m14 · Biopotential front ends: instrumentation amplifiers, CMRR, noise, Wilson central terminal, ADCs
Education only. Not for diagnosis, treatment or any clinical decision.
In 60 seconds
A biopotential front end must amplify microvolt-to-millivolt differences while the whole body floats at a much larger common-mode voltage picked up from the mains. A differential amplifier passes the difference and rejects the common mode; the ratio of the two gains is the common-mode rejection ratio (CMRR). Resistor tolerance limits it: \((1 + G)/(4t)\) for a difference stage, improved by the input-stage gain of an instrumentation amplifier. Electrode impedances that differ convert common mode to difference at the rate \(\Delta Z/Z_{cm}\), however good the amplifier. A driven right leg shrinks the common mode itself by \(1 + G\). Noise adds in quadrature: the source’s thermal noise \(\sqrt{4kTR}\), the amplifier’s voltage noise, and its current noise through the source. The Wilson central terminal is a resistor average that electrode impedances bias. The converter needs enough bits for offsets plus detail, and averaging gains half a bit per doubling of the sample rate.
From the field
- Common mode is the enemy you cannot see. Mains interference and its cures have a long engineering literature, including Huhta and Webster’s paper on 60-Hz interference in electrocardiography (Huhta and Webster 1973) and Winter and Webster’s on driven-right-leg circuit design (Winter and Webster 1983).
- The central terminal is not a true zero. Measured with a true unipolar recorder, the Wilson central terminal varied unpredictably through the cardiac cycle, and depends on balanced contact of at least four electrodes (Gargiulo 2015).
- Front-end quality shows in the signals. In a multicenter blinded comparison of 218 signal pairs from 51 patients, reviewers rated a new signal-processing platform superior to conventional sources for 75.2% of signals (Al-Ahmad et al. 2021). That is a reader-preference study, not a measurement of noise.
- Know the acquisition chain. A review for electrophysiologists covers electrogram acquisition (Venkatachalam et al. 2011).
Derivations
Symbols used throughout. \(v_+\), \(v_-\) input voltages; \(v_d\) differential and \(v_{cm}\) common-mode voltage; \(A_d\), \(A_{cm}\) their gains; \(R_1\)–\(R_4\) difference-stage resistors; \(G = R_2/R_1\); \(t\) resistor tolerance; \(R\), \(R_G\) input-stage resistors; \(G_1\) input-stage gain; \(Z_1\), \(Z_2\) electrode impedances; \(\Delta Z\) their difference; \(Z_{cm}\) common-mode input impedance; \(i_d\) displacement current into the body; \(Z_{RL}\) right-leg electrode impedance; \(G_{RL}\) driven-right-leg loop gain; \(k\) Boltzmann’s constant; \(T\) temperature; \(R_s\) source resistance; \(e_n\), \(i_n\) amplifier voltage and current noise densities; \(B\) bandwidth; \(\Phi_i\) limb potentials; \(R_W\) Wilson resistors; \(N\) converter bits; \(m\) samples averaged.
Differential and common mode
Any pair of inputs splits into a difference and a mean; an amplifier treats them with different gains:
\[ v_d = v_+ - v_-, \quad v_{cm} = \frac{v_+ + v_-}{2}, \qquad v_{out} = A_d v_d + A_{cm} v_{cm}, \qquad \text{CMRR} = \left\lvert\frac{A_d}{A_{cm}}\right\rvert \tag{1}\]
Expressed in decibels, CMRR is \(20\log_{10}\) of that ratio. A common mode of \(v_{cm}\) appears at the output as if it were a difference of \(v_{cm}/\text{CMRR}\).
