This is about Tapeshnik. My slow-burn project that I'm working on for literally years, but most of that time it's just resting in a box, waiting for inspiration or parts. 

Here's the latest video from the series.

The idea is to make a fun-looking data drive that would be something you could find in a late 70s or early 80s computer. Maybe with random access. And with the spools doing fun things. And so that you could build 4 of them and run mergesort. Anyway... 

I started with a commercially available tape mechanism and developed some test program to read/write data. Those mechanisms are cool for a while, but then you realise that they still have belts, even the new ones are actually old, they can be hard to buy. Not that you need many in your life but there's grave digging and then there's grave digging. Also their command-wheel / solenoid control scheme is incredibly clever but slow and I don't think it's designed for random-access-tape kind of abuse.

Notes on the first design:

https://github.com/svofski/tapeshnik/blob/master/docs/NOTES.md

I grew bored of that version too soon and switched to developing a custom mechanism. I was inspired by some streamers from the 70s. 

 * TU60 DECassette https://gunkies.org/wiki/TU60_DECasetteTape_Transport
 * TEAC MT-2ST https://oldcomputer.info/media/teac/index.htm

There are also older versions of TEAC data recorder that have slow data rates and tape travel speed.

These drives are interesting because they don't have a capstan. They use biphase mark encoding, which is "self-clocking" to some extent and they are relatively insensitive to some variations in tape speed. Something I could try and replicate myself.

Notes on the second design:

https://github.com/svofski/tapeshnik/blob/master/docs/FUTURISTIC.md

To summarise my story so far: not so easily. My expectations of using SimpleFOC with BLDC motors on two spindles were probably inadequate. I can achieve a pretty stable general velocity, but the detonation/wow/flutter are terrible. And I think this happens because my hub motors still have cogging, there's no mass to smooth it out. The tiniest cogging at the axle gets amplified by the radius of the spool times 2 and this becomes a very funny space chipmunks audio filter if you try to listen to it.

When I came to that conclusion last time I sighed, carefully put the project in its home plastic container and stashed it away. The old time drives used massive motors with analog control so they probably were free from this problem.

I'm writing this because I'm getting some new inspo from yet another forgotten capstanless gem:

 * HP 82161A Digital Cassette Drive https://literature.hpcalc.org/items/2150

It's yet another capstanless drive, technically random access. And it only has two motors and a clever analog control circuit, which is documented in this brilliant service manual. It's worth a read even if you don't care about analog electronics, tape or calculators. Just get a breath of old school engineers' spirit.

New ideas

I still don't know what to do about my drive though. I don't think I'm ready to switch to analog control like 70s HP. I have several paths to explore:

• just ignore flutter and use slow super resilient encoding. after all it's a toy and I don't need to achieve high capacity or speed
• add a second clock track, compensate the flutter. I have a 4-track head.
• what if the control loop is just slow and can be tuned up?
• I use one rp2040 to control 2 motors, this could technically be a problem for SimpleFOC.
• also there is no current sensing, only magnetic position sensor AS5047: 14-bit sounds good, but it could be slow and readings can be skewed by misplaced magnets for example. SimpleFOC community seems to be skeptical about magnetic digital sensors for sensitive applications.
• use massive flywheels and belts to even out the flutter

If you're reading this and have ideas don't hesitate to reach out and comment even if you think they're silly.