You are on page 1of 112

Short Course On Phase-Locked Loops and Their Applications Day 2, AM Lecture Basic Building Blocks Voltage-Controlled

Oscillators
Michael Perrott August 12, 2008 Copyright © 2008 by Michael H. Perrott All rights reserved

VCO Design for Wireless Systems
Zin

From Antenna and Bandpass Filter

PC board trace

Mixer Package Interface RF in LNA LO signal VCO IF out To Filter

Zo

Reference Frequency

Frequency Synthesizer

Design Issues

- Tuning Range – need to cover all frequency channels - Noise – impacts receiver blocking and sensitivity performance - Power – want low power dissipation - Isolation – want to minimize noise pathways into VCO - Sensitivity to process/temp variations – need to make it
manufacturable in high volume
2

M.H. Perrott

VCO Design For High Speed Data Links
Zin

From Broadband Transmitter

PC board trace

Data

Zo

Package Interface

In Amp

Clock and Data Recovery

Clk

Data Out

Data In

Phase Detector

Loop Filter VCO

Clk Out

Design Issues

- Same as wireless, but:
Required noise performance is often less stringent Tuning range is often narrower
3

M.H. Perrott

Impedance transformation . Perrott 4 .Negative feedback topologies Voltage controlled oscillators M.Outline of Talk Common oscillator implementations Barkhausen’s criterion of oscillation One-port view of resonance based oscillators .H.

H.Popular VCO Structures LC oscillator VCO Amp Vout C L Rp -Ramp Vin Ring oscillator Vout Vin -1 LC Oscillator: low phase noise. large area Ring Oscillator: easy to integrate. higher phase noise M. Perrott 5 .

1. Perrott 6 .2.H.Amounts to two conditions: Gain = 1 at frequency wo Phase = n360 degrees (n = 0.…) at frequency wo M.Barkhausen’s Criteria for Oscillation x=0 e H(jw) y Barkhausen Criteria e(t) Closed loop transfer function Asin(wot) H(jwo) = 1 Self-sustaining oscillation at frequency wo if Asin(wot) y(t) .

Assume N stages each with phase shift ΔΦ .H.Alternately. N stages with delay Δt M.Example 1: Ring Oscillator Δt (or ΔΦ) A B C A A B C Gain is set to 1 by saturating characteristic of inverters Phase equals 360 degrees at frequency of oscillation A T . Perrott 7 .

JSSC.Further Info on Ring Oscillators Due to their relatively poor phase noise performance. Perrott 8 .berkeley.. Dec 1993 (look at pp 127128 for delay cell description) Todd Weigandt’s PhD thesis – http://kabuki.They are used quite often in high speed data links.edu/~weigandt/ M. though We will focus on LC oscillators in this lecture Some useful info on CMOS ring oscillators .Maneatis et.H. ring oscillators are rarely used in RF systems . “Precise Delay Generation Using Coupled Oscillators”.eecs. al.

Assuming G M.Example 2: Resonator-Based Oscillator Z(jw) -1 Vx GmVx Rp Cp Lp Vout 0 Vx -1 -Gm Z(jw) Vout Barkhausen Criteria for oscillation at frequency wo: . Perrott m is purely real.H. Z(jwo) must also be purely real 9 .

Perrott . purely real) at resonance Phase condition is satisfied Magnitude condition achieved by setting GmRp = 1 10 M.Looks like resistor (i.H.A Closer Look At Resonator-Based Oscillator Rp 0 Gm Z(jw) Vout |Z(jw)| 0 90o 0o -90o wo 10 wo 10wo Z(jw) w For parallel resonator at resonance ..e.

As gain (G R ) increases. Perrott .Impact of Different Gm Values jw S-plane Increasing GmRp Open Loop Resonator Poles and Zero σ Locus of Closed Loop Pole Locations Root locus plot allows us to view closed loop pole locations as a function of open loop poles/zero and open loop gain (GmRp) m p .H. closed loop poles move into right half S-plane 11 M.

Perrott 12 .Impact of Setting Gm too low GmRp < 1 jw S-plane Closed Loop Step Response Open Loop Resonator Poles and Zero σ Locus of Closed Loop Pole Locations Closed loop poles end up in the left half S-plane .Underdamped response occurs Oscillation dies out M.H.

Unstable response occurs Waveform blows up! M. Perrott 13 .H.Impact of Setting Gm too High jw S-plane GmRp > 1 Closed Loop Step Response Open Loop Resonator Poles and Zero σ Locus of Closed Loop Pole Locations Closed loop poles end up in the right half S-plane .

How do we achieve this in practice? M.Setting Gm To Just the Right Value jw GmRp = 1 Open Loop Resonator Poles and Zero σ Locus of Closed Loop Pole Locations S-plane Closed Loop Step Response Closed loop poles end up on jw axis . Perrott 14 .Oscillation maintained Issue – GmRp needs to exactly equal 1 .H.

and adds noise 15 M.H.Complex. we can precisely achieve G R m p =1 Issues . Perrott .Amplitude Feedback Loop Output Oscillator Adjustment of Gm Peak Detector Desired Peak Value One thought is to detect oscillator amplitude. requires power. and then adjust Gm so that it equals a desired value .By using feedback.

Harmonics created. Perrott 16 .Leveraging Amplifier Nonlinearity as Feedback -Gm 0 -1 Vx Ix Z(jw) Vout |Vx(w)| A wo GmA W 0 |Ix(w)| 0 wo 2wo 3wo W Practical transconductance amplifiers have saturating characteristics .H.Our interest is in the relationship between the input and the fundamental of the output M. but filtered out by resonator .