The difference amplifier
One op amp and four resistors subtract: \(v_-\) enters through \(R_1\) with feedback \(R_2\), and \(v_+\) through the divider \(R_3\), \(R_4\). With an ideal op amp:
\[ v_{out} = v_+\frac{R_4}{R_3 + R_4}\left(1 + \frac{R_2}{R_1}\right) - v_-\frac{R_2}{R_1}, \qquad \text{CMRR}_{\text{worst}} \approx \frac{1 + G}{4t} \tag{2}\]
Matched ratios (\(R_4/R_3 = R_2/R_1 = G\)) give \(G\,v_d\) and no common-mode gain. Resistors within tolerance \(t\) in the worst pattern leave \(A_{cm} \approx 4tG/(1 + G)\), so 1% resistors and \(G\) = 10 give 48.8 dB, and 0.1% gives 68.8 dB. Random resistors rarely hit the worst case: drawn uniformly within tolerance, their median is about 14 dB better (lab part A).
The instrumentation amplifier
Put two op amps in front, joined through \(R_G\). The difference sees gain \(G_1\) while the common mode passes at unity:
\[ G_1 = 1 + \frac{2R}{R_G}, \qquad \text{CMRR}_{\text{total}} = G_1 \times \text{CMRR}_{\text{difference stage}} \tag{3}\]
So gain in the first stage buys rejection: with \(G_1\) = 101 the 68.8 dB stage becomes 108.9 dB (lab part B). The input stage also presents a high input impedance, which matters next.
Key takeaways
- CMRR is set by matching; gain before the difference stage multiplies it.
Electrode impedance mismatch
Each input has a common-mode input impedance \(Z_{cm}\) to ground, and the electrodes in series with it form two dividers. If the electrodes differ, the common mode reaches the two inputs unequally:
\[ v_{d,\text{error}} = v_{cm}\left(\frac{Z_{cm}}{Z_{cm} + Z_2} - \frac{Z_{cm}}{Z_{cm} + Z_1}\right) \approx v_{cm}\frac{Z_1 - Z_2}{Z_{cm}}, \qquad \text{CMRR}_{\text{imbalance}} = \left\lvert\frac{Z_{cm}}{\Delta Z}\right\rvert \tag{4}\]
This conversion happens before the amplifier, so no amplifier CMRR removes it; the two add as \(1/\text{CMRR}\) values, and the weaker path wins. A high \(Z_{cm}\) (small input capacitance) and matched electrodes are the cures; see also Huhta and Webster (Huhta and Webster 1973).
Exact
Assumptions
- Model: worst-case resistor mismatch in a gain-10 difference stage (Equation 2) after an input stage of gain \(G_1\) (Equation 3), and a resistive electrode mismatch into a real common-mode input impedance (Equation 4); conversions add in the worst case. Tested against
epml.frontend. - Parameters: all values are adjustable illustrative choices, not those of any product.
- What it ignores: op-amp CMRR, input capacitance mismatch and phase, frequency dependence, cable shields.
The driven right leg
The body picks up a displacement current \(i_d\) from the mains through stray capacitance; it returns through the right-leg electrode. A driven right leg senses the common mode, inverts and amplifies it by \(G_{RL}\), and drives the right leg with it:
\[ v_{cm} = i_d Z_{RL} - G_{RL}\,v_{cm} \;\Longrightarrow\; v_{cm} = \frac{i_d Z_{RL}}{1 + G_{RL}} \tag{5}\]
With a loop gain of 100, the common mode, and with it every conversion path above, falls by about 40 dB (lab part C). The loop must be stable and current-limited; see Winter and Webster on its design (Winter and Webster 1983). Patient safety for the drive circuit belongs to m15.
Key takeaways
- Mismatched electrodes convert common mode before the amplifier; only a high \(Z_{cm}\) or a smaller \(v_{cm}\) helps.
- A driven right leg attacks the common mode at its source.
Noise
Independent noise sources add in power. Referred to the input, with source resistance \(R_s\):
\[ e_{\text{total}}^2 = 4kTR_s + e_n^2 + (i_n R_s)^2, \qquad v_{\text{rms}} = e_{\text{total}}\sqrt{B}, \qquad R_{s,\text{opt}} = \frac{e_n}{i_n} \tag{6}\]
The first term is thermal noise, derived by Nyquist (Nyquist 1928): about 4.1 nV/√Hz for 1 kΩ at 37 °C, with \(k\) exact in the SI (Tiesinga et al. 2021). The noise factor, total power over the source’s own, is smallest at \(R_s = e_n/i_n\). Small electrodes raise \(R_s\) (m13), so they push the thermal and current-noise terms up: halving the radius doubles \(R_s\) and raises thermal noise by \(\sqrt2\). The electrode–electrolyte interface adds its own frequency-dependent impedance and noise, which this model leaves out.
