VYGER turntable, VISION tonearm, TITAN vacuum controller, and ME

[BjörnÖsten]

For the last several months I have been working in partnership with Mr. Giuseppe Viola at VYGER in Italy.

My role is to add a new technical dimension to the research, design and manufacturing capabilities of the company, particularly with respect to electronics and compressor design. Mr. Viola is still the designer and the technical lead on, and the builder of, the turntables and the tonearms.

My background is in defence-based technologies with military application, including the use of artificial intelligence and low earth orbit surveillance techniques. Previously I worked for Lockheed Martin, among other leading defence companies. I then moved on to work for some leading brands in the high-end audio industry.

I hold a PhD in physics, two bachelor's degrees and two masters degrees spread over physics, engineering and electronics (both analogue and digital). Over the past few months I have been busy designing a new controller and compressor system for the VYGER family of products.

We are proud to announce the pre-launch of the TITAN, a 32 bit dual core machine learning capable control system. Our advanced compressor control algorithm learns how to provide the precise amount of air for your specific VYGER turntable system based on multiple factors.

This new system will achieve reliable and consistent operation. Furthermore, the user interface is very intuitive and the user experience is improved -- a button press will increase or decrease the compressed air and vacuum to the desired amount, allowing the user to test different settings based on his/her setup to achieve the best sound reproduction.

This new system will also be available as a direct upgrade to your existing VYGER system. A new controller and compressor can be deployed and directly coupled to your existing system using the same tubing, and the system will automatically configure itself to your turntable.

The system is also a connected device, following all the protocols of IoT. If a wi-fi connection is available the system becomes even more powerful with its ability to writeback to VYGER with any issues, and also take data from the server to ensure optimal operation. The system works without wi-fi as well; the network connection simply adds additional capabilities.

Giuseppe and I are excited to announce these developments!

Please feel free to ask any questions you may have.

 

[Ron Resnick]

This is very exciting, Björn!

What are some of the factors that the compressor algorithm will "learn" related to a user's specific Vyger? How does that work?

Does this system affect only the air pressure in the Vision tonearm, or does it affect the air pressure of the air-bearing system for the platter, as well?

 

[BjörnÖsten]

This is very exciting, Björn!

What are some of the factors that the compressor algorithm will "learn" related to a user's specific Vyger?

Does this system affect only the air pressure in the Vision tonearm, or on the air-bearing system for the platter, as well?

Hej Ron,

So the algorithm takes vectors from the environment and then adjusts. The concept as with any form of machine learning is to reduce errors to zero, as the system uses a very advanced PID controller with multiple sensors we are able to use the input in order to maintain great accuracy with respect to the set point.

However just getting and maintaining the set point is not by any means where it ends. We also want to achieve this in the most efficient way possible, so the system is also learning how to tune the PID and motors to provide the most efficient and sustainable delivery, which preserves the mechanical parts and ensures the longest life.

Once the system has a high degree of repeatable confidence it will then lock these setting and add them to a lookup table that the system can dynamically query as it feels necessary to ensure the error correction is robust.

The approach is not to look specifically at the arm and platter in isolation, it is to look at how the system is performing holistically and then tune accordingly as an integrated system, for instance, the platter has 3 forces (massive over simplification); gravity and vacuum which are almost acting on the platter in the same way, and then we have the air supplying the fluid layer in the frictionless bearing system. We need to look at all those forces and how they are interacting at a nano second level to ensure that the platter, arm and vacuum are working with unity to ensure our users have the best listening experience.

Björn

 

[Ron Resnick]

Thank you for your speedy reply!

What is "PID"?

When you write: "learning how to tune the PID and motors" what motors are referring to? Motors in the air compressor?

After the system takes various factors into account and achieves a setting and locks the setting, then is that the continuing air pressure setting? There is no on-going, servo-style, air pressure adjustment, correct?

 

[BjörnÖsten]

PID or proportional integral derivative, is a form of control theory used within engineering where you have to precisely control a set point. Your car when you engage cruise control is operated by PID.

There are no servos or moving parts controlling the motors, there is a PID algorithm that provides a signal to the control of the motor that will control the motor to a set point.

A simple PID controller however in this application is not sufficient, we need to take other factors into account and therefore we look at other methods in order to achieve this.

I do not want to sound pompous, however the algorithm is highly complex and quite difficult for me to explain the working of it, you may well not actually be asking that, but my autistic brain will automatically revert to the way I can understand it and I am terrible at times explaining things in a layman manner.

Bottom line is we want to avoid having to make millions of corrections, we want it it learn how to achieve the target in a stable way and then repeat this method, in doing so we reduce the error and can make very minor corrections only when the external or independent variables change.

Sorry if I went hyperbolic!

Björn

 

[BjörnÖsten]

And yes, to answer your other question, the motors in the compressor as the air pressure and flow is proportional to the RPM of the motors.

