Biomedical Filters: ECG and EEG Signal Filtering

Biomedical signals, like electrocardiograms (ECG) and electroencephalograms (EEG), are usually low-frequency, low-amplitude signals that are susceptible to noise contamination. Muscle activity (EMG), patient movement (tremors, respiration), inadequate electrode contact, and ambient electromagnetic noise are examples of common interferences.[1][2]. Filtering is essential to attenuate these artifacts and reveal the true cardiac or neural waveform. Filters work by selectively passing desired frequency bands while attenuating unwanted frequencies. For example, analog filters use continuous-time circuits (capacitors, inductors, op-amps) with transfer functions

in the Laplace domain[3], whereas digital filters operate on sampled data using H(z) in the z-domain. In ECG/EEG processing, analog filtering is often applied in the front end for tasks like antialiasing and baseline drift removal, followed by digital filtering for precision and flexibility. This document reviews analog and digital biomedical filtering – their definitions, mathematics, advantages, applications, and future trends – focusing on ECG and EEG cases.

Analog Filters

Analog filters are physical circuits that “allow certain band of frequencies to pass and attenuate other frequencies”[3]. In the Laplace (s-)domain, a filter’s transfer function is

where N(s) and D(s) are polynomials in s.

Common analog filter types include low-pass (passes low frequencies), high-pass (passes high frequencies), band-pass, and band-stop/notch filters. For example, second-order (biquad) filters have standard forms

These analog networks are implemented using resistor-capacitor (RC) or active (op-amp) circuits such as Sallen–Key or multiple-feedback topologies.

· Low-pass filters (LPF) pass frequencies below a cutoff Ω and attenuate above it. In biomedical ECG, LPFs are typically set around 100–150 Hz because “clinically relevant info in the ECG falls below [150 Hz]”[8]. LPFs smooth out high-frequency noise (e.g. muscle artifact), but excessive attenuation can distort QRS amplitudes[9].

· High-pass filters (HPF) block low-frequency drift. They remove baseline wander from breathing or movement, which often appears below ~0.5 Hz[10]. For ECG, a 0.05–0.5 Hz HPF is common (lower cutoff for diagnostic fidelity)[11]. However, steep HPF roll-off or high cutoff can distort slow ECG features (ST/T waves)[11][12].

· Band-pass and band-stop filters isolate specific bands. An ECG or EEG system might use a band-pass to reject both DC and ultra-high frequencies. A notch filter (very narrow band-stop) is widely used to eliminate power-line interference (50 or 60 Hz)[13][14]. In EEG, where α waves are ~8–13 Hz, narrowband filters can target such rhythms or reject 50/60 Hz interference.

Image of generic low-pass frequency response

Figure: Example frequency response of an analog low-pass filter. The passband (left) is flat up to the cutoff, after which the response rolls off, attenuating higher frequencies.

Active analog filters are often employed in biomedical front ends. For instance, a Sallen–Key or multiple-feedback (MFB) topology can implement a second-order LPF. The pole frequency and quality factor Q determine the cutoff and bandwidth. A Butterworth design (maximally flat passband) is frequently used for biomedical smoothing, since “their gain response is maximally flat in the passband, and the roll-off rate is adequate”. Analog filters are typically located before the analog-to-digital converter (ADC) to prevent aliasing by attenuating frequencies above Nyquist. For example, a 4th-order analog LPF with cutoff near half the sampling rate will eliminate aliased signals.

Digital Filters

Once signals are digitized, digital filters process the sampled data. These filters are defined by difference equations and have transfer functions H(z). Two major classes are finite impulse response (FIR) filters and infinite impulse response (IIR) filters. An FIR filter’s output is a weighted sum of current and past inputs:

Because FIR filters have no feedback, they are inherently stable and can be designed to have exact linear phase (no phase distortion)[18]. For example, a simple digital FIR low-pass might implement a moving average. IIR filters include feedback (poles in H(z)) and can achieve sharper responses with fewer coefficients, but phase response is generally nonlinear. Both FIR and IIR filters are readily implemented in DSP hardware or software with high precision.

Digital filters offer great flexibility and stability. They are “more advanced in precision and stability than analog filters”[19], and filter parameters (cutoff, shape) can be easily modified in software. However, a key limitation is that digital filtering cannot retroactively remove aliasing: if high-frequency noise is sampled without an analog anti-alias filter, unwanted components will appear within the passband[20]. Therefore, practical biomedical systems often use a modest analog pre-filter before ADC (to block gross out-of-band noise[15]) and then employ digital filters for fine-tuned processing.

