Short Course for Qualcomm -- Analysis and Design of LC VCOs -- by B Razavi 2010 [good]

  • 格式:pdf
  • 大小:2.15 MB
  • 文档页数:95
Short Course for Qualcomm
Analysis and Design of LC VCOs
April 19, 2010
Behzad Razavi Electrical Engineering Department University of California, Los Angeles
33
Analysis of Phase Noise
Need to answer these questions: • • How does noise injected by a device corrupts the phase? How much noise does each device inject? Tens of papers have been published on phase noise in oscillators. Many mechanisms result in phase noise. No single approach has been sufficient to give insight into all mechanisms. We follow two approaches here: - Approach I: based on time averages (a) the average spectrum of noise of a device while the noise spectrum varies with time. (b) the “average resistance” - Approach II: based on phase response of an oscillator to an injected impulse in the time domain [Hajimiri & Lee, JSSC, Feb. 98].
The spectrum at ω0 is thus given by:
37
Linear Model?
Find the noise injected by transistors at equilibrium:
But by what impedance should this noise be multiplied?
13
Cross-Coupled Oscillator
• •
Looks like a diff pair with positive feedback. Oscillation freq is given by:
14
Problem of Swings

Peak Vds must not stress the transistors.

But on-resistance of switches lowers tank Q:
28
Trade-Off Between Ron and Tuning Range

It can be shown that for :
[Razavi, ISCAS07]
29
Use of “Floating” Switch
31
Basic Concepts
• Random abberations in zero crossings:

Or random fluctuations in period (i.e., frequency).
32
Q of An Oscillator

Q is not infinite here!
AM
Nonlinear Circuit
PM
PM
The AM component is suppressed by the loop nonlinearity (and subsequent stages).
36
Derivation
Alternatively, we can express noise in terms of its quadrature components:
3
Basics
4
MOS Varactors
• •
Simpler to use than pn junctions. C/V characteristic scales with technology.
5
Q-Range Trade-Off
6
Symmetric Inductors

Inductors driven differentially have a higher Q.

Now include the varactor:
22
VCO Type II

Select device dimension to set the output CM level to about Vdd/2.
23
Varactor Modulation by IDD
• •
Noise of current mirror becomes the dominant source. Does this effect exist in Type I VCO?
17
NMOS vs. PMOS
18
Frequency Tuning (Type I)
• •
To maximize tuning range, we wish to minimize C1. But C1 is given by: - Caps of M1 and M2 (including 4Cgd) - Cap of L1 - Input cap of next stage
7
Feedback Oscillator
8
Output Swing

Peak differential output voltage swing is given by:
9
One-Port View

Example of negative resistance:
10
3-Point Oscillator
34
Cyclostationary Noise
Periodically-switched white noise is white:
35
Conversion of Additive Noise to Phase Noise
Noise injected into an oscillator introduces both phase and amplitude modulation:
30
Study of Phase Noise
• Basic Concepts • Conversion of Additive Noise to Phase Noise • Approach I • Effect of Tail Noise • Class-C Oscillator • Approach II • AM/PM Conversion
Program
• • • • • •
Basic Design of LC VCOs VCO Topologies Phase Noise Quadrature Oscillators Injection Locking and Pulling Case Studies
2
Voltage-Controlled Oscillators
15
Supply Sensitivity

Voltage-dependent Cdb results in a finite Kvco from Vdd to output frequency:
16
One-Port View

Oscillation condition easier to meet than in 3point topologies:
40
Simulations
Freq. = 10 GHz, Tail Current 1 mA, 65 –nm tech. Vdd = 1.2 V, Q = 10, W/L = 10 um / 60 nm. When comparing reported oscillators, note the following: - Do internal swings stress the devices? - How are the inductors designed? Do they use thick metal? - Always normalize to power consumption and center frequency.
26
Oscillation Amplitude vs. Frequency
• Suppose the tank inductor has only a series resistance:

Oscillation amplitude falls as freq is lowered.
27
Discrete Tuning
38
Time-Varying Resistance
What happens if the tail current is increased?
Average transconductance must exactly cancel Rp.
39
Overall Model
Impedance of lossless LC tank:
24
VCO Type III

Tuning range:

With 5% bottom-plate parasitic cap:
25
VCO Type IV
• • • Select device dimension to set the output CM level to about Vdd/2. Output swing twice that of previous topologies. But tail noise modulates varactors.