+ All Categories
Home > Documents > HgCdTe Noise from µHz to kHz

HgCdTe Noise from µHz to kHz

Date post: 07-Jan-2016
Category:
Upload: shamus
View: 30 times
Download: 0 times
Share this document with a friend
Description:
Detectors for Astronomy Garching, 2009-10-14. HgCdTe Noise from µHz to kHz. Roger Smith, Gustavo Rahmer, David Hale, Elliott Koch Caltech. Measure it the way you will use it. Conclusion. In theory , there is no difference between theory and practice , - PowerPoint PPT Presentation
Popular Tags:
29
HgCdTe Noise from µHz to kHz Roger Smith, Gustavo Rahmer, David Hale, Elliott Koch Caltech Detectors for Astronomy Garching, 2009-10-14
Transcript
Page 1: HgCdTe Noise from  µHz to kHz

HgCdTe Noise from µHz to kHz

Roger Smith,

Gustavo Rahmer, David Hale, Elliott Koch

Caltech

Detectors for AstronomyGarching, 2009-10-14

Page 2: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

3

Conclusion

In theory, there is no difference

between theory and practice,

but, in practice, there is. Jan L. A. van de Snepscheut or Yogi Berra ?

Measure it the way you will use it.

Page 3: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

4

Noise studies in progress

• We are particularly interested in the noise floor where many samples are combined.

• At the extremes of exposure time, are the causes for the noise floor the same ?

Applications Exp. t Window Processing

Wavefront Sensing millisec 4x4 Extreme Fowler

Imaging minutes Full frame Moderate Fowler

Spectroscopy hours 300x500 Extreme, least squares fit

Page 4: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

5

First, optimize pixel timing

• 10 µs/pixel is standard. We had 3µs dwell.• We reduced overheads to 2.16µs, and overlapped this with signal settling.• For 3µs dwell, pixel time is halved: sample twice as often with same noise bandwidth.

For Astronomical Research Cameras Inc. 8ch IR video card

Page 5: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

6

Settle and Dwell Time Optimization6

Will Signal-to-Noise ratio be improved more by:– increasing settling time above 2µs, or

– adding more dwell time (noise BW limiting), or

– coadding more frames ?

More coadds are better than more settling.

More dwell is better at high frequency, with most gain by 4us; slightly worse at low frequency.

6µs/pixel is good compromise.

Small window for fast readout

Page 6: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

8

Possible causes of noise floor ?

• Dark currentIdark < 0.002 e-/s for 1.7µm @120K, < 0.004 e-/s for 2.5µm @80K.

• Mux glow ?Iglow < 0.0034 e-/read for 5µs/pixel.

Keep sample rate << 1.7s/read, so Iglow << Idark

• 1/f noise in detector material.

• RTS noise in mux (on small number of pixels)

• Bias variationsStabilize biases; remove common mode with ref pixels.

• Thermal variationsGood temperature control (~0.8e-/mK)Constant cadence clocking for uniform self heating.Could use metal trace on mux to track its temperature better then apply correction based on per pixel temperature coefficient.

Page 7: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

9

Dark signal … Is this mux glow? 9

For SUR at 2s/sample, Idark = 0.008 e-/s

For small fast windows, 0.0034e-/read at 6µs/pixel

Frame number Time (s)

Page 8: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

10

Self-heating masquerades as mux glow

As window size is reduced same power is concentrated in smaller area so temperature rises: dark current increases with number of reads rather like mux glow.

8x8 windowAfter160,000 frame SUR in 75s

32x32 windowAfter 10,000 frame SUR in 75s

8x8 Hot spot in next readout

Page 9: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

12

Spatial variation in noise

Noise histogram has high tail.

Why worry?..

• Wavefront sensing: don’t want small guide window to land on a bad pixel.

• Spectrocopy: don’t want key spectral feature on a bad pixel.

RTS noise in mux?

Page 10: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

13

Raw pixel values vs Time (no coadding)

• Noisiest pixels exhibit “Random Telegraph Signal” a bimodal noise distribution due to single traps in or channel near buffer FET.

• Number of such traps and distance from channel produce a spectrum of amplitudes.

