On the 1/f noise in ultra-stable quartz oscillators Enrico Rubiola - - PowerPoint PPT Presentation

On the 1/f noise in ultra-stable quartz oscillators

Amplifier noise Leeson effect Interpretation of Sφ(f) Examples Enrico Rubiola and Vincent Giordano FEMTO-ST Institute, Besançon, France

(CNRS and Université de Franche Comté)

Outline

☹ ☺ ☺

Amplifier white noise

2

b0 = FkT0 P0

white phase noise

Sϕ =

• i=−4

bif i

power law f Sφ(f) low P0 high P0 P0

∑

V0 cos ω0t nrf(t) Noise figure F Input power P0 g Cascaded amplifiers (Friis formula) N = F1kT0 + (F2 − 1)kT0 g2

1

+ . . . As a consequence, (phase) noise is chiefly that of the 1st stage

Amplifier flicker noise

3

near-dc flicker

no carrier

S(f) f

t

S(f) f

noise up-conversion

t a

parametric up-conversion of the near-dc noise

expand and select the ω0 terms carrier + near-dc noise

vi(t) = Vi ejω0t + n′(t) + jn′′(t)

non-linear amplifier

vo(t) = Vi

• a1 + 2a2
• n′(t) + jn′′(t)
• ejω0t

get AM and PM noise

m cascaded amplifiers

In practice, each stage contributes ≈ equally

α(t) = 2 a2 a1 n′(t) ϕ(t) = 2 a2 a1 n′′(t) independent of Vi (!) (b−1)cascade =

m

• i=1

(b−1)i

f Sφ(f) b–1 ≈ independent of P0

Sϕ =

• i=−4

bif i

vo(t) = a1vi(t) + a2v2

i (t) + . . .

substitute

(careful, this hides the down-conversion)

the parametric nature of 1/f noise is hidden in n’ and n”

ω0 = ?

no flicker

ω0

Resonator in the phase space

4

h(t) U(t) H(s) h(t) 1 H(s) b(t) = 1 τ e− t

τ

τ = 2Q ω0 B(s) = 1 1 + sτ δ(t) h(t)

• h(t) dt

H(s) 1/s (1/s) H(s)

1 – theory of linear systems

Laplace t Dirac Heaviside

cos

• ω0t +
• b(t)dt
• cos
• ω0t + U(t)
• resonator

2 – resonator phase response

transform

b(t)

3 – the resonator phase response is a low-pass function

Laplace transform

t=0 ω0 cos[ t] ω0 cos[ t+ κ ] ω0 cos[ t+ κ U(t)] κ −> 0 linearize

this is easy –>

cos[ω0t + b(t)]

resonator

☹☠☈≹∞⁈

cos[ω0t + δ(t)] B(s) U(t)

• b(t) dt

1/s (1/s) B(s)

The Leeson effect

5

resonator

β(s) A

noise free random phase

Σ

+ + (input)

random phase

1

noise free

• scill
• ut

main

• ut

ejΨ

j

e

b

Ψ

resonator

1

Σ

+ + 1 B(s)

• scill
• ut

main

• ut

Ψb(s) Ψ(s) Φ(s) Φo(s)

Σ

+ +

|H(jω)|2 = 1 + ω2τ 2 ω2τ 2 Sϕ o(f) =

• 1 + 1

f 2 ν2 4Q2

• Sψ(f) + Sψ b(f)

H(s) = Φ(s) Ψ(s) = 1 1 − B(s) = 1 + sτ sτ

f Sφ(f) Sψ(f) fL

phase response – use the linear-feedback theory

basic feedback theory

1/f2

H(s) = A 1 − Aβ(s) τ = 2Q ω0

Interpretation of Sφ(f) [1]

6

f

ϕ(f)

b f

−1 −1 f

x −2

fc fL ’ b0f0 b f

−3 −3 FkT0 b0 P0 = check b−3 ν2 σ2

y

2ln(2) = check f2 1 ν0 2Q S (f)

ϕ

Sψ(f) + 1

2

= Leeson effect?

S

real phase-noise spectrum

Sanity check: – power P0 at amplifier input – Allan deviation σy (floor) after parametric estimation

(f) b0f0

Leeson effect?

f−1 b−1a

buffer + sust.ampli

b f

−3 −3

take away the buffer 1/f noise ν0

L

2f’’ Qs= evaluate fL ’’ estimate fL ’ fc 6dB ~ ~ fL ’’

sustaining ampli

S f

ϕ

Interpretation of Sφ(f) [2]

7

2–3 buffer stages => the sustaining amplifier contributes ≲ 25% of the total 1/f noise

(f)

f

x −2

fc fL fL ν0 = 2Qt fL ’ technology => Q t fL ’’ b f

−3 −3

resonator 1/f

• freq. noise

the Leeson effect is hidden f

ϕ s u s t a i n i n g a m p l i

S

Interpretation of Sφ(f) [3]

