Testing a Faster Wheel - Is the Music Tighter? 

Next year: I'm giving up on the Marbel Machine 3 because I couldn't get tightness below 2 nanoseconds.

Testing your theories in the prototyping phase will increase the odds of successfully creating a functional marble machine! Great job Martin! The only thing I might add to the conversation is DOUBLING the speed of the programming wheel ONLY decreased the std dev by 1 ms total. This slight advantage in std dev seems too costly from a music programming perspective. You may need to create a decision matrix to determine which route really is the best option to achieve your goals.

What you missed (or didn't mention) was, that there are diminishing returns. Look at your results: 2x1.41=2.82 > 2.37 . This means that in absolute terms, yes it is tighter. But the gain is not linear, it has diminishing returns. So you might find a point where working on the tolerances or increasing the spring strength (make the mechanisms more snappy and settle faster) might have more of an impact than pure programming wheel speed. on the other hand, lessening the spring strenght might reduce the drag on the programming wheel (it will flex in itself and the gears if it has to exert power, that might lead to increased tolerances that you can't see while the parts are not running if you look only at the machining), so there might be another tradeoff that is worth looking at.

The older marble machines have shorter, smaller pins and the wheel turns much slower. If you double speed your loosing half your playable duration for a minute change in accuracy. Good to know but i think as you know its probably a better idea to just make it slow. Especially since you can improve the tolerance on your final tour ready programming wheels instead of 3d printed clip ons on a 3d printed wheel. Imagine in that case how much smaller the timing difference will be.

I think you might be close to the sweet spot already. The original speed with spiral/stepped tracks looks very promising

I really enjoy all the testing Videos! But I wanted to give my thoughts on this one. Variances in the range of 2 ms are practically nothing to the human ears and brain. In my experience, very few people Start noticing the delay at around 3-5 ms, most at around 10. So doubling the speed of the programming wheel (and halfing the Song time) just to get an already amazing timing from 2 ms down to 1 is nowhere close to being worth it. Noone would notice the benefit of having the fast programming wheel, but only half the length of the song will have a pretty big impact on the possibilities of the entire machine.

I think that you should test if the drive is tight with variable load. Since the amount of note played and so the marble that get lifted is not constant the drive will receive different load, depending on where you are on the song. On the mmx this was not an issue because it was hand cranked and so you compensate the different load with a higher torc from your wrist. This is not the case on the new machine, so it MAY behave differently

How you've approached this challenge this year makes me super excited for what's to come. There is a noticeable shift, which seems more sustainable, realistic and constructive in the long term. I verifiably believe!

Faster wheel will give tighter timings, but cause you to need a larger wheel making manufacture more difficult and less programming time.

I have loved your experiment videos perhaps a bit more than the CAD videos. That being said I am always super excited whenever I see a new wintergaten video on my feed. Thank you so much for sharing the process with us it always gets my mind racing with ideas of weird contraptions.

If the marble is being dropped twice as quickly, wouldn't any measurement of variance half due to it being spread between two pulses rather than one? The slow test should have had two triggers so that the resultant drop rate was the same. I think the result indicated no significant variance.

May I suggest using bike parts like a brake disk and brake pads for the governor. This way you can have a "more constant" pressure, force, and timing from it with easily changeable replacements for parts that WILL wear down and mess the timing.

Thanks for sharing Martin! Always a joy to follow along in your endeavors! Happy New Year!

I love the testing and project prototyping videos and explanations :D

Be careful with going fast; mechanical parts have inertia and will "float". Will you have to worry about this? Probably not, but do keep it in mind.

Happy New Year Martin, look forward to seeing all of this hard work and research pay off in a beautiful new gen Marble Machine!

5:30 These videos are less all-over-the-place than then MMX videos. The way Martin worked haphazardly on things that took his fancy on that project was always bemusing to me. This new method of working - real engineering - is far superior, and much more interesting to watch.

I love this series so much. Hope you never give up!!! Happy new year!

Very nice progress! I'm looking forward to the future!

Thanks for the thought provoking video as always Martin!

Thanks for video. Always a pleasure to watch.

You've said it already, but it's worth saying again - the biggest lesson from these tests has to be, don't try to reinvent the wheel. Lots of very smart people in the past have looked at many of these problems and solved them.

i love , more than your results , your process , your approach, your curiosity, your enthusiasm. thanks for continuing to share your journey

These design prototyping videos have been super super interesting!!! Please continue with them in the new year

HAppy New Year Martin!! Upward and onward!

I want to see a test with more marbles per turn, the load of the marbles and mechanism has to stay below the pre-load on the governor friction plate or things will begin to slow down. The power side of the system will need to be quite beefy if there is going to be a lot of load on the machine. Given that the governor is friction and heat based it's paramount that you find a material that will remain consistent even if it heats up during a quiet part of the program. If your braking material changes coefficient through the operating range this will shift the tempo depending on the program load.

