47
Week 7 “Midterm” exam in class next Thursday, March 23 Evening sessions this week are Python data analysis package oriented in support of Lab 3 writeups. Given the interruption for the exam the due date will be March 27. Problem Set 4 will be available shortly (and inpart intended for midterm exam preparation). Lab 4 (making beautiful threecolor calibrated images with the Fan Mountain RRRT) will be out early next week. Observing happens the week after midterm. Topics through the midterm: Photon detection/Imaging devices Poisson statistics/noise/background Determining detector “gain” with Poisson statistics Astronomical photometry and filters Aperture vs. PSF fit photometry

Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Week  7

• “Midterm”  exam  in  class  next  Thursday,  March  23• Evening  sessions  this  week  are  Python  data  analysis  package  

oriented  in  support  of  Lab  3  writeups.– Given  the  interruption  for  the  exam  the  due  date  will  be  March  27.

• Problem  Set  4  will  be  available  shortly  (and  in-­‐part  intended  for  midterm  exam  preparation).

• Lab  4  (making  beautiful  three-­‐color  calibrated  images  with  the  Fan  Mountain  RRRT)  will  be  out  early  next  week.– Observing  happens  the  week  after  midterm.

• Topics  through  the  midterm:– Photon  detection/Imaging  devices– Poisson  statistics/noise/background– Determining  detector  “gain”  with  Poisson  statistics– Astronomical  photometry  and  filters

• Aperture  vs.  PSF  fit  photometry

Page 2: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Analysis  in  the  Frequency  Domain

• Any  time  series  signal  can  be  reconstructed  from  the  sum  of  a  continuum  of  sine  waves  of  different  frequencies  and  phases.

• The  “Fourier  Transform”  provides  a  means  of  calculating  the  frequency  spectrum  decomposition  of  a  time-­‐domain  signal.

• |S(f)|2   represents  the  “power  spectrum”  of  the  signal  – the  amount  of  power  in  the  time  series  at  every  frequency.

Page 3: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

The  Ideal  Imaging  DeviceYou  tell        me…

Page 4: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

The  Ideal  Imaging  Device

• Reports  light  incident  on  cells  • Detects  only  photon  events  – no  noise

– No  random  spontaneous  counts  – No  continuous  leakage  of  “fake”  signal

• Does  not  “miss”  incident  photons– Perfect  “quantum  efficiency”– 100%  “fill  factor

• Does  not  saturate– Infinite  “well”  capacity

• Uniform  sensitivity  from  pixel  to  pixel– Perfect  “flat  field”

Page 5: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Real-­‐world  Imaging  Devices• Reports  light  incident  on  cells

– Measures  electronic  signal  ~proportional  to  counts.  

• Detects  only  photon  events  – no  noise– No  random  counts  – Electronic  measurement  is  susceptible  to  noise  

– fake  counts…– No  leakage  of  “fake”  signal

• “Dark  current”  creates  counts  not  originating  from  photons.

• Does  not  “miss”  incident  photons– Perfect  “quantum  efficiency”– 60-­‐80%  typical…very  wavelength  dependent– 100%  “fill  factor– Nope…  but  close

• Does  not  saturate– Infinite  “well”  capacity– Maximum  count  capacity  in  any  cell.    If  the  

source  is  too  bright  the  measurement  fails.• Uniform  sensitivity  pixel  to  pixel

– Nope…  but  close  again.    Pixel  to  pixel  variability  of  a  few  percent.

Page 6: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Quantifying  Light:  Photon  Detection

• Three  methods  for  converting  photons  into  “data”1. Direct  electromagnetic  detection

• Electromagnetic  waves  drive  currents  in  an  electrical  conductor.    Amplifiers  enable  direct  measurement  of  amplitude  (signal)  vs.  frequency  of  oscillation.

à Radio…  save  that  method  for  a  different  class2. Photon  counting  (conversion  of  photons  into  free  electrons  in  free  space  or  

within  an  otherwise  electrically  “insulating”  solid  making  it  more  conductive)• The  photoelectric  effect  describes  the  ability  for  a  photon  to  liberate  an  electron  from  a  metal  if  the  photon  carries  enough  energy  to  overcome  the  potential  barrier  binding  the  electron  to  the  metal.– Direct  evidence  for  the  quantization  of  photon  energy.