However air is not equally proportional, which the algorithm is also working with as a vector, compress anything requires force and the force required is not inversely proportionate to the effort, especially with air which will be acting on the mechanical parts of the compressor it self, the more rigid the air becomes the more effort is requires to increase the compression.

You can as an experiment try this with a sponge, at first you will be able to compress the sponge with large amounts, however there will be a point where an order of magnitude more effort is required to achieve a very small amount of incremental compression.

 

[Ron Resnick]

I think I understand. So there is no constant, servo-style air pressure adjustment or correction going on in real time? (I'm thinking of direct drive turntables that use a servo feedback system to constantly "hunt and peck" to make small corrections to the speed of the platter.)

 

[morricab]

Hej Ron,

So the algorithm takes vectors from the environment and then adjusts. The concept as with any form of machine learning is to reduce errors to zero, as the system uses a very advanced PID controller with multiple sensors we are able to use the input in order to maintain great accuracy with respect to the set point.

However just getting and maintaining the set point is not by any means where it ends. We also want to achieve this in the most efficient way possible, so the system is also learning how to tune the PID and motors to provide the most efficient and sustainable delivery, which preserves the mechanical parts and ensures the longest life.

Once the system has a high degree of repeatable confidence it will then lock these setting and add them to a lookup table that the system can dynamically query as it feels necessary to ensure the error correction is robust.

The approach is not to look specifically at the arm and platter in isolation, it is to look at how the system is performing holistically and then tune accordingly as an integrated system, for instance, the platter has 3 forces (massive over simplification); gravity and vacuum which are almost acting on the platter in the same way, and then we have the air supplying the fluid layer in the frictionless bearing system. We need to look at all those forces and how they are interacting at a nano second level to ensure that the platter, arm and vacuum are working with unity to ensure our users have the best listening experience.

Björn

And once your turntable becomes self-aware does it seek and destroy all digital in your house?

 

[BjörnÖsten]

And once your turntable becomes self-aware does it seek and destroy all digital in your house?

Ha..

I am sure it is just a coincidence that we have called the backend system "Skynet" and also labelled all digital devices "Sarah Conor" to the algorithm.

Nothing to see here at all!

 

[morricab]

PID or proportional integral derivative, is a form of control theory used within engineering where you have to precisely control a set point. Your car when you engage cruise control is operated by PID.

There are no servos or moving parts controlling the motors, there is a PID algorithm that provides a signal to the control of the motor that will control the motor to a set point.

A simple PID controller however in this application is not sufficient, we need to take other factors into account and therefore we look at other methods in order to achieve this.

I do not want to sound pompous, however the algorithm is highly complex and quite difficult for me to explain the working of it, you may well not actually be asking that, but my autistic brain will automatically revert to the way I can understand it and I am terrible at times explaining things in a layman manner.

Bottom line is we want to avoid having to make millions of corrections, we want it it learn how to achieve the target in a stable way and then repeat this method, in doing so we reduce the error and can make very minor corrections only when the external or independent variables change.

Sorry if I went hyperbolic!

Björn

Serious question: how is then able to achieve optimal results when things like needle drag are completely arbitrary and variable from recording to recording? I can see how it might come to learn the specifics of the motor/electronics interface (torque and PSU pulsations are perhaps predictable, for example) but how does it "learn" something that is highly variable and for all practical purposes random? I could understand that if you played the record through the first time and had record of the groove modulation through that record then you could 'learn" to anticipate the need for more or less torque to maintain the most precise speed. But this would need to be done with each record played if done that way.

Anyway, interested to learn more if you are willing to share.

 

[morricab]

Ha..

I am sure it is just a coincidence that we have called the backend system "Skynet" and also labelled all digital devices "Sarah Conor" to the algorithm.

Nothing to see here at all!

My worst fears confirmed...death by analog!

 

[BjörnÖsten]

Serious question: how is then able to achieve optimal results when things like needle drag are completely arbitrary and variable from recording to recording? I can see how it might come to learn the specifics of the motor/electronics interface (torque and PSU pulsations are perhaps predictable, for example) but how does it "learn" something that is highly variable and for all practical purposes random? I could understand that if you played the record through the first time and had record of the groove modulation through that record then you could 'learn" to anticipate the need for more or less torque to maintain the most precise speed. But this would need to be done with each record played if done that way.

Anyway, interested to learn more if you are willing to share.

VYGER is a high mass system, plus the issue with such vectors as needle drag would inherently attributable to the platter drive motor and not the compressor, as there is not way the friction moment and down forces of the entire arm assembly and weight of the vinyl being played would be adequate to have any material affect on the performance.

