The Fly’s Eye: Instrumentation for Detection of Radio
Ephemeron
Andrew P. V. Siemion1,3 , Dan Werthimer2,3,4 and Geoff Marcy1,5
University of California, Berkeley, CA 94720
The Fly’s Eye Team: Geoff Bower1 , Griffin Foster1,3 , Peter McMahon3 , Joeri van
Leeuwen1 , Mark Wagner3
Received
;
accepted
1
Department of Astronomy
2
Space Sciences Laboratory
3
Center for Astronomy Signal Processing and Electronics Research
4
Research Advisor
5
Faculty Advisor
–2–
ABSTRACT
Here we present a preliminary report on the design, construction, deployment
and testing of the “Fly’s Eye,” a 44 input, 128 channel, 209 MHz bandwidth,
600 microsecond accumulation time spectrometer and analysis system designed
to detect powerful dispersed radio transients using the Allen Telescope Array
(ATA). The Fly’s Eye has been successfully installed at the ATA, and to-date
over 24 hours of observations have been performed using the instrument. Of the
extant 24 hours of observation time, approximately 18 hours has used the antenna
beams configured in a close-pack hexagon (‘fly’s eye’) pattern, with the remaining observations using a single-position pointing of several diagnostic sources.
Although analysis of the close-pack hexagon-configuration data is still at a very
early stage, cursory examination of the diagnostic data indicates the system is
functioning within expectations. Here we explore the background and motivation
for the experiment and describe development details, including specifications of
the instrument hardware and testing procedures. Early analysis of diagnostic
observations are also included. Future analysis of Fly’s Eye data is outlined.
1.
Background and Motivation
1.1.
The Lorimer Pulse
The recent discovery by Lorimer et al. (5) of a powerful (∼ 30Jy) and highly dispersed
(DM ∼ 375 pc cm−3 ) radio pulse has dramatically renewed interest in transient radio
phenomena. The Lorimer Pulse was found during a re-examination of Parkes Pulsar Survey
data with a source direction of approximately 3 degrees south of the center of the SMC.
The ionized portion of the ISM (and IGM) introduces a frequency dependence on the
–3–
group velocity of an otherwise broadband radio pulse according to the relation:
vlo −2 vhigh −2 DM
−
∆t = 4.15ms
GHz
GHz
cm−3 pc
(1)
Where ∆t is the time delay between the portions of the pulse at frequency vlo and vhigh .
With the dispersion measure, DM given in units of cm−3 pc, defined as:
DM =
Zd
ne dl
(2)
0
Where ne is electron density and d the distance to the source. This allows us to infer a
distance to the source based on estimates of the various contributions to the integrated
electron column density along the line of sight to the source.
The extraordinarily large dispersion measure of the Lorimer Pulse, and the apparent
absence of any ISM or SMC contribution that could have generated it, led Lorimer to
conclude that the pulse may have originated well outside our galaxy ( 500Mpc.) Analysis
by S. R. Kulkarni et al. (4) of other potential sources of the large dispersion measure, such
as a suitably arranged ionized nebula, has been unable to account for the large dispersion
of the pulse and frequency dependence of the pulse width via any other mechanism than
the pulse’s extra-galactic origin.
The strongest known sources of radio pulses, RRATs and pulsar giant pulses, are
incapable of producing a pulse of the power of the Lorimer Pulse at such a great distance.
Thus, if we accept that the Lorimer Pulse is indeed extra-galactic, this pulse and others
like it hint at the existence of a previously unobserved, highly energetic, transient radio
phenomena that could provide an invaluable means of probing the IGM.
–4–
1.2.
