Noise, Filtering, Shielding and Grounding
CSIR NET Physical Sciences: Johnson, shot and 1/f noise, SNR, noise figure and Friis formula, averaging, filters, shielding, ground loops, with practice.
- 19 min
On this page · 15 sections
Prerequisites
- Canonical and Grand Canonical Ensembles: equipartition, which is where \( k_BT \) per mode and Johnson noise come from.
- Operational Amplifiers: first-order active filters and their cut-off frequencies.
- Magnetostatics and Electromagnetic Induction: Faraday’s law explains magnetic pickup and why loop area matters.
- Measurement, Signal Conditioning, Impedance Matching and Feedback: the chain this page cleans up.
The Hook: The Resistor That Is Never Quiet
Describing Noise
Random noise has zero mean, so it is described by its rms value or, better, its power spectral density \( S(f) \) (V\( ^2 \)/Hz), whose square root is quoted in V/\( \sqrt{\text{Hz}} \). The rms noise in a bandwidth \( B \) is
$$v_{\text{rms}} = \sqrt{\int S(f)\,df} \;\to\; \sqrt{S\,B}\ \text{ for white noise.}$$Independent noise sources add in quadrature (powers add), never linearly:
$$v_{\text{total}} = \sqrt{v_1^2 + v_2^2 + \cdots}$$Signal-to-noise ratio: \( \text{SNR} = P_s/P_n \), or in decibels \( 10\log_{10}(P_s/P_n) = 20\log_{10}(V_s/V_n) \).
Fundamental Noise Sources
Johnson–Nyquist (thermal) noise
Any resistance \( R \) at temperature \( T \) has an open-circuit noise voltage
$$\boxed{v_n = \sqrt{4k_BTRB}} \qquad \text{or a noise current } i_n = \sqrt{4k_BTB/R}$$It is white (flat in frequency up to frequencies where \( hf \sim k_BT \)), Gaussian, and present with no current flowing. Useful anchor: a 1 kΩ resistor at 300 K gives about 4 nV/\( \sqrt{\text{Hz}} \); scale by \( \sqrt{R/1\ \text{k}\Omega} \).
The available noise power delivered to a matched load is \( k_BTB \), independent of \( R \). At the standard reference temperature 290 K this is \( -174 \) dBm/Hz, the floor every radio receiver is compared with.
An RC circuit gives a total noise voltage on the capacitor, integrated over all frequencies, of
$$\boxed{v_{\text{rms}} = \sqrt{k_BT/C}}$$independent of \( R \) (larger \( R \) gives more noise density but less bandwidth). This kTC noise is about 64 µV for 1 pF at 300 K and sets the floor of sampling circuits and CCD pixels. It is equipartition: \( \frac{1}{2}C\langle v^2\rangle = \frac{1}{2}k_BT \).
Reactances (ideal \( L \), \( C \)) generate no thermal noise; only dissipation does (fluctuation–dissipation theorem).
Shot noise
A current made of discrete charges \( q \) arriving at random (Poisson) times fluctuates:
$$\boxed{i_n = \sqrt{2qIB}}$$It needs a DC current crossing a barrier (photodiode, vacuum tube, PN junction); a current in a plain metallic resistor shows no full shot noise. Anchor: 1 µA gives about 0.57 pA/\( \sqrt{\text{Hz}} \). Relative fluctuation \( i_n/I = \sqrt{2qB/I} \) falls as current rises, which is the electrical form of \( \sqrt N \) counting statistics.
Shot noise of a current \( I \) equals the Johnson noise current of a load \( R \) when \( IR = 2k_BT/q \approx 52 \) mV at 300 K. A photodiode whose photocurrent develops more than about 50 mV across its load is shot-noise limited, the best a photodetector can do.