8

Technology suggests a merit factor Qt. In all xtal oscillators we find Qt ≫ Qs

Example – CMAC Pharao

9

F=1dB b0 => P0=–20.5 dBm (b–3)osc => σy=5.9x10–14, Q=8.4x105 (too low) Q≟2x106 => σy=2.5x10–14 Leeson (too low)

(f) dBrad 2/Hz

ϕ

10−1 102 103 104 b0=−152.5dB (b−3)tot=−132dB f’

L=1.5Hz

f’

c =50Hz L

f"=3Hz

c

f =13Hz (b−1)osc =−141.5dB (b−1)tot=−135.5dB

L

f =1.25Hz

6?

Q=2x10 technology => 10

5

S 10 Fourier frequency, Hz −120 −130 −90 −100 −110 −140 −150 −160 −170 1

Courtesy of CMAC. Interpretation and mistakes are of the authors.

Example – Oscilloquartz 8607

10

F=1dB b0 => P0=–20 dBm (b–3)osc => σy=8.8x10–14, Q=7.8x105 (too low) Q≟2x106 => σy=3.5x10–14 Leeson (too low)

(f) dBrad 2/Hz

ϕ

b0=−153dB f’

L=1.6Hz L

f =1.25Hz (b−1 )osc =−138.1dB

6?

Q=2x10 <= Oscilloquartz 8600

2

103 104 105 106 S Q=7.9x10 5 => (b−1 )tot=−132.5dB 10 (specifications) 1 10 −147 −127 −107 −87 −67 Fourier frequency, Hz −167 (b−3)tot =−128.5dB

L

f"=3.2Hz

Courtesy of Oscilloquartz. Interpretation and mistakes are of the authors.

Example – Wenzel 501-04623

11

F=1dB b0 => P0=0 dBm (b–3)osc => σy=5.3x10–12 Q=1.4x104 Q≟8x104 => σy=9.3x10–13 (Leeson)

5

101 102 103 104 10 −180 −170 −160 −150 −140 −120 −110 −130 −100

Phase noise, dBrad Fourier frequency, Hz /Hz

2

ampli noise (?) Leeson effect (hidden) is about here

L

’’ =3.5kHz

/Hz −30dB/dec dBrad2/Hz b0=−173 specifications

Wenzel 501−04623

f

L =625Hz

f Q=8x104 => guess b−3 =−67 dBrad2

Estimating (b–1)ampli is difficult because there is no visible 1/f region

Data are from the manufacturer web site. Interpretation and mistakes are of the authors.

Other oscillators

12

R = (σy)oscill (σy)Leeson

• floor

=

• (b−3)tot

(b−3)L = Qt Qs = f ′′

L

fL

Oscillator ν0 (b−3)tot (b−1)tot (b−1)amp f ′

L

f ′′

L

Qs Qt fL (b−3)L R Note Oscilloquartz 8600 5 −124.0 −131.0 −137.0 2.24 4.5 5.6×105 1.8×106 1.4 −134.1 10.1 (1) Oscilloquartz 8607 5 −128.5 −132.5 −138.5 1.6 3.2 7.9×105 2×106 1.25 −136.5 8.1 (1) CMAC Pharao 5 −132.0 −135.5 −141.1 1.5 3 8.4×105 2×106 1.25 −139.6 7.6 (2) FEMTO-ST LD prot. 10 −116.6 −130.0 −136.0 4.7 9.3 5.4×105 1.15×106 4.3 −123.2 6.6 (3) Agilent 10811 10 −103.0 −131.0 −137.0 25 50 1×105 7×105 7.1 −119.9 16.9 (4) Agilent prototype 10 −102.0 −126.0 −132.0 16 32 1.6×105 7×105 7.1 −114.9 12.9 (5) Wenzel 501-04623 100 −67.0 −132 ? −138 ? 1800 3500 1.4×104 8×104 625 −79.1 15.1 (6) unit MHz dB rad2/Hz dB rad2/Hz dB rad2/Hz Hz Hz (none) (none) Hz dB rad2/Hz dB Notes (1) Data are from specifications, full options about low noise and high stability. (2) Measured by CMAC on a sample. CMAC confirmed that 2×106 < Q < 2.2×106 in actual conditions. (3) LD cut, built and measured in our laboratory, yet by a different team. Qt is known. (4) Measured by Hewlett Packard (now Agilent) on a sample. (5) Implements a bridge scheme for the degeneration of the amplifier noise. Same resonator of the Agilent 10811. (6) Data are from specifications.

