Closed-loop latency and stability limits

Any closed-loop neural system, whether a self-paced motor BCI, adaptive DBS, or a bidirectional prosthesis, is a control loop, and control theory imposes limits that no amount of decoder tuning can escape. Loop delay, from sensing, computation, transmission, and neural conduction, sets a bandwidth beyond which feedback destabilizes rather than helps; and Bode's sensitivity integral, the waterbed effect, guarantees that suppressing error in one frequency band amplifies it in another. These are structural bounds, not implementation shortcomings.

For BCI this means on-implant, low-latency computation is not a luxury but a stability requirement, and it caps how tightly a co-adaptive human-machine loop can be closed. It also means adaptive stimulation that chases a fast biomarker can oscillate or entrain pathological rhythms if the loop is mistuned, a safety-relevant frontier for psychiatric and motor neuromodulation.