Design of a Continuous-Output Switch-Access Interface for Users Whose Structured Motor Signature Is Sustained Engagement with Modulation: Detection Framework and Channel-Capacity Argument

Abstract

Switch-access interfaces for augmentative and alternative communication (AAC) assume the user can produce discrete tap-like activations. When a candidate user cannot, the standard clinical inference is that voluntary motor control sufficient for switch access is absent. This paper addresses a different structured motor signature—sustained engagement with within-engagement modulation (canonically, sustained finger-pad contact with within-contact force modulation)—as a viable control signal for a redesigned interface, under the taxonomy in which sustained-hold-with-modulation is distinguishable from pure-tap activation, undifferentiated resting contact, and rhythmic-tap (Morse-style) sequences. Methods. The paper specifies a detection framework for identifying the sustained-contact-with-modulation signature from video-and-audio recordings of pad interaction: MediaPipe-based fingertip landmark extraction yields a continuous contact-state estimate; two independent operationalizations of "press" are defined (velocity-reversal kinematic; contact-interval geometric); librosa spectral-flux onset detection provides an independent acoustic channel; and the sustained-contact ratio R_SC = N_onsets / N_presses quantifies how many acoustic events occur per press, with R_SC ≈ 1 the tap-like null and R_SC >> 1 the sustained-hold-with-modulation regime. Proposed detection framework. For recordings made while a device emits its own periodic audio (a drum-machine loop, a metronome, any stereotyped acoustic pattern), a loop-subtraction technique is specified that partitions in-contact audio onsets against the observed loop grid and yields a strict lower bound on subject-initiated events. Channel capacity. A structural argument grounds the interface specification: A binary-closure switch transmits at most one bit per contact event, whereas a force-sensing pad sampling at f_s with b-bit resolution encodes b · f_s · T bits per contact of duration T. At 8 bits, 100 Hz, and T = 1 s, the nominal capacity ratio is 800:1; after Shannon–Hartley correction for realistic noise (20 dB SNR) and motor bandwidth (10 Hz) the effective ratio is ≈ 66:1, falling to ≈ 17:1 only under aggressively pessimistic assumptions (10 dB SNR, 5 Hz bandwidth). Interface contribution. A continuous-output interface is specified, with the force-sensing pad (capacitive or piezoresistive) as the canonical transducer, to encode engagement onset, engagement duration, and within-engagement modulation, with a signal-processing pipeline that maps these dimensions onto a small AAC command alphabet and a proposed binary (yes/no) bootstrap protocol derived from modulated-vs-silent sustained holds. The non-thresholded output (≥ 100 Hz sampling, no in-device binarization) is the critical architectural choice. Deployment scenarios. Deployment is discussed separately for static-etiology populations (hypoxic-anoxic brain injury, post-status-epilepticus syndromes, severe cerebral palsy with preserved tactile response), where the interface may be the first channel with sufficient bandwidth to register the user's motor signature across any assessment in their lifespan, and progressive populations (notably late-stage amyotrophic lateral sclerosis), where it serves as a bridge channel that maintains direct-selection AAC usability during the period when tap fidelity degrades before sustained-contact modulation does. Additional deployment scenarios address users whose primary constraint is distal upper-extremity weakness or atypical wrist kinematics (contractures, hypermobility, congenital variants) and integration with standard clinical assessment batteries (CRS-R, Motor Behavior Tool, AAC timing protocols). The framework is stated in sensor-agnostic form: The force-sensing pad is the canonical transducer, but the channel-capacity argument and detection pipeline transfer to non-contact transducers (capacitive proximity, infrared / time-of-flight optical, millimeter-wave radar) for users for whom physical contact is contraindicated by tactile hypersensitivity, skin compromise, or pressure-sore risk, and to bio-signal transducers (surface electromyography from residual musculature or targeted muscle reinnervation sites) for users with upper-extremity amputation or prosthetic-limb deployment, for whom the multi-channel structure of the EMG record expands the achievable symbol alphabet by one axis per electrode. Combat blast-injury survivors (burn scarring, upper-extremity amputation, and blast-induced TBI frequently co-occurring in the same patient) are discussed as a cross-cutting clinical population for whom preserved visuomotor coordination typically renders the non-contact variant's distance-dimension vocabulary gain immediately usable. Epilepsy and rhythmic-involuntary-motor populations are treated as a first-class deployment scenario, including the severe case of continuous subclinical (electrographic) seizure activity in which involuntary micro-twitches are superimposed on every sustained-contact interval; frequency-domain separation of involuntary rhythmic contamination from intentional modulation, supported by a mandatory per-session diagnostic baseline and a ≥ 25 Hz Nyquist sampling floor that covers the full 1–12 Hz involuntary band, is the operational key to serving this cohort. Scope. The paper is a design and methods contribution; empirical validation of the detection framework and the bootstrap protocol on a recruited cohort is flagged as future work, and prevalence of the target signature in candidate populations is an open empirical problem.