SIMPLE PULSE BLANKING TECHNIQUE and
IMPLEMENTATION in DIGITAL RADIOMETER
A Thesis
Presented in Partial Fulfillment of the Requirements for
the Degree Master of Science in the
Graduate School of The Ohio State University
By
Noppasin Niamsuwan, B.Eng.
* * * * *
The Ohio State University
2005
Master’s Examination Committee:
Prof. Joel T. Johnson, Adviser
Prof. Ronald M. Reano
Approved by
Adviser
Department of Electricaland Computer Engineering
ABSTRACT
Radio Frequency Interference (RFI) can adversely impact microwave passive re-
mote sensing measurements, and prohibits passive observations outside protected por-
tions of the radio frequency spectrum. A microwave radiometer including a digital
back-end with a simple real-time RFI mitigation technique for reducing pulsed RFI
has been developed to address this issue; the process is termed “asynchronous pulse
blanking” (APB). The idea of this technique is to remove incoming data whose power
exceeds the mean power by a specified number of standard deviations. Although suc-
cessful performance of this algorithm has been qualitatively demonstrated through
local experiments with the digital radiometer, a detailed quantitative assessment of
its performance in a variety of RFI environments has not been reported.
To address this issue, a simulation study of the APB algorithm was initiated using
data obtained from the L-Band Interference Surveyor/Analyzer (LISA), an airborne
instrument developed for observing the RFI in the region 1200-1700 MHz. LISA was
deployed on NASA’s P3-B aircraft to observe RFI in flights in the US and Japan.
This data set is very useful for assessing the APB algorithm, since many RFI envi-
ronments were observed that include multiple sources of interference. Several aspects
of algorithm use and performance have been studied, including means for choosing
the algorithm’s parameters and the robustness of the method in a realistic RFI envi-
ronment. Effects of the blanking process on the final output are also examined.
ii
In order to further quantitatively demonstrate the performance of the system, a se-
ries of L-band sky observations was performed. These experiments were based on the
predictable brightness of the cold sky background, so that long-term measurements
of instrument performance are possible. The results reported in this thesis confirm
stability of the system for six hours of observation. In addition, post-processing tech-
niques proposed for removing other types of RFI have been tested.
While the algorithm is conceptually simple, developing a process that is both
automatic and robust requires careful consideration of the implementation details.
Issues regarding implementation in field-programmable gate array (FPGA) devices
are also discussed in this thesis.
iii
To my Mom and Dad,
iv
ACKNOWLEDGMENTS
Grateful acknowledgement is made to my adviser Prof. Joel T. Johnson, for his
genuine support and beneficial suggestions throughout this research. It is his efforts
and supports that make this research possible.
This work was supported by NASA Instrument Incubator Program project NAS5-
02001.
v
VITA
January 17, 1981 . . . . . . . . . . . . . . . . . . . . . . . . . . . Born - Bangkok, Thailand
September 1999 . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.Eng., Electrical and ElectronicEngineering, Asian University ofScience and Technology, Thailand
1999-present . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Graduate Research Associate,The Ohio State University.
PUBLICATIONS
Research Publications
N. Niamsuwan and J. T. Johnson, “Sky Observations at L-band with an InterferenceSuppressing Microwave Radiometer”. IEEE IGARSS 2005 International ConferenceProceeding, Jul 2005.
N. Niamsuwan, J. T. Johnson, and S. .W. Ellingson, “Examination of a simple pulse-blanking technique for radio frequency interference mitigation”. Radio Science, Vol.40, No. 3, Jun 2005.
N. Niamsuwan, J. T. Johnson, and S. .W. Ellingson, “Examination of a simple pulse-blanking technique for radio frequency interference mitigation”. Workshop in Miti-gation of Radio Frequency Interference in Radio Astronomy, conference proceedings,Jul 2004.
J. T. Johnson, N. Niamsuwan, and S. .W. Ellingson, “Digital Receiver for InterferenceSuppression in Microwave Radiometry”. NASA’s ESTO Conference Proceeding, Jun2005.