Possible Pulse Source
One of the most intriguing possible sources of the Lorimer Pulse can be found in a
suggestion by Martin Rees in 1977 (9) that primordial black holes, evaporating via the
Hawking Process, could release a large electromagnetic pulse of short duration. According
to Hawking (2), a black hole of mass M emits radiation like a blackbody with a temperature
TBH given by:
TBH
~c3
=
= 10−6
8πkGM
M⊙
M
K
(3)
This radiation emanates from the black hole event horizon and comes completely from the
black hole’s mass. For a non-accreting black hole, Stefan-Boltzmann yields a lifetime of:
τBH
M
= 10 years
1012 kg
10
3
(4)
For stellar mass black holes, this theory predicts lifetimes of order 1034 years, much too long
to ever expect to observe. However, some cosmologies predict the creation of numerous
small (M ∼ 1012 g) primordial black holes in the early universe, which according to theory
should be evaporating now (3).
The specific mechanism by which the evaporating black hole produces a strong radio
pulse has not been fully elucidated, but in short, it is thought that the process is similar to
the EMP that accompanies supernova explosions. In such a process, a highly conductive
plasma fireball expanding into the ambient magnetic field of space can exclude the field and
create an electromagnetic pulse. For typical values of the interstellar magnetic field, this
pulse would be peaked near 1GHz (1).
An observation of these pulses would not only provide a significant confirmation of
Hawking radiation, but would also give strong evidence of the existence of primordial black
holes.
–5–
1.3.
The Allen Telescope Array
The ATA is an ideal instrument to search for transient radio pulses. It provides
the ability to cover a large fraction of the sky and also offers the opportunity to easily
search multiple frequency regimes via its tunable IF. In the nominal close pack hexagon
configuration, ATA-42 can cover a total of 150 square degrees on the sky (at L-band).
Although the Parkes-multibeam system used to discover the Lorimer Pulse has a
much larger collecting area than ATA-42, the potential narrowness of the intrinsic pulse
width may allow detection of a comparable pulse with a single ATA dish. The hardware
used in the Parkes system could only limit the Lorimer Pulse width to 5 ms. If the pulse
were narrower, which it may very well have been, and the instrument integration time
commensurately shortened, a similar pulse could indeed be observed by an individual
ATA-42 dish.
Fig. 1.— Artists Rendition of ATA-42 in Fly’s Eye Pattern (Courtesy Joeri van Leeuwen)
–6–
2.
The Fly’s Eye Instrument System
2.1.
Preliminary Design Criteria
In order to be able to detect a pulse of similar power to the Lorimer Pulse, we desired
to attain a spectral integration time of no more than 1 millisecond. It was also desired to
be able to utilize at least one polarization for each of the 42 antennas currently in place
at the ATA. We were also constrained (on a per-spectrometer-basis) by the available 209.5
MHz bandwidth provided by the ATA IF system. The final short-list called for computing
auto-correlations on approximately 128 spectral channels over a bandwidth of at least
100MHz for at least 42 separate IFs. It was planned to perform de-dispersion and threshold
calculations subsequently using a workstation cluster.
2.2.
CASPER-Based Design
2.2.1. The CASPER Group
The Center for Astronomy Signal Processing and Electronics Research (CASPER),
seeks to speed the development of radio astronomy signal processing instrumentation by
designing and demonstrating a scalable, upgradable, FPGA-based computing platform and
software design methodology that targets a range of real-time signal processing applications.
To-date, CASPER-designed boards have been used for numerous beam-forming, correlation
and spectroscopy projects all over the world, including the University of North Carolina’s
PARI Observatory, Harvard/CfA’s SMA and the NASA/JPL Deep Space Network (DSN)
(8).
With the ready availability of CASPER hardware, and the author’s familiarity with
the CASPER toolflow, selection of CASPER hardware for the Fly’s Eye Instrument was
–7–
an easy decision. Additionally, rapid construction of the system could be facilitated by the
numerous existing CASPER instrument designs similar to the proposed Fly’s Eye System.
2.3.
FPGA Hardware
The core of the Fly’s Eye Instrument is made up of eleven CASPER Internet Break
Out Boards (IBOBs Fig. 2) each of which is connected to two CASPER iADCs via the
IBOB Z-DOK interface.