Effective gain from input to fundamental of output drops . Perrott 17 .Amplitude feedback occurs! (G R = 1 in steady-state) m p M.H.Leveraging Amplifier Nonlinearity as Feedback -Gm 0 -1 Vx Ix Z(jw) Vout |Vx(w)| GmA A wo GmA W 0 |Ix(w)| A Gm W 0 wo 2wo 3wo GmRp=1 A As input amplitude is increased .

One-Port View of Resonator-Based Oscillators Zactive Active Negative Resistance Generator Zres Vout Resonator Active Negative Resistance Zactive 1 = -Rp -Gm Zres Resonator Vout Rp Cp Lp Convenient for intuitive analysis Here we seek to cancel out loss in tank with a negative resistance element . we must have 18 M. Perrott .H.To achieve sustained oscillation.

RLC networks can have secondary (or more) resonant frequencies.H. we can always do series to parallel transformations to achieve a parallel network for analysis Cs RsC Ls Rp RsL Cp Lp . Perrott 19 .Warning – in practice.One-Port Modeling Requires Parallel RLC Network Since VCO operates over a very narrow band of frequencies. which cause undesirable behavior Equivalent parallel network masks this problem in hand analysis Simulation will reveal the problem M.

H. Perrott 20 .Understanding Narrowband Impedance Transformation Series Resonant Circuit Cs Zin Ls Zin Parallel Resonant Circuit Rs Cp Lp Rp Note: resonance allows Zin to be purely real despite the presence of reactive elements M.

Comparison of Series and Parallel RL Circuits Series RL Circuit Ls Zin Rs Zin Lp Rp Parallel RL Circuit Equate real and imaginary parts of the left and right expressions (so that Zin is the same for both) . Perrott 21 .H.Also equate Q values M.

Also equate Q values M.Comparison of Series and Parallel RC Circuits Series RC Circuit Cs Zin Parallel RC Circuit Rs Zin Cp Rp Equate real and imaginary parts of the left and right expressions (so that Zin is the same for both) . Perrott 22 .H.

Perrott 23 .H.Example Transformation to Parallel RLC Assume Q >> 1 Ls Zin Series to Parallel Transformation Rs Zin C Lp Rp C Note at resonance: M.

Perrott 24 .Tapped Capacitor as a Transformer C1 Zin L C2 RL To first order: We will see this used in Colpitts oscillator M.H.

Advantages Simple topology Differential implementation (good for feeding differential circuits) Good phase noise performance can be achieved M.Negative Resistance Oscillator Include loss in inductors and capacitors Vout Vout C1 M1 Vs M2 C2 C1 M1 Vs M2 Vout C2 RL1 L1 RL2 L2 L1 Vout L2 RC1 Ibias Ibias RC2 This type of oscillator structure is quite popular in current CMOS implementations .H. Perrott 25 .

Typically.H. such loss is dominated by series resistance in the inductor M.Analysis of Negative Resistance Oscillator (Step 1) RL1 L1 Vout C1 M1 Vs M2 RL2 L2 Rp1 Cp1 Vout Lp1 Lp2 Cp2 Vout M2 Rp2 Vout C2 RC1 Ibias Narrowband parallel RLC model for tank M1 Vs RC2 Ibias Derive a parallel RLC network that includes the loss of the tank inductor and capacitor . Perrott 26 .

H. but close enough) s Recognize that we have a diode connected device with a negative transconductance value Note: Gm is large signal transconductance value M. Perrott 27 .Analysis of Negative Resistance Oscillator (Step 2) Rp1 Cp1 Vout M1 Vs Lp1 Vout M2 Rp1 Cp1 Vout M1 Lp1 -1 Rp1 Cp1 Vout Lp1 1 -Gm1 Ibias Split oscillator circuit into half circuits to simplify analysis .Replace with negative resistor as being incremental ground (this is not quite true.Leverages the fact that we can approximate V .

Resulting R p value is as large as possible Choose bias current (Ibias) for large swing (without going far into saturation) .We’ll estimate swing as a function of I shortly Choose transistor size to achieve adequately large g .H.Design of Negative Resistance Oscillator Rp1 Cp1 Vout M1 Vs Lp1 Lp2 Cp2 Vout M2 Rp2 Rp1 Cp1 Vout Lp1 1 -Gm1 A A Gm1 gm1 Gm1Rp1=1 Ibias A Design tank components to achieve high Q . Perrott .Usually twice as large as 1/R to guarantee startup bias p1 m1 28 M.

Usually twice as large as 1/R to guarantee startup bias p1 m1 29 M.We’ll estimate swing as a function of I in next slide Choose transistor size to achieve adequately large g .Calculation of Oscillator Swing Rp1 Cp1 Vout M1 Lp1 Lp2 Cp2 Vout M2 Rp2 A I1(t) Vs I2(t) A Ibias Design tank components to achieve high Q .H. Perrott .Resulting R p value is as large as possible Choose bias current (Ibias) for large swing (without going far into saturation) .

assume I1(t) is a square wave (DC and harmonics filtered by tank) I1(t) |I1(f)| Ibias/2 W=T/2 1 I π bias 1 I 3π bias t 1 1 T W f .Resulting oscillator amplitude M. Perrott 30 .We are interested in determining fundamental component Ibias Ibias/2 T T .Calculation of Oscillator Swing as a Function of Ibias By symmetry.H.Fundamental component is .

PMOS. 2000 (pp 286-287) M.H. “Design Issues in CMOS Differential LC Oscillators”. JSSC. Perrott .Variations on a Theme Bottom-biased NMOS Top-biased NMOS Top-biased NMOS and PMOS Ibias L1 Vout C1 M1 Vs M2 L2 Vout C2 Vout C1 M1 M2 L1 L2 Vout C2 C1 Vout M1 M2 Ibias M3 Ld Vout C2 M4 Ibias Biasing can come from top or bottom Can use either NMOS. May 1999 and Feb. al. or both for transconductor achieve better phase noise at a given power dissipation See Hajimiri et.Use of both NMOS and PMOS for coupled pair would appear to 31 .