Flicker (1/f) noise
Power spectral density rising as \( 1/f^\alpha \) with \( \alpha \approx 1 \) at low frequency. Found in nearly every device (carbon resistors, transistors, carrier trapping in semiconductors) and in many non-electrical systems. It dominates below the 1/f corner frequency, where it equals the white noise. Since each decade of frequency carries equal 1/f noise power, a DC measurement averaged for longer does not keep improving. The remedy is to move the measurement to a higher frequency by modulation (chopping), then detect synchronously: the reason lock-in amplifiers exist.
Other sources
Generation–recombination noise (fluctuating carrier number in semiconductors), burst (popcorn) noise (random switching between levels in faulty junctions), avalanche noise (APDs, Zener diodes), quantisation noise of ADCs (\( \text{LSB}/\sqrt{12} \)), and in optics photon noise (shot noise of the photon stream).
| Noise | Spectrum | Depends on | Reduce by |
|---|---|---|---|
| Johnson | white | \( T \), \( R \), \( B \) | cool, lower \( R \) (for voltage), narrow \( B \) |
| Shot | white | \( I \), \( B \) | narrow \( B \); relative noise falls with \( I \) |
| 1/f | \( \propto 1/f \) | device quality | modulate to above the corner frequency |
| Interference (mains, RF) | lines and spikes | coupling path | shield, ground, twist, filter, separate |
Amplifier Noise and Noise Figure
An amplifier’s noise is represented by a voltage source \( e_n \) and a current source \( i_n \) at its input. With source resistance \( R_s \), total input noise density:
$$e_{\text{tot}}^2 = e_n^2 + (i_nR_s)^2 + 4k_BTR_s$$The noise figure compares this with the source’s own thermal noise:
$$F = \frac{\text{SNR}_{\text{in}}}{\text{SNR}_{\text{out}}} = \frac{e_{\text{tot}}^2}{4k_BTR_s}, \qquad \text{NF} = 10\log_{10}F\ \text{dB}$$\( F = 1 \) (0 dB) is a noiseless amplifier. \( F \) is minimised when \( R_s = R_{\text{opt}} = e_n/i_n \) (noise matching); a transformer can bring a low-impedance source up to \( R_{\text{opt}} \). Low-impedance sources suit bipolar input stages (small \( e_n \)); high-impedance sources suit FET inputs (tiny \( i_n \)).
Noise temperature: \( T_e = (F - 1)T_0 \) with \( T_0 = 290 \) K. A 3 dB amplifier (\( F = 2 \)) has \( T_e = 290 \) K.
Cascaded stages: Friis formula
$$\boxed{F_{\text{total}} = F_1 + \frac{F_2 - 1}{G_1} + \frac{F_3 - 1}{G_1G_2} + \cdots}$$with \( G \) the linear power gains. The first stage dominates if its gain is large, so put the lowest-noise, high-gain amplifier first, as close to the sensor as possible.
Signal Averaging
If a repetitive signal is recorded \( N \) times and averaged, the signal adds coherently (\( \propto N \)) and uncorrelated noise adds in quadrature (\( \propto\sqrt N \)), so
$$\boxed{\text{SNR}_N = \sqrt N\,\text{SNR}_1}$$Improving SNR from 2 to 20 needs \( N = 100 \) averages. Averaging is equivalent to narrowing the bandwidth: time is traded for noise. It works only if the signal is locked to a trigger (repeatable) and the noise is uncorrelated between sweeps; it does nothing for drift or for 1/f noise at the sweep rate. Box-car and multichannel averagers are hardware versions; see the lock-in and box-car page.
Filtering
First-order RC filters
$$\text{Low-pass: } H = \frac{1}{1 + j\omega RC}, \qquad \text{High-pass: } H = \frac{j\omega RC}{1 + j\omega RC}, \qquad f_c = \frac{1}{2\pi RC}$$At \( f_c \): gain \( 1/\sqrt2 \) (\( -3 \) dB), phase \( \mp45^\circ \). Far beyond: 20 dB/decade (6 dB/octave). The equivalent noise bandwidth of a first-order low-pass is \( (\pi/2)f_c = 1/(4RC) \), larger than \( f_c \) because of the gradual skirt. For example, \( R = 10 \) kΩ and \( C = 16 \) nF give \( f_c \approx 995 \) Hz.
