Coherent Detection

with Asynchronous

GmAPD

Joseph C. Marron, Maurice J. Halmos and Brian F. Boland

Raytheon Space and Airborne Systems

El Segundo, CA 90245

Presentation Outline

Overview and motivation

Derivation of CNR, blocking loss relation

Simulation experiments

Laboratory experiments

Summary

What is a Photon Counting Receiver:

Used when return signal is comprised of

single photon events

– Low source power, distant objects

– Signal consists of 2D photo-event locations

and times

Multiple pulses are typically integrated for

reliable detection. Poisson statistics are

key aspect of signals.

Detector properties such as reset time

and dark count rate are important

considerations

– Reset time- detector pixel cannot record

another event until reset

– Dark count rate sets lower limit on the signal

level

SensL Inc. 2D GmAPD Array

(macro-pixel)

Comparison of Direct and Coherent Detection:

GMAPD arrays are typically used for direct detection ladar

Here we consider their use for coherent detection

Amp GMAPD

Array

Target

LO

x

z

Angle-A

ngle

2040

60

204060

X

Time

Time-Angle Top

2040

60

20406080

Time

Z

Time-Angle Front

2040

6080

204060

Time

X

Amp GMAPD

Array

Target

MO

Time

X

MO

X

Z

Angle-Angle

10 20 30

5

10

15

20

25

30

X

Y (

tim

e/ra

ng

e)

Range-Angle Top

10 20 30

50

100

150

Y (time/range)

Z

Range-Angle Front

50 100 150

5

10

15

20

25

30

Photo-Events

Photo-event rate is

proportional to Intensity

Direct Detection

Coherent Detection

Why Geiger Mode Detection for Coherent Applications

Linear coherent detection: 2-D linear detector arrays for coherent detection are complicated

– ROICS for high-density, high-bandwidth linear detection are difficult to implement

– Each detector (or small cluster) requires a high-speed A/D converter

– Volume of data is large

Linear detectors do perform better than GmAPDs- no blocking loss

GmAPD coherent detection: GMAPD detector arrays already exist in 2-D array format

The output of the GMAPD is already digital, so no additional A/D required

GMAPD will typically NOT perform as well as the linear counterparts

Next-generation asynchronous readout architecture overcomes many of the

shortfalls of the previous frame-synchronous devices

Coherent Detection with Asynchronous GmAPD Detector

Next generation of GMAPD detectors will operate in

asynchronous mode (as opposed to frame synchronous)

Result is improved reset time

10 msec (for example)

Frame Synchronous (one photo-event per frame)

Asynchronous

Frame Time

300 nsec (for example)

Reset Time

Blocked Photo-events

Blocked Photo-events

GmAPD Macro-Pixel Saturation or Blocking Loss

Detectors are comprised of macro-pixels to increase dynamic range

Traditional frame-synchronous continues to loose sensitivity as the number of

detections continue until the array is reset (every ~ 50 ms)

Asynchronous arrays reset pixels individually ~ 0.5 ms after a detection event

– Creates a flow of pixels being reactivated countering saturation

Frame Synchronous:

Infrequent reset results in

blocking efficiency that

approaches 0

Asynchronous:

Reaches a balance- detectors

reset at rate near photon flux

Macro-pixel Macro-pixel

Time Time

GmAPD Function Described by Differential Equation

Expanding on the work of Luu and Jiang (Appl. Optics, v45, No. 16, p3798 2006),

we write the expression for the rate of change of the active pixels in the macro-

pixel,

Rate Count DarkDCR

Interval Processing CoherentCPI

time reset pixel SingleT

pixels active of NumberN(t)

macropixel the in pixels of NumberN

R

0

efficiency mixing Heterodyne

electron-photo a creating photon a ofy ProbabilitPDE

CPI per photons LO of NumberN

CPI per photons signal of NumberN

HET

LO

S

where (t) is the photon flux for the heterodyne detection given by,

and fIF is the heterodyne beat frequency

Key term for

Asynchronous

GmAPD

𝑁′ 𝑡 = −𝑃𝐷𝐸 ∙ 𝜆 𝑡 ∙𝑁 𝑡

𝑁0−𝑁 𝑡 ∙ 𝐷𝐶𝑅 +

𝑁0 − 𝑁(𝑡)

𝑇𝑅

𝜆 𝑡 =𝑁𝑆+𝑁𝐿𝑂

𝐶𝑃𝐼1 +

2 𝜂𝐻𝐸𝑇𝑁𝑆𝑁𝐿𝑂

𝑁𝑆+𝑁𝐿𝑂𝑐𝑜𝑠 2𝜋𝑓𝐼𝐹𝑡 + 𝜙

0 2 4 6 8 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

N z( )

N0

N2 z( )

N0

z

Blocking Efficiency Obtained by Solving Diff. Eq.