Differential structure can also be implemented Good phase noise can be achieved. but not apparent there is an advantage of this design over negative resistance design for CMOS applications M. Perrott 32 .Colpitts Oscillator L Vout Vbias M1 C1 V1 Ibias C2 Carryover from discrete designs in which single-ended approaches were preferred for simplicity .H.Achieves negative resistance with only one transistor .

H.Analysis of Cap Transformer used in Colpitts Rin= 1 N 2 RL C1 Vout L C2 V1 RL Vout L C1||C2 1:N RL V1=NVout Vout L C1||C2 1 N2 RL C 1C 2 C1||C2 = C1+C2 N= C1 C1+C2 Voltage drop across RL is reduced by capacitive voltage divider .Assume that impedances of caps are less than R resonant frequency of tank (simplifies analysis) Ratio of V1 to Vout set by caps and not RL L at Power conservation leads to transformer relationship shown M. Perrott 33 .

Reduces swing at source node (important for bipolar version) -1/Gm N Transformer ratio set to achieve best noise performance M.H.Reduces loading on tank .Simplified Model of Colpitts L L Vout Rp Include loss in tank vout C1||C2 L vout Rp|| 1/Gm N2 Vbias M1 1/Gm id1 C1 V1 M1 v1 C1||C2 1/Gm N2 -GmNvout Ibias C2 C1||C2 = Nvout C 1C 2 C1 N= C1+C2 C1+C2 C1||C2 L vout Purpose of cap transformer Rp|| 1/Gm N2 . Perrott 34 .

Design of Colpitts Oscillator L I1(t) Vbias M1 1/Gm Vout A C1||C2 L vout Req= Rp|| 1/Gm N2 C1 V1 Gm1 -1/Gm N Ibias C2 gm1 NGm1Req=1 A Design tank for high Q Choose bias current (Ibias) for large swing (without going far into saturation) Choose transformer ratio for best noise .Rule of thumb: choose N = 1/5 according to Tom Lee Choose transistor size to achieve adequately large gm1 M.H. Perrott 35 .

Perrott 36 .Calculation of Oscillator Swing as a Function of Ibias I1(t) consists of pulses whose shape and width are a function of the transistor behavior and transformer ratio .Resulting oscillator amplitude M.Fundamental component is .H.Approximate as narrow square wave pulses with width W I1(t) Ibias W |I1(f)| Ibias average = Ibias T T t 1 T 1 W f .

H. Perrott 37 .Allows inductor voltage to achieve a larger amplitude without exceeded the max allowable voltage at the drain Good for achieving lower phase noise M.Clapp Oscillator L C3 Llarge Vout Vbias M1 1/Gm C1 V1 Ibias C2 Same as Colpitts except that inductor portion of tank is isolated from the drain of the device .

Perrott 38 .Hartley Oscillator C Vout Vbias M1 L2 L1 V1 Cbig Ibias Same as Colpitts.Not popular for IC implementations due to the fact that M. but uses a tapped inductor rather than series capacitors to implement the transformer portion of the circuit capacitors are easier to realize than inductors .H.

Perrott 39 .Integrated Resonator Structures Inductor and capacitor tank .H.Spiral inductors have moderate Q (5 to 10).Bondwire inductors have high Q (> 40). but not as “integrated” and have poor tolerance (> § 20%) Lateral Capacitor A A Spiral Inductor Bondwire Inductor package die A A B C1 B B Lm B M. but completely integrated and have tight tolerance (< § 10%) .Lateral caps have high Q (> 50) .

7 GHz CMOS VCO”. “A low-noise transformer-based 1.. pp 286-287 A B Cpar1 k C D L1 L2 C Cpar2 D A B M. ISSCC 2002. al. Perrott 40 .H.Leverages self and mutual inductance for resonance to achieve higher Q .Integrated Resonator Structures Integrated transformer .See Straayer et.

Looks like parallel LC tank! Benefit – very high Q can be achieved with fancy dielectric Negative – relatively large area (external implementation in the past).Quarter Wave Resonator λ0/4 ZL x y Z(λ0/4) z z L 0 Impedance calculation (from Lecture 4) .H. but getting smaller with higher frequencies! M. Perrott 41 .

for resonant frequency . but non-tunable and haven’t made it to the GHz range.High frequency. yet.More on this topic in the last lecture this week 42 MEMS devices M. but non-tunable and poor accuracy .Very high Q.Cantilever beams – promise high Q. and very accurate and stable resonant frequency Confined to low frequencies (< 200 MHz) Non-integrated Used to create low noise.Other Types of Resonators Quartz crystal . Perrott .H.FBAR – Q > 1000. but poor accuracy (for resonant frequency) . accurate. “reference” oscillators SAW devices .

Perrott 43 .Typically use a variable capacitor (varactor) (transistor junctions. interconnect.) Fixed cap lowers frequency tuning range Varactor incorporated by replacing fixed capacitance .Note that much fixed capacitance cannot be removed M.Voltage Controlled Oscillators (VCO’s) L Vout M1 Vs M2 Vout L1 Vout Cvar Vcont L2 Vbias Cvar M1 C1 V1 Ibias Vcont Ibias Cvar Include a tuning element to adjust oscillation frequency . etc.H.

