Electronic Noise Physics116C, 6/2/06 D. Pellett Reference: Horowitz and Hill, The Art of Electronics, 2nd Ed., Ch. 7 Any original material copyright D. Pellett 2006 Electronic Noise • • Distinguish between Noise and Interference Resistor at temperature T is a source of a random noise voltage due to thermal fluctuations of charge carriers • • • Called “Johnson” or “Thermal” noise Spectral distribution is flat in frequency (“white noise”) Other types of noise: • Shot noise arises when charges cross a barrier • • Poisson statistics – also has flat frequency distribution Flicker noise, approx. 1/f spectrum • • Has fractal properties suggesting chaotic origin f α: α may not be exactly -1 Johnson Noise Derivation (1) Thermal equilibrium at temperature T • • • Noise signal from left R travel right, absorbed in right R Noise signal from right R travels left, absorbed in left R Resistors come to thermal equilibrium, both R’s at same T, power flowing left = power flowing right (If not in equilibrium, hotter resistor sends more power until equilibrium reached) J.B. Johnson, Phys. Rev. 29, 367 (1927) H. Nyquist, Phys. Rev. 32, 110 (1928) Johnson Noise Derivation (2) In the frequency interval ∆f, let Pav = available power from each resistor. Assume PavL = PavR = universal function of T: Pav = <Vn2>av/R The voltage source produces a wave propagating at velocity v on the line. Energy on line from left resistor: E = ∫ Pav dt = Pav ∆t = Pav L/v ∆t (where ∆t is the time to traverse the line, L = v∆t) Total energy on line = 2Pav L/v (from both R’s) Johnson Noise Derivation (3) Now suddenly short-circuit both ends of the line. Energy is trapped in standing waves on the line. Normal modes (the standing waves) within the frequency range ∆f are excited. Natural frequencies are f = nv/2L Ends are now shorted – get voltage nodes at each end Density of states in frequency: dn = (2L/v) df Number of states in the frequency interval ∆f : Nf = (2L/v) ∆f This is a quantum harmonic oscillator. • Equally spaced levels, level spacing hf • E = hf (n + ½) per photon in each level. nλ/2 = L f = v/λ = nv/2L Johnson Noise Derivation (4) Find average energy per mode*, assuming hf /kT 1 av. occupancy = 1/(exp(hf /kT) - 1) (Bose-Einstein Dist’n) E = hf per photon in mode (ignoring zero point energy) <E>av = average energy <E>av = hf /(exp(hf /kT) - 1) ≈ kT (if hf /kT 1). (Get classical thermodynamic result in this approximation – independent of f). Total energy on line: Nf <E>av = (2L/v) ∆f <E>av = 2Pav L/v or Pav ≈ kT∆f . *See, for example, Feynman’s Lectures on Physics Johnson Noise derivation (5) • Relate to Vn: • Since Vn2 just depends on ∆f and not directly on f, the spectrum is flat in frequency – “white” noise (as long as hf /kT 1) • For a 10 kΩ resistor at T = 300 K, ∆f = 10 kHz <Vn>rms = (4 x 1.38x10-23 x 300 x 10000 x 10000)1/2 = 1.29 μV Johnson Noise Vn Distribution (B = bandwidth, ∆f ) Resistor Johnson Noise Data Gaussian Vn dist’n Waveform (short and long time scales) • Flat frequency spectrum (fb = 16 kHz for antialias filter) Data collected using Johnson Noise Test Fixture from 116C Lab Shot Noise • • Due to fluctuations in the number of independent charge carriers crossing a boundary - random process following Poisson distribution, quantized by carrier charge: <In>rms = (2qIDCB)1/2 Gaussian In dist’n, uniform freq. dist’n (like Johnson noise) Example: current in a vacuum diode (thermionic emission of carriers in hot cathode at ground potential, collection on anode at +HV) Also photodiode or silicon strip leakage current, BJT collector, base current, FET gate leakage current, etc. Not current crossing imaginary plane in conductor or bulk semiconductor - carriers are correlated (or shot noise effects averaged over many collisions in short mean free paths, reducing rms fluctuations - see paper by Landauer) • • • Flicker (1/f ) Noise • • Present in various electronic components, e.g., • Carbon composite resistors – have higher 1/f noise than metal film or wire-wound • Amplifiers, particularly evident at low frequencies – see Op Amp spectra in right-hand column A.K.A. “Pink noise” (emphasizes low frequencies) LT1793 1/f Noise Data Gaussian Vn dist’n Waveform – note self-similar time dist’ns on different scales (fractal-like) Frequency Spectrum (log-log scales) • Source: pseudorandom 1/f distribution from computer waveform generator. (Note antialias filter cutoff in freq. spectrum near 20 kHz.) BJT Noise Model Base spreading resistance ≈ 40 Ω - 400 Ω (decreases as current increases) JFET Noise Model Non-Inverting Amplifier Noise Model • LT1793 (first stage in Johnson Noise Test Fixture): typical en = 6 nV Hz-1/2, in = 1 fA Hz-1/2 (f ≈ 1 kHz) • AD797 (second stage amplifier): typical en = 0.9 nV Hz-1/2, in = 2 pA Hz-1/2 (f ≈ 1 kHz) LT1793 Noise Performance • en and in are measured in noninverting configuration – evidently can use “as is” in estimating noise of your amplifier AD797 Noise Performance • This is the second stage amplifier. In good design, its noise contribution should be negligible relative to the amplified noise from the first stage – the case here. • One would use this amplifier as the first stage if the signal source impedance was < 1kΩ