Blocking efficiency, B, is the relative # of active detectors in the macro-pixel

𝐵 𝑡 =𝑁(𝑡)

𝑁0

Plot below shows numerical solution

As expected, the Blocking Eff. for the synchronous case goes to zero, but for the

asynchronous case, it reaches a nonzero steady state

t (ms)

B(t)

Asynchronous

Synchronous

Values

N0 = 25

CPI = 10 msec

NS = 100

NLO = 1000

DCR= 105

PDE = 0.30

HET = 0.30

f = 5 MHz

TR = 500 nsec

Steady State Operation: 𝒅𝑵(𝒕)/𝒅𝒕 = 𝟎

• Steady state Blocking Efficiency 𝑵(𝒕)/𝑵𝟎

s5.0T

MHz5f

%30

%30PDE

kHz100DCR

1000N

100N

s10CPI

25N

R

IF

HET

LO

S

0

m

m

Given,

10 100 1 103

1 104

1 105

0.01

0.1

1

B vs LO Power

B 100 x, 0, ( )

B 1000 x, 0, ( )

B 10000 x, 0, ( )

x

Parameter Value

Detectors in Macro Pixel N0 = 25

Coherent Processing Interval CPI = 10 msec

Incident Signal Photons NS = 102, 103,

104

Incident LO Photons NLO = variable

Dark Count Rate for Macro

Pixel

0 kHz

Photon Detection Efficiency PDE = 0.30

Heterodyne Efficiency HET = 0.30

Interference Frequency f = 5 MHz

Asynchronous Reset Time TR = 500 nsec

Blocking Factor B = plotted

𝐵(𝑁𝑆, 𝑁𝐿𝑂, 𝐷𝐶𝑅)

10 100 1 103

1 104

1 105

0.01

0.1

1

B vs LO Power

B 100 x, 0, ( )

B 1000 x, 0, ( )

B 10000 x, 0, ( )

x

𝐵 =1

1 + 𝑇𝑅𝑁𝑆 + 𝑁𝐿𝑂 𝑃𝐷𝐸

𝐶𝑃𝐼 ∙ 𝑁0+ 𝐷𝐶𝑅

B=0.6

Steady State Operation: CNR

CNR plotted vs LO power and signal photons

10 100 1 103

1 104

1 105

0.1

1

10

100

CNR vs LO Power

SNR 100 x, 0, ( )

SNR 1000 x, 0, ( )

SNR 10000 x, 0, ( )

x

10 100 1 103

1 104

1 105

0.1

1

10

100

CNR vs Signal Photons

SNR x 700, .1, ( )

SNR x 2000, .1, ( )

SNR x 10000, .1, ( )

x

Values

N0 = 25

CPI = 10 msec

NS = 102, 103,

104

NLO = 700,

2000, 10000

DCR= 0, 105

PDE = 0.30

HET = 0.30

f = 5 MHz

TR = 500 nsec

B = calculated

𝐶𝑁𝑅 =𝑃𝐷𝐸 ∙ 𝐵 ∙ 𝜂𝐻𝐸𝑇𝑁𝑆𝑁𝐿𝑂

𝑁𝑆 + 𝑁𝐿𝑂 + 𝑁0 ∙ 𝐵 ∙ 𝐶𝑃𝐼 ∙ 𝐷𝐶𝑅/𝑃𝐷𝐸

CNR=4.9 (DCR=0)

Steady State Operation: CNR

For N0=100 signal photons hitting the detector – Linear mode CNR is ~9 (9 signal photo-electrons per CPI)

– CNR ≅ 4.8 for asynchronous GmAPD with DCR = 100 kHz

– CNR value > 0.5 x Ideal

0.1 1 10 100 1 103

0.1

1

10

100

CNR vs DCR

SNR 100 700, x, ( )

SNR 1000 2000, x, ( )

SNR 10000 10000, x, ( )

x

DCR (MHz)

Values

N0 = 25

CPI = 10 msec

NS = 102, 103, 104

NLO = 700, 2000, 10000

DCR= variable

PDE = 0.30

HET = 0.30

f = 5 MHz

TR = 500 nsec

B = calculated

CNR = plotted

CNR

CNR=4.8

(DCR=100 kHz)

Simulation Results- Asynchronous

Simulations performed by applying blocking rule to pixel signals

Blocking factor B = 0.58

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x 10-5

1

1.5

2x 10

-3

Sig

na

l In

ten

sit

y

Time (sec)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x 10-5

0

0.5

1

Pho

to E

ven

ts

Time (sec)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x 10-5

0

0.5

1

Acc

ep

ted

Ph

oto

Eve

nts

Time (sec)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x 10-5

0

0.5

1

Tota

l P

ho

to E

ven

ts

Time (sec)

-1.5 -1 -0.5 0 0.5 1 1.5

x 107

0

1000

2000

Spe

ctr

al

Inte

nsi

ty

Frequency (Hz)

Signal

Intensity

Spectral

Intensity

Photo-

Events

Unblocked

Photo-Events

Macro-Pixel

Photo-Events

Blocked

Simulation Results- Asynchronous

Frequency content retained even with TR = 500 nsec

– Conventional wisdom would indicate cutoff of 2 MHz

– Information retained via clock precision 0 1 2 3 4 5 6 7 8 9 10

0

2

4x 10

-3

Time (usec)