Model for Voltage to Frequency Mapping of VCO T=1/Fvco L1 Vout Cvar Vcont Vin Vbias M1 Vs L2 Vout M2 Cvar VCO frequency versus Vcont Fvco Fout slope=Kv Vin Vbias Vcont fo Ibias Model VCO in a small signal manner by looking at deviations in frequency about the bias point output frequency . Perrott .H.Assume linear relationship between input voltage and 44 M.

The two are related through an integral relationship Intuition of integral relationship between frequency and phase 1/Fvco= α out(t) out(t) 1/Fvco= α+ε M. Perrott 45 .Model for Voltage to Phase Mapping of VCO Phase is more convenient than frequency for analysis .H.

Perrott v is in units of Hz/V L2 Vout M2 Vs Cvar Frequency Domain VCO Model M1 vin 2πKv s Φout Ibias 46 .H.Note that K T=1/Fvco L1 Vout Cvar Vcont Vin Vbias M.Frequency Domain Model of VCO Take Laplace Transform of phase relationship .

Varactor Implementation – Diode Version
Consists of a reverse biased diode junction

- Variable capacitor formed by depletion capacitance - Capacitance drops as roughly the square root of the
bias voltage

Advantage – can be fully integrated in CMOS Disadvantages – low Q (often < 20), and low tuning range (§ 20%)
V+ V+ VP+ N+ N- n-well P- substrate

Cvar

V-

Depletion Region

V+-V-

M.H. Perrott

47

A Recently Popular Approach – The MOS Varactor
Consists of a MOS transistor (NMOS or PMOS) with drain and source connected together Advantage – easily integrated in CMOS Disadvantage – Q is relatively low in the transition region
region will be swept across each VCO cycle
V+ V+ W/L VVCox
N+ PDepletion Region N+

- Abrupt shift in capacitance as inversion channel forms - Note that large signal is applied to varactor – transition
Cvar

Cdep VT V+-V-

M.H. Perrott

48

A Method To Increase Q of MOS Varactor
to VCO

Cvar

C W/L LSB

2C 2W/L

4C W/L 4W/L

MSB

Vcontrol

Overall Capacitance

000 001 010 011 Coarse 100 Control 101 110 111

Vcontrol Fine Control

Coarse Control

Fine Control

High Q metal caps are switched in to provide coarse tuning Low Q MOS varactor used to obtain fine tuning See Hegazi et. al., “A Filtering Technique to Lower LC Oscillator Phase Noise”, JSSC, Dec 2001, pp 1921-1930
M.H. Perrott 49

thereby causing a shift in its capacitance .Supply Pulling and Pushing L1 Vout Cvar Vcont M1 Vs M2 L2 Vout L Vout Vbias Cvar M1 C1 V1 Ibias Vcont Ibias Cvar Supply voltage has an impact on the VCO frequency .H.Voltage across varactor will vary.Voltage across transistor drain junctions will vary. Perrott 50 . thereby causing a shift in its depletion capacitance This problem is addressed by building a supply regulator specifically for the VCO M.

H. Perrott 51 .Injection Locking Vnoise Vin VCO Vout |Vnoise(w)| W |Vout(w)| W wo Δw wo -Δw Δw |Vout(w)| Noise close in frequency to VCO resonant frequency can cause VCO frequency to shift when its amplitude becomes high enough |Vnoise(w)| W wo Δw |Vnoise(w)| wo -Δw Δw |Vout(w)| W wo Δw |Vnoise(w)| W wo -Δw Δw |Vout(w)| W wo Δw W wo -Δw Δw W M.

H. Perrott 52 .Injection locking can happen if inadequate isolation from mixer RF input to LO port Follow VCO with a buffer stage with high reverse isolation to alleviate this problem M. VCO frequency can be very close to that of interferers RF in(w) Interferer Desired Narrowband Signal W LNA RF in Mixer 0 wint wo LO frequency LO signal Vin .Example of Injection Locking For homodyne systems.

Increased range achieved by using switched capacitor .Summary Several concepts are useful for understanding LC oscillators .Impedance transformations Voltage-controlled oscillators incorporate a tunable element such as varactor network for coarse tuning Improves varactor Q. as well . coupling Several things to watch out for M. injection locking.Barkhausen criterion .Supply pulling.H. Perrott 53 .

Noise in Voltage Controlled Oscillators .

Receiver – lower sensitivity.VCO Noise in Wireless Systems Zin From Antenna and Bandpass Filter PC board trace Mixer Package Interface RF in LNA LO signal VCO IF out To Filter Zo Reference Frequency Frequency Synthesizer Phase Noise f fo VCO noise has a negative impact on system performance . Perrott .Transmitter – increased spectral emissions (output spectrum must meet a mask requirement) 55 Noise is characterized in frequency domain M. poorer blocking performance .H.

H.Receiver – increases bit error rate (BER) .VCO Noise in High Speed Data Links Zin From Broadband Transmitter PC board trace Data Zo Package Interface In Amp Clock and Data Recovery Clk Data Out Data In Phase Detector Loop Filter VCO Clk Out Jitter VCO noise also has a negative impact on data links .Transmitter – increases jitter on data stream (transmitter must have jitter below a specified level) 56 Noise is characterized in the time domain M. Perrott .

Noise figure .Equipartition theorem .Outline of Talk System level view of VCO and PLL noise Linearized model of VCO noise .Hajimiri model Cyclo-stationary view of VCO noise Back to Leeson’s formula M.H.Leeson’s formula . Perrott 57 .

Noise Sources Impacting VCO Time-domain view Jitter out(t) t Frequency-domain view Sout(f) PLL dynamics set VCO carrier frequency (assume noiseless for now) vc(t) Extrinsic noise vn(t) vin(t) Intrinsic noise out(t) Phase Noise fo f Extrinsic noise Intrinsic noise M.Noise from other circuits (including PLL) .H. Perrott .Noise due to the VCO circuitry 58 .

