Definition, architecture, and system boundary
A PSG acquisition box is the patient-proximal signal interface and acquisition hardware in a polysomnography system. It converts weak potentials, analog voltages, or digital measurements from electrodes and sensors into synchronized data streams that can be recorded. Product names include headbox, patient module, amplifier, input box, and acquisition unit. One device may only provide connectors and impedance checking, while another integrates preamplification, patient isolation, pressure transducers, oximetry, analog-to-digital conversion, wireless transport, a battery, and local storage. The label on the enclosure does not establish its functions.
The boundary should be defined from the actual signal path: which unit receives the patient connection, where amplification, filtering, and digitization occur, which clock is authoritative, when samples reach nonvolatile storage, and which component continues recording if the workstation disconnects. The acquisition box, base station, recording computer, audio-video system, and scoring software may be separate or partly combined. Procurement and acceptance therefore require a system block diagram, a list of compatible accessories, and software versions, not merely a count of sockets on the enclosure.
An acquisition box supplies the signal foundation; it does not automatically make a PSG technically adequate and does not itself diagnose a disorder. Electrode placement, sensor selection, patient hookup, montage configuration, overnight observation, event scoring, and physician interpretation remain parts of the complete workflow. A live waveform is also not equivalent to vital-sign surveillance. Unless a system has a separately defined and validated alarm function, a PSG acquisition box does not replace an apnea alarm, intensive-care monitor, or bedside monitor intended to trigger immediate intervention.
Input types, connections, and channel configuration
Biopotential inputs serve microvolt-to-millivolt differential signals such as EEG, EOG, chin and limb EMG, and ECG, usually with active, reference, and common or bias electrodes. Respiratory effort belts, thermal airflow, nasal pressure, snoring, body position, photoplethysmography, oxygen saturation, and carbon dioxide may instead use auxiliary DC inputs, dedicated transducers, isolated serial interfaces, or digital ports. Each class has its own range, offset tolerance, bandwidth, excitation power, and safety path. A plug that fits mechanically does not establish electrical compatibility.
“Channel count” should be decomposed into differential biopotential channels, reference inputs, auxiliary analog channels, built-in pressure ports, oximeter ports, digital sensors, and event inputs, with simultaneous-use limits stated. A shared reference, multiplexed converter, optional module, or software license may constrain concurrent recording. A channel called multifunctional may also be unable to serve a slowly varying pressure signal and a high-frequency EMG under one configuration. Channel labels, units, sensitivity, sampling, and sensor serial numbers should travel with the study so that an export does not reduce them to ambiguous names such as CH1 or AUX2.
Connector design should reduce the risk of attaching an EEG electrode to a stimulation, DC, or nonisolated accessory port and of swapping left and right limbs, thoracic and abdominal belts, reference electrodes, or backup leads. Color, shape, keying, and screen prompts should agree, but color alone is insufficient. Home hookup also needs clear diagrams, tactile orientation, and positive confirmation. Adapter cables, splitters, third-party sensors, and service replacements can change leakage current, noise, range, or calibration. Only combinations validated as a system should be treated as compatible.
Analog front end, impedance, and calibration
A biopotential front end must recover a weak differential signal in the presence of much larger electrode DC offsets and common-mode interference. Input impedance, common-mode rejection, noise, dynamic range, patient isolation, and lead geometry interact; no single specification guarantees the acquired waveform. Unequal electrode contact converts common-mode mains interference into differential artifact. Drying gel, a sweat bridge, lead motion, or a loose common electrode can contaminate several channels at once.
Impedance checking helps identify poor contact, an open circuit, or imbalance, but a satisfactory number does not prove correct anatomical placement and does not exclude an electrode bridge, crossed leads, or intermittent movement artifact. The system applies a controlled test signal during a measurement, so the procedure must follow the instructions and return the channel to its recording state afterward. Test frequency, reference method, and displayed upper limit vary among devices. Values from two instruments are not necessarily equivalent, and the pursuit of an extremely low value should not cause repeated abrasion or skin injury.
System calibration should verify channel identity, polarity, gain or sensitivity, time constants, displayed filters, and response to a known input. An internal square wave or calibration signal can test consistency in the recording chain, but it does not prove patient electrode location or the physical conversion of every sensor. Pressure, oximetry, and carbon-dioxide modules still require their own functional checks. Any bedside zeroing, gain change, channel inversion, or reference reconstruction should carry a timestamp and audit history; raw acquired data must not be silently overwritten.
Digitization, filtering, and time synchronization
Analog-to-digital sampling rate, effective resolution, and input range should match the bandwidth and amplitude of the target signal. Analog antialias filtering before conversion prevents high-frequency energy from folding into lower frequencies. Later software magnification or interpolation cannot recover a component that was clipped, saturated, undersampled, or removed before digitization. Nominal bit depth describes coding steps rather than effective noise performance. A very high sample rate likewise cannot repair poor electrodes or a noisy front end and increases storage and transmission load.
Low- and high-frequency filters can improve viewing, but aggressive settings can attenuate slow waves, sharp transients, EMG, or respiratory contours and can create ringing or phase distortion. Acquisition filters, display filters, and processing applied after export should be distinguished. A reviewer needs to reproduce an approved viewing state and know which irreversible processing affected the stored signal. A notch filter can reduce mains-frequency interference but is not a substitute for repairing an electrode, grounding, cabling, or nearby-equipment problem; indiscriminate use may hide genuine content near that frequency.