Fig. 2.— CASPER Internet Breakout Board and ADCs
In addition to the Z-DOK interface, each IBOB provides two CX4 connectors, a
10/100 Mb Ethernet interface and a RS-232 serial connection. The computation engine of
each IBOB board is a Xilinx XC2VP50 FPGA which provides 232 18x18-bit multipliers,
two PowerPC CPU cores and over 53,000 logic cells. In addition to the FPGA, the
board includes 36Mbit of on-board ZBT SRAM. Fig. 3 shows a block diagram of IBOB
components.
Each of the iADCs contains an Atmel AT84AAD001B dual 8-bit ADC chip, capable of
simultaneously digitizing two analog streams at 1 Gsample/sec.
–8–
Fig. 3.— IBOB Block Diagram, Courtesy of UC Berkeley SETI Group
2.4.
FPGA Gateware
Each of the eleven IBOB / iADC systems in the Fly’s Eye processes 4 of the 44 total
inputs to the instrument. Four parallel time samples per input, acquired at four times the
FPGA clock rate (838.8608 MHz), are passed from the ADC to a digital down converter
where the signal is mixed at 209.7152 MHz and then low-passed filtered to a bandwidth
equal to 209.7152 MHz. The signal is then decimated to a sample rate equal to 209.7152
MHz. The resultant signal is complex, 8 bits I and 8 bits Q, representing the signals from
104.8576 MHz to 314.5728 MHz.
The down converted and decimated digitized data is passed into a Polyphase Filter
Bank (PFB) / Fast Fourier Transform logic block which together implements a 128 point 4
tap Polyphase Filter Bank on each of the 4 inputs (labeled internally as A, B, C, D). The
frequency domain data is then passed into an equalization block which can selectively allow
each frequency channel to be scaled by an individual coefficient. This allows a non-flat
passband to be flattened digitally and provides for dynamic gain control over the pre-power
–9–
spectra. A schematic diagram of the Fly’s Eye signal path is shown in Fig. 4.
2.5.
Instrument Operating System
Individual IBOBs in the Fly’s Eye Instrument use a highly modified version of
the open-source TinySH operating system for debug and testing (11). The TinySH
interface is accessible via a telnet server running on each IBOB. TinySH allows probing
of FPGA-PowerPC shared memory regions, access to network configuration information
and execution of a variety of debug commands. Scripted operation of the instrument
(i.e. boot-up sequences) is accomplished by automated interaction with the TinySH
interface. Equalized spectra from each of the four inputs are passed into individual 64 bit
A
ADC
DDC
PFB/
FFT
Accumulation
B
ADC
DDC
PFB/
FFT
Accumulation
Packetization
C
ADC
DDC
PFB/
FFT
Accumulation
Out to Ethernet
D
ADC
DDC
PFB/
FFT
Accumulation
Fig. 4.— Fly’s Eye signal path schematic diagram.
accumulators, where spectral values are accumulated for a period of time given by the value
of an accumulation counter software register.
A non-standard and novel method was devised for storing accumulated spectra to
– 10 –
allow individual byte-boundary octets of the accumulated values from each input to be
output at high speed (∼ 7 Mb/sec.) Each 64-bit accumulation is split into 8 byte-boundary
bytes, which are concatenated with corresponding bytes from other inputs into a 32 bit
word and stored, in channel order, in one of eight BRAMs See Fig. 5. This allows 8-bit
resolution spectra to be read contiguously by the FPGA’s integrated PowerPC processor,
nearly cutting in half the minimum accumulation time. For temporally narrow signals (less
than our minimum accumulation length), this effectively doubles our observed SNR.
