Station Overview, ARA Trigger & Digitizer Gary S. Varner ARA Workshop in Honolulu, 17-AUG-10 • Station geometry • Triggering Overview • Trigger Simulation • Geometrical constraints • Trigger rates • Digitization & • Data rates
Station Overview, ARA Trigger & Digitizer
Gary S. VarnerARA Workshop in Honolulu, 17-AUG-10
• Station geometry• Triggering Overview• Trigger Simulation• Geometrical constraints• Trigger rates• Digitization & • Data rates
Basic Station Geometry -- Initial
ARA Readout Electronics
• Defer general discussion of architecture– Trigger update
– ASIC (IRS) update
Basic Station Geometry -- Revised
ARA Readout Electronics: Triggering
• Maximize local and global sensitivity– Station (few 100ns window) [local]– Array prompt (10’s of us) [global, subthreshold]– High level (100’s of seconds) [global, WF low threshold]
Geometric Considerations
Top View
Consider simple coincidence – as a function of string spacing
Arriving radio front
Single Antenna “singles” rates
• Raw rates
Trigger Rates versus Trigger Threshold
1000
10000
100000
1000000
10000000
3 3.5 4 4.5
Trigger Threshold [sigma noise]
Rat
e [H
z]
Ant singles
Station coincidence
• 5 m “tight” spacing (50ns window)
Trigger Rates versus Trigger Threshold
0.00000010.000001
0.000010.0001
0.0010.01
0.11
10100
100010000
100000
3 3.5 4 4.5
Trigger Threshold [sigma noise]
Rat
e [H
z]
5 of 16
~3.8
Trigger Rates versus String Distance [5 of 16]
0.00000010.000001
0.000010.0001
0.0010.01
0.11
10100
100010000
100000
3 3.5 4 4.5
Trigger Threshold [sigma noise]
Rat
e [H
z]
5m
10m
20m
50m
Station coincidence
~3.9
Additional constraint: causality
Arriving radio front
Use temporal/spatial constraints to reduce incoherent thermal accidentals and reject pathological directions (e.g. surface noise)
Implemented as a 2D sliding window
Simplified coordinatesTop View
Antenna hits quantized with 4ns resolution (250MHz pipeline)
Arriving radio front
++
Side View
16
antennas
N time bins (32 for 128ns)
Example – strong nu signal
Arriving radio front
+
Side View
= 10o
Top View
+
= 45o
A1A2A3A4
Arriving radio front
+
Side View
= -10o
Example – strong signal (e.g. ICL)Top View
+
= 30o
A1A2A3A4
Example – at threshold nu signal
Arriving radio front
+
Side View
= 15o
Top View
+
= 67o
A1A2A3A4
Thermal Noise (~10MHz singles)Top View
++
Side View
Arriving radio front ??
A1A2A3A4
Thermal Noise (~3.x sigma)Top View
Arriving radio front ??
++
Side View
Combinatorics are enormous (C[512,5]=512!/(5!*(512-5)!))
How to implement?
1. Track “road” search? (computationally intensive)2. Step time thread logic3. Fit to plane wave (again CPU heavy)4. Brute force pattern match?
– 2^(16+32) ~ 280 Terabits (very sparse)
Direct logic search – programmable logic good at this
Use “5th” (last hit as seed)
Divide up sky into arrival directionsTop View
Arriving radio front
++
Side View
Many downgoing directions pathological
With quantization, something likeSomething like 5o
Something like 10o
36
361296 equations
Example – at threshold nu signal
Arriving radio front
+
Side View
= 15o
Top View
+
= 67o
Hit search seed
Example – at threshold nu signalHit search seed
Term of equation:
Hit = S1A3[0]*S2A1[1]*S3A4[19]*S4A1[9]*S4A2[10]
Build terms from MC
Since degenerate – some OR termsHit search seed
Reduce Terms of equation:
Hit = S1A3[0]*S2A1[1]*S3A4[19]*S4A1[9]*(S4A2[10]+S4A2[9])
Needs detailed study, but can guesstimate:
16 ant seeds * (16 theta * 8 phi) * 32 patterns ~ 65k terms
Thermal Noise (~3.x sigma)
One way to think of this: can tolerate a larger number of spurious hits Effectively raise coincidence level in the window
Combinatorics are enormous
seed Physically impossible
Hit predicts allowed other times
Trigger Rates versus Effective Threshold [10m]
0.00000010.000001
0.000010.0001
0.0010.01
0.11
10100
100010000
100000
3 3.5 4 4.5
Trigger Threshold [sigma noise]
Rat
e [H
z]
N=5
N=6
N=8
N=10
Station coincidence
~3.3
Looks promising – Lisa to continue…
ARA Readout Electronics: ASIC
• Build on experience with “next generation” ASICs– Deeper storage depth, higher bandwidth?