J. T. Johnson, A. J. Gasiewski, B. Guner, G. A. Hampson, S. W. Ellingson, R.Krishnamachari, N. Niamsuwan, E. McIntyre, M. Klein, and V. Leuski, “Airborne
vi
radio frequency interference studies at C-band using a digital receiver”. IEEE Trans.Geosc. Remote Sens., preprint
vii
FIELDS OF STUDY
Major Field: Electrical and Computer Engineering
Studies in:
Remote Sensing Prof. Joel T. JohnsonRandom Media and Rough Surface Scattering Prof. Joel T. Johnson
viii
TABLE OF CONTENTS
Page
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Chapters:
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Digital Radiometer Design . . . . . . . . . . . . . . . . . . . . . . . 2
2. Examination of Asynchronous Pulse Blanking Technique . . . . . . . . . 7
2.1 Principle of asynchronous pulse blanking . . . . . . . . . . . . . . . 72.2 Hardware Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3 L-band Interference Surveyor/Analyzer (LISA) . . . . . . . . . . . 19
2.3.1 The Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.3.2 Deployment . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Simulation and Performance Evaluation . . . . . . . . . . . . . . . 252.4.1 Simulating the hardware process . . . . . . . . . . . . . . . 252.4.2 Choosing β2, Nblank, and Nwait . . . . . . . . . . . . . . . . 262.4.3 Output χ2 test . . . . . . . . . . . . . . . . . . . . . . . . . 322.4.4 Effect of blanking on integrated spectra . . . . . . . . . . . 362.4.5 Frequency domain blanking . . . . . . . . . . . . . . . . . . 41
ix
2.4.6 χ2-filter: alternative pulse blanking technique . . . . . . . . 422.4.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3. Sky observation with digital radiometer . . . . . . . . . . . . . . . . . . . 44
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443.2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . 443.3 Noise diode calibration . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3.1 Experimental setup for noise-diode calibration . . . . . . . . 473.4 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503.6 Post-processing mitigation technique . . . . . . . . . . . . . . . . . 51
3.6.1 Cross-time blanking . . . . . . . . . . . . . . . . . . . . . . 543.6.2 Cross-frequency blanking . . . . . . . . . . . . . . . . . . . 553.6.3 Temperature control and stability of the system . . . . . . . 593.6.4 Narrow-band interferences at 1400 and 1403 MHz . . . . . . 603.6.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4. Hardware issues for implementation in FPGA Components . . . . . . . . 65
4.1 Clock management for LISR3’s FPGA . . . . . . . . . . . . . . . . 674.2 Floorplan and Signal-Routing . . . . . . . . . . . . . . . . . . . . . 704.3 Implementation of slow scaling process . . . . . . . . . . . . . . . . 784.4 Thermal Control for Analog Down-converter . . . . . . . . . . . . . 79
4.4.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Appendices:
A. LISA’s frequency channels . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
x
LIST OF TABLES
Table Page
3.1 Contribution of narrow-band RFI to the average noise power (orderedby contributed error) . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
A.1 Frequency range of LISA’s channel 1 to 14 . . . . . . . . . . . . . . . 91
xi
LIST OF FIGURES
Figure Page
1.1 Block diagram of radiometer . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 Simulated sampled signal. The blanker is triggered whenever the powerof the input data exceeds the threshold. It is possible that anotherpulse may be received while the blanker is occupied. Extra blankersare required to ensure that the whole pulse can be completely removed. 8
2.2 FIFO buffer and Nwait to control pre-pulse blanking region. . . . . . . 11
2.3 Definition of Nblank and Nwait: note time is incresing from left to right 12
2.4 Simulated case for too small Nblank. For Nsep clock cycles after eachblanking process is initiated, new trigger would not be recognized.Small blanking region would fail to suppress successive pulses receivedduring Nsep period. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Though Nblank is sufficiency large to blank all samples during Nsep
period, Eq. 2.6 can not guarantee that successive pulse occurs withinNsep period can be completely suppressed (circled area). . . . . . . . 18
2.6 Block diagram of an L-band Interference Surveyor/Analyzer (LISA) . 20
2.7 Map of January 3rd transit fight from Wallops Island, VA to Monterey,CA, including known locations of ARSR systems (marked with ’×’).Edge markings are latitude and longitude. . . . . . . . . . . . . . . . 23
2.8 Pulse interference at 1340 MHz during transit flight over Illinois. . . . 23
xii
2.9 Navigation path of LISA flight. The location of the aircraft (lat. 39.12◦