Higher order and filter families
An \( n \)th-order filter falls at \( 20n \) dB/decade. Families trade flatness, sharpness and phase:
| Family | Passband | Transition | Phase / step response |
|---|---|---|---|
| Butterworth | maximally flat, no ripple | moderate | moderate overshoot |
| Chebyshev | ripple in passband (type I) | sharper | more overshoot and ringing |
| Bessel | gently drooping | slowest | nearly linear phase: preserves pulse shape |
| Elliptic (Cauer) | ripple in pass- and stopband | sharpest | worst phase |
Butterworth magnitude: \( \lvert H\rvert = 1/\sqrt{1 + (f/f_c)^{2n}} \). At \( 2f_c \) a fourth-order Butterworth is down by 24 dB (factor 0.062); a first-order filter only 7 dB.
Band-pass and notch filters
A band-pass filter centred at \( f_0 \) with \( -3 \) dB bandwidth \( \Delta f \) has quality factor \( Q = f_0/\Delta f \). A notch (band-stop) filter removes one frequency, typically mains interference (50 Hz in India, with harmonics); the passive twin-T network and the Wien bridge are classical notch circuits. Notching is a last resort: it also removes any signal at that frequency, and harmonics need their own notches.
Anti-aliasing and digital filters
Before an ADC, an analogue low-pass must remove everything above half the sampling rate, because aliased noise cannot be separated afterwards. After digitisation, digital filters (moving average, FIR, IIR) can be sharp and exactly reproducible. A moving average of \( N \) points is a low-pass that improves white-noise SNR by \( \sqrt N \).
Interference: Coupling Paths
Interference needs a source, a coupling path and a receiver. Identify the path, then break it.
| Coupling | Mechanism | Grows with | Cure |
|---|---|---|---|
| Conductive (common impedance) | shared ground wire carries another circuit’s current, \( V = IZ_{\text{ground}} \) | shared return current | separate returns, star (single-point) ground |
| Capacitive (electric) | stray capacitance from a high-voltage node to a high-impedance input, \( i = C\,dV/dt \) | frequency, source voltage, victim impedance | grounded electrostatic shield (Faraday cage), distance, low impedance |
| Inductive (magnetic) | changing flux through a circuit loop, \( \mathcal E = -d\Phi/dt \) | frequency, loop area, current in source | minimise loop area, twisted pair, distance, magnetic shielding |
| Radiative | electromagnetic waves | frequency | conducting enclosure, filtered feedthroughs |
Electrostatic shielding
A conducting shield connected to the circuit’s reference intercepts the stray capacitance: the displacement current flows to ground instead of into the input. The shield must be grounded (a floating shield can couple noise in) and, for a cable at low frequency, connected at one end only, at the signal reference, so that no current flows along it. A guard driven at the same potential as a high-impedance input eliminates leakage currents and effective cable capacitance (used in electrometers and with triaxial cable).
Magnetic shielding
At low frequency (mains, 50 Hz) a non-magnetic conductor is ineffective because its skin depth is large (about 9 mm for copper at 50 Hz). Use:
- Reduced loop area: route signal and return together; twisted pair makes the induced emfs in successive twists cancel.
- High-permeability shields (mu-metal) that divert flux around the circuit.
- Distance and orientation from transformers and motors.
At high frequency a conductive shield works well against magnetic fields too, because eddy currents oppose the field. Absorption loss through a shield of thickness \( t \) is
$$A = 8.69\,\frac{t}{\delta}\ \text{dB}, \qquad \delta = \sqrt{\frac{2\rho}{\omega\mu}}$$For copper, \( \delta \approx 65 \) µm at 1 MHz, so a 0.5 mm sheet gives about 67 dB of absorption there but almost nothing at 50 Hz. Holes and seams leak: an aperture is transparent to wavelengths short compared with its size and attenuates well only when it is much smaller than the wavelength.