• Similar  behavior  can  occur  in  the  solid  state.3. Integrated  photon  response

• Instead  of  liberating  electrons,  photons  can  “warm  up”  a  lump  of  material.– The  smaller  the  lump  the  bigger  the  effect.– The  altered  temperature  leads  to  a  change  in  the  lump’s  properties,  for  

example  altered  electrical  conductivity.

Page 7: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Making  “Free”  Electrons  with  Photons

• A  freed  electron  is  a  detectable  electron  (via  voltage  or  current)– an  electron  can  be  free  in  space  -­‐-­‐ photoelectric  effect– or  it  can  be  ''free''  within  a  crystal  lattice  -­‐-­‐ solid  state  detection

The Photoelectric Effect• Metals  are  characterized  by  a  “work  function”  that  is  the  energy  difference  between  the  highest  energy  state  for  an  electron  within  the  metal  and  the  energy  of  an  electron  in  free  space.

• A  photon  with  energy  in  excess  of  this  work  function  will  liberate  a  free,  detectable,  electron  -­-­ the  photoelectric  effect

.

Page 8: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Making  “Free”  Electrons  with  Photons

• A  freed  electron  is  a  detectable  electron  (via  voltage  or  current)– an  electron  can  be  free  in  space  -­‐-­‐ photoelectric  effect– or  it  can  be  ''free''  within  a  crystal  lattice  -­‐-­‐ solid  state  detection

Dark  Current  

• Warm  metals  will  emit  free  electrons,  those  with  thermal  energy  in  excess  of  the  material's  work  function  

The Photoelectric Effect

• Metals  are  characterized  by  a  work  function  which  determines  the  energy  difference  between  the  highest  energy  state  for  an  electron  within  the  metal  and  the  energy  of  an  electron  in  free  space.

• A  photon  with  energy  in  excess  of  this  work  function  will  liberate  a  free,  detectable  electron  -­-­ the  photoelectric  effect.

Page 9: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

The  Photoelectric  Effect

• Photomultipliers  are  based  on  the  cascade  amplification  of  individual  electrons  liberated  from  a  photocathode  by  the  photoelectric  effect

• Work  functions  for  metals  are  typically  a  few  electron  volts– 1  eV 1240  nm

Page 10: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

http://hyperphysics.phy-astr.gsu.edu/hbase/tables/photoelec.html

Work  Functions  of  Metals

• These  numbers  sure  don’t  look  too  interesting  if  you  goal  is  to  detect  low  energy  photons.

Page 11: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Photomultiplier  Shortcomings

● Poor  wavelength  coverage  due  to  large  work  function  of  materials    (limited  to  visible  operating  wavelength  with  some  exceptions)

● Poor  quantum  efficiency  (<20%  conversion  of  photons  to  electrons)● Thermally  emitted  electrons    (known  as  dark  current,  requiring  cooling  to  suppress)

● Large  single-­‐detector  area.One  big  advantage  →    fast  photon  counting

One  big  disadvantage    à one  tube  =  one  measurement

Page 12: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Solid  State  Detection:  Metals  vs.  Insulators

• At  T=0K,  the  world  contains  only  conductors  and  insulators.• Conductivity  (or  not)  depends  on  how  atomic  energy  levels  shift  and  

spread  as  interatomic  distance  decrease  going  from  a  gas  to  a  solid.• Materials  with  energy  gaps  (as  illustrated  below)  are  insulators.  

Silicon  atoms(energy  levels)

Solid  silicon(energy  bands)

Page 13: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Solid  State  Detection:  Semiconductors

• A  photon  with  energy  (hn)  greater  than  the  gap  energy  (Eg)  can  transform  a  “stuck”  electron  in  the  valence  (insulator)  band  to  a  mobile  electron  in  the  conduction  band.

For  silicon,  specifically,  photons  with  energy  greater  than  1.1  eV can  lift  an  electron  up  to  the  conduction  band  (wavelength  shorter  than  about  1  micron)

hν = hcλ> Eg

Page 14: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Semiconductor  Detectors:  Bandgaps• Photoexcitation only  occurs  if  hn >  bandgap energy.• Different  materials  have  different  “cutoff”  wavelengths.