Now if we expand this to controlling the velocity and speed of the platter, this is a very technical piece of theoretical work, as with the VYGER for example the vinyl is coupled to the platter via vacuum and then can assume it to have the same high mass as the patter. A high mass platter is not solely reliant on the coupling of the drive motor to maintain the speed and velocity, and the inertial reference in respect to the arm is dynamic.

Any needle drag acting upon the system would be infinitesimal when we compare this to the angular kinetic force of the platter, such in the boon of a high mass system.

 

[spiritofmusic]

Ha..

I am sure it is just a coincidence that we have called the backend system "Skynet" and also labelled all digital devices "Sarah Conor" to the algorithm.

Nothing to see here at all!

Just don't go past a T2 version, the third onwards Terminator films really are awful.

 

[Rhapsody]

Bjorn, Giuseppe and myself are working on a daily basis with respect to many new VYGER developments which will unfold over the next 6-12 months.

By June 2023, the new Titan interface control box along with the new pump/vacuum system will be available for demonstration in the US at the Rhapsody Dallas (driving M9s), Portland and Brooklyn locations.

Stay tuned....lot's of interesting happenings at VYGER and the combination of Giuseppe and Bjorn future developments will be interesting for sure.

 

[tima]

I think I understand. So there is no constant, servo-style air pressure adjustment or correction going on in real time? ...

I'd like to hear the answer to this question.

 

[Rhapsody]

Bjorn will advise with respect to this question if he chooses, although due to pending patent activity the specific engineering details of how the VYGER control system works is not being publicly discussed at this time.

Minute details of how the new VYGER control system operates can be and will be explained in private conversations to interested VYGER potential clients.

 

[ALF]

…with a NDA, of course

vbw,
-a

 

[BjörnÖsten]

I'd like to hear the answer to this question.

Hi Tima,

@Ron Resnick thanks for the question, I thought I had kind of answered it but I may have carried way with the detail a little and it became lost in this.

So there is no servo or mechanical parts at work for control, the system you described it quite linear and based on IFTTT technology, where as our system is anticipating the outcome and learning from that.

So as an analogy, your are sitting in your car and you have to reach a set point of 100 KPH, you could floor it all the way to 100 then release the throttle, but this would lead to overshoot, once you fell behind the 100 mark you could floor it again and so on, this would cause a huge oscillation, not that we do not have this method of control in engineering, your oven uses a form of this method called bang-bang control, but this is not adequate for our application.

So instead if we got from 0 - 60 at full acceleration, then our error goes from 100 to 40... we can then adjust the accelerating force (your foot on the pedal) and anticipate what force we would need to apply in order to get to 100.. without overshooting.

Once we have attained our speed we will want to keep that constant, so we may encounter a hill, going up the hill we will need to apply more force on the accelerator pedal, we have already gained information from earlier, and we can produce a dynamic acceleration based on the speed in which our error is decreasing which we can map to the force of the pedal.

The machine learning aspect is for the system to understand based on a certain condition what is occurring and the best method to apply.

There is constant monitoring via the air pressure sensor, however this has a dual purpose, it not only provides input to the PID controller and machine learning model but it also displays to the user what the current pressure is. Once the system has stabilised it will only change if there are any movements to the stability, which will be infrequent as the system is robust, this negates any form of oscillation in the air supply.

The entire system is Microcontroller controlled, via a dual core module.

Please let me know if this has answered you question, if not I am happy to elaborate further.

Björn

 

[Rhapsody]

…with a NDA, of course

vbw,
-a

Possibly depending on the requested detail, but I think Bjorn has explained how VYGER approaches the control of the VYGER TT's adequately for a user to understand, without digging further into specific design details.

 

[strappoacustico]

The idea that the details of how the new VYGER control system works can and will be explained in private conversations to interested potential VYGER customers is rather "curious".
From what BjörnÖsten hinted this Titan is supposed to be an advanced compressor control algorithm that learns how to deliver the precise amount of air for the system based on multiple factors.
I hope we can clearly understand in the future what the factors are and how they affect the operation of the machine.

From his words we can understand the effort put into making the purpose of this implementation be reliable, coherent and able to guarantee the functioning of the turntable over time.

However, it is important that when everything is ready, the operation can be explained clearly and exhaustively to the audiophile user. Now if it is true that not all users have to be physicists or engineers, it is equally true that even very complex topics can be explained. There are many examples in this sense and I mention one that is in history or the famous publication "Relativity: Divulgative Exposure" by Einstein. With it, despite the topics covered, he addressed a wider audience, and not his fellow scientists, and surprisingly managed to make everyone understand the meaning of his studies.

Now, given the due differences, I believe that in this context the same can and must be done not only to understand the reasons and effects of these implementations but also to avoid increasing doubts about the ease of use that has always accompanied the fame of this turntable.

I think the undoubted qualities of this product deserve this attention. I wish everyone involved a good job.