Model as phase signal in output sine waveform M.VCO Model for Noise Analysis Time-domain view Jitter out(t) t Note: Kv units are Hz/V PLL dynamics set VCO carrier frequency (assume noiseless for now) vc(t) Extrinsic noise vn(t) vin(t) Intrinsic noise 2πKv s Φvn(t) Φout Frequency-domain view Sout(f) Phase Noise fo 2cos(2πfot+Φout(t)) out(t) f We will focus on phase noise (and its associated jitter) .H. Perrott 59 .

Simplified Relationship Between Φout and Output PLL dynamics set VCO carrier frequency (assume noiseless for now) vc(t) Extrinsic noise vn(t) vin(t) Intrinsic noise 2πKv s Φvn(t) Φout 2cos(2πfot+Φout(t)) out(t) Using a familiar trigonometric identity Given that the phase noise is small M. Perrott 60 .H.

phase spectral density can be placed into one of two categories .H.Calculation of Output Spectral Density Calculate autocorrelation Take Fourier transform to get spectrum . Perrott 61 .Φ (t) is periodic out out M.Phase noise – Φ (t) is non-periodic .Spurious noise .Note that * symbol corresponds to convolution In general.

H.Output Spectrum with Phase Noise Suppose input noise to VCO (vn(t)) is bandlimited.In practice. Perrott 62 . non-periodic noise with spectrum Svn(f) . derive phase spectrum as Resulting output spectrum Ssin(f) 1 f SΦout(f) -fo fo * Sout(f) 1 dBc/Hz SΦout(f) f fo f -fo M.

Perrott is small in deviation (i. as currently assumed) 63 out(t) .Units are dBc/Hz For this case .e. when carrier is not modulated.Valid when Φ M.Measurement of Phase Noise in dBc/Hz Sout(f) 1 dBc/Hz -fo fo SΦout(f) f Definition of L(f) ..H.

H.Units are dBc/Hz For this case . Perrott 64 .Single-Sided Version Sout(f) 1 dBc/Hz fo SΦout(f) f Definition of L(f) remains the same . we can work with either one-sided or two-sided spectral densities since L(f) is set by ratio of noise density to carrier power M.So.

Perrott fspur 65 .H.Output Spectrum with Spurious Noise Suppose input noise to VCO is Resulting output spectrum Ssin(f) 1 f SΦout(f) -fo fo * Sout(f) 1 dBc fo 1 dspur 2 2 fspur f -fspur fspur 1 dspur 2 2 fspur f -fo M.

Perrott 66 .We are assuming double sided spectra. so integrate over positive and negative frequencies to get power Either single or double-sided spectra can be used in practice For this case M.H.Measurement of Spurious Noise in dBc Sout(f) 1 dBc -fo fo fspur f 1 dspur 2 2 fspur Definition of dBc .

Perrott 67 .Calculation of Intrinsic Phase Noise in Oscillators Zactive Active Negative Resistance Generator Zres Vout Resonator Active Negative Resistance Zactive 1 = -Rp -Gm Zres Resonator inRn Vout Rp inRp Cp Lp Noise sources in oscillators are put in two categories .H.Noise due to tank loss .Noise due to active negative resistance We want to determine how these noise sources influence the phase noise of the oscillator M.

Perrott 68 .Equivalent Model for Noise Calculations Active Negative Resistance 1 = -Rp -Gm Zactive Zres Resonator inRn Vout Rp inRp Cp Lp Noise Due to Active Negative Resistance Compensated Resonator with Noise from Tank 1 = -Rp -Gm inRn Vout Rp inRp Cp Lp Noise Due to Active Negative Resistance Noise from Tank Ztank Ideal Tank inRn inRp Vout Cp Lp M.H.

H. Perrott 69 .Calculate Impedance Across Ideal LC Tank Circuit Ztank Cp Lp Calculate input impedance about resonance =0 negligible M.

H.A Convenient Parameterization of LC Tank Impedance Ztank Cp Lp Actual tank has loss that is modeled with Rp . Perrott 70 .Define Q according to actual tank Parameterize ideal tank impedance in terms of Q of actual tank M.

Overall Noise Output Spectral Density Noise Due to Active Negative Resistance Noise from Tank Ztank Ideal Tank inRn inRp Vout Cp Lp Assume noise from active negative resistance element and tank are uncorrelated .H. Perrott 71 .Note that the above expression represents total noise that impacts both amplitude and phase of oscillator output M.

Perrott 72 .Parameterize Noise Output Spectral Density Noise Due to Active Negative Resistance Noise from Tank Ztank Ideal Tank inRn inRp Vout Cp Lp From previous slide F(Δf) F(Δf) is defined as M.H.

Perrott 73 .Fill in Expressions Noise Due to Active Negative Resistance Noise from Tank Ztank Ideal Tank inRn inRp Vout Cp Lp Noise from tank is due to resistor Rp Ztank(Δf) found previously Output noise spectral density expression (single-sided) M.H.

H.Amplitude variations suppressed by feedback in oscillator 74 M. Perrott .Separation into Amplitude and Phase Noise Noise Due to Active Negative Resistance Noise from Tank Ztank Ideal Tank inRn inRp Vout Cp Lp Amplitude Noise Vsig A t vout t Vout A Phase Noise t Equipartition theorem states that noise impact splits evenly between amplitude and phase for Vsig being a sine wave .

H. Perrott 75 .Output Phase Noise Spectrum (Leeson’s Formula) Output Spectrum Noise Due to Active Negative Resistance Noise from Tank Ztank Ideal Tank SVsig(f) Carrier impulse area normalized to a value of one inRn inRp Vout Cp Lp L(Δf) fo Δf f All power calculations are referenced to the tank loss resistance. Rp M.