Material Bandgap Cutoff  WavelengthSilicon 1.1  eV 1.05  micronsGermanium 0.67  eV 1.8  micronsPbS 0.37  eV 3.6  micronsInSb 0.23  eV You  Tell  Me

Carbon 5.5  eV !

Bandgap varies  slightly  with  temperature  because  the  crystalline  lattice  spacing  changes  as  the  temperature  changes.

Q:    based  on  previous  slide,  would  you  expect  greater  or  smaller  gap  in  material  with  more  tightly  packed  atoms?

Page 15: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Semiconductor  Detectors:  Bandgaps• Photoexcitation only  occurs  if  hn >  bandgap energy.• Different  materials  have  different  “cutoff”  wavelengths.

Q:  is  there  a  downside?  Why  not  always  use  a  small  gap?

Page 16: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Semiconductors  and  Cooling

• Semiconductors  are  insulators  with  bandgaps so  small  that  thermal  energy  can  maintain  some  population  of  electrons  in  the  conduction  band  at  room  temperature  making  them  weak  (semi)  conductors.– Of  course  these  same  small  bandgaps make  a  material  interesting  from  a  photon  

detection  perspective.– These  materials  would  suffer  from  significant  dark  current  if  maintained  at  room  

temperature.

• The  solution… operate  the  detectors  at  low  temperature.– Most  inexpensive  CCD  cameras  use  thermoelectric  coolers  (another  cool  but  

complicated  semiconductor  effect)  to  keep  the  detectors  at  -­‐20C/253K  or  cooler.– Professional  CCD’s  operate  at  near  liquid  nitrogen  temperature  (<100K)– Infrared  detectors  (small  bandgap materials)  may  have  to  be  cooled  to  liquid  

helium  temperature  (4K)  if  the  bandgap is  small  enough.

Page 17: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Cryogenics• Since  dark  current  is  the  result  of  thermal  excitation,  cool  the  

detector  so  that  kT <<  bandgap energy.

Longer  wavelength  =  smaller  bandgap =  lower  operating  temperature.

Page 18: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Imaging  Devices• Because  the  solid  state  detector  materials  are  crystalline  (e.g.  silicon)  the  

same  crystal  growth  techniques  used  to  make  integrated  electronic  circuits  (computer  chips)  apply  to  the  detectors  themselves.– Enables  the  construction  of  precision  structures  on  the  submicron  scale  containing  

both  electronics  and  detectors  – Arrays!

Page 19: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Solid  State  Photon  Detection  in  Semiconductors

• Technically  the  electron  is  not  so  much  “free”  in  a  semiconductor  as  it  is  “borrowed”.    

• Photon  excitation  creates  a  unbound  electron  and  a  corresponding  “hole”  within  the  crystal  lattice.    Both  are  mobile.    – If  they  find  one  another  they  recombine  =  no  detection.– As  long  as  they  are  kept  separate  (typically  via  an  electric  field)    they  can  be  

detected…• as  a  current  if  they  change  the  resistance  of  the  material.• as  a  voltage  if  they  are  collected  on  a  capacitor.

Page 20: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

The  MOS  Capacitor  – The  Pixel

• The  Charge  Coupled  Device’s  (CCD’s)  unit  cell,  the  pixel,  is  based  on  a  silicon  structure  that  permits  the  collection  of  electrons  from  photon-­‐created  electron-­‐hole  pairs  at  a  positively  charged  “gate”.

• Additional  gates  permit  the  dragging  of  the  accumulate  charge  across  the  device  and  ultimately  to  a  readout  circuit  that  converts  the  electrons  into  a  measurable  voltage.

Page 21: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

CCD  Architecture

Test open  shutter

closed  shutter

Note  that  bad  things  can  happen  when  buckets  overflow  (saturation).

Page 22: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

HST  F656N  CTE  issues  

Page 23: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

CCD  vs.  CMOS• CCD’s  drag  charge  to  a  destination  amplifier.

– Good:  the  few  amplifiers  on  the  chip  can  be  engineered  to  be  very  sensitive.– Bad:  charge  can  be  lost  and  smeared  along  the  way,  each  “bucketfull”  has  a  

long  journey  with  potential  pitfalls.    Also,  readout  takes  a  long  time

• CMOS  arrays  “x:y”  address  each  pixel.    The  charge  stays  “local”– Good:  fast  readout,  non-­‐destructive  readout  (you  can  “peek”  at  the  

accumulating  image  without  destroying  it).– Bad:  millions  of  amplifiers,  but  today  their  sensitivity  is  comparable  to  or  

better  than  CCD’s.