Grounding
- Ground loops: if a signal source and an amplifier are each grounded at different points, any potential difference between those points (from return currents, or flux from a nearby transformer threading the loop) appears in series with the signal. A 0.1 m\( ^2 \) loop in a 1 µT field at 50 Hz picks up about 31 µV peak, enough to swamp a thermocouple.
- Cures: ground at a single point (star ground) at low frequency; use differential inputs (in-amp) so that ground-potential differences are common mode; break the loop with isolation amplifiers, optocouplers or transformers; keep analogue and digital returns separate, joined at one point.
- High frequency: stray capacitance already closes loops, and long ground wires are inductive, so multipoint grounding to a low-impedance ground plane is used above roughly a megahertz.
- Safety earth must never be removed to “cure” a ground loop.
graph TD
A[Interference seen on the signal] --> B{Changes when the cable is moved or touched?}
B -- yes --> C[Capacitive pickup: shield, ground the shield at one end, lower impedance]
B -- no --> D{Strong at mains frequency and harmonics?}
D -- yes --> E{Changes with loop area or nearby transformers?}
E -- yes --> F[Magnetic pickup: twisted pair, smaller loops, mu-metal]
E -- no --> G[Ground loop: single-point ground, differential input, isolation]
D -- no --> H[Broadband: filter, narrow bandwidth, check digital returns and RF]Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| Noise: random, intrinsic, add in quadrature | Interference: coupled from outside, coherent, removable | |
| Johnson noise: any resistance, no current needed | Shot noise: needs a DC current crossing a barrier | |
| White noise: improves with averaging | 1/f noise: needs modulation to escape | |
| Noise figure in dB (\( 10\log F \)) | Noise temperature in K (\( (F - 1)290 \)) | |
| \( f_c \) of a first-order low-pass | Its noise bandwidth \( (\pi/2)f_c \) | |
| Electrostatic shield: any good conductor, grounded | Low-frequency magnetic shield: high-permeability material | |
| Single-point ground at low frequency | Multipoint ground plane at high frequency |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper.
Level 1: Foundation
- By what factor does Johnson noise voltage change if both \( R \) and \( B \) are quadrupled?
- What happens to the SNR when 400 sweeps are averaged?
- Why is a twisted pair used for low-level signals?
Answers
- Factor 4. \( v_n \propto \sqrt{RB} = \sqrt{16} \).
- Improves 20 times (\( \sqrt{400} \)), for uncorrelated noise.
- It minimises the loop area and alternates its orientation, so magnetically induced emfs cancel between twists.
Level 2: Part B Style
Practice Q1. The rms Johnson noise voltage of a 1 MΩ resistor at 300 K over a 10 kHz bandwidth is about:
(a) 0.4 µV (b) 4 µV (c) 13 µV (d) 130 µV
Answer: (c) 13 µV
\( v = \sqrt{4 \times 1.38 \times 10^{-23} \times 300 \times 10^6 \times 10^4} = \sqrt{1.66 \times 10^{-10}} = 1.29 \times 10^{-5} \) V. Shortcut: 4 nV/\( \sqrt{\text{Hz}} \) for 1 kΩ, times \( \sqrt{1000} \) for 1 MΩ, times \( \sqrt{10^4} = 100 \) for the bandwidth: \( 4 \times 31.6 \times 100 \) nV \( \approx 12.6 \) µV.
Practice Q2. Two amplifiers are cascaded: the first has noise figure 3 dB and gain 20 dB; the second has noise figure 10 dB. The overall noise figure is closest to:
(a) 3.2 dB (b) 6.5 dB (c) 10 dB (d) 13 dB
Answer: (a) 3.2 dB
Linear: \( F_1 = 2 \), \( G_1 = 100 \), \( F_2 = 10 \). \( F = 2 + 9/100 = 2.09 \), \( 10\log_{10}2.09 = 3.2 \) dB. Option (d) adds the dB values; dB figures never add in Friis.
