World Sleep Industry ENCYCLOPEDIA

PSG amplifiers

2.2.3

Definition, function, and device boundary

A PSG amplifier is the front end of a polysomnography signal chain that receives and conditions patient biopotentials, principally the small voltages used for EEG, EOG, chin or limb EMG, and ECG. Through differential inputs it extracts the voltage difference between two electrodes and provides buffering, gain, common-mode rejection, bandwidth limitation, patient isolation, and usually analog-to-digital conversion. Some systems place the amplifier inside a patient-worn box; others separate the connector headbox, amplifier, and recorder.

The amplifier boundary should be defined by circuit function rather than a product label. An enclosure called a headbox or acquisition box may merely connect electrodes or may already contain preamplifiers. Respiratory pressure, position, oximetry, and digital sensors may bypass the biopotential amplification path entirely. System documentation should identify where each channel class is amplified, isolated, filtered, and digitized and which specifications apply only to biopotential inputs. A whole-system bit depth or sample rate must not be assigned indiscriminately to every sensor.

Amplifier fidelity is one necessary part of a technically adequate PSG, but it cannot correct misplaced electrodes, incorrect channel labels, missing sensors, clock misalignment, or faulty scoring. Impressive specifications also do not establish diagnostic or alarm capability; the output still has to be recorded by a qualified system and interpreted by trained professionals. Unless a particular product has a separately validated monitoring indication, a PSG amplifier is not a substitute for intensive-care ECG, long-term epilepsy monitoring, or a real-time apnea alarm.

Differential front end, common-mode rejection, and loading

A biopotential channel measures the difference between an active and a reference electrode. Similar voltages induced at both inputs by mains electricity and nearby equipment are common-mode components. Common-mode rejection ratio describes rejection under specified test conditions, but practical performance also depends on frequency, source-impedance balance, lead geometry, the patient bias or common circuit, and equality of electrode impedances. The best CMRR number on a data sheet does not directly describe interference rejection throughout an individual patient’s night.

High amplifier input impedance reduces loading of the electrode–skin interface, so signal amplitude is less attenuated when contact impedance is relatively high; it does not eliminate the need for electrode preparation. An impedance imbalance converts common-mode interference into differential artifact. Sweat bridges, drying gel, hair, cable motion, and a loose common electrode can still reduce signal-to-noise ratio. An abnormally low impedance can also represent a conductive-paste or sweat bridge between electrodes rather than exceptionally good contact.

The front end must accommodate electrode DC offset and movement transients in addition to the target waveform. With a narrow range or excessive first-stage gain, turning, electrode pop, and offset drift can saturate the amplifier, and recovery may take seconds or longer. A wide range accompanied by excessive input noise may instead obscure weak cerebral activity. Design and acceptance should therefore consider allowable offset, dynamic range, recovery time, input protection, and noise together rather than pursuing maximum gain or the smallest numerical quantization step in isolation.

Noise, gain, linearity, and calibration

Input-referred noise represents random fluctuation added by the amplifier and should be evaluated with a stated bandwidth, input termination, and statistic. Peak-to-peak, root-mean-square, and typical values are not interchangeable, and a wideband result cannot be compared directly with a narrowband result. Low electronic noise helps preserve weak EEG, but it cannot remove eye movement, cardiac, muscular, pulse, or lead-motion activity that originates in the patient or interface. Once these components reach the front end, they are amplified along with the target signal.

Gain maps an input potential into the converter range and should remain linear across relevant amplitudes and polarities. If hardware gain is fixed, a software sensitivity control normally changes only display scaling. If front-end gain is switchable, a change may alter available range, noise, or saturation behavior. The system should state whether stored values are calibrated physical units or screen pixels, and viewing adjustments should never rewrite the underlying samples. Channel inversion and rereferencing likewise belong as traceable display operations rather than destructive changes to the original record.

An internal calibration signal can test polarity, gain consistency, time constants, and transmission from the internal front end to the file. It may, however, be injected after the patient-electrode input and therefore cannot validate the patient cable, contact, or anatomical location. An external traceable signal is more suitable for acceptance of amplitude, frequency, and phase response across complete channels. Actual hookup still needs physiologic calibration to confirm that eye movements, jaw contraction, leg movement, and cardiac activity appear in the expected leads. Settings, results, operator, and time should be preserved with maintenance records.

Frequency response, filtering, and digitization

A high-pass or low-frequency cutoff determines transmission of slow change, while a low-pass or high-frequency cutoff limits rapid components and noise. Cutoff frequency, filter order, phase response, and definition method together determine the real response; one frequency value is insufficient. An aggressive high-pass setting can attenuate slow waves and baseline shifts, a low low-pass setting can blunt fast activity or EMG, and a steep filter can ring. Hardware filtering before storage is irreversible, whereas display filtering should be adjustable and reproducible within the content of the saved data.