Perrott 76 .Example: Active Noise Same as Tank Noise Active Negative Resistance 1 = -Rp Vout -Gm Resonator inRn Rp inRp Cp Lp Noise factor for oscillator in this case is L(Δf) Resulting phase noise -20 dB/decade log(Δf) M.H.

Perrott 77 .Noise from M and M is modulated on and off .H. transistors can degrade Q of tank M.Noise from M is modulated before influencing V .Transistors have 1/f noise 1 3 2 out Also.The Actual Situation is Much More Complicated inRp1 Tank generated noise Rp1 Cp1 Vout Lp1 Lp2 Cp2 Vout Rp2 inRp2 Tank generated noise A A inM1 Transistor generated noise M1 Vs M2 inM2 Transistor generated noise Ibias Vbias M3 inM3 Impact of tank generated noise easy to assess Impact of transistor generated noise is complicated .

Phase Noise of A Practical Oscillator L(Δf) -30 dB/decade -20 dB/decade 10log ( Δf1/f 3 fo 2Q 2FkT Psig ( log(Δf) Phase noise drops at -20 dB/decade over a wide frequency range. Perrott .H.Low frequencies – slope increases (often -30 dB/decade) . but deviates from this at: .High frequencies – slope flattens out (oscillator tank does not filter all noise sources) 78 Frequency breakpoints and magnitude scaling are not readily predicted by the analysis approach taken so far M.

Perrott 79 .Phase Noise of A Practical Oscillator L(Δf) -30 dB/decade -20 dB/decade 10log ( Δf1/f 3 fo 2Q 2FkT Psig ( log(Δf) Leeson proposed an ad hoc modification of the phase noise expression to capture the above noise profile .H.Note: he assumed that F(Δf) was constant over frequency M.

Perrott .Amplitude noise is suppressed through feedback (or by amplitude limiting in following buffer stages) Our main concern is phase noise We argued that impact of noise divides equally between amplitude and phase for sine wave outputs M.A More Sophisticated Analysis Method Ideal Tank Amplitude Noise Vout iin Vout Cp Lp Phase Noise A t Our concern is what happens when noise current produces a voltage across the tank .H.What happens when we have a non-sine wave output? 80 .Such voltage deviations give rise to both amplitude and phase noise .

Amplitude deviations are suppressed M.Modeling of Phase and Amplitude Perturbations Ideal Tank Amplitude Noise Vout iin Vout Cp Lp Phase Noise iin(t) A t Phase iin(t) τo t hΦ(t.H.τ) Amplitude iin(t) τo t A(t) hA(t.Phase deviations are accumulated .τ) Φout(t) Φout(t) τo A(t) τo t t iin(t) Characterize impact of current noise on amplitude and phase through their associated impulse responses . Perrott 81 .

τ) τo τ1 t If we vary the time at which the current impulse is injected.Impact of Noise Current is Time-Varying Ideal Tank Amplitude Noise Vout iin Vout Cp Lp Phase Noise iin(t) A t Phase iin(t) τo τ1 t Φout(t) hΦ(t. Perrott 82 .τ) Φout(t) τo τ1 A(t) t iin(t) Amplitude iin(t) τo τ1 t A(t) hA(t. its impact on phase and amplitude changes .Need a time-varying model M.H.

Perrott 83 .Illustration of Time-Varying Impact of Noise on Phase T iin(t) qmax t0 iin(t) qmax t1 iin(t) qmax t2 iin(t) qmax t3 iin(t) qmax t4 Vout(t) ΔΦ t Vout(t) ΔΦ t t Vout(t) ΔΦ=0 t t Vout(t) ΔΦ t t Vout(t) ΔΦ t t t High impact on phase when impulse occurs close to the zero crossing of the VCO output Low impact on phase when impulse occurs at peak of output M.H.

Observe that it is periodic with same period as VCO output 84 .Define Impulse Sensitivity Function (ISF) – Γ(2πfot) T iin(t) qmax t0 iin(t) qmax t1 iin(t) qmax t2 iin(t) qmax t3 iin(t) qmax t4 Vout(t) ΔΦ t Vout(t) ΔΦ t t Vout(t) ΔΦ=0 t Γ(2πfot) T t Vout(t) ΔΦ t t t Vout(t) ΔΦ t t t ISF constructed by calculating phase deviations as impulse position is varied M.H. Perrott .

H. Perrott t hΦ(t.Parameterize Phase Impulse Response in Terms of ISF T iin(t) qmax t0 iin(t) qmax t1 iin(t) qmax t2 iin(t) qmax t3 iin(t) qmax t4 Vout(t) ΔΦ t Vout(t) ΔΦ t t Vout(t) ΔΦ=0 t Γ(2πfot) T t Vout(t) ΔΦ t t t Vout(t) ΔΦ t t t iin(t) τo M.τ) Φout(t) τo t 85 .

Perrott 86 .Its magnitude indicates where VCO waveform is most sensitive to noise current into tank with respect to creating phase noise ISF is periodic In practice. Γ) is approximately proportional to derivative of VCO output waveform .H.e. derive it from simulation of the VCO M.Examples of ISF for Different VCO Output Waveforms Example 1 Vout(t) Vout(t) Example 2 t t Γ(2πfot) Γ(2πfot) t t ISF (i..

Perrott 87 .H.τ) Φout(t) Computation of phase deviation for an arbitrary noise current input Analysis simplified if we describe ISF in terms of its Fourier series (note: co here is different than book) M.Phase Noise Analysis Using LTV Framework in(t) hΦ(t.