N, long. 90.87◦ W), marked with dot and circle, where interference inFigure 2.8 is received. The nearest ARSR site (lat. 38.70◦N, long.90.39◦ W) is highlighted. . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.10 Estimated power (solid line) transmitted from the ARSR site in Fig-ure 2.9. Free-space model has been used assuming all power is receivedfrom the main-lobe of the radar’s antenna. The measured receivedpower is also included (dotted line). . . . . . . . . . . . . . . . . . . . 24
2.11 Initialize stage of APB process . . . . . . . . . . . . . . . . . . . . . . 26
2.12 Initialization process for the simulation software. The process is slightlydifferent than that implemented in the actual hardware due to theamount of data that is available. Several jumps in the first stage (forceupdate) corresponds to the presence of RFI. After 20480 samples (5captures), the mean and variance converge. The second stage (blankeron) of the process repeats the mean/variance computation with theblanker turned on to reduce the effect of pulsed RFI. . . . . . . . . . 27
2.13 Percentage of samples exceeding threshold for given β2. All 145 cap-tures of LISA’s channel 7 data are used. . . . . . . . . . . . . . . . . 29
2.14 Detection Threshold (for detecting pulses) and Reference Threshold(for estimating number of pulses). Nwait=0, Nblank=1536 are used inthe APB output results shown . . . . . . . . . . . . . . . . . . . . . . 30
2.15 (Pin−Pout)/Pin×100%, where Pin and Pout are the number of samplesexceeding the “reference threshold” at input and output of the APBalgorithm, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.16 Percentage of blanked samples ((Pin − Pout)/Pin × 100%) of LISA’schannel 13 data. The pulses shorter than 100 ns were ignored in thiscomputation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.17 Percentage of blanked samples of LISA’s channels 7–12. . . . . . . . . 33
xiii
2.18 Top: Noise power from 5 captures sucessively taken in the vicinityof Wheeling, IL (approx. 39.87◦ N lat, 80.67◦ W long) at 22,000 ft.LISA is tuned to channel 7 (1324 - 1344 MHz). Several examples ofpulsed RFI are observed. Bottom: χ2 of APB output as Nblank is varied(β2=40, Nwait=0). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.19 Definition of BLANK, PARTIAL BLANK and NO BLANK frames . . . . . . 37
2.20 Frequency spectrum of an example PARTIAL BLANK frame, comparedto the entire NO BLANK average for the corresponding capture . . . . . 39
2.21 Impact of PARTIAL BLANK on the final output. Top: Method 2: Instan-taneous Scaled averages compared to method 1. Bottom: The error ofmethods 2 (inst. scaled) and 3 (slow scaled) . . . . . . . . . . . . . . 40
2.22 Mean error in total power estimate from methods 2 and 3 in LISAchannels 7-13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.23 Frequency-domain APB algorithm results. . . . . . . . . . . . . . . . 42
3.1 Top: 10ft-diameter parabolic reflector. Bottom: Schematic diagram ofantenna front-end (AFE) . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2 Experimental Setup for noise-diode calibration . . . . . . . . . . . . . 48
3.3 Estimated ENR vs Lx pad . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4 APB-OFF: Top–Spectrogram, color scale represents sky brightness tem-perature in Kelvin. Bottom–Average brightness temperature VS Time.Without RFI mitigation some of data are corrupted. Pulsed RFI at1330 MHz is expected and responsible for 3-Kelvin oscillation in theaverage temperature plot. . . . . . . . . . . . . . . . . . . . . . . . . 52
3.5 APB-ON: Spectrogram and average noise temperature plot observedwith the blanker turned on. . . . . . . . . . . . . . . . . . . . . . . . 53
3.6 APB-ON with cross-time blanking applied . . . . . . . . . . . . . . . 54
xiv
3.7 Xfreq-I cross-frequency RFI detection algorithm. A simple thresholdis derived from the average noise temperature over the 100 MHz spec-trum. Th possible lowest threshold (dashed line) is limited by the gainpattern of the instrument. . . . . . . . . . . . . . . . . . . . . . . . . 56
3.8 Xfreq-II cross-frquency RFI detection algorithm. The threshold isderived from the derivative of the data. This allows weak RFI to bedetected. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.9 Xfreq-III cross-frequency RFI detection algorithm. Gain pattern isfirst estimated by interpolating local minima of the frequency spectrum(Left panel). This allows the contribution of uncalibrated instrumentgain pattern to be considered in the detection process. . . . . . . . . 58
3.10 APB-ON with cross-time and cross-frequency blanking applied. . . . 59
3.11 A survey of 1400 MHz RFI. The antenna was focused to different di-rections. Strong RFI is found from the East while there was no otherRFI received from other directions. . . . . . . . . . . . . . . . . . . . 61
3.12 Multi-window plot show the spectrum centered around 700 MHz and1400 MHz measure at the same time. The antenna is directed into theradiometer back-end rack. . . . . . . . . . . . . . . . . . . . . . . . . 62