Warning: an effect not accounted for still remains

13

fluctuating impedance

This does not fit general experience on amplifiers, yet it is to be reported A fluctuating impedance that affects the input without participating to the gain

14

Conclusions

The analysis of Sφ(f) provides insight in the oscillator The oscillator 1/f3 phase noise (Allan variance floor)

• riginates from:

– amplifier 1/f noise, via the Leeson effect – resonator instability In actual oscillators, the resonator instability turns out to be the dominant effect

☹ ☺ ☺

qed

Full text available on http://arxiv.org/abs/physics/0602110 Talk slides and full text (20 pages, pdf) available on http://rubiola.org

We owe gratitude to J.-P. Aubry (Oscilloquartz), V. Candelier (CMAC), G.J. Dick (JPL),

• J. Grolambert (FEMTO-ST), L. Maleki (JPL), R. Brendel (FEMTO-ST)

t=0 ω0 cos[ t] ω0 cos[ t+ κ ] ω0 cos[ t+ κ U(t)] κ −> 0 linearize

Details

(t)

O(t)

v’

I (t)

v’’

O

v’ v’’

I (t)

τ τ initial phase κ t v(t) v(t) envelope envelope t t t

cos(ω0t) e−t/τ cos(ω0t + κ)

• 1 − e−t/τ

τ = 2Q ω0

15

t=0 ω0 cos[ t] ω0 cos[ t+ κ ] ω0 cos[ t+ κ U(t)] κ −> 0 linearize

Details [2]

Details [3]

Summary of the amplifier phase noise

18

depends on ϕ(f) b f

−1 −1

b0f0 FkT0 P0 b = b−1~ ~ constant P0

high

P0

low

fc fc P0 P0 f

S

• White PM noise is inversely proportional to P0
• Flicker PM noise is about independent P0
• The corner frequency fc follows

The Leeson effect

19

(microw.)

c

b0f0 b f

−1 −1

fL

f

x −2

Sϕ(f) b0f0

f

x −2

b f

−2 −2

b f

−3 −3 Leeson effect

fL Sϕ(f)

Leeson effect

b f

−3 −3

fc ν0 ν0 f f A − High Q, low (xtal) B − Low Q, high f

two typical patterns

Example – Oscilloquartz 8600

20

F=1dB b0 => P0=–18 dBm (b–3)osc => σy=1.5x10–13, Q=5.6x105 (too low) Q≟1.8x106 => σy=4.6x10–14 Leeson (too low)

−1 =−131dBrad /Hz 2

• sust. ampli + buffer

Leeson effect (hidden)

L

f =1.4Hz b−3 =−124dBrad /Hz

2

b−1 =−137dBrad /Hz

2

f =63Hz

c L

f ’’ =4.5Hz (guessed) /Hz

2

=−155dBrad b0 b resonator instability sustaining amplifier

Oscilloquartz OCXO 8600

Courtesy of Oscilloquartz. Interpretation and mistakes are of the authors

Example – FEMTO-ST prototype

21

F=1dB b0 => P0=–26 dBm (b–3)osc => σy=1.7x10–13, Q=5.4x105 (too low) Q=1.15x106 => σy=8.1x10–14 Leeson (too low)

(there is a problem)

2/Hz

b−1=−130dBrad

• sust. ampli + buffer

L=9.3Hz

f’ ’ fL=4.3Hz 3dB 3dB S (f) dBrad 2Hz

ϕ 2/Hz

b0=−147dBrad apparent

2/Hz

b−3=−116.6dBrad Fourier frequency, Hz /Hz b−3=−123.2dBrad

2/Hz

b−1=−136dBrad sustaining amplifier LD cut, 10 MHz, Q=1.15E6

2

resonator instability Leeson effect (hidden) apparent calculated FEMTO−ST prototype

Example – Agilent 10811

22

F=1dB b0 => P0=–11 dBm (b–3)osc => σy=8.3x10–13, Q=1x105 (too low) Q≟7x105 => σy=1.2x10–13 Leeson (too low)

/Hz dBrad2

ϕ

S (f) 102 103 104 f =320Hz

c

f’

L=50Hz

’ fL 7Hz ~ ~ frequency, Hz 1 10 −107 −117 −127 −137 −147 −167 −157 (guess!) b0 = −162 dB

−1

= −131 dB b

• sust. ampli + buf

b−3 = −103 dB b−1 = −137 dB

• sust. amplifier

Example – Agilent prototype

23

103 104 105 106 dBrad2/Hz 102 −160 −140 −120 −100 −80 1 10 −180 frequency, Hz (guess!)

ϕ

S (f) fL 7Hz ~ ~ fc=400Hz f’

L=32Hz

’

0 = −158dB −1

= −126 dB b

• sust. ampli + buf

b−1 = −132 dB

• sust. amplifier

−3

= −102 dB b b

F=1dB b0 => P0=–12 dBm (b–3)osc => σy=9.3x10–13 Q=1.6x105 Q≟7x105 => σy=2.1x10–13 (Leeson)