Page 24: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

“Sandwich”  Infrared  Arrays

http://gruppo3.ca.infn.it/usai/cmsimple3_0/images/PixelAssembly.png

http://www.flipchips.com/tutorial10.html

• Silicon  is  a  terrific  material  because  it  not  only  makes  great  detectors,  but  it  is  the  basis  of  nearly  all  integrated  circuit  electronics.    Silicon  CCD  arrays  can  be  “grown”.

• Infrared  detector  material  (e.g.  InSb)  must  be  attached  to  silicon  integrated  circuits,  typically  through  mechanical  means  

• metallic  bumps  of  elemental  indium  here• differential  thermal  expansion  here  is  a  nightmare!

Page 25: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

What  Do  You  Actually  Measure?

Photons  make  electrons,  but  electronics  of  some  sort  must  convert  that  signal  into  a  detectable  voltage.

Electrons  →  Voltage

Analog  to  DigitalConverter  (ADC) Digital  “counts”  

proportional  to  thevoltage

For  example   5  Volts  might  correspond  to  4096  counts,  in  which  case  measuring  1640  counts  corresponds  to  2  Volts.

Page 26: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

What  Do  You  Actually  Measure?

Photons  make  electrons,  but  electronics  of  some  sort  must  convert  that  signal  into  a  detectable  voltage.

Electrons  →  Voltage

Analog  to  DigitalConverter  (ADC) Digital  “counts”  

proportional  to  thevoltage

For  example  5V  might  correspond  to    4096  countscounts = 4096* actual voltage

5V

Page 27: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Where  Do  The  Volts  Come  From?

Circuitry  converts  collected  electrons  into  electronically  quantified  information.

Electrons  →  Voltage

Analog  to  DigitalConverter  (ADC) Digital  “counts”  

proportional  to  thevoltage  

Drive  a  photon-­produced  current  through  a  resistor  (Ohm's  Law).

Collect  electrons  in  a  (very  small)  capacitor,  “C”.

Page 28: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

CCD  Gain

• Gain  is  the  number  of  electrons  that  yield  one  analog-­‐to-­‐digital  count.• Gain  is  an  electronics  dependent  quantity

Volts / electron = 1.6x10−19 coulombs / electron

Creadout

(capacitance  measured  in  Farads)

Volts / count = 5 Volts4096 counts full range

Combine  and  get  electrons/count.

Page 29: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

An  Image!

17          22        14            19          16          18        21          20          17          15        

22          15          15        18          25          26          15          19          21          11

19          18          27          14        13          18          16          20          12          15

12          15        23            17          15        19          22          21          14          18

15          17          11        24          54          30          21          15          14          19

24          20          13        17          15          21          15          18        21            17

19        12          18          24          15          19          14          22        22            18

17        288        11          20          15          13          18          19          21          22

20          19        18          15          22          14          15        17          20            14

20          14          21        32        102        44          25        17            14          21          

Page 30: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

FITS  Format:        Behind  the  Curtain

• Image  storage  and  “representation”  are  two  different  things.      • A  series  of  numbers  represents  a  two  dimensional  image  if  you  have  

the  format  (pixel  grid  x  by  y)  and  other  “metadata”available.• FITS  files  consist of  a  metadata  text  “header”  followed  by  data  values.

– The  header  consists  of  an  integral  number  of  2880  character  blocks.  • Each  block  contains  a  series  of  80  character  “keyword”  parameters• The  last  keyword  of  the  last  block  is  “END”    padded  out  by  blanks• The  first  bytes  (how  many  and  what  sort  depend  on  the  header  information)  of  the  next  block  is  the  first  pixel  of  the  image.