– Fewer timing alignment constants
Ice Radio Sampler (IRS)
• Actually a fairly generic part– Follow-on evaluation of deeper storage [TARGET, others]
(LABRADOR technology now >half decade old)
– “2 stage” transfer mechanism (reduced calibration)
– No amplifier on the input
– Self-trigger capability (not useful this application)
Collaborative effort with NTU
Ice Radio Sampler (IRS) Specifications
32768 samples/chan (16-32us trig latency)8 channels/IRS ASIC8 Trigger channels
~9 bits resolution (12-bits logging)64 samples convert window (~32-64ns)
1-2 GSa/s1 word (RAM) chan, sample readout
16 us to read all samples100's Hz sustained readout (multibuffer)
• Strictly only 5 channels necessary– 4x antenna, 1x reference channels
– Could interleave for twice depth, or multiple reference channels
IRS Floorplan
8x RF inputs(die upside down)
5.82mm
7.62mm
32k storage cells per channel
(512 groups of 64)
IRS Single Channel
• Storage: 64 x 512 (512 = 8 * 64)
• Sampling: 128 (2x 64 separate transfer lanes
Recording in one set 64, transferring other (“ping-pong”)
• Wilkinson (32x2): 64 conv/channel
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0Sampling Simulation with full parasitic Extraction
0.000
0.500
1.000
1.500
2.000
2.500
3.000
3.500
4.000
4.500
5.000
0 0.5 1 1.5 2 2.5
RCObias [V]
Sam
plin
g R
ate
[GSa
/s]
Extracted
2 stage sampling speed sim
“RCObias” VadjP1,2 = RCObias; VadjN1,2 = VDD-RCObias
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
sampling speed measurement
Delta V RCObias Matches expectation, but….
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Measurement via RF sine
Samples much faster, but at higher sampling rateWrite strobe width problem
Measurements by Chih-Ching
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Measurement via RF sine
Analog BW~1GHz
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Input coupling sim (35fF sample)
~1 GHz input signal ABW
Onto chip (flip chip)
Magnitude [dB
]
From IRS Design Review
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Measurement via RF sine
Samples much faster, but at higher sampling rateWrite strobe width problem (know how to fix)
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Linearity Calibration
Comparator bias parameters NOT optimized
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Noise Measurement
mV
~ 2mV
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Need dT calibrations
Conversion/readout speed
• Assume 8 channel (5 needed)
– 5us/ADC cycle (8*64 samples/channel in parallel)
– Transfer at 50MHz (20ns/sample) to FPGA
– 1 conversion cycle ~ 5us (ADC) + 10us (transfer)
– 256ns window (512 samples @ 2GSa/s) = 8 conv cycles
– Total ~ 120us [CF: 1kHz trigger]
– Deadtimeless: 256ns (512 samples) of 16us (32k samples) held – sampling continues on others
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Station Data Reduction (self-trigger)
Raw Signals
16 RF channels@ 1.5By * 2GSa/s
= 48 GBytes/s
Level-1
Antenna
Fullband
A ~MHz (L1L)A ~ 0.1MHz (L1H)
Level-2
Station
5-of-16
100’s kHz (L2L)100’s Hz (L2H)
Level-3Phi
Patternmatch
10-50Hz@ 8kBy/evt
= 60-300kBy/s
Prioritizer?(+compress)
10H
z W
F ev
ents
/link
WF data = 80%HK/trigger timestamp = 10%
High-level req = 10%
ARA Readout Electronics – system discussion
• Uplink bandwidth (~1Mbit/s [wireless])– Multi-tier trigger
– Deeper sampling allows for “array” trigger (subthreshold)
IRS AARDVARC Specifications ?
262144 samples/chan (130us trig latency)1 channel/ASIC-- Trigger channels
~9 bits resolution (12-bits logging)64 samples convert window (32ns)2 GSa/s1 word (RAM) chan, sample readout
<10 us to read all window samples10k Hz sustained readout (multibuffer)
• Avoids issue of channel-channel cross-talk– Slave sampling all ASICs together
– Plenty depth for multi-hit buffering
Summary• Station design evolving
– Build sample station for firmware/cal testbed development
– Initially test with thermals (servo-loop software/firmware)
• Key technology decisions– Tunnel diode versus RF power mon– IRS AARDVARC– Data and fast trigger links
• Proposed architecture– Rather flexible – Optimize as we go
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Back-up slides
Askaryan Radio Array (ARA)
Askaryan Radio Array
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Buffered LABRADOR (BLAB1) ASIC• 10 real bits of dynamic range, single-shot
Measured Noise
1.45mV
1.6V dynamic range
Wilkinson Clock Generation
• Strictly only 5 channels necessary– 4x antenna, 1x reference channels
– Could interleave for twice depth, or multiple reference channels
Wilkinson Recording
Start = start 0.5-8GHz Clock Ripple counter (run as fast as can)
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Wilkinson speed measurement
0.7 GHz 1.4us conversion to 10 bits
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Output Bus Settling Time
~100MHz bus operation should be possible
~8.5ns (10-90%)
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Diode detector Response
σ
Voltageσ
Exponential distribution
<P>
Power:
P/<P>
P/<P>
~7ns integration
Gaussian distribution
Tunnel Diode DetectorLNA
Quad−ridgehorn antenna
Needs amplification!