EEG, respiration, oximetry, body position, audio, and video from one night require a common timeline. A multi-module system should define its master clock, sample counters, latency compensation, resampling, and dropped-packet markers. Wireless or network transport must not assemble streams merely from each module’s wall-clock label. Alignment should be checked at the beginning, end, and after a long recording, especially for external oximetry, carbon dioxide, PAP flow, and video. Daylight-saving changes, computer time correction, and network reconnection must not disrupt the monotonic order of acquired samples.
Installation, online quality, and artifact handling
Before hookup, staff should verify patient identity, protocol, channel template, accessory models, cleaning status, battery, storage, and clock. Connections are then checked with appropriate physiologic calibration or functional maneuvers. Eye opening and closure, eye movements, jaw clench, leg movement, breathing, breath holding, or position changes can confirm channel behavior when clinically safe and within the patient’s ability. The technologist should verify that each maneuver appears on the expected channel with plausible polarity; the mere appearance of any waveform is not enough.
Online quality review must distinguish real physiology, contact artifact, and system failure. Frequent problems include mains interference, sweat drift, electrode pop, lead movement, channel saturation, condensation or kinking in pressure tubing, reversed effort belts, low-perfusion oximetry, wireless packet loss, and base-station disconnection. Similar activity across many channels may indicate a common or reference electrode problem. A flat channel can represent quiet physiology, an open or short circuit, a frozen display, or the wrong range, and should be assessed with impedance, device status, and related channels.
The system should distinguish “acquiring,” “buffered only,” “committed to storage,” “transmitting,” and “completely uploaded.” If the network fails, a patient unit with local storage may continue recording, but remaining capacity, battery, disconnect time, and recovery outcome should remain visible. A system without local redundancy needs a prompt indication of data risk. Reconnection must avoid duplication, gaps, reordering, and presentation of an old waveform as live. The technical log should mark interventions, sensor replacement, time out of bed, and unrecoverable intervals so that the scorer knows when the evidentiary conditions changed.
Safety, data governance, and operational management
The patient-connected portion must control leakage current, insulation failure, foreseeable energy from defibrillation or electrosurgery, liquid ingress, overheating, battery faults, and mechanical traction. Electrode leads should not create strangulation or trip hazards, and a worn module should not lie under the patient or press connectors into the skin. When the unit connects to nonmedical computers, displays, network hardware, or chargers, isolation and protective-earth boundaries of the whole medical electrical system require assessment. Certification of one component does not establish safety for an arbitrary combination.
The acquired record can contain identifiable EEG, ECG, respiration, video, audio, and a detailed behavioral timeline. Device-to-base-station, workstation, and server links need identity control, least privilege, encryption, integrity protection, audit logging, and secure updating. Offline cache, export media, and service logs require equivalent governance. Cybersecurity controls must not destroy availability of the clinical record. An update should be tested for drivers, synchronization, channel mapping, and access to older data, with a safe degraded mode and manual recovery route for an attack or prolonged service outage.
Operational management includes asset and serial numbers, firmware, software, an accessory compatibility matrix, calibration and preventive-maintenance intervals, cleaning, and recall or field-correction tracking. Routine inspection can cover connector wear, cable shielding, enclosure cracks, pressure ports, radio function, battery capacity, internal clock, and storage media. After a board, converter, radio, cable, or driver is repaired or replaced, a power-on check is not sufficient. A regression test should cover the affected signal, safety, and data paths.
Acceptance, failure criteria, and future directions
Acceptance should derive from system requirements and use traceable signal sources to test channel mapping, polarity, gain, noise, common mode, frequency response, dynamic range, sampling, timing, and saturation recovery for both biopotential and auxiliary inputs. It should then test real sensors, audio-video synchronization, all-night storage, and report replay. Maximum configuration, shared references, module disconnect, wireless range limits, battery depletion, network outage, nearly full storage, and abnormal shutdown also need challenge testing. Seeing an ideal waveform at each socket does not demonstrate adequacy of the clinical configuration.
Technical failure criteria should be specified in advance, such as prolonged loss of essential sleep leads, simultaneous loss of respiratory channels, uncorrectable clock offset, incorrect sampling or units, file corruption, ambiguous channel identity, or failure to persist the data. Whether a brief nonessential channel loss still permits scoring depends on the study purpose, backup derivations, and valid duration. Software may label a gap or reconstruct a display, but it cannot recreate samples that were never acquired. A failure report should describe extent, cause, remedy, and affected conclusions, with repeat hookup, repeat testing, or a different device when needed.
Future designs may combine higher patient-side integration, wireless or optical isolation, modular digital sensors, edge-based quality assessment, automated connection verification, and interoperable data models. Algorithms may alert to rising impedance, swapped channels, saturation, or inconsistent respiratory signals, but they need transparent thresholds and human confirmation; “intelligent repair” must not conceal the original fault. A meaningful improvement reduces hookup error and missing data, preserves fidelity and synchronization, supports secure updating and long-term reproducibility, and is validated in the intended population and environment—not merely by adding channels, a wireless icon, or another computed score.