Practice Q3. A fourth-order Butterworth low-pass filter with cut-off \( f_c \) attenuates a signal at \( 2f_c \) by about:
(a) 6 dB (b) 12 dB (c) 24 dB (d) 80 dB
Answer: (c) 24 dB
\( \lvert H\rvert = 1/\sqrt{1 + 2^8} = 1/\sqrt{257} = 0.062 \), i.e. \( -24.1 \) dB. Asymptotically \( 80 \) dB/decade is \( 24 \) dB/octave. Option (d) uses a full decade.
Practice Q4. A thermocouple amplifier shows 50 Hz hum that grows when the sensor and the amplifier are plugged into earthed sockets in different rooms. The most effective cure is:
(a) a grounded copper box round the amplifier (b) a differential (instrumentation) input or an isolation amplifier (c) a higher-gain amplifier (d) a longer averaging time
Answer: (b)
The symptom is a ground loop: two earth points at different potentials put the difference in series with the signal. A differential input turns it into common mode (rejected by the CMRR); isolation breaks the loop. A copper box (a) stops electric fields, not ground-potential differences; gain (c) amplifies the hum too; averaging (d) does not remove a coherent 50 Hz component unless synchronised to reject it.
Level 3: Part C Style
Practice Q5. A photodiode delivering a photocurrent of 10 nA (dark current 1 nA) feeds a transimpedance amplifier with \( R_f = 10 \) MΩ at 300 K; the measurement bandwidth is 1 kHz. Neglect amplifier voltage noise. Consider:
- The output signal is 0.1 V.
- The shot-noise current is about 1.9 pA rms.
- The Johnson noise current of \( R_f \) is about 1.3 pA rms, so the measurement is not yet shot-noise limited.
- The SNR is about 4400 (about 73 dB).
Which are correct?
(a) 1 and 2 only (b) 1, 2 and 4 only (c) 1, 2, 3 and 4 (d) 2 and 3 only
Answer: (b) 1, 2 and 4 only
Statement 1: \( V = IR_f = 10^{-8} \times 10^7 = 0.1 \) V. True.
Statement 2: \( i_{sh} = \sqrt{2q(I_{ph} + I_d)B} = \sqrt{2 \times 1.6 \times 10^{-19} \times 1.1 \times 10^{-8} \times 10^3} = 1.88 \) pA. True.
Statement 3: \( i_J = \sqrt{4k_BTB/R_f} = \sqrt{4 \times 1.38 \times 10^{-23} \times 300 \times 10^3/10^7} = 1.29 \) pA. That is smaller than the shot noise, so the measurement is shot-noise limited. (Equivalent check: \( IR_f = 0.11 \) V exceeds \( 2k_BT/q \approx 52 \) mV.) False.
Statement 4: total \( \sqrt{1.88^2 + 1.29^2} = 2.28 \) pA; SNR \( = 10\ \text{nA}/2.28\ \text{pA} \approx 4400 \), \( 20\log_{10}4400 \approx 73 \) dB. True.
Practice Q6. An amplifier has input voltage noise 5 nV/\( \sqrt{\text{Hz}} \) and input current noise 1 pA/\( \sqrt{\text{Hz}} \), at 300 K. Consider:
- The optimum source resistance for noise is 5 kΩ.
- With a 1 kΩ source, the total input noise density is about 6.5 nV/\( \sqrt{\text{Hz}} \).
- With a 1 kΩ source, the noise figure is about 4 dB.
- Cooling the source resistor to 77 K would lower the noise figure as defined at the source temperature.
Which are correct?
(a) 1 and 2 only (b) 1, 2 and 3 only (c) 2 and 3 only (d) 1, 2, 3 and 4
Answer: (b) 1, 2 and 3 only
Statement 1: \( R_{\text{opt}} = e_n/i_n = 5 \times 10^{-9}/10^{-12} = 5 \) kΩ. True.