3.13 1403 MHz radiation from spectrum analyzer. Observation at 1403MHz with a spectrum analyzer also found the RFI leaked from theinstrument itself . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.1 LISR2 digital backend; the vertical cascade of three circuit boards nearthe left hand side contains the dual ADC sections (upper and lowerboards) and the digital channel combination and filtering (DIF) sec-tion (center board). The APB section for removing temporal pulses isalso implemented on the center board. Following the vertical cascadeto the right is the FFT processor, then the SDP section for power com-putation and integration operations. Finally a “capture card” providesthe interface to the PC. Microcontrollers are also included on each card(the smaller attached circuit boards with ethernet cables) to enable PCsetting of FPGA parameters through an ethernet interface. . . . . . . 66
xv
4.2 LISR3 digital backend; underneath a heat-sink on the center board isthe Stratix FPGA which contains the entire processor. The boardson the left and right of the Stratix board are the ADC cards. Thesystem clock used to run the Stratix processor is also derived fromone of these ADC cards. Above the Stratix card is a “capture card”providing a 256K-FIFO buffer and the interface to the PC. Only asingle Microcontroller card (the smaller card attached to the bottomof the Stratix board) is needed in this prototype. Except for the Stratixcard, the other parts of the hardware are the same as in LISR1 andLISR2. The clock-generator and power-supply (top-left and bottom-right respectively) are also the same as in previous prototype. . . . . 68
4.3 Clock and data signal at the output of ADC cards (dashed line) beforecrossing the SCSI-3 connector to the Stratix card (solid line). The min-imum high (1.7 V) and maximum low (0.7 V) voltage level accordingto LVTTL/LVCMOS requirement are also shown as a reference. Thelower plots demonstrate the expected logic level to be recognized bythe Stratix FPGA according to these reference levels. . . . . . . . . . 69
4.4 Dual-clock FIFO. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.5 Clock regeneration and synchronization scheme for Stratix FPGA . . 71
4.6 Final layout of Stratix FPGA. Yellow-shaded area shows occupied re-sources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.7 Floorplan of the Stratix FPGA. All marked/labeled areas are assignedas Logic-Lock region. During “fitting” process, the simulator will at-tempt to allocate resources according to these given criteria. The con-nections to the rest of the system are also shown. . . . . . . . . . . . 73
4.8 DSPs and M4K-RAM locations in Stratix FPGA. The location of PLLused to generate the “master clock” for the entire chip is also shown. 75
4.9 M512-RAM and M-RAM locations in Stratix FPGA . . . . . . . . . 76
4.10 Detailed block diagram of APB and SDP. . . . . . . . . . . . . . . . . 78
4.11 Slow-scaling test results . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.12 Temperature of down-converter compared to the ambient temperature. 80
xvi
4.13 Power at distict ports of the front-end switch: shown are Noise Diode,Terminator and Antenna (H-Pol). The front-end temperature is set to45 ◦C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.14 Comparison of the power seen at antenna port and the tempearutureof down-converter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.15 Thermal insulation plan and location of the temperature probes . . . 84
4.16 Temperature of down-converter enclosure for sky-observation in Chap-ter 3. The temperature measured from “Control” probe is used as aninput for controlling feed-back loop. . . . . . . . . . . . . . . . . . . . 86
xvii
CHAPTER 1
Introduction
Radio Frequency Interference (RFI) mitigation is very important for passive re-
mote sensing observation. At L-band, since only a 27 MHz spectral window (1400–
1427 MHz) is preserved for passive operations, the presence of numerous RFI sources,
including strong pulsed RFI from ground-based aviation radars (GBARs) [1, 2], cause
spectroscopy outside this protected band can be adversely impacted. To address this
issue, a digital radiometer with a capability of detecting and rejecting pulse-like RFI
was developed at the Ohio State University ElectroScience Laboratory. This RFI
mitigation technique relies on the high temporal sample rate of the system. For ra-
diometers operating at a sufficiently high sample rate, a simple strategy for reducing
pulsed RFI is to remove incoming data whose power exceeds the mean power by a
specified number of standard deviations. It may also be advantageous to remove data
within a specified time region before and after this “trigger” data, to ensure that any
pre- and post-detection “pulse” information is successfully removed. The process is
termed “asynchronous pulse blanking” (APB) because no periodic properties of the
interference source are assumed.