Exact
Assumptions
- Model: Equation 6 with white noise densities and a resistive source at 37 °C; Boltzmann’s constant from CODATA (Tiesinga et al. 2021). Tested against
epml.frontend. - Parameters: amplifier noise densities, source resistance and bandwidth are adjustable illustrative choices.
- What it ignores: 1/f noise, the interface’s capacitive impedance and its noise, filter shapes and quantization.
Key takeaways
- Noise adds in quadrature; the largest term dominates.
- Match the amplifier to the electrode: the noise factor is lowest at \(e_n/i_n\).
The Wilson central terminal
The central terminal joins the three limb electrodes through equal resistors \(R_W\). Each electrode adds its impedance \(Z_i\), so the node settles at a weighted mean:
\[ \Phi_{\text{WCT}} = \frac{\sum_i \Phi_i/(R_W + Z_i)}{\sum_i 1/(R_W + Z_i)} \tag{7}\]
With equal impedances it is the plain mean of m12 (eq-m12-augmented). One poor electrode pulls it off: with \(R_W\) = 5 kΩ and one 50-kΩ electrode the error was 0.24 (relative units, for limb potentials spanning 1.2), and it falls as \(R_W\) grows or buffers isolate the electrodes (lab part E). Even an ideal terminal moves with the heart’s activity (Gargiulo 2015).
The converter
An \(N\)-bit converter of full scale \(V_{FS}\) after gain \(G\) resolves steps of \(V_{FS}/(G\,2^N)\); its quantization noise is the step over \(\sqrt{12}\) (m04, eq-m04-quantization). Averaging \(m\) samples of white noise and decimating improves the signal-to-noise ratio by:
\[ q_{\text{in}} = \frac{V_{FS}}{G\,2^N}, \qquad N \ge \log_2\frac{\text{span}}{\text{step}}, \qquad \Delta\text{SNR} = 10\log_{10} m \text{ dB} \tag{8}\]
That is half a bit per doubling. A DC-coupled input must span the electrodes’ offset potentials as well as the signal, which drives the bit count: spanning an illustrative 200 mV in 1-µV steps needs 18 bits.
Key takeaways
- The reference is part of the measurement; electrode impedances bias it.
- Bits must cover offsets and detail; oversampling trades rate for resolution.
Clinical correlate
This section describes published findings only. It is not guidance for patient care, and it has not yet been reviewed by an EP physician.
Where this shows up in EP. Every electrogram passes through a front end. Mains interference, baseline wander and noise shape what can be annotated, especially low-amplitude signals in scar (m22, m23). The Wilson central terminal references every unipolar chest lead and is not fixed (Gargiulo 2015). Signal quality differs between recording platforms (Al-Ahmad et al. 2021). The standards for ECG technology are reviewed in a scientific statement (Kligfield et al. 2007), and an EP-oriented review covers electrogram acquisition (Venkatachalam et al. 2011).
What the evidence is. Engineering analyses (Huhta and Webster 1973; Winter and Webster 1983; Nyquist 1928), a device study (Gargiulo 2015), a blinded reader study (Al-Ahmad et al. 2021) and reviews (Kligfield et al. 2007; Venkatachalam et al. 2011).
What a reader should ask:
- What were the electrode impedances, and were they matched? (Equation 4)
- What is the noise floor in microvolts over the stated bandwidth? (Equation 6)
- What reference did the unipolar signals use? (Equation 7)
- Was a filter or notch applied, and where? (m04)
Visual summary
Schematic
Assumptions
- Schematic: not to scale, no quantitative claim. Labels follow Equation 4, Equation 3, Equation 5 and Equation 8.