A bits 0-7
A bits 8-15
A bits 16-23
A bits 24-31
A bits 32-39
A bits 40-47
A bits 48-55
A bits 56-63
Channel A 64 bits accumulation
B bits 0-7
B bits 8-15
B bits 16-23
B bits 24-31
B bits 32-39
B bits 40-47
B bits 48-55
B bits 56-63
Channel B 64 bits accumulation
C bits 0-7
C bits 8-15
C bits 16-23
C bits 24-31
C bits 32-39
C bits 40-47
C bits 48-55
C bits 56-63
Channel C 64 bits accumulation
D bits 0-7
D bits 8-15
D bits 16-23
D bits 24-31
D bits 32-39
D bits 40-47
D bits 48-55
D bits 56-63
Channel D 64 bits accumulation
A bits 0-7
B bits 0-7
C bits 0-7
D bits 0-7
Channel 0
A bits 56-63
B bits 56-63
C bits 56-63
D bits 56-63
Channel 0
A bits 0-7
B bits 0-7
C bits 0-7
D bits 0-7
Channel 1
A bits 56-63
B bits 56-63
C bits 56-63
D bits 56-63
Channel 1
C bits 56-63
D bits 56-63
Channel 127
BRAM 0-8
BRAM 56-63
....
....
A bits 0-7
B bits 0-7
....
C bits 0-7
D bits 0-7
Channel 127
A bits 56-63
B bits 56-63
Fig. 5.— Byte arrangement system used in the Fly’s Eye Instrument to enable fast UDP
data output.
The complete top-level diagram for the Fly’s Eye gateware is depicted in Appendix
Fig. 17.
2.6.
The Instrument System
The eleven IBOB/ADC systems are housed in two modified rack-mountable
CompactPCI chassis, 6 in chassis ‘ALPHA’ and 5 in chassis ‘ZULU.’ The cabling diagram
for an individual IBOB is depicted in Fig. 6.
At the conclusion of a spectra integration sequence, a user selectable byte-boundary
8-bit portion of the accumulated spectra from each of the four channels on an individual
– 11 –
Fig. 6.— IBOB Cabling Diagram, Courtesy of Matt Dexter, UCB RAL
IBOB is packetized via the FPGA PowerPC processor and output using UDP protocol over
the IBOB’s 10/100Mbit Ethernet port on a closed network. The packets from all eleven
IBOBs are captured en masse on a single data collection machine. The total aggregate data
rate for the entire system is approximately 7 Mb/sec x 11 IBOBs ∼ 80Mb/sec. High rate
data capture and output to fixed disk is accomplished using an open-source tool, ‘gulp’
(10), designed expressly for this purpose,.
An additional network interface card on the data collection machine allows connection
to a 12 Terabyte gigabit-network attached RAID system for medium term data storage.
All critical components of the Fly’s Eye system are fed power through an Ethernet
controlled power switch to enable remote power cycling and monitoring. Fig. 7 depicts a
diagram of the complete Fly’s Eye System.
– 12 –
IBOB Rack
ALPHA
g38 - g42
IBOB Rack
ZULU
g43 - g47
10/100/1000 Mb Ethernet Switch
2.0 GHz AMD Athlon
Machine Name: "flytrap"
ATA-42
1000 Mb Ethernet Switch
Ethernet Controlled Power Strip
Machine Name: "flypower"
1000 Mb Ethernet
100 Mb Ethernet
IP Controlled 110V AC Power
ATA Analog IF
Dual G4 X-Serve
Machine Name: "lordoftheflies"
Nexsan 10 TB Storage Array
Volume: "jack"
SCSI Control by lordoftheflies
Fig. 7.— Fly’s Eye Instrument System Schematic Diagram Not shown is remote power
connection to ATA-42 Walsh Switching system.
– 13 –
3.
Preliminary Results
3.1.
Installation
The Fly’s Eye Instrument was installed at the ATA on December 20, 2007. Fig. 8
shows an image of the Fly’s Eye instrument installed in the ATA Correlator Room. At the
time of installation, only 26 of the 42 built dishes were available. The best polarization
of each of the 26 available dishes was connected to the Fly’s Eye instrument, with the
remaining 18 inputs filled by the highest quality opposite polarization feeds. Although we
had initially sought to use all 42 dishes, the opportunity to make polarization measurements
of a candidate pulse is scientifically interesting in its own right.
3.2.
Diagnostic Sources
To date, approximately six hours of Fly’s Eye observation time has been dedicated to
observation of several bright pulsars, shown in Table 1. Our first attempts at data analysis
have consisted of reconstituting each of the 44 individual IFs from the gulp-produced data
files and performing folding analysis at barycentric corrected periods. Figs. 9 - 14 show
images generated through folding analysis. Note, these images represent a composite sum
of 36 polarizations12 .