1
Successful performance of this algorithm has been qualitatively demonstrated
through local experiments with the digital radiometer [3]. However, a detailed quan-
titative assessment of its performance in a variety of RFI environments has not been
reported.
This thesis involves an examination of APB technique. The next section serves as
a preview of the digital radiometer’s design. Detailed discussion on this mitigation
technique can be found in Chapter 2 to 4.
1.1 Digital Radiometer Design
Under the support of the NASA “Instrument Incubator Program” (IIP), devel-
opment of a digital-receiver based “L-band Interference Suppressing Radiometer”
(LISR) was initiated in December 2001. The project is primarily focused in L-band
and C-band given the received strong RFI sources at L-band and limited protected
spectrum, as well as the absence of protected spectrum at C-band. The project is
focused on passive microwave observation of the Earth in these bands; however, the
radiometer can be applied for L-band radio astronomy, and both applications are
severely impacted by RFI.
A block diagram of the L-band radiometer is shown in Figure 1.1. The analog
front end downconverts an 80 MHz swath of spectrum from L-band to 150 MHz, and
samples this signal at 200 MSPS using 10 bits. Note the system can also be operated
at frequencies other than L-band simply by modifying the analog low-noise front-end
and downconverter sections. Using a large number of ADC bits opens the possibility
of reduced gain in the radiometer frontend and downconverter sections, as higher
dynamic range is typically achieved in ADC’s by increasing sensitivity (i.e. lowering
2
L o w � n o i s eF r o n t � e n d A D C2 � c h a n n e l o f8 0 M H z B W@ 1 5 0 M H z D i r e c t ( C o n v e r s i o n R e c e i v e r
A n a l o gD o w n C o n v e r t e rD i g i t a l I F( D I F )
A s y n c h r o n o u sP u l s e B l a n k e r( A P B )1 K F F TI n t e g r a t i o n ( S D P )C a p t u r e b o a r d( 2 5 6 K F I F O )
A D C D i g i t a l I F( D I F )2 0 0 M S P S @ 1 0 b i t s 5 0 M H z B W @ + 2 5 M H z , 1 0 0 M S P S 1 6 + 1 6 b i t s
5 0 M H z B W @ d 2 5 M H z , 1 0 0 M S P S 1 6 + 1 6 b i t s
Figure 1.1: Block diagram of radiometer
the power required to toggle the lowest ADC bit.) Reduced gain is desirable in general
in order to improve receiver stability, and thereby potentially reduce thermal control
requirements throughout the entire system. Trade-off studies of receiver stability
versus gain were not performed under the IIP project however.
Although resolving 100 MHz with one such ADC would be possible, it was deemed
preferable to utilize 2 ADC’s sampling 50 MHz each so that digital filtering could be
incorporated into the RFI processor to limit receiver bandwidth digitally. Use of
3
digital filters is desirable due to possible stability issues of analog filters with temper-
ature and other environmental variations, particularly near the cutoff frequencies of
the filter response. The design developed keeps the analog filter passband wider than
that ultimately set by the digital filters, so that the near cutoff regions of the analog
filters do not contribute significantly to the observed bandwidth.
Accordingly, the 100 MHz bandwidth is split into 2 50 MHz channels, and each
channel is then sampled at 200 MSPS using 10 bits. The resulting digital data of
each channel is centered at 50 MHz and is spectrally reversed due to the use of the
second ADC Nyquist zone. The “Digital IF” (DIF) FPGA module downconverts each
channel to