The videos have been VERY illuminating of the inside thought process of the project. Keep them coming (You could make the videos longer and more indept ;) )

I am still confused why music has to be tight? It just needs to be in sync with itself rather than to a bpm. And if it's slightly out well that's where musicians can make up for it. Maybe all you really need to incorporate is a conductor's baton on the machine to keep the beat.

Happy New Year Wintergatan, you are my favorite mad scientist!

Happy New Year, Martin!

Great video as always! It might be worth considering long-term and short-term tightness separately. It could be that a faster programming wheel is tighter for individual notes because inaccuracies in pin placements matter less, but not tighter for maintaining BPM over a long duration!

awesome research and testing to build! Happy New year!🦌💌❤️👍🎼🙏

Love the videos Martin! So happy I’m early!

Impressive work! Great success 🎉

Super excited for the processing. It's great to see the prototyping as it happens.

Once again we are all with you Martin! Keep it up

I am so very excited for this project. Watching the experiments and seeing the results is so satisfying! This is the kind of stuff they should be doing in schools to awaken interest and light the fire of inquisitiveness in young minds!

I am so pumped up for this series. It's a bit like watching you grind XP and level up in real time.

I'm not a musician, so maybe I'm way off base here. But I wonder if "tightness" really is the be-all-end-all. I know when it comes to jazz for example, there's a lot of expressiveness that comes from playing music just *slightly* "behind" the beat. Some people even do it deliberately. I wonder if "tight = good" is essentially a bias that comes from music made with a program rather than being played by a real person. (Of course I know you're a very accomplished musician -- it always blows my mind how you can pick up an instrument and make something up when testing these assemblies). I guess my thinking is like...even if the machine is a little "loose", maybe that's fine? The wheel obviously loops, and if I'm not mistaken, all the instruments are on the same wheel, so it's not like the programming will get progressively out of sync. I guess the worst case I can imagine is that the music might drift slowly across the course of the song, but I wonder if it's actually a problem (maybe it gives the song character ;) or adds a human element to what is otherwise a very mechanical result). Is there a particular (cascading?) problem that is solved by tightness? Could that problem be solved in a way that doesn't require computer-level precision in the music? Just curious about what you see as the endgame in this particular area.

Solution : Make the circumference of the programming wheel proportionally larger. In mathematical terms, you'd need to double the radius of the programming wheel if you double its rotational speed, but you can compromise : You gained about 55% in precision when doubling the speed, which means that you can also make your programming pins proportionally shorter without losing precision from the original radius.

Happy new year Martin ❤🎉

Feels like more and more procrastination with stuff that probably don't matter ... I might not have the full picture, but been seeing more and more "tightness" and "divergence" and other mathematical terms that honestly just makes me feel like it's chasing the wrong problem

Have been following for a long time. The approach that you are taking to mm3 is much better and I’m really enjoying that you are taking your time to explore all the elements of the machine before trying to create the machine. Every video I’m feeling more confident that you are really making good progress and will end up with much more versatile machine than before. I’m quietly hoping that you’ll end up making each instrument a self enclosed module, just plug into the power and timing mechanism, could end up with a whole orchestra. Makes me wonder just how many channels the flywheel can support and if adding more channels will alter the tightness.

Love to follow your designs and experimenting for the marble machine. Little tip for working with data sets, you had a diff. column which is great, but you should use it to find outliers and then try to find the cause of this. Would have allowed you to find the error in your initial results😊

Happy new year!

after checking on you after 7 years when you made your Wintergaten machine, its good to see you still uploading

I am super excited about your project - in fact I have been, since the first iteration of the marble machine. I think 1.38 ms is beyond insane in terms of tightness. In my DAW i have to live with 128ms latency, and up until now I never ever had an issue with it.OK, granted, it is a steady delay, so no fluctuations here. But being away a tenth of a second (me) in regards to a thousandth of a second (you) from ultimate tightness is - mind boggeling :)

Fun fact: this is also not a new problem. Something you could draw on for inspiration for finding the best compromise is the speed of recording media like tape. Tape recording of analog signals has this exact same problem the faster you pull it past the tighter the signal capture can be. Modern tape has digital signals recorded onto it. Imo this is closer to your drum design than analog tape recording. It may be possible to mathematically find the point where you get the best return for the minimum speed increase.