Page 31: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$
Page 32: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Uncertainty = counts

Poisson  Statistics

• The  uncertainty  in  a  measurement  in  a  counting  experiment  (detecting  photons  in  this  case)  is  equal  to  the  square  root  of  the  number  of  counts  (you’ve  seen  this  before  – now  it’s  serious…).– Quantization  of  light  as  photons  makes  astronomical  detection  a  counting  

experiment– Even  with  a  perfect  detection  system  with  no  noise  and  no  interfering  light  

from  background,  if  you  detect  100  photons  from  a  star,  the  measurement  is  uncertain  by  10  photons,  or  10%.  

Whatever  is  being  counted

Page 33: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Poisson  Statistics

• The  uncertainty  in  a  measurement  in  a  counting  experiment  (detecting  photons  in  this  case)  is  equal  to  the  square  root  of  the  number  of  counts.– Quantization  of  light  as  photons  makes  astronomical  detection  a  counting  

experiment– Even  with  a  perfect  detection  system  with  no  noise  and  no  interfering  light  

from  background,  if  you  detect  100  photons  from  a  star,  the  measurement  is  uncertain  by  10  photons,  or  10%.  

– You  can't  measure  a  star  to  a  precision  of  1%  until  you  have  detected  10,000  photons  from  that  star.

– detection  systems  aren't  perfect  (dark  current)  and  there  are  contaminating  sources  of  light  such  as  the  glow  of  the  sky  (and  glow  of  the  telescope  in  the  thermal  infrared)  • Not  to  mention  extraneous  sources  of  noise  (detector  “read  noise”  in  particular)  that  masquerades  as  additional  unwanted  counts.

Page 34: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Signal  to  Noise  Ratio

• Traditionally,  astronomers  like  to  express  the  quality  of  the  detection  of  a  star  or  spectral  line  in  terms  of  the  ratio  of  signal  to  noise  (signal-­‐to-­‐noise  ratio  or  SNR).    – simplest  terms:    #  signal  counts  /  uncertainty.– S/N=10      is  a  measurement  with  10%  precision

• 100  electrons  gets  you  there  if  there  is  no  source  of  contaminating  light.– S/N=100  is  a  measurement  with    1%  precision

• 10,000  electrons  without  contamination.

• In  general,  if    the  star  is  the  only  source  of  counts,  N:

Page 35: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

• Sources  of  background  add  to  the  detected  photons.– These  unwanted  counts  thus  add  additional  Poisson  noise.– Reducing  these  backgrounds  improve  signal-­‐to-­‐noise

• sharper  images  (landing  on  fewer  background-­‐containing  pixels)• selecting  filter  bandpasses to  avoid  skyglow and  maximize  signal• cooling  telescopes  used  in  the  thermal  infrared

• If  N  is  the  number  of  counts  from  the  star  and  B  is  the  number  of  counts  from  the  background  in  each  pixel  in  the  measurement

• Consider  a  star  that  covers  four  pixels,  each  containing  contaminating  background,  vs the  same  star  covering  only  one  pixel.– Same  “N”  but  npix is  4  times  smaller  leading  to  4  times  lower  total    

background.    If  B  is  large  compared  with  N  sensitivity  is  improved  substantially.

Accounting  for  Background  Contamination

SNR =N fromstar

N fromstar + npixBper pixel

Page 36: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Background  and  Photometry  Footprint

• A  telescope  collects  only  so  many  photons  from  a  given  star  per  unit  time.

• These  star  photons,  depending  on  the  optical  system  and  the  array  pixel  size,  can  land  on  either  a  few  or  a  whole  lot  of  pixels.– Each  pixel  carries  a  background  penalty,  so  your  choice  of  how  many  pixels  to  

use  when  trying  to  measure  “all”  of  the  collected  light  from  a  star  has  signal-­‐to-­‐noise  consequences.

SNR =N fromstar

N fromstar + npixBper pixel

Page 37: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

“Read  Noise”  from  a  Poisson  Perspective

• The  act  of  measuring  the  counts,  in  a  CCD  pixel  for  example,  can  be  (usually  is)  inherently  noisy.

– This  noise  tends  to  be  random/Gaussian  (i.e.  the  value  being  drawn  from  a  Gaussian  distribution  of  probability)

– It  is  therefore  characterized  by  the  “width”  of  the  distribution  of  these  random  counts,  “s”.

– Recall  that  the  Poisson  noise  for  an  actual  “count”  of  N  electrons  is  sqrt(N),  which  also  behaves  in  a  Gaussian  manner  for  large  N.

