Master thesis : Head phantoms: Design and implementation of a physiological signal generator on a PCB for the validation of new polysomnography systems

Abstract

Polysomnography (PSG) is the gold standard for diagnosing sleep disorders and relies on the simultaneous overnight acquisition of several physiological signals. Validating new PSG acquisition systems before they reach patients requires a testbench capable of reproducing realistic multimodal signals under controlled conditions; yet existing solutions either target a single modality, lack configurable artifact injection, or remain inaccessible because of cost. This work presents the design, implementation, and characterization of an accessible, multi modal physiological signal generator on a printed circuit board for the validation of new PSG systems. The signals of interest in the scope of this work are EEG, EOG, ECG and EMG. The device replays real recordings from the Haaglanden Medisch Centrum sleep staging database through eight channels: seven single-ended (3 EEG, 2 EOG, EMG) and one differ ential (ECG). Cardiac and ocular artifacts are first extracted through a method-per-artifact pipeline, allowing them to be re-injected at user-defined gains. The hardware combines a 12 bit eight-channel DAC, a digitally controlled two-stage resistive divider sized per modality, and precision buffers. Three playback modes (raw replay, artifact mixing, and on-demand artifact injection), each with optional sleep-stage selection, are controllable from a PC ter minal without reflashing. Signals are stored on an SD card, and dry-electrode connection is supported through ten on-board conductive pads. The complete device costs approximately 200 euros. The generator presents an output range from 62 µV to 79 mV with a best-case noise floor near 10 µV, a faithful reproduction on the ECG differential channel across all configurations. On the single-ended channels, inter-channel coupling through the shared analog reference node limits faithful simultaneous multichannel replay below the millivolt range; the mechanism is analyzed and mitigation strategies are proposed. The supply noise is shown to be rejected by the differential output topology, making the generator effectively independent of the choice of power source. Overall, the work demonstrates that accessible, multimodal, hardware-level PSG validation with configurable artifact injection is achievable, and provides a documented basis for future versions.