Technical writing from the Ruten engineering team on closed-loop BMI, real-time signal processing, and neuro-rehab device development.
A walkthrough of the four-layer closed-loop middleware stack: from raw electrode acquisition through spike decoding to stimulation command dispatch, and why each boundary was drawn where it was.
Read article
When your stimulator fires through the same electrode array you are recording from, the artifact problem becomes the whole problem. This post covers the rejection strategies built into the decoder layer.
Running threshold-crossing detection and principal component clustering on a Cortex-M4 with 256KB SRAM requires different tradeoffs than cloud-side sorting. We walk through our embedded-optimized pipeline.
Device makers moving from research prototype to clinical evaluation face a protocol documentation and validation bottleneck that firmware infrastructure can either widen or narrow. Here is how middleware shortens that gap.
The safety limits for neural stimulation are well-established in the literature. What is less discussed is how those limits propagate from IEC 60601 into firmware constraints and why middleware is the right place to enforce them.
Every BMI research group ships a proprietary data format. The cost compounds at every integration point. We look at NWB, BrainFlow, and what a common transport layer for closed-loop middleware could look like.
Functional electrical stimulation for upper-limb rehab needs residual EMG as the intent signal. This post covers the signal chain: EMG amplification, envelope extraction, threshold gating, and FES command generation.
Electrode arrays from three different vendors, two amplifier generations, one stimulator. The adapter pattern that makes the Ruten SDK hardware-agnostic and why we chose it over a unified driver model.
A closed-loop BMI is only useful if the total round-trip latency from electrode to stimulation command stays under the physiological window. We break down every stage in the pipeline and what contributes to the budget.
Electrocorticography offers spatial resolution that scalp EEG cannot match, but the firmware requirements for implanted ECoG amplifiers introduce constraints around power budget, packet loss recovery, and safe-state handling.
The problem of inferring movement intention from neural population activity is the core decoding challenge in motor BMI. This post covers the signal features, decoder architectures, and update rate tradeoffs relevant to rehabilitation devices.
Static stimulation protocols plateau. Closed-loop systems that adapt stimulation parameters based on decoded neural activity have a fundamentally different interaction with Hebbian plasticity mechanisms. This post outlines the theoretical basis and practical firmware implications.