I love seeing the prototyping and experimentation phase. Looking forward to more in 24.

your videos haven't been all over the place. you're experimenting and trying to tick off all the boxes to improve your current 2 marble machines to make an even better version. i've enjoyed watching every little random vid you have posted. as i can see you are making progress, keep it up ^.^

I think to solve the programming length, you could substitute a program Belt for the wheel. This would allow for very long sequences (think player piano). I also think devising a 4 way chain link for that 'belt' that you could 3d print would be a creative solution. I'm imagining two types, a blank space (no note) and a play marker (play note). Linked on the 4 sides to the neighboring links. Thus programming (and changing) would be as simple as clicking the links together to form the belt. I envision that a 10m long belt could have a complete song encoded before repeating; even at the higher drive wheel speeds. While 10m sounds long and large amount of space, a clever arrangement of idler wheels would reduce the footprint that the belt needs. Cheers and Happy New Year!

i think you should test with asymetric placement/timing of notes. so you will see any problems if the symetry is hiding a problem.

Just having a little fun recording my own hypothesis before watching the rest. I suspect that the faster wheel will make negligible difference as it is the start and the end of the programming features that has the effect so it would seem the only error could be in its length. Now the fun part. God I love these videos!

Love your marble machine

Gott nytt år!

No worries about being all over the place. I just assumed you were building out the prototype and following your ears and intuition. It’s been a fun journey!

An happy new year man:

I think Martin's goal is not to finish the marble machine but the marble machine is just an excuse to do unnecessary tests that won't matter at the end. 😂

And still no idea what the power needs / losses from the instrumental section of the machine will be... Hint: it'll be huge seeing some of Martin's previously shared thoughts. Never seen an engineer start a project by sizing and tuning an engine without any idea what the application requirements are (torque, power speed, weight, etc) 🙄 Power / energy in this day and age is a solved problem. Bazillion mechanical solutions developed in last 200 years. But a marble machine with instruments ? Not solved and really cool. That's what he should be starting with, then design the power system accordingly, thanks to the modular-sections approach he presented a while back. Oh well, entertaining 8(!) years series anyway. Will be rooting for the 4th incarnation of this mythical beast 😂

Good grief that prototype setup is rickety as heck. Stoked that you’re pushing through Martin. Can’t wait to take my unborn children’s kids to your world tour premier

I don't understand why there should be variation due to the programming pin/bar when it's always the same? From my understanding: if there is variance, it is because the speed of the wheel is not constant. Even the governor will react with a delay introduced by friction and dynamics. A faster wheel will be tighter because the variation in speed is not tightly coupled to the speed itself.

Happy New Year🎉

Consider having more then one programming wheel, that turn at different speeds. By turning the playing of wheels on and off, and using different multiples you can get longer compositions out of the same length of wheels. You could even have a track on one wheel that starts and stops playing of other wheels.

Martin, you really shoud test usage of programing cards, like in the programmable looms like "Jacquard Design Machine" where you can have linked cards going thought a wheel and making contact with the machine, this way you could have much more music lenght to play

suggestion that have already been covered but what about multiple programing wheels? most songs are composed of paterns that repeat at various points in the song. while I dont know how complicated it could be; having multiple programming wheels which each have a patern being governed by some form of control programming wheel. potentially this would allow for longer songs as it would break it down into components which could be triggered by the control programming wheel.

Happy New Year.

i love the short update format

Methodical, piece by piece development wins the day, Martin. Always remember the race between the hare and the tortoise ... slow but steady wins the race. 🐢 Happy New Year to you! Looking very much forward to your progress in 2024.

I can't wait for the final result. And Happy NEW YEAR!! 2024!

Two Wintergatan Wednesdays in a row! Am I dreaming?!? Are we back to regular weekly updates?!? If so, that gives me massive hope that you are feeling more confident this design you're working on is going to be the winner!

Until now, I have not been physically sick for a very long time, but now my days are running together because of this. I see there's a new video here, so I must realize that today is my favorite day, Wednesday. This is the best calendar. 🤗

I know recently you had checked out music boxes to see how they could play longer songs. Could you do something similar to that, but also add in special notches along the wheel that you could line up easily? This way you can play the whole wheel, but also move it to certain spots for songs as needed.

great explaination

Hey, when are you gonna make more songs. Your album was amazing and sommarfågel is one of my all time favorite songs and I need more.

You could try experiementing with lower friction materials on the programming part, like metal against plastic, oiled, laquered wood, oiled wood, a bearing that is rolling, etc

Hey I know I'm like the 500th guy giving a suggestion. What I'd figure would really improve the music tightness, is instead of direct drive, some sort of temporary wind up mechanism that stores and keeps the tension more or less constant. Yk kinda like those old music boxes

Man the people with milliseconds hearing 20k years in the future are gonna be so stoked ! Just kidding, I'm super excited for the next machine and tremendously enjoying the ride so far, the prototyping is super interesting. Thanks for sharing and happy holidays to y'all

Hey Martin, you're an absolute inspiration for never giving up on your dream! I have a question for you. I'm in love with the arrangement of La Valse d'Amelie from MMX which is the only one song produced with MMX which I know. Why don't you use the MMX for what she can do? I'd love, and I think so everyone here on comment section, to hear more from it! Even if she has limitations she is still a capable wanderful piece of art!

keep up the great work cant wait to see the final product!

this is where the change from a programming "wheel" to a programming "tank tread" starts and it becomes closer to a player piano (which is a mechanized music playing device) Have you looked at player pianos and seen what lessons they have?