On careful inspection, the well described triple peak profile of PSR 0329+54 is visible
in Figs. 9 - 11 (Seiradakis et al.). Dispersion profiles for pulsars PSR 0329+54 and PSR
0950+08 are consistent with published values. Pulsars PSR 0355+54 and PSR 0450+55 are
1
Only 36 polarizations were summed due to instrumental malfunction caused by inclement
weather
2
Amplitude scale is arbitrary
– 14 –
Table 1: List of Bright Pulsars Observed as Diagnostic Sourcesa
Name
a
RAJ2000(hms) DECJ2000(dms) Period(s)
DM(cm3 pc) S1400(mJy)
B0329+54
03:32:59.368
+54:34:43.57
0.7145
26.833
203
B0355+54
03:58: 53.7165
+54:13:13.727
0.1563
57.1420
23
B0450+55
04:54:07.709
+55:43:41.51
0.3406
14.495
13
B0531+21
05:34:31.973
+22:00:52.06
0.0330
56.791
14
B0950+08
09:53:09.3097
+07:55:35.75
0.2530
2.958
84
B0329+54
03:32:59.368
+54:34:43.57
0.7145
26.833
203
Values from (7)
not readily distinguishable in similarly produced images, likely owing to their much reduced
flux.
Appendix Fig. 18 shows a plot of averaged spectra for all 44 inputs to the Fly’s Eye
Instrument over a 18 minute observation of PSR 0329+54. The 21-cm line is clearly visible,
as are moderate RFI features near the edge of the band.
3.3.
Proposed Analysis
Analysis of the Fly’s Eye diagnostic data has increased our confidence in the Fly’s Eye
Instrument and enabled us to begin preparations for reducing the close to twenty hours of
observation data taken with the ATA antennas pointed in a close pack hexagon pattern.
While all of the analysis presented here has been completed manually using in-house
code, we expect to begin using the open-source SigProc (6) tools for exhaustive searches of
our hexagon-pattern data. Fig. 16 shows a proposed processing pipeline for our first-pass
analysis.
– 15 –
Fig. 8.— Fly’s Eye Instrument Installed in ATA Rack
– 16 –
Fig. 9.— PSR 0329+54 Observation 1 - 18 Minutes Folded at Barycentric Corrected Period
Fig. 10.— PSR 0329+54 Observation 2 - 18 Minutes Folded at Barycentric Corrected Period
– 17 –
Fig. 11.— PSR 0329+54 Observation 3 - 18 Minutes Folded at Barycentric Corrected Period
Fig. 12.— PSR 0450+55 - 18 Minutes Folded at Barycentric Corrected Period
– 18 –
Fig. 13.— PSR 0355+54 - 18 Minutes Folded at Barycentric Corrected Period
Fig. 14.— PSR 0950+08 - 18 Minutes Folded at Barycentric Corrected Period
– 19 –
Fig. 15.— PSR 0329+54 Pulse Profile at 1.4 GHz, Courtesy of (Seiradakis et al.)
– 20 –
Raw gulp
Datafile
IF 0
Error Analysis / Dropped
Packet Correction
IF 1
IF 2
IF Seperation
....
IF 43
IF SUM
Average
Computation
....
....
Equalization /
Normalization
....
....
De-dispersion
DM 0
Frequency
Collapse
De-dispersion
DM 1
De-dispersion
DM 2
....
....
Compute RMS /
Threshold
....
De-dispersion
DM 1000
....
Log Candidate
Pulses
Decimate
Fig. 16.— Proposed Analysis Pipeline for Close-Pack Hexagon Data
Disk
– 21 –
Acknowledgments
The Fly’s Eye Experiment benefited enormously from the generous assistance and
advice offered by a number of people, including Don Backer (UCB Astronomy, RAL,
CASPER), Colby Craybill (RAL), Matt Dexter (RAL, CASPER), David McMahon (RAL,
CASPER) and Mel Wright (UCB Astronomy, RAL, CASPER.)
