Skip to content
Telecom Electronics

Oscillators: Colpitts, stability and phase noise

Frequency and start-up condition of the Colpitts proved, frequency stability versus Q proved, phase noise and Leeson's model, crystals, TCXO and OCXO.

Every radio has at least one oscillator: it sets the transmit frequency and the conversion frequency in the receiver. It is not enough that it oscillates: it must have the right frequency, stable, and a clean spectrum. In this lesson we look at the oscillation condition, the Colpitts oscillator with the proof of its frequency and start-up condition, the link between quality factor and frequency stability proved, phase noise and its consequences, and crystals.

The oscillation condition

An oscillator is an amplifier with frequency-selective feedback. By the Barkhausen criterion (Applied Electronics lesson on oscillators), a sinusoidal oscillation is sustained at the frequency where the loop gain has magnitude 1 and phase a multiple of 360°. At switch-on the gain must be somewhat greater than 1, so that noise grows into an oscillation; then non-linearity (amplifier compression, lesson on non-linearity) reduces the average gain to 1 and stabilises the amplitude.

The Colpitts oscillator

The Colpitts uses a π network with one inductor and two capacitors, and an inverting amplifier of transconductance g_m; the network provides the other 180° of phase shift.

Colpitts oscillator
An inverting amplifier of transconductance g m has its input on capacitor C 1 and its output on capacitor C 2 and resistor R, which represents the load and losses. Inductor L connects the output to the input. The oscillation frequency is 1 over 2 pi root of L times the series combination of C 1 and C 2.

Theorem. The circuit oscillates at the frequency

    \[ \omega_0 = \frac{1}{\sqrt{LC_s}},\qquad C_s = \frac{C_1C_2}{C_1 + C_2} \]

and starts if g_mR \ge C_1/C_2.

Proof. Let v_1 be the input voltage (on C_1) and v_2 the output voltage (on C_2); the amplifier injects the current -g_mv_1 into the output node. Current law at node 1, where the amplifier input draws no current: j\omega C_1v_1 + (v_1 - v_2)/(j\omega L) = 0, hence v_2 = v_1(1 - \omega^2LC_1) and (v_2 - v_1)/(j\omega L) = j\omega C_1v_1. At node 2:

    \[ v_2\left(\frac{1}{R} + j\omega C_2\right) + j\omega C_1v_1 + g_mv_1 = 0 \]

Substituting v_2 and dividing by v_1 \ne 0:

    \[ (1 - \omega^2LC_1)\left(\frac{1}{R} + j\omega C_2\right) + j\omega C_1 + g_m = 0 \]

The imaginary part vanishes if C_2(1 - \omega^2LC_1) + C_1 = 0, i.e. \omega^2LC_1 = 1 + C_1/C_2, which gives \omega^2 = (C_1 + C_2)/(LC_1C_2) = 1/(LC_s). The real part then gives g_m = -(1 - \omega^2LC_1)/R = C_1/(C_2R): this is the gain that exactly sustains the oscillation, and start-up needs a larger value. ∎

Example. L = 1 µH, C_1 = C_2 = 200 pF: C_s = 100 pF and f_0 = 1/(2\pi\sqrt{10^{-6}\cdot 10^{-10}}) \approx 15.9 MHz. With R = 1 kΩ we need g_m \ge 1 mS; the design uses a margin of 3-5 times to guarantee start-up over tolerances and temperature. The ratio C_1/C_2 sets how much signal returns to the input.

Quality factor and stability

An oscillator’s frequency shifts when something adds phase to the loop: the transistor capacitance changing with temperature, the supply voltage, the load. A resonator with a high quality factor makes the frequency insensitive to these phase shifts.

Theorem. If a spurious phase shift \Delta\varphi appears in the loop, the frequency shifts by about

    \[ \frac{\Delta\omega}{\omega_0} = \frac{\Delta\varphi}{2Q} \]

Proof. Near resonance the impedance phase of a parallel resonator is \varphi(\omega) = -\arctan(2Q\,\Delta\omega/\omega_0) \approx -2Q\,\Delta\omega/\omega_0, where \Delta\omega = \omega - \omega_0 (lesson on resonant circuits). To keep the total loop phase at zero, the resonator must compensate the spurious phase shift, i.e. work at the frequency where 2Q\,\Delta\omega/\omega_0 = \Delta\varphi. ∎

Example. A phase shift of 0.01 rad with an LC resonator of Q = 50 shifts the frequency by 10^{-4}, 100 ppm: at 1 GHz that is 100 kHz, unacceptable for a radio channel. With a crystal of Q = 10^5 the shift drops to 5\cdot 10^{-8}.

Phase noise

Component noise also makes the oscillator’s phase fluctuate: the spectrum is not a line, but a line with two “skirts” on either side. It is measured as phase noise \mathcal{L}(\Delta f): the power in a 1 Hz bandwidth at an offset \Delta f from the carrier, relative to the carrier, in dBc/Hz. Leeson’s model estimates it as:

    \[ \mathcal{L}(\Delta f) \approx 10\log_{10}\left[\frac{2FkT}{P_s}\left(1 + \left(\frac{f_0}{2Q\,\Delta f}\right)^2\right)\right] \]

where F is the amplifier’s noise figure and P_s the power in the resonator. The resonator converts the amplifier’s phase noise into frequency noise close to the carrier, falling as 1/\Delta f^2; again, a high Q and a high power reduce it.

The consequences are practical. In the receiver, a strong signal on an adjacent channel mixes with the local oscillator’s skirts and falls into the wanted channel (reciprocal mixing); in dense digital modulations such as 256-QAM phase noise rotates the constellation points and increases errors (lesson on digital modulation performance).

Crystals

A quartz crystal cut into a plate vibrates mechanically, and through the piezoelectric effect behaves like a resonant circuit with Q from 10^4 to 10^6, very stable with temperature. It is the reference of almost all oscillators: temperature-compensated crystal oscillators (TCXO) reach a stability of 0.5 ppm, oven-controlled ones (OCXO) 10^{-9}. High, variable frequencies are obtained by locking a voltage-controlled oscillator to the crystal with a phase-locked loop (Electrical Communications lesson on FM and PLLs), which within its bandwidth inherits its stability and low phase noise.

Subscribe
Notify of
guest
0 Comments
Oldest
Newest Most Voted
0
Would love your thoughts, please comment.x
()
x