also, if everyone wants to continue to discuss this completely off topic subject, I suggest that someone start a separate thread.

I understand Todd's point. The side discussion does make for a tremendously interesting thread though.
So would one of you guys be up for starting a new thread like "Digital Feedback Possible"?

Great stuff guys!

Bryan

I have sent a PM to the list owner to split the thread that way both may flow much better and keep the content of this thread more pure. You will find that another thread has been moved from the Treehouse as it was in the wrong place and that this index has a had a rename.


Steve in the UK.