A notch filter reduces 50- or 60-Hz mains interference but should not replace correction of impedance imbalance, a common electrode, shielding, grounding, or a nearby electrical source. It changes amplitude and phase around the rejected band and may make a contaminated trace look deceptively clean. The record should state whether notching occurred in hardware, in the stored data, or only on display. If a system selects frequency automatically from regional information, location or time-zone errors must not choose the wrong configuration.

Analog-to-digital conversion requires a sample rate, antialias filter, input range, and effective resolution suited to the analog bandwidth. A nominal 24-bit converter does not mean that every bit rises above electronic noise and does not prevent saturation when the front-end range is narrow. A high sample rate similarly cannot restore information already lost in the analog path. When channels are multiplexed, interchannel timing offset should be assessed. When an amplifier is combined with independent oximetry, respiration, or video modules, a common timeline must be verified rather than inferred from matching file start times.

Montages, references, and all-night quality control

PSG montages usually use software to combine an active electrode and a reference into a displayed derivation, so physical inputs, stored electrodes, and visible channels are different concepts. A contaminated shared reference can affect several EEG or EOG derivations, while one failed electrode can recur in multiple displays. A backup electrode helps only when it was independently connected and retained adequate signal. Offline rereferencing can recombine acquired potentials but cannot recreate an input that was never connected or that remained saturated.

Before recording, staff should verify ports, labels, reference and common electrodes, impedance, and physiologic calibration, then inspect baseline, polarity, and spatial distribution of each maneuver. During the night, sweat, turning, traction, and drying gel alter the interface; one impedance reading at hookup cannot represent eight hours. Online quality tools may flag flatline, saturation, excessive noise, or contact change, but the algorithm must distinguish genuine high-amplitude activity from failure and allow review of raw and related channels before action.

Typical faults include sinusoidal mains contamination, slow drift, electrode pop, step-and-recovery behavior, channel crosstalk, shared-reference contamination, flat-topped clipping, and a frozen trace. Strong filtering may hide their appearance temporarily while sacrificing clinically relevant information. Software “repair” or interpolation must not fabricate samples. Electrode reapplication, lead replacement, cable rerouting, removal of an interference source, or switching to a backup derivation should be time-marked so that a scorer knows the signal conditions differ before and after the intervention.

Electrical safety, integration, and data governance

The patient-applied portion must control leakage current, insulation failure, overvoltage, liquid ingress, electrostatic discharge, electromagnetic disturbance, and foreseeable external energy. Electrodes, auxiliary devices, chargers, nonmedical computers, and network connections together form a medical electrical system. Compliance of one amplifier does not establish safety of an arbitrary combination. Patient-contact parts should also minimize skin injury, lead entanglement, and pressure from a worn unit. Home use adds wrong power supplies, pets, children, moisture, and the absence of continuous professional observation.

A wired, wireless, or network path between amplifier and host must preserve sample identity, order, timing, and integrity. Dropped packets should be explicitly marked rather than filled silently by holding the last value or interpolation. Buffering, durable storage, and upload are distinct states. On disconnect, the system should state whether patient-side recording continues, for how long, and how recovered data are merged. Driver, firmware, and protocol updates can change channel mapping, scaling, and latency and therefore require regression testing before clinical deployment.

PSG waveforms become highly sensitive when combined with identity, video, audio, and clinical notes. Transmission, remote service, and export need authentication, least privilege, encryption, integrity checks, audit logging, version control, and secure deletion. Service access to raw signals or settings should be governed as well, and update policy must not make older studies impossible to reproduce. During a network or server failure, the system should degrade safely and preserve clinical data rather than render “unavailable” as a normal event-free recording.

Acceptance, maintenance, and future directions

Amplifier acceptance should use traceable signal sources and representative source impedances to test every channel for polarity, gain accuracy, linearity, noise, frequency response, common-mode rejection, input range, offset tolerance, saturation recovery, crosstalk, sampling, and timing. It should also cover actual electrodes and cables, shared references, impedance measurement, calibration paths, maximum channel configuration, and filter options. Passing a component data sheet does not demonstrate the complete system; testing must extend through stored files, replay displays, and exported units.

Routine maintenance includes the enclosure, connectors, patient cables, shielding, isolation components, battery, power supply, calibration, self-test, clock, firmware, and compatible software. Replacement of a front-end board, converter, cable, or supply requires verification of the affected functions, not merely a successful boot. Gradual noise increase, interchannel discrepancy, repeated saturation, or abnormal impedance readings warrant trend review. Recalls, field corrections, and configuration changes should be traceable to serial numbers and study dates.

Future designs include active electronics closer to the electrode, wider-dynamic-range conversion, optical or wireless isolation, continuous contact-quality estimation, automated lead-swap recognition, and edge artifact alerts. These features can reduce cable and hookup burden but introduce battery, clock, radio, algorithmic, and hidden-preprocessing risks. Evaluation should emphasize real all-night fidelity, visible failure, synchronization, safety, and long-term reproducibility in the intended population and setting—not only nominal bit depth, sample rate, or one laboratory CMRR figure.