How experts think
When a recording looks noisy, ask in this order:
- Is it common mode leaking through? Check electrode impedances and their match (Equation 4).
- Is the amplifier’s CMRR the limit? Gain before the difference stage helps (Equation 3).
- Can the common mode itself be reduced? Driven right leg, cabling, grounding (Equation 5).
- Which noise term dominates? Thermal, voltage or current noise (Equation 6).
- Is the reference stable? (Equation 7)
- Is the converter the floor? Bits, offsets and oversampling (Equation 8).
One sentence to keep: matching and impedance decide what the amplifier ever gets to reject.
Cheat sheet
| Quantity | Result |
|---|---|
| Modes (Equation 1) | \(v_d = v_+ - v_-\); \(v_{cm} = (v_+ + v_-)/2\); CMRR \(= \lvert A_d/A_{cm}\rvert\) |
| Difference stage (Equation 2) | gain \(R_2/R_1\); worst-case CMRR \((1 + G)/(4t)\) |
| Instrumentation amplifier (Equation 3) | \(G_1 = 1 + 2R/R_G\); CMRR \(\times G_1\) |
| Electrode mismatch (Equation 4) | error \(\approx v_{cm}\,\Delta Z/Z_{cm}\); CMRR \(= \lvert Z_{cm}/\Delta Z\rvert\) |
| Driven right leg (Equation 5) | \(v_{cm} = i_d Z_{RL}/(1 + G_{RL})\) |
| Noise (Equation 6) | \(e^2 = 4kTR_s + e_n^2 + i_n^2R_s^2\); optimum \(R_s = e_n/i_n\) |
| Wilson central terminal (Equation 7) | weighted mean, weights \(1/(R_W + Z_i)\) |
| Converter (Equation 8) | step \(V_{FS}/(G2^N)\); averaging \(m\): \(+10\log_{10}m\) dB |
Values used in this module
| Value | Source |
|---|---|
| Boltzmann’s constant \(1.380649\times10^{-23}\) J/K | Tiesinga et al. (2021) |
| 218 signal pairs, 51 patients, 75.2% rated superior | Al-Ahmad et al. (2021) |
| Resistor tolerances, impedances, noise densities, 50 Hz, 10 mV common mode, 200 mV span | illustrative |
| CMRR values, noise totals, WCT errors, bit counts | our own code |
Interview questions
Try each question before you open its model answer. In a real interview, say which assumptions you are making.
Q1. Define CMRR and say why biopotential amplifiers need a high one.
Model answer
The ratio of differential to common-mode gain (Equation 1); the body’s common-mode voltage from the mains can be far larger than the signal, so even a small common-mode gain swamps it.
Q2. What limits the CMRR of a four-resistor difference amplifier?
Model answer
Resistor matching: worst case \((1 + G)/(4t)\) for tolerance \(t\) (Equation 2).
Q3. Why does an instrumentation amplifier’s input stage improve CMRR?
Model answer
It amplifies the difference by \(G_1\) but passes the common mode at unity, so the difference stage sees a ratio \(G_1\) times better (Equation 3).
Q4. The amplifier’s CMRR is 120 dB, yet mains interference is strong. What else could it be?
Model answer
Electrode impedance mismatch converting common mode before the amplifier, at \(\Delta Z/Z_{cm}\) (Equation 4); also cable pickup or a poor right-leg connection.
Q5. How does a driven right leg work?
Model answer
It senses the common mode, inverts and amplifies it, and drives the body through the right-leg electrode, dividing the common mode by \(1 + G\) (Equation 5) (Winter and Webster 1983).
Q6. Write the input-referred noise of an amplifier with source resistance \(R_s\).
Model answer
\(\sqrt{4kTR_s + e_n^2 + i_n^2R_s^2}\) per root hertz, times \(\sqrt{B}\) for the rms (Equation 6) (Nyquist 1928).
Q7. At what source resistance is the noise factor lowest?