• Characteristic time constants vary widely.

• All silicon transistors suffer from this to some extent. In big transistors many traps are in play and it accounts for 1/f noise. In small transistors one or a few traps produce RTS noise.

• Cooling increases the time constant. Slow traps become so slow they become invisible, but fast traps which would average to zero now move into signal passband.

Quiet pixel

Excess noise is due to RTS in mux

Raw

val

ue m

inus

1st

fram

e (A

DU

)

Frame number

Noisy pixel

Page 11: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

14

Histogram of RTS noisefor the nasty case of two traps about the same size

Time series

Page 12: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

15

Same after coadd and subtract (100 coadds)

• For time series on previous slide

• Differencing turns steps into spikes.

• Coadding helps but noise is still

• Better to reject outliers than try to average them away

Page 13: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

16Spatial distribution of Noisedifferent processing of same data in each case

Fowler 1 Fowler 16 Fowler 256

Page 14: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

17

Noise vs exposure time

Same SUR data in both cases:

• For CDS use samples n sec apart.

• For CDS sum or n sec, then subtract from sum of next n sec.

Maximum number of fowler samples at 0.5Hz fitting into each exposure time, for this data point fowler 50.

Fowler 5

Page 15: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

18

Power Spectral Distribution

From SUR data used in previous slide

Page 16: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

20

Raw, CDS with alternate samples

SUR at 0.5Hz:

CDS frames synthesized from alternate samples

Page 17: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

21

Raw, CDS with alternate samples

SUR at 0.5Hz:

CDS frames synthesized every 2nd sample.

Page 18: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

22

Raw, CDS with alternate samples

Page 19: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

23

Raw, CDS with alternate samples

Page 20: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

24

PSD for CDS and Fowler 100

Page 21: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

25

Noise vs frame rate for small windows, deep sampling

Turn up due to dark current + mux glow

Latest low noise 2.5µm recipe

Frame rate after fowler sampling

Noise floor due to 1/f noise.

Kink

Fixed by excluding hot pixels not present in smaller windows

Page 22: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

26

Power spectra vs Sample rate26

1/f noise causes floor at low frequencies

If noise power spectrum is a property of the detector, why does the 1/f corner and white noise floor change with SUR sample rate (window size) ?

Nyquist ~ 2.1kHz 1/f corner ~ 3.5Hz

Page 23: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

27

PSD, sampling at 0.5Hz

Frequency range for previous slide

Nyquist =0.25 Hz

1/f corner = 0.0035Hz

Isn’t the power spectrum a property of the detector?

How can it change with sample rate ?

Page 24: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

28

Aliasing “101”P

ower

Den

sity

Sample rate/2

Sample rate

Sample rate*3/2

BW ~ 1/pixel time

~ 1/ frame time

Page 25: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

29

Aliasing “101”P

ower

Den

sity

Sample rate/2

Sample rate

Sample rate*3/2

Page 26: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

30

Simulated Aliasing of 1/f + white noise

White noise above nyquist shows up in baseband due to aliasing.

Nyquist frequency

No aliasing

Elevated noise floor due to aliases

With alias

Without alias

Page 27: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

31

Aliasing of pure 1/f noise

Even pure 1/f looks like it has a white noise floor after aliasing.

Nyquist frequency

No aliasing

Flattening due to aliases

With aliasing

Without

Page 28: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

32

Why does 1/f corner move ?

• Noise BW ~ 2/pixel_T– For CCD, sample rate = 1/pixel_T

– For mulitplexed detector, sample rate = 1/frame_T

……most of the noise BW is above Nyquist.

• White noise floor is raised by aliasing …illustrated in next slides … This lowers the 1/f corner.

This explains how fowler sampling can still work even when one expects 1/f noise to dominate.

Page 29: HgCdTe Noise from  µHz to kHz

Garching, 2009-10-14HgCdTe Noise from µHz to kHz

33

Conclusion

In theory, there is no difference

between theory and practice,

but, in practice, there is. Jan L. A. van de Snepscheut or Yogi Berra ?

Measure it the way you will use it.

PS: Data comparing noise spectra for 1.7µm and 2.5µm materials will be submitted on the web site.


Recommended