128 places to be touched

The usual way to build a vibrotactile array is one driver channel per actuator, fanned out from a central board. It works beautifully at four actuators and becomes the whole project at forty: the wiring harness dominates the garment, the pin count dominates the board, and adding one more actuator means revisiting both.

VibraForge inverts it. Each actuator is sealed into a self-contained vibration unit with its own microcontroller, its own driver and an input and an output connector. Units daisy-chain: the control unit talks to the head of the chain, and each unit passes the stream along, acting on the messages addressed to it and forwarding the rest.

The consequences are all in the wiring. A chain of sixteen is one cable from the control unit, not sixteen. Adding a unit means plugging it into the end. The control unit carries eight connectors, so the system tops out at 8 × 16 = 128 independently addressed actuators — and the limit is the address space and the link budget, not the pin count.

The cost is that every actuator now contains a microcontroller, which is a real per-unit cost and a real per-unit failure mode. It is the right trade for a research toolkit, where the number of actuators is the variable you most want to change and the unit cost is the one you least mind.

The command path is split by what each layer is good at: BLE from the host for high-level patterns, because it is wireless and a wearable cannot be tethered, and UART down the chain for distribution, because it is deterministic and cheap. Measured end to end, that is 14 ms of BLE plus 2 ms of UART, with a new command admissible every 5 ms.

Those two numbers are the ones the editor above spends. Sixteen milliseconds is comfortably inside what feels simultaneous with a visual event; 200 commands a second sounds like a lot until you ask a pattern to change all sixteen units at once every 50 ms, which needs 320.

The eight frequencies are 123, 145, 170, 200, 235, 275, 322 and 384 Hz, and the spacing is the interesting part: each is about 1.18× the one below it. That is not an arbitrary ladder — it is geometric, eight roughly equal steps across 1.6 octaves, which is how a channel gets divided when the perception on the other end is closer to logarithmic than linear. Sixteen intensity levels sit on top of that, per unit, independently.

The evaluation is done on the right side of the transducer. It is easy to demonstrate that a driver commanded 235 Hz emits a 235 Hz signal; what matters is whether the mass on the end of it, pressed against a body, actually moves at 235 Hz — so the measurement is an accelerometer on the unit, and the answer is a clean peak in every condition.

Where the design pressure really is

The technical numbers make more sense read as a budget than as specifications. A vibrotactile pattern's information rate is (units that change) × (how often they change). The link gives you 200 changes per second to spend, total, across the chain. So:

• All sixteen units changing at once every 100 ms peaks at 160 cmd/s — fits, just.
• The same thing every 50 ms peaks at 320 — does not fit on one chain, and has to be split across two.
• The travelling wave preset, which overlaps two or three units so the sensation reads as continuous, peaks at 100 cmd/s and averages 79 — comfortably inside, and it feels smoother than the one-at-a-time sweep that peaks at only 40.
• The phoneme burst averages 9 cmd/s and peaks at 100. Averages are the wrong thing to size a link on; a pattern is a queue problem, not a throughput one.

Which is the useful design lesson, and the editor above makes it tangible: the patterns that feel best are usually not the ones that hammer the link. Continuity on skin comes from overlapping a few units, not from updating all of them quickly.

The assembly is deliberately short, and the toolkit ships a manual for it. In outline:

1. Power the control unit. Li-Po into the battery connector; a green LED confirms it. Chain into connector 1 to start.
2. Mind the direction. Every unit has an input and an output, and the arrow on the cap points input → output. Reversing it can short the driver board.
3. Chain them. Up to sixteen per chain, addressed 0–15 in order.
4. Attach. Slide each unit onto the compression sleeve, press-fit ring underneath, press until it locks. No sewing.
5. Run it. Python_Test.py for a quick check (needs bleak), or the GUI editor at GUI_Editor/main_app/app.py.

The one step worth pausing on is the second. A daisy chain with a direction is a footgun, and the mitigation here is a moulded arrow rather than a keyed connector — a reasonable call for a research kit built in small numbers, and the first thing I would change if it were going to be built in large ones.