Perrott 88 .Block Diagram of LTV Phase Noise Expression Input Current Normalization in(t) 1 qmax ISF Fourier Series Coefficients 2 co 2 Integrator Φout(t) 1 j2πf Phase to Output Voltage out(t) 2cos(2πfot+Φout(t)) 2cos(2πfot + θ1) c1 2 2cos(2(2πfo)t + θ2) c2 2 2cos(n2πfot + θn) cn 2 Noise from current source is mixed down from different frequency bands and scaled according to ISF coefficients M.H.

H. Perrott 1 -3fo 1 f 3fo ( ( 2 i2 0 SD(f) fo f -fo 0 fo f 89 .Phase Noise Calculation for White Noise Input (Part 1) Note that 2 in is the single-sided noise spectral density of in(t) Δf (q ( 2Δf n 1 2 i2 SX(f) -3fo -2fo -fo 0 fo 2 i2 max 2fo SA(f) 3fo f in(t) 1 qmax X 2 A 1 f 0 n 2 q max 2Δf (1( 1 -fo n 2 q max 2Δf 2cos(2πfot + θ1) B 2cos(2(2πfo)t + θ2) 1 -fo 1 f fo ( ( 1 2 i2 0 SB(f) fo f -fo n 2 q max 2Δf C 2cos(3(2π)fot + θn) 1 -2fo 1 f 2fo ( ( 1 2 i2 0 SC(f) fo f -fo n 2 q max 2Δf D M.

Phase Noise Calculation for White Noise Input (Part 2) 2 i2 n 2 q max 2Δf (1( 1 SA(f) f ISF Fourier Series Coefficients A 0 SB(f) fo co 2 Integrator Φout(t) 1 j2πf Phase to Output Voltage out(t) 2cos(2πfot+Φout(t)) -fo n 2 q max 2Δf ( ( 1 2 i2 B 0 SC(f) fo f c1 2 -fo n 2 q max 2Δf ( ( 1 2 i2 C 0 SD(f) fo f c2 2 -fo n 2 q max 2Δf ( ( 2 i2 D 0 fo f -fo c3 2 M.H. Perrott 90 .

etc. SB(f).Spectral Density of Phase Signal From the previous slide Substitute in for SA(f). Perrott 91 .H. Resulting expression M.

H. Perrott 92 .Output Phase Noise SΦout(f) Sout(f) 1 dBc/Hz f 0 -fo 0 fo Δf SΦout(f) f We now know Resulting phase noise M.

The Impact of 1/f Noise in Input Current (Part 1) Note that 2 in is the single-sided noise spectral density of in(t) Δf (q ( 2Δf n 1 2 i2 SX(f) 1/f noise -3fo -2fo -fo 0 fo 2 i2 max 2fo SA(f) 3fo f in(t) 1 qmax X 2 A 1 f 0 n 2 q max 2Δf (1( 1 -fo n 2 q max 2Δf 2cos(2πfot + θ1) B 2cos(2(2πfo)t + θ2) 1 -fo 1 f fo ( ( 1 2 i2 0 SB(f) fo f -fo n 2 q max 2Δf C 2cos(3(2π)fot + θn) 1 -2fo 1 f 2fo ( ( 1 2 i2 0 SC(f) fo f -fo n 2 q max 2Δf D 1 -3fo 1 f 3fo ( ( 2 i2 0 SD(f) fo f -fo 0 fo f 93 M. Perrott .H.

The Impact of 1/f Noise in Input Current (Part 2) SA(f) ISF Fourier Series Coefficients Integrator Φout(t) 1 j2πf Phase to Output Voltage out(t) 2cos(2πfot+Φout(t)) n 2 q max 2Δf (1( 1 2 i2 A fo f B fo f C co 2 -fo n 2 q max 2Δf ( ( 1 2 i2 0 SB(f) c1 2 -fo n 2 q max 2Δf ( ( 1 2 i2 0 SC(f) -fo n 2 q max 2Δf ( ( 2 i2 0 SD(f) fo f D c2 2 -fo 0 fo f c3 2 M. Perrott 94 .H.

H. Perrott 95 .Calculation of Output Phase Noise in 1/f3 region From the previous slide Assume that input current has 1/f noise with corner frequency f1/f Corresponding output phase noise M.

Calculation of 1/f3 Corner Frequency L(Δf) -30 dB/decade (A) -20 dB/decade (B) Δf1/f 3 10log ( fo 2Q 2FkT Psig ( log(Δf) (A) (B) (A) = (B) at: M.H. Perrott 96 .

e.If DC value of ISF is zero. c o is also zero For symmetric oscillator output waveform .DC value of ISF is zero – no upconversion of flicker noise! (i. output phase noise does not have 1/f3 region) For asymmetric oscillator output waveform M.DC value of ISF is nonzero – flicker noise has impact 97 .H. Perrott .Impact of Oscillator Waveform on 1/f3 Phase Noise ISF for Symmetric Waveform Vout(t) ISF for Asymmetric Waveform Vout(t) t Γ(2πfot) Γ(2πfot) t t t Key Fourier series coefficient of ISF for 1/f3 noise is co .

Issue – We Have Ignored Modulation of Current Noise inRp1 Tank generated noise Rp1 Cp1 Vout Lp1 Lp2 Cp2 Vout Rp2 inRp2 Tank generated noise A A inM1 Transistor generated noise M1 Vs M2 inM2 Transistor generated noise Ibias Vbias M3 inM3 In practice.H. Perrott 98 . its associated noise changes dramatically Can we include this issue in the LTV framework? M. transistor generated noise is modulated by the varying bias conditions of its associated transistor .As transistor goes from saturation to triode to cutoff.

Perrott 99 .Inclusion of Current Noise Modulation iin(t) in(t) hΦ(t.H.τ) T = 1/fo Φout(t) α(2πfot) α(2πfot) Recall 0 t By inspection of figure We therefore apply previous framework with ISF as M.