Instrument installation was gracefully shepherded by the entire ATA staff, especially
Rick Forster.
The original idea for the Fly’s Eye Experiment was suggested by Jim Cordes (Cornell
University)
My own involvement in the Fly’s Eye project and the UC Berkeley Undergraduate
Honors Program would not have been possible without the patient tutelage I have been
honored to receive from both Mr. Dan Werthimer and Professor Geoff Marcy.
Professor Geoff Marcy is and has been an incredible advisor and friend. Our many
discussions regarding life, love and the pursuit of astronomical awareness have been
wonderfully enlightening.
Dan Werthimer is and has been a brilliant and fantastic research supervisor and friend.
I am utterly humbled to have had the opportunity to work so closely with him and the
SETI and CASPER research groups.
Part of this research has made use of the data base of published pulse profiles maintained by the European Pulsar Network, available at: http://www.mpifrbonn.mpg.de/pulsar/data/
Financial support for this research has been provided, in-part, by the Josephine de
Karman Fellowship Trust
– 22 –
A.
Appendix material
Four parallel time samples (acquired at 4x the
FPGA clk rate) are passed from the ADC to the
Digital Down Converter, where the signal is mixed
at 1/4 of the ADC clock rate (200 MHz if sampling
at 800 Msps), and then low-pass filtered to a
bandwidth of 1/4 the ADC clock rate, and decimated
to a sample rate of 1/4 the ADC clock rate. The
resultant signal is 8 bits I, 8 bits Q, representing the
band of signals from 1/8 the ADC clock rate to
3/8 the ADC clock rate (100 MHz to 300 MHz at
800 Msps).
This design is set for a maximum
clock rate of 210 MHz (sampling
from the ADCs at 800 Mxps). It
runs on an IBOB with 2 ADC boards,
and uses ModelSim to simulate an
SRAM interface.
System
Generator
ModelSim
xlhdlcosim
XSG core config
In1
sim_i
Sine Wave
sim_q
Out2
din2
Out3
din3
Out4
din4
i0
In1
i1
In2
i2
In3
i3
In4
q0
In1
q1
In2
q2
In3
q3
In4
In1
[ant1_pol2_adc]
Out1
[ant1_pol1_adc]
Out1
din1
Out2
din2
Out3
din3
Out4
din4
sync2
sync3
data_valid
a
xlslice
[a:b]
b
xlslice
[a:b]
[ant1_pol2_adc]
In1
[ant2_pol1_adc]
Out1
din1
Out2
din2
Out3
din3
Out4
din4
pol1_out1
din
c
pol2_in1
In1
[ant2_pol2_adc]
out0
d
pol2_out1
din
dout
pol1
z -2
sync
sync_out
shift
pol1_in1
pol1_out1
pol2_in1
pol2_out1
din
dout
downshift2
Out2
din2
Out3
din3
Out4
din4
dout
pol1
downshift3
out0
[ant2_pol1_fft]
out1
[ant2_pol2_fft]
fft1
pfb_fir1
sync_out
sync_in
decat3 [sync_adc]
dout
[ant1_pol1_fft]
a1p1
[ant1_pol2_fft]
a1p2
[ant2_pol1_fft]
a2p1
[ant2_pol2_fft]
a2p2
[sync_ant1_fft]
sync
[coeff]
coeff
reg_out
out
sim_out
en
acc_num
in
out
pulse_ext3
gpio_out
sim_out
led_new_acc
"acc_num" is a counter that is
incremented every new accumulation.
This allows a computer to know when
a new accumulation is available.
dout
sync_out
[coeff_en]
[fft_shift]
[coeff_addr]
DDC0_3
Logical3
rst
Counter
[ant12_pol12_rnd]
pol0
din1
din
dir_x
[OnePPS]
The "equalizer" allows each frequency channel of the
FFT to be scaled by a different number. This allows passbands which are
not flat to be flattened digitally to allow for optimal quantization to 4 bits.
