Source-aware forward modelling and detection limits of acoustoelectric neural recording

Objective. Acoustoelectric recording uses focused ultrasound to select the region sampled by scalp or implanted electrodes. We relate its recorded voltage to source structure, terminal conditions and the carrier phase of the focus, and identify configurations that favour neural recording. Approach. We compare measured and simulated acoustoelectric fields in saline at seven incidence angles. We derive the first-order model from conservation of impressed and conduction current and relate it to conventional reciprocity. A finite-source benchmark examines how the travelling-wave focus weights distributed sources. Anatomical head-model simulations with six active cortical columns test source selection and detection limits. Main results. The interaction coefficient of saline is k_ae= (0.95 ± 0.11) × 10^-9Pa^-1, consistent with theoretical predictions, and predicts a second phantom's response without refitting. For fixed impressed currents the complete first-order model recovers the conventional lead-field integral exactly. The carrier phase favours sources whose sensitivity kernel changes sign within half a wavelength along the beam. In the lead-field phantom the simulated peak response under current-clamped terminals, the analogue of a fixed impressed source, is an order of magnitude smaller than under voltage-clamped terminals. In the head model a focus selects one of six active columns, and the conventional lead-field integral reproduces the complete solution's scalp topography. At the FDA intensity reference an idealised model of contacts 2 to 5 mm from the cortex gives 3.4 to 53 nV at 500 kHz, whereas the scalp signal of the 1 MHz focus is 3.6 to 17 pV. Significance. Carrier frequency, beam orientation and contact position can be chosen to favour laminar cortical sources, including configurations that cancel substantially in conventional recordings. Scalp recording trades selectivity for sensitivity, whereas implanted contacts at 500 kHz are predicted to detect steady-state activity within 0.02 s to a few minutes, so focused ultrasound could add steerable sampling around them.