Perrott 100 .Placement of Current Modulation for Best Phase Noise Best Placement of Current Modulation for Phase Noise T = 1/fo α(2πfot) 0 Γ(2πfot) α(2πfot) Worst Placement of Current Modulation for Phase Noise T = 1/fo t 0 Γ(2πfot) t t t Phase noise expression (ignoring 1/f noise) Minimum phase noise achieved by minimizing sum of square of Fourier series coefficients (i. rms value of Γeff) M.H.e.

Colpitts Oscillator Provides Optimal Placement of α T = 1/fo Vout(t) L I1(t) Vbias M1 Vout(t) α(2πfot) t I1(t) 0 Γ(2πfot) C1 t Ibias C2 t Current is injected into tank at bottom portion of VCO swing .H. Perrott 101 .Current noise accompanying current has minimal impact on VCO output phase M.

Summary of LTV Phase Noise Analysis Method Step 1: calculate the impulse sensitivity function of each oscillator noise source using a simulator Step 2: calculate the noise current modulation waveform for each oscillator noise source using a simulator Step 3: combine above results to obtain Γeff(2πfot) for each oscillator noise source Step 4: calculate Fourier series coefficients for each Γeff(2πfot) Step 5: calculate spectral density of each oscillator noise source (before modulation) Step 6: calculate overall output phase noise using the results from Step 4 and 5 and the phase noise expressions derived in this lecture (or the book) M. Perrott 102 .H.

Key question: how do you determine F(Δf)? M.Alternate Approach for Negative Resistance Oscillator Rp1 Cp1 Vout M1 Vs Lp1 Lp2 Cp2 Vout M2 Rp2 A A Ibias Vbias M3 Recall Leeson’s formula .H. Perrott 103 .

F(Δf) Has Been Determined for This Topology
Rp1 Cp1
Vout M1 Vs

Lp1

Lp2

Cp2
Vout M2

Rp2

A

A

Ibias Vbias
M3

Rael et. al. have come up with a closed form expression for F(Δf) for the above topology In the region where phase noise falls at -20 dB/dec:

M.H. Perrott

104

References to Rael Work
Phase noise analysis

- J.J. Rael and A.A. Abidi, “Physical Processes of Phase
Noise in Differential LC Oscillators”, Custom Integrated Circuits Conference, 2000, pp 569-572

Implementation

- Emad Hegazi et. al., “A Filtering Technique to Lower LC
Oscillator Phase Noise”, JSSC, Dec 2001, pp 1921-1930

M.H. Perrott

105

Designing for Minimum Phase Noise
Rp1 Cp1
Vout M1 Vs

Lp1
Vout M2

A

(A)

(B)

(C)

(A) Noise from tank resistance (B) Noise from M1 and M2

Ibias Vbias
M3

(C) Noise from M3

To achieve minimum phase noise, we’d like to minimize F(Δf) The above formulation provides insight of how to do this

- Key observation: (C) is often quite significant

M.H. Perrott

106

H.Component (C) is eliminated! Issue – impedance at node Vs is very low .Elimination of Component (C) in F(Δf) Capacitor Cf shunts noise from M3 away from tank Vout M2 Vs Rp1 Cp1 Vout M1 Lp1 A .Causes M Ibias Vbias M3 inM3 Cf and M2 to present a low impedance to tank during portions of the VCO cycle Q of tank is degraded 1 M. Perrott 107 .

Q preserved! M. Perrott 108 .Fundamental component is at twice the oscillation frequency A T = 1/fo Lf Ibias High impedance at frequency 2fo Place inductor between Vs and current source .Use Inductor to Increase Impedance at Node Vs Voltage at node Vs is a rectified version of oscillator output Vout M2 Vs Rp1 Cp1 Vout M1 Lp1 .H.Choose value to resonate with Cf and parasitic source capacitance at frequency 2fo Vbias M3 inM3 Cf Impedance of tank not degraded by M1 and M2 .

Designing for Minimum Phase Noise – Next Part Rp1 Cp1 Vout M1 Lp1 Vout M2 Vs (A) (B) (C) A (A) Noise from tank resistance (B) Noise from M1 and M2 T = 1/fo Lf Ibias High impedance at frequency 2fo (C) Noise from M3 Vbias M3 inM3 Cf Let’s now focus on component (B) M. Perrott .Depends on bias current and oscillation amplitude 109 .H.

as seen by: sig) to get best phase M.e. P noise.Not true – want to maximize A (i. it would seem that Ibias has no effect! . Perrott 110 .Minimization of Component (B) in F(Δf) Rp1 Cp1 Vout M1 Lp1 Vout M2 Vs (B) Recall from Lecture 11 A Ibias So.H.

Perrott 111 .H. cannot be increased above supply imposed limits If Ibias is increased above the point that A saturates. then (B) increases A Ibias Current-limited regime: amplitude given by Voltage-limited regime: amplitude saturated Best phase noise achieved at boundary between these regimes! M.Current-Limited Versus Voltage-Limited Regimes Rp1 Cp1 Vout M1 Lp1 Vout M2 Vs (B) Oscillation amplitude. A.

Spectre RF from Cadence. FastSpice from Berkeley Design Automation is often utilized to estimate phase noise for integrated oscillators M.Provides a great deal of design insight Practical VCO phase noise analysis is done through simulation these days .Summary Leeson’s model is outcome of linearized VCO noise analysis Hajimiri method provides insights into cyclostationary behavior.H. Perrott 112 . 1/f noise upconversion and impact of noise current modulation Rael method useful for CMOS negative-resistance topology .Closed form solution of phase noise! .