Statement 2: \( e_{\text{tot}}^2 = 25 + 1 + 16.6 \) (nV)\( ^2 \)/Hz (the last is \( 4k_BT \times 1\ \text{k}\Omega \)), so \( e_{\text{tot}} = 6.5 \) nV/\( \sqrt{\text{Hz}} \). True.
Statement 3: \( F = 42.6/16.6 = 2.57 \), NF \( = 4.1 \) dB. True.
Statement 4: cooling the source lowers its own thermal noise, so the amplifier’s (unchanged) noise becomes a larger fraction: \( F \) relative to a colder source rises, even though the absolute SNR improves. False. This is why noise figure must always be stated with its reference temperature.
Practice Q7. Consider these statements about shielding and grounding:
- A grounded thin aluminium enclosure is effective against capacitively coupled interference from a nearby high-voltage line.
- At 1 MHz a 0.5 mm copper sheet provides tens of decibels of absorption loss because it is several skin depths thick.
- A cable shield for a low-frequency microvolt signal should be grounded at both ends to give the lowest impedance.
- Reducing the area enclosed between a signal wire and its return reduces 50 Hz magnetic pickup in proportion to the area.
Which are correct?
(a) 1, 2 and 4 only (b) 1 and 3 only (c) 2, 3 and 4 only (d) 1, 2, 3 and 4
Answer: (a) 1, 2 and 4 only
Statement 1: electric-field shielding needs only a grounded conductor. True.
Statement 2: \( \delta \approx 65 \) µm, so \( t/\delta \approx 7.7 \) and \( A = 8.69 \times 7.7 \approx 67 \) dB. True.
Statement 3: at low frequency, grounding both ends makes the shield part of a ground loop. False.
Statement 4: \( \mathcal E = -d\Phi/dt = -A\,dB/dt \), proportional to loop area. True.
Quick Revision Table
| Item | Result |
|---|---|
| Johnson noise | \( \sqrt{4k_BTRB} \); 1 kΩ, 300 K: 4.07 nV/\( \sqrt{\text{Hz}} \) |
| Available thermal power | \( k_BTB \); \( -174 \) dBm/Hz at 290 K |
| kTC noise | \( \sqrt{k_BT/C} \); 64 µV for 1 pF at 300 K |
| Shot noise | \( \sqrt{2qIB} \); 1 µA: 0.57 pA/\( \sqrt{\text{Hz}} \) |
| Shot-limited when | \( IR > 2k_BT/q \approx 52 \) mV |
| 1/f noise | below the corner; escape by modulation |
| Quadrature sum | \( \sqrt{v_1^2 + v_2^2} \) |
| Noise figure | \( F = \text{SNR}_{\text{in}}/\text{SNR}_{\text{out}} \); \( T_e = (F - 1)290 \) K |
| Optimum source | \( R_{\text{opt}} = e_n/i_n \) |
| Friis | \( F_1 + (F_2 - 1)/G_1 + \cdots \) |
| Averaging | SNR \( \times\sqrt N \) |
| RC filter | \( f_c = 1/2\pi RC \); ENBW \( = (\pi/2)f_c \) |
| \( n \)th order | \( 20n \) dB/decade; Butterworth \( 1/\sqrt{1 + (f/f_c)^{2n}} \) |
| Shield absorption | \( 8.69\,t/\delta \) dB |
| Grounding | single point (LF), ground plane (HF), differential inputs, isolation |
Key Takeaways
What’s Next
- Fourier Transforms, Lock-in Detection, Box-car Integration and Modulation: narrowing the bandwidth to fractions of a hertz around a modulated signal.
- High-Frequency Devices: skin depth, transmission lines and the receivers where noise figure is the headline specification.
- Linear and Nonlinear Curve Fitting and the Chi-Square Test: how the noise becomes the \( \sigma_i \) of a fit.
- Electronics and Experimental Methods (Advanced): Quick Revision.