Have you looked at the structural integrity of the rotating parts at all? As an aerospace engineer, I can't help but get nervous for you as your speeds increase a bit. Keep up the good work. Loving this new process and approach you are taking. I hope you get into some analysis of the physics, stress, strain, fatigue creep/wear, life of the parts, and such things at some point. It is easier than most people think to do a quick check of such things, particularly if you are serious about the world tour hardware. Best of luck!

The prototypes are beautiful machines.

it is slowly coming together

To allow for a slow wheel speed and tighter music you could implement triggers. So that a single peg hits the trigger and spring force then drops the marble. This makes the dropper not care about wheel speed as it would be determined by the spring force. The main downside is in the adding a mechanism to reset the triggers. One way that this could be accomplished is by adding sections of teath that interface with a gear in the trigger to reset the trigger.

Great video. One small thing. In the beginning I think you mean Hypothesis. Theory is for once you have experimental data to backup your original hypothesis.

Increasing the speed gave the test wheel more momentum, so it was also affected less by the resistive force (no matter how small) of the triggering mechanism. I would test a heavier wheel at the lower speed to compare to. But the heavier the wheel, the more carefully it will need to be balanced.

it feels like Martin has the expertise to do incredible mechanical musical machines. Of course a marble machine is like the ultimate challenge so good luck with it

Suggestion: decouple the marble release from the music wheel, so the wheel determines which marble gets released, but the actual release is done by another mechanism, analogous to a clock in electronics. This would mean the wheel could have an error approaching half a beat (half a clock) whilst still having perfect timing. Because the (mechanical) clock now controls the timing, it should be possible to adjust the relative release points of each instrument to achieve close to zero smearing because of latency reduction. Happy New Year!

To increase the length of music, make the programming wheel a spiral instead of a circle. To address it, you will need a corkscrew read head which rotates next to it, reading further sideways as it turns and engaging the beginning of the roll again when it has made a full rotation. The control signals would have to be linked down the center of the read head tube and out the end. Perhaps this much additional mechanism would introduce too much error of its own to be worthwhile. But there are other possibilities for extending what is essentially a physical "piano roll".

You said that a faster programming wheel would decrease the timing error from the imprecision in the programming profile manufacturing. But that is not what you measure with this test. I think that a better approach for this is to just calculate the timing error. If the wheel rotates with period `T`, have a radius `r`, and the programming profiles have a standard deviation of `Δx`, the timing standard deviation should be `Δt = T*Δx/(2*r*π)`. Assuming `T = 11.3 s`, `r = 300 mm`(guessed this one), `Δx = 1mm` (for reference), the timing error should be `Δt = 0.006s`, so about 6ms of error for each mm of error in the profile.

I think you should consider what factors in the programming wheel interaction dominate the "tightness" and how can you best control them. For example, you've shown that rotating the wheel faster has a positive effect, but is this because the tolerances on your parts can be greater, the variance in wheel speed is lower, or some other factor? You may be accidentally uncovering some other factor that could help you in a much greater way.

Great video, like every time. However - there is one thing that is floating around in my mind: Suppose you reach the heaven of tightness music. Wouldn't the music then sound artificial? In general, in addition to crescendo/decrescendo, you need accelerando and decelerando to be able to really shape the music. So you need to be able to accelerate/decelerate in addition to loud/soft. Just to add one more element to your to-do list, you would need a lever to be able to apply a delta factor to the basic speed...

At this point I don’t even care about the tightness of the music as long as it’s roughly as accurate as a person it should be ok

A thinner reading head and pins, and that spiral (helicoidal) design you talked about in the other video, would allow for more music to still fit in a faster perhaps?

How exiting!

This Project will never end, it has become its own purpose....

I know this has been discussed before, but I think it’s worth emphasising again. I think your measure of tightness should take into account the BPM of the muic. Essentially you doubled the speed of the BPM here, so obviously the absolute time of the tightness will improve with a factor of two. That’s totally expected. But if you normalise it against the BPM the tightness hasn’t changed at all. I don’t know, for me at least such a BPM normalisation makes sense and should always be applied. Any thoughts?

Halving the play time for that tiny improvement seems like a bad trade off, I think a faster wheel could cause more problems elsewhere.

I have never once heard of faster things being more precise

@Wintergatan You can also play with the resolution of the programming wheel....