Sig

na

l In

ten

sit

y

-1.5 -1 -0.5 0 0.5 1 1.5

x 107

0

500

1000

1500

2000

Spe

ctr

al

Inte

nsi

ty

Frequency (Hz)

0 1 2 3 4 5 6 7 8 9 100

1

2

Reco

rde

d P

ho

to E

ven

ts

0 1 2 3 4 5 6 7 8 9 100

2

4x 10

-3

Time (usec)

Sig

na

l In

ten

sit

y

-1.5 -1 -0.5 0 0.5 1 1.5

x 107

0

500

1000

1500

2000

Spe

ctr

al

Inte

nsi

ty

Frequency (Hz)

0 1 2 3 4 5 6 7 8 9 100

1

2

Reco

rde

d P

ho

to E

ven

ts

Values

N0 = 25

CPI = 10

msec

NS = 300

NLO = 1000

DCR= 105

PDE = 0.30

HET = 0.30

f = 5, 10 MHz

TR = 500

nsec

B = 0.55

CNR = 11.3

f = 5 MHz

f = 10 MHz

Simulation Results- Synchronous

Frame-synchronous detector has high blocking loss and

does not retain frequency information

– Realized blocking loss B = 0.0627

Values

N0 = 25

CPI = 10

msec

NS = 300

NLO = 1000

DCR= 105

PDE = 0.30

HET = 0.30

f = 5 MHz

TR = NA

B = 0.0627

0 1 2 3 4 5 6 7 8 9 100

2

4x 10

-3

Time (usec)

Sig

na

l In

ten

sit

y

-1.5 -1 -0.5 0 0.5 1 1.5

x 107

0

200

400

600

Spe

ctr

al

Inte

nsi

ty

Frequency (Hz)

0 1 2 3 4 5 6 7 8 9 100

1

2

Reco

rde

d P

ho

to E

ven

ts

f = 5 MHz

Peak diminished and broadened

Sig

na

l In

ten

sit

y

Sp

ec

tra

l In

ten

sit

y

Rec

ord

ed

PE

Comparison of Simulation to Theory

Simulation matches blocking and gives lower CNR values than theory

Theory is for steady state, simulation includes transients

Values

N0 = 25

CPI = 10 msec

NS = 100

NLO = Variable

DCR= 105

PDE = 0.30

HET = 0.30

f = 5 MHz

TR = 500 nsec

0 100 200 300 400 500 600 700 800 900 10000

0.2

0.4

0.6

0.8

1

LO Photons

Blo

ckin

g L

oss

0 100 200 300 400 500 600 700 800 900 10000

1

2

3

4

5

LO Photons

CN

RB

lockin

g L

oss

CN

R

LO Photons

Blocking Loss Theory

CNR Theory

Experiments- Silicon Photo Multiplier (SPM)

SPMs are used in Time-of-Flight Positron-Emission-

Tomography (ToF-PET) to detect entangled beta particles

V. C. Spanoudaki and C. S. Levin, “Photo-Detectors for Time of Flight Positron Emission Tomography (ToF-PET),”

Sensors, 10, 2010.

Biomolecule labeled with a radioactive tracer

emits two temporally coincident b+ particles

that travel in opposite directions

SPM

Radiation

Detectors Lesion of Interest

Silicon Photo-Multiplier

Lesion location is determined by recording time-of-flight difference for scintillated photo-events

ToF-PET (and particle physics) is driving the development of advanced SPMs as alternative to photomultiplier tubes

Size of SPM pixels is mm rather than 10s of microns for GMAPD arrays.

Silicon

Photo-Multiplier

Silicon

Photo-Multiplier

Center of

FOV Annihilation

Point

DX Scintillating

Crystal

Scintillating

Crystal

Silicon Photo-Multiplier

Technical Highlights: – 260 nsec reset time

– 0.250 nsec time-of-flight resolution

– 0.100 nsec signal rise time Gain and optical response uniformity <+-10%

– Available from Sensl Inc. in surface mount (SMT) package

0.25mm, 1mm, 3mm, and 6mm active detector area

Device from: Sensl Inc.

1 pixel = 576 (24 x 24),

parallel GMAPDs

1 mm

Experimental Setup

Evaluated detector response from temporally modulated LED

With a dead time of 260 ns, the fastest Nyquist sampling rate is 1.9 MHz.

Macropixel detects modulation at ~50 times the Nyquist frequency of a single pixel.

)2sin()( 21 tfIItI

Thorlabs Modulated LED

405 nm: 10 to 90 MHz

Sensl Detector:

260 nsec reset time

Modulation Depth

0 20 40 60 80 100

MHz

1

0.8

0.6

0.4

0.2

0.0

Magnitude of Fourier Coefficient

-100 -60 -20 20 60 100

MHz

1

0.8

0.6

0.4

0.2

0.0

IRIS

Summary

Asynchronous GmAPD detectors show promise for coherent detection

Macro-pixel composed of several individual GmAPDs

Formulas for blocking loss and CNR derived

– Performance approaches that of a linear detector

– Allows coherent detection with array of detectors

– Optimal LO level ≅ Signal level

Frequency response >> 1/reset time

Simulations in good agreement with theory

Experiments with Silicon Photo-Multipliers are underway