– Although  read  noise,  characterized  by  an  RMS  uncertainty,  “RN”,  is  not  Poisson  noise,  one  can  pretend  that  the  noise  RN  is  caused  by  the  collection  of  RN2 counts.

Page 38: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Read  Noise  and  SNR

● If  source  photons  are  the  only  source  of  noise

● unwanted  background  photons  B  add  to  the  Poisson  noise,  

● Read  noise  is  the  random  fluctuation  (measured  in  units  of  electrons)  in  the  measurement  (readout)  of  each  pixel.      To  convert  the  read  noise  into  the  equivalent  number  of  electrons  that  would  produce  equivalent  noise  one  has  to  square  read  noise  and,  like  with  background,  account  for  the  number  of  pixels  contributing  read  noise.

SNR = N fromstar

N fromstar

SNR =N fromstar

N fromstar + npixBper pixel

SNR =N fromstar

N fromstar + npixBper pixel + npixRN2

Page 39: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Back  to  CCD  “Gain”  – A  Poisson  Perspective• Electronic  “gain”  (a.k.a.  amplification)  accounts  for  the  difference  

between  measured  digital  counts  and  collected  electrons.– 10  electrons  may  end  up  on  the  output  capacitor,  but  the  analog  to  digital  

converter  may  read  these  10  electrons  as  4  digital  counts.• In  this  example  the  “gain”  is  2.5  electrons  per  analog  to  digital  unit:                        2.5  e-­‐/ADU

• Poisson  noise  provides  a  tool  to  determine  this  CCD  gain.– Consider  a  system  with  a  gain  of  100  e-­‐/ADU  that  makes  multiple  

measurements  of  a  signal  of  10,000  counts.– Given  the  gain,  1,000,000  electrons  were  collected.– The  Poisson  noise  resulting  from  those  million  collected  electrons  is  1000  

electrons,  but  since  the  gain  is  such  that  it  take  100  electrons  to  make  one  ADU  count  the  measured  RMS  noise  in  the  counts  will  be  1000/100  =  10  ADU.

– So,  in  this  situation  you  have  a  signal  of  10,000  counts  resulting  in  an  RMS  noise  of  10  counts  – clearly  not  Poisson,  but  this  mismatch  is  a  clue  to  how  to  calculate  the  gain.      

Page 40: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Statistically  Estimating  CCD  Gain• If  the  gain  were  unknown  in  the  example  on  the  previous  page  one  

could  reverse  engineer  the  value  under  the  assumption  that  the  noise  was  Poisson.– You  illuminate  your  CCD  uniformly  and  make  a  bunch  of  measurements  of  the  

scene,  each  one  illuminated  to  a  mean  level  of  10,000  ADU  counts.• At  this  point  you  have  no  idea  how  many  electrons  10,000  ADU  counts  represents.

– Given  that  you  have  a  number  of  exposures  you  punch  the  exact  ADU  value  of  a  given  pixel  in  each  of  the  frames  into  your  calculator  and  find  the  standard  deviation.• Your  calculator  spits  out  that  the  standard  deviation  is  10  ADU,  so  clearly  not  Poisson  (it  would  have  been  100  ADU  if  it  was  Poisson).

– You  can  now  ask  (in  equation  form  below)  what  does  the  gain,  g,  have  to  be  in  order  to  make  the  measurements  agree  with  Poisson  statistics.

σ observed =g*ADUaverage

gσ observed =

Number of collected electronsgain

Page 41: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Statistically  Estimating  CCD  Gain• If  the  gain  were  unknown  in  the  example  on  the  previous  page  one  

could  reverse  engineer  the  value  under  the  assumption  that  the  noise  was  Poisson.– You  illuminate  your  CCD  uniformly  and  make  a  bunch  of  measurements  of  the  

scene,  each  one  illuminated  to  a  mean  level  of  10,000  ADU  counts.• At  this  point  you  have  no  idea  how  many  electrons  10,000  ADU  counts  represents.

– Given  that  you  have  a  number  of  exposures  you  punch  the  exact  ADU  value  of  a  given  pixel  in  each  of  the  frames  into  your  calculator  and  find  the  standard  deviation.• Your  calculator  spits  out  that  the  standard  deviation  is  10  ADU,  so  clearly  not  Poisson  (it  would  have  been  100  ADU  if  it  was  Poisson).

