AMPS — Asymmetric modular pulse synthesizer

MPS-TMS works, and its limitation is structural: output fidelity scales with the number of distinct voltage levels available, and levels come from modules. Each module is a capacitor, a bridge, gate drive, and isolation. Wanting a cleaner waveform means wanting a bigger, costlier, less movable machine, and that is the reason instruments like this stay in laboratories.

AMPS starts from an observation that in hindsight looks obvious: there is no reason the modules should be identical.

The idea

A conventional modular synthesizer gives every module the same operating voltage, so the reachable output levels are evenly spaced and adding a level requires adding a module. Assign different voltages, and the same three modules span many more combinations — the classic version of this is binary or ternary weighting, where levels scale as powers of two or three.

The step here is not to use a fixed weighting pattern at all. Module voltages are optimized for the target stimulation waveform, because the waveform you actually need to produce is known in advance and is not uniformly distributed over its range. Spacing the levels to match where the waveform spends its time closes the gaps that matter and tolerates gaps where nothing happens.

Result

For a Gaussian polyphasic waveform, a three-module AMPS prototype cut total voltage distortion by 13.4 percentage points against a symmetric three-module system — and by 4.5 percentage points against a symmetric six-module system. Half the modules, better fidelity.

What it cost

Two things had to be handled that symmetric designs do not face.

Charging modules to unequal voltages. The prototype uses a cascaded-bridge topology with switched-capacitor charging, which lets the modules reach different voltages without a separate supply per module.

Transistor selection under very low duty cycle. Neurostimulation pulses are brief and infrequent, so devices operate far outside their continuous ratings. Sizing them by datasheet steady-state numbers wastes a great deal of capability, so the work includes characterizing transient overperformance at the duty cycles this application actually uses.

Status

Manuscript under review; the prototype is three modules on the bench, not a complete stimulator. The claim being tested is narrow and specific — fidelity per module — rather than a full system comparison against MPS-TMS.