These coefficients can also be updated dynamically for gain control.
[fft_shift]
sync_out
new_acc
acc_len
[ant1_pol2_fft]
fft
pfb_fir
sync
xlslice
[a:b]
acc_phs: bits 0-10
acc_len: bits 16-18
cross_bit_sel: bits 24-25
[ant1_pol1_fft]
out1
downshift1
sync_out
sync_out
reg_in
x_config
pol0
dout
dout
sim_in
[sync_ant1_fft]
xlslice
[a:b]
dout
downshift
DDC0_2
Out1
sync_out
shift
pol1_in1
sync_in
decat2 [sync_adc]
xllogical
or
-1
z
sync
sync_out
DDC0_1
Out1
0
z -2
sync_out
sync
sync_in
decat1 [sync_adc]
sync0
sync1
dout
sync_in
[sync_adc]
outofrangeq0
sim_data_valid
xlslice
[a:b]
sync
new_acc
acc_len
DDC0_0
outofrangei1
– 23 –
din1
ibob_lwip
outofrangeq1
1
Out1
decat
outofrangei0
sim_sync
xlslice
[a:b]
[sync_rnd]
coeff_we
coeff_addr
equalizer
adc0
sim_i
sim_q
i0
In1
i1
In2
i2
In3
i3
In4
q0
In1
q1
In2
q2
In3
q3
In4
Out1
[ant2_pol1_adc]
Out1
[ant2_pol2_adc]
gpio_out
led_1pps
outofrangei0
outofrangei1
Coefficients can be dynamically written into the "equalizer" by
specifying the channel ("coeff_addr") and the value of the
coefficient ("coeff"), and then using "coeff_en" to write the
new coefficient into the equalization table. In this design,
this can be done either through the "eq_coeff" software
register, or through the XAUI.
This, the "control" portion of the design, uses
32 bits of "eq_coeff" to program the equalization
coefficients, and 32 bits of "ctrl" to set fft shifting
and choose which ADC to use for calculating
input power (see "adc_sum_sq").
sim_out
[OnePPS]
outofrangeq0
sim_sync
outofrangeq1
sync0
sync1
sync2
sim_data_valid
sync3
data_valid
fpt
[sync_rst]
xllogical
or
z-1
xllogical
or
z-1
rst sync
[sync_adc]
sync_gen
Logical4
fpt
adc1
sim_in
1
Constant5
xlslice
[a:b]
reg_in
xlslice
[a:b]
eq_coeff
dbl
Gateway Out
To provide a means for measuring the magnitude of the signals
coming in to each ADC channel, we have a "calc_adc_sum_sq"
block which uses a single multiplier and a regsiter to calculate
the sum of the square of 2^16 samples. To do this with one
multiplier, only a single sample of the 4 provided is used
per clock, and then different samples are used each clock (in
a non-periodic order) to avoid favoring certain frequencies.
Of the output 32 bits, 16 represent "whole bits", and 16
are sub-bit resolution.
eq_coeff: bits 0-16
coeff_en: bit 17
coeff_addr: bits 20-30
[coeff]
llogica
and
xlslice
[a:b]
[coeff_en]
[coeff_addr]
dbl
Gateway Out2
fpt
dbl
Gateway Out1
1
Scope
fpt
fft_shift: bits 0-15
sync_rst: bit 16
arm_rst: bit 17
use_fft_tvg: bit 18
dbl
Gateway Out3
fpt
65535
Constant18
sim_in
dbl
Gateway Out4
reg_in
fpt
ctrl_sw
Scope1
dbl
Gateway Out5
z -1
fpt
[input_sel]
sel
[ant1_pol1_adc]
d0
[ant1_pol2_adc]
xlmux
d1-2
z
[ant2_pol1_adc]
d2
In1
Out1
In1
Out2
In2
Out3
Out4
decat4
[ant2_pol2_adc]
In3
sum_sq
In4
calc_adc_sum_sq
reg_out
sim_out
adc_sum_sq
xlslice
[a:b]
[input_sel]
xlslice
[a:b]
[use_sram_tvg]
xlslice
[a:b]
[use_fft_tvg]
xlslice
[a:b]
[arm_rst]
xlslice
[a:b]
[sync_rst]
xlslice
[a:b]
[fft_shift]
d3
dbl
Gateway Out6
Fig. 17.— Top-level FPGA Gateware Diagram
0
IBOB
LWIP
ModelSim
Band-Limited
White Noise
[ant12_pol12_rnd]
[sync_rnd]
[ant1_pol1_adc]
MSSGE
This is the "Fly's Eye" specific portion. 64 total bits are accumulated from each auto-correlation,
but 8 bit octets from each of the four input streams are concatenated and stored as 32-bit words in blockrams
labeled 0-7, 8-15, 16-23.....56-63. This allows very rapid
data dumps over 100Mbit ethernet of all four input streams at 8bit resolution.