Tunnel Diode Output Single Channel Trigger Rate
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
3 3.2 3.4 3.6 3.8 4 4.2 4.4 4.6 4.8 5
Power/<Power>
Co
un
t R
ate
[M
Hz]
singles
2.3 ~= 3.9 P/<P>
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Log-amp, tunnel diode test
• Can fast log-amps give same SNR as TD trigger?
• Log-amp: V proportional to power
• Uses multi-stage switching to get wide “linear” dynamic range, good stability
• Tunnel-diode: square-law detector with long history in radio astronomomy & physics
– But they are fussy to use!
100ps Pulsegen
0.2‐1.2GHz receiver Hybrid
splitter
TekTDS784C scope
AD8318 test boardCoax tunnel diode detector5ns rise time
trig
CH3
CH2DC block
CH1
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Log-amp vs. tunnel diode SNR test
• Look at Vpeak to Vrms ratio for each device
• Log-amp:– saturation evident– Loss of SNR fidelity below
SNR~3 • TD: square-law behavior evident• Conclusions: log-amps may be
problematic• We really need a true trigger
efficiency test
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Design Basis: Buffered LABRADOR (BLAB1) ASIC
• Single channel• 64k samples deep,
same SCA technique as LAB, no ripple pointer
• Multi-MSa/s to Multi-GSa/s
• 12-64us to form Global trigger
3mm x 2.8mm, TSMC 0.25umArranged as 128 x 512 samples
Simultaneous Write/Read
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
BLAB1 Architecture
200ps/sample
FPGA-based TDC: 10-bits in 1us (300ps resolution)
BLAB1 Sampling Speed
200ps/sample Single sample:200/SQRT(12)~ 58ps
In practice, treat each row of 512 samples as independent
Can store 13us at 5GSa/s (before wrapping around)
BLAB1 Analog Bandwidth
• A few fixes (lower power, higher BW)• Multi-channel desired for BLAB2
-3dB ~300MHz
Buffer amps
LAB3 ~ 900MHz
IRS Input Coupling
• Input bandwidth depends on 2x terms– f3dB[input] = [2**Z*Ctot]-1
– f3dB[storage] = [2**Ron*Cstore]-1
Input Coupling versus total input Capacitance
0
0.5
1
1.5
2
2.5
3
3.5
0 500 1000 1500 2000 2500 3000
Total input Capacitance [fF]
Anal
og B
andw
idth
[-3d
B fre
quen
cy]
R_S = 50Ohm
Input coupling versus frequency
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
0.1 1 10 100
Frequency [GHz]R
elat
ive
ampl
itude
[dB
]
C=15fF,Ron=1kC=15fF,Ron=5kC=25fF,Ron=1kC=25fF,Ron=5k
IRS Input Coupling
• Role of inductance
Input inductance impedance versus frequency
0
20
40
60
80
100
120
140
160
180
200
0.1 1 10 100
Frequency [GHz]
Impe
danc
e [O
hms]
Bond-wireBump-bond
Input coupling versus frequency
-10
-8
-6
-4
-2
0
2
4
6
8
10
0.1 1 10 100
Frequency [GHz]
Rela
tive
ampl
itude
[dB]
Bond-wireBump-bond
Sample Cell
• Main element is buffer amp (OTA)– Relatively low current (10’s uA) operation possible
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Constraint: kTC Noise
Desire small C for better Input Coupling
Storage Cell
• Diff. Pair as comparator– Only power on selected block
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Another Constraint: Leakage Current
Can Improve? (readout faster)
Need small C for Input Coupling
Sample channel-channel variation ~ fA leakage typically
AR
A T
rig/D
ig E
lect
roni
cs -
17-A
UG
-201
0
Temperature Dependence
0.2%/degree C(servo-loop width)
6GSa/sReference for BLAB1 ASIC
Matches SPICE simulation