– You  can  now  ask  (in  equation  form  below)  what  does  the  gain,  g,  have  to  be  in  order  to  make  the  measurements  agree  with  Poisson  statistics.

σ observed =g*ADUaverage

gσ observed =

Number of collected electronsgain

g =ADUavg

σ 2

Page 42: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Why  Do  You  Need  to  Know  the  Gain??• Assigning  a  proper  uncertainty  to  the  measurement  of  a  star’s  flux  is  

possibly  as  important  as  measuring  the  flux  itself.      – A  measurement  is  meaningless  if  it  does  not  have  a  reliably  assigned  statistical  

significance.– For  a  stellar  flux  measurement  extracted  from  a  single  image  frame  proper  

quantification  of  the  Poisson  noise  is  the  only  means  of  assigning  an  appropriate  uncertainty.    Knowing  the  gain  you  can  calculate  the  Poisson  noise.

Page 43: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

CCD  Quantum  Efficiency

• Quantum  efficiency  is  a  measure  of  what  fraction  of  incoming  photons  actually  make  a  detectable  electron  (QE  =  1  is  100%)

• Just  how  efficiently  vs.  wavelength  depends  on  detector  structure.    – In  simplest  terms,  light  must  penetrate  to  and  interact  in  a  region  where  it  

can  produce  electron-­‐hole  pairs  that  ultimately  survive  to  yield  a  collected  electron.

• Electrodes  may  absorb  photons  on  the  way  in  (short  wavelengths).

• Photons  may  penetrate  too  far  before  being  absorbed  (long  wavelengths).

• Photons  may  reflect  off  the  detector  surface.

Page 44: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

CCD  Quantum  Efficiency  vs.  Wavelength

• The  active,  photon-­‐detecting  layer  in  a  CCD  lies  within  about  10  microns  of  the  silicon  surface  where  the  readout  structures  are  grown  (brown  at  right).

• Typically  light  shines  on  this  layer  through  the  gate  structures  used  to  shuffle  the  charge.– These  structures  are  transparent  at  longer  

wavelengths  but  become  opaque  in  the  blue  and  ultraviolet.

– Simple  CCD  architectures  have  poor  blue/UV  response.

http://hamamatsu.magnet.fsu.edu/articles/quantumefficiency.html

Page 45: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Improving  Blue  Response  via  “Thinning”

http://hamamatsu.magnet.fsu.edu/articles/quantumefficiency.html

• “Backside  illuminated”  CCD’s  take  advantage  of  mechanical  thinning  of  the  original  silicon  substrate  on  which  the  device  is  grown  to  permit  illumination  from  the  side  opposite  the  electrodes.

Fluorescent  coatings  that  convert  ultraviolet  photons  to  longer  wavelength  photons  can  also  enhance  ultraviolet  quantum  efficiency.

Page 46: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Optimizing  CCD  Infrared  Response

• Detecting  infrared  photons  requires  a  thicker  “active”  layer  in  the  CCD.      A  special  class  of  “deep  depletion”  devices  optimize  quantum  efficiency  for  the  infrared.

F  =  front  illuminatedB  =  back  illuminated  (thinned)DD  =  deep  depletion

Page 47: Week7faculty.virginia.edu › skrutskie › ASTR3130A › notes › astr3130... · 2017-03-21 · Week7 • “Midterm”exam$in$class$next$Thursday,$March$23 • Eveningsessions$this$week$are$Python$data$analysis$package$

Quantum  Efficiency  and  SNR

• For  the  same  photon  flux  level  in  a  fixed,  say  10  second,  integration,  the  total  Poisson  noise  (square  root  of  collected  electrons)  is  lower  for  a  low  quantum  efficiency  detector  compared  with  a  high  quantum  efficiency  detector  because  the  incident  photons  produce  fewer  electrons.  

– BUT  the  high  quantum  efficiency  detector  will  make  the  detection  at  a  higher  signal  to  noise  ratio,  which  is  what  counts  since  SNR  depends  on  the  square  root  of  N  – the  total  number  of  electrons  collected.

– If  the  sentences  above  make  sense  you  “get  it”.      If  you  don’t  “get  it”  start  asking  questions.