Model answer
\(e_n/i_n\): differentiate the noise factor with respect to \(R_s\) (Equation 6).
Q8. How does electrode size affect noise?
Model answer
Smaller electrodes have higher spreading resistance (m13), raising thermal noise as \(\sqrt{R_s}\) and current noise as \(R_s\) (Equation 6).
Q9. Why is the Wilson central terminal not an exact average of the limb potentials?
Model answer
Electrode impedances add to the averaging resistors, weighting the mean (Equation 7); and even an ideal average moves with the heart’s activity (Gargiulo 2015).
Q10. How many bits does a DC-coupled front end need?
Model answer
Enough to span offsets plus signal at the needed step: \(\log_2(\text{span}/\text{step})\) (Equation 8).
Q11. What does oversampling by 16 buy, and under what assumption?
Model answer
12 dB, two bits, if the noise is white and uncorrelated between samples (Equation 8).
Q12. Reviewers preferred one recording system’s signals. What would you measure to confirm it?
Model answer
Input-referred noise over a stated band, CMRR with realistic electrode mismatch, filter responses and the reference; a preference rating is not a measurement (Al-Ahmad et al. 2021).
Code lab
The lab builds a front end’s error budget.
| Part | What you do | Data |
|---|---|---|
| A | The difference amplifier and its CMRR (Equation 1, Equation 2) | synthetic |
| B | The instrumentation amplifier (Equation 3) | synthetic |
| C | Mains interference: electrode mismatch and the driven right leg (Equation 4, Equation 5) | synthetic |
| D | A noise budget (Equation 6) | synthetic |
| E | The Wilson central terminal and the converter (Equation 7, Equation 8) | synthetic |
- Python notebook: labs/python/m14_lab.ipynb. Run it from the course repository after
pip install -r requirements.txt, or open it in Google Colab. - Data: none to download.
Each part ends with checks: lines that pass with the code as given. If you change something and a check fails, that is information, not an error to hide.
Self-check
Short checks you can do in your head. Open each answer after you have tried it.
S1. \(v_+\) = 1.002 V and \(v_-\) = 0.998 V. Differential and common-mode voltages?
Answer
4 mV and 1 V (Equation 1).
S2. CMRR is 100 dB and the common mode is 1 V. Equivalent input error?
Answer
10 µV (Equation 1).
S3. Gain-10 difference stage, 1% resistors. Worst-case CMRR?
Answer
\(11/0.04 = 275\), about 49 dB (Equation 2).
S4. \(\Delta Z\) = 10 kΩ and \(Z_{cm}\) = 100 MΩ. Equivalent CMRR?
Answer
\(10^4\), 80 dB (Equation 4).
S5. A driven right leg with loop gain 99. Common mode reduced by?
Answer
A factor of 100, 40 dB (Equation 5).
S6. Averaging 4 samples of white noise. SNR gain?
Answer
About 6 dB, one bit (Equation 8).
Literature figures
These figures are worth looking at alongside this module. We link to them instead of copying them: some licenses do not allow reuse here, and others we could not confirm. Each entry says what to look for and where this module explains it.
- Interference paths. Huhta and Webster’s paper on 60-Hz interference in electrocardiography (Huhta and Webster 1973). This module: electrode impedance mismatch.
- The driven right leg. Winter and Webster’s paper on driven-right-leg circuit design (Winter and Webster 1983). This module: the driven right leg.
- The central terminal over a beat. Gargiulo’s recordings of the Wilson central terminal with a true unipolar device (Gargiulo 2015). This module: the Wilson central terminal.
Further reading
- Interference and the driven right leg. (Huhta and Webster 1973; Winter and Webster 1983).
- Noise. Thermal noise (Nyquist 1928).
- The reference. (Gargiulo 2015).
- EP acquisition. A review (Venkatachalam et al. 2011); ECG technology standards (Kligfield et al. 2007); a platform comparison (Al-Ahmad et al. 2021).
- Related modules. m04 (sampling and filtering), m12 (the surface ECG), m13 (electrodes), m15 (patient safety), m22 (annotation).