The "pfb_fir" and "fft" blocks together implement a 128 point,
4 tap Polyphase Filter Bank. Between the two blocks, a "downshift"
divides the output of the "pfb_fir" by 8 so that the first 2 stages of
the FFT cannot overflow (see FFT documenation for details on why
this is necessary). Because the effective signal resolution out of
the "pfb_fir" is 8 bits, we are free to choose the placement of
these 8 bits within the 18 bits of the FFT without adversely affecting
our signal. However, one should avoid placing the signal in the LSB
of the FFT word because of rounding effects. Thus, we downshift
only 3 bits.
– 24 –
Fig. 18.— 44 Averaged Spectra from 18 Minute PSR 0329+54 Observation
– 25 –
REFERENCES
Blandford, R. D. 1977, “Spectrum of a radio pulse from an exploding black hole,” MNRAS,
181, 489
Hawking, S. W. 1974, Nature, 248, 30
Hawking, S. W. 1971, ”Gravitationally collapsed objects of very low mass,” MNRAS, 152,
75
S. R. Kulkarni, E. O. Ofek, J. D. Neill, M. Juric & Z. Zheng, ”Giant Sparks at Cosmological
Distances?” pre-print, http://www.astro.caltech.edu/ srk/sparker.pdf
Lorimer, D. R., Bailes, M., McLaughlin, M. A., Narkevic, D. J., and Crawford, F. 2007,
ArXiv e-prints, 709
D. R. Lorimer SIGPROC pulsar data analysis tools available online at
http://sigproc.sourceforge.net.
Manchester, R. N., Hobbs, G. B., Teoh, A. & Hobbs, M., AJ, 129, 1993-2006 (2005),
http://www.atnf.csiro.au/research/pulsar/psrcat/
Aaron Parsons, Donald Backer, Chen Chang, Daniel Chapman, Henry Chen, Patrick
Crescini, Christina de Jesus, Chris Dick, Pierre Droz, David MacMahon, Kirsten
Meder, Jeff Mock, Vinayak Nagpal, Borivoje Nikolic, Arash Parsa, Brian Richards,
Andrew Siemion, John Wawrzynek, Dan Werthimer & Mel Wright, 2005,
“PetaOp/Second FPGA Signal Processing for SETI and Radio Astronomy, Asilomar
Conference on Signals, Systems and Computers, November 2006
Rees, M. J. 1977, “A Better Way of Searching for Black-Hole Explosions?” Nature, 266, 333
Satten, Corey 2007, “Lossless Gigabit Remote Packet Capture With Linux,”
http://staff.washington.edu/corey/gulp/
– 26 –
Seriadakis, J. H., Gil, J. A., Graham, D. A. Jessnet, A., Kramer, M., Malofeev, V. M.,
Sieber, W. & Wielebinski, R., 1995. Pulsar profiles at high frequencies. I. The data.
Astro. Astrophys. Suppl. Ser. 111, 205
Siemion, A. & Chapman, D. 2007, “TinySH Video Lecture Series,”
http://casper.berkeley.edu/documentation
This manuscript was prepared with the AAS LATEX macros v5.2.