Digital filter design often uses established methods (windowing, bilinear transform, Parks–McClellan, etc.) to meet specifications. In ECG/EEG applications, digital filters can implement precise FIR notch filters at 50/60 Hz, zero-phase (forward-backward) filtering to avoid phase shifts, and adaptive or multistage filters. For instance, IIR sections can be used for effective elliptic or Chebyshev responses, or FIR low-pass and high-pass can be cascaded to create a band-pass. Word-length effects are the primary limitation on the extremely steep roll-off and high stop-band attenuation that contemporary digital filters, regardless of type, can achieve.

Filter Categories and Design Considerations

Biomedical filters can be broadly classified by their function or domain: time-domain linear vs nonlinear, and frequency-domain low-/high-pass, etc. Linear filters produce an output that is a linear combination of inputs; nonlinear filters (e.g. median filters, adaptive cancellation) can suppress artifacts within signal bands without removing physiological content[21]. However, nonlinear methods may unpredictably distort waveforms if not carefully designed[22]. In most ECG/EEG devices, linear filters (analog or digital) are preferred for predictable behavior.

Key design points for analog filters: the cutoff frequencies are chosen based on signal content. For ECG, AHA/ACC guidelines recommend analog HPF cutoff ~0.05 Hz (to preserve ST-segment) and LPF ~100–150 Hz[11][23]. (Ambulatory monitors may raise HPF to 0.5 Hz to reduce baseline drift[11].) For EEG, typical analog band-pass is ~0.5–70 Hz[24] (0.3 Hz HPF to remove DC, 70 Hz LPF to remove EMG), plus a 50/60 Hz notch. Filter orders are chosen to balance roll-off and phase distortion: higher order yields steeper slopes but more phase lag. Op-amps in biomedical front ends should have low noise, low offset, and sufficient bandwidth (e.g. 100× cutoff[25]).

Key design points for digital filters: sampling rate is chosen (e.g. 250–2000 Hz for EEG); then anti-alias LPF in analog ensures Nyquist criteria. Digital cutoff frequencies are normalized (e.g. a 50 Hz notch at Fs=250 Hz). FIR filters may require many taps for sharp response, while IIR can achieve similar selectivity with fewer coefficients. Zero-phase filtering (applying filter forward and reverse) is often used offline to avoid waveform distortion. Adaptive filtering (e.g. LMS algorithm using a reference noise signal) is another approach for removing correlated noise.

ECG Signal Filtering

In ECG systems, filters target specific artifacts while preserving cardiac waveform integrity. Baseline wander (slow drift due to respiration or electrode movement) is mitigated by a high-pass filter (0.05–0.5 Hz). Muscle and EMG noise (100+ Hz) are attenuated by a low-pass filter (commonly 40–100 Hz)[23]. Power-line interference (50/60 Hz) is addressed by notch filters. ECG signals typically have content between ~0.05 and 150 Hz[8], so filters outside this range are straightforward to design. For example, an ECG analog front end might use a 0.5 Hz HPF and 40 Hz LPF for continuous monitoring[26][27], then digital filtering can further refine the signal.

Clinically, filtering must be done carefully. Excessive HPF roll-off can distort the ST-segment and T-wave, potentially simulating ischemia[11][12]. Likewise, overly aggressive LPF can blunt QRS amplitude or width. Thus, filters must meet standards: the IEC 60601-2-25 specifies that diagnostic ECG bandwidth should be ~0.05–150 Hz[11][23]. When line noise falls within the passband (e.g. 50 Hz), a narrowband filter is used instead of a broad LPF, since a linear filter removing 50 Hz would also remove parts of the ECG. The GE guide notes that power-line filters “remove the fuzz in the middle of the ECG signal”[14] without affecting other components.

In practice, many ECG devices digitize the signal and then apply digital filtering. FIR notch filters at 50/60 Hz and FIR band-pass filters tailored to the chosen sample rate are common. Advanced methods (adaptive filters, wavelet denoising) have been explored for baseline and motion artifacts[28]. Nevertheless, the simplest and most reliable improvements often come from proper electrode placement and grounding[29]; filtering is used when those measures are insufficient.

Figure: ECG signals before filtering v/s after filtering.

EEG Signal Filtering

EEG signals, representing brain activity, are also band-limited. Typical EEG rhythms range from ~0.5 Hz (delta waves) to ~30 Hz (beta waves), with important clinical bands defined as delta (0.5–4 Hz), theta (4–8), alpha (8–13), beta (13–30), etc. Standard EEG systems thus use a band-pass filter around 0.5–70 Hz[24]. High-pass filtering above ~0.5 Hz removes slow drifts and DC offset (including electrogalvanic skin potentials), while low-pass filtering below ~70 Hz removes muscle (EMG) and high-frequency noise[24]. For example, the default “band-pass” on many EEG machines is ~0.5 Hz to 70 Hz. In addition, a notch filter at 50 or 60 Hz is routinely applied to reject line noise[30].

Subtle waveform shapes, such as sharp transients in epileptiform discharges, must be preserved during EEG filtering. In order to prevent phase distortion, linear-phase FIR filters or bidirectional IIR filters (zero-phase) are frequently chosen. Digital multiband filtering is another technique used in modern EEG analysis to extract particular frequencies (e.g. filtering raw EEG into alpha or gamma bands for analysis). Adaptive and blind-source-separation methods (ICA, etc.) supplement classical filtering to remove artifacts like eye blinks or muscle bursts, but those are beyond simple filter design.

Figure: EEG signals before filtering v/s after filtering.

Filter Advantages and Considerations

Filters improve signal quality but introduce trade-offs.

Advantages: They increase signal-to-noise ratio, making diagnosis or automated detection more reliable. For example, low-pass filtering an ECG eases detection of QRS complexes by smoothing high-frequency clutter[9]. Digital filters can be finely tuned, programmable, and can achieve ideal characteristics (e.g. exact notches or linear phase) that analog cannot[19][18]. Filters like Butterworth provide smooth passbands (maximally flat) that are often chosen in biomedical applications[16].

Drawbacks: Any filter can distort the waveform. Phase shifts from IIR filters or transient ringing from sharp cutoffs can alter physiologic features. For instance, as noted above, an ECG HPF can shift the ST segment[11][12]. In EEG, steep LPF may distort fast-spike morphology. Hence filtering must be chosen judiciously: use the lowest necessary cutoff and consider zero-phase designs. Clinicians are warned that filters should be “the exception rather than a convenient way” to clean ECGs[29].

Analog vs. Digital: Analog filters (pre-ADC) effectively remove out-of-band noise immediately, preventing aliasing[15]. Digital filters offer accuracy and stability (no component drift)[19]. In practice, analog front-end filters are often low-order (to minimize distortion and power), and the “workhorse” filtering is done digitally. The IEEE (AHA/ACC) guidance reflects this: “nearly all current ECG machines convert the analog ECG signal to digital form before further processing” (with analog limits per IEC standards)[31][11].

Future Developments

Biomedical filtering continues to evolve. On the analog side, research targets ultra-low-power tunable filters for wearable monitors. For example, a recent CMOS filter design achieved a programmable cutoff from 0.5 Hz to 150 Hz with only ~6 nW power for the filter core[32]. Such adaptive analog filters can save power while maintaining biomedical bandwidth. In the digital domain, advances include highly selective multirate filters, machine-learning-based artifact suppression, and real-time adaptive filtering using reference channels. Additionally, system-level innovation (e.g. combined hardware-software co-design) is reducing noise at the source and enabling more sophisticated filtering in implantable or ambulatory EEG/ECG devices.

Conclusion

Filtering is a fundamental step in biomedical signal acquisition and processing. Properly designed analog filters remove gross noise (DC drift, RF interference) and prevent aliasing, while digital filters refine the signal (suppress specific bands or artifacts) without bulky hardware. In ECG and EEG applications, standard practice uses band-limited filtering (e.g. 0.05–150 Hz for ECG; 0.5–70 Hz for EEG) with notch filters for power-line interference[11][24]. The choice of filter type and cutoff critically affects signal fidelity: a Butterworth or linear-phase design is often chosen to minimize waveform distortion[16]. As highlighted by expert guidelines, filters should be applied carefully only when needed[29], since over-filtering can obscure diagnostic features. Ongoing research into low-power and adaptive filters promises to improve biomedical monitoring by providing clean signals in challenging environments.

References:

Authoritative reviews and recent research on ECG/EEG filtering, including GE Healthcare clinical insights[1][8], IEEE/IEC standards commentary[11][23], and filter design literature[15][16], inform this summary. These works cover analog filter theory and implementation as well as state-of-the-art digital filtering methods and their applications in biomedical signal processing.

[1] [8] [9] [10] [12] [14] [21] [22] [29] [31] A Guide to ECG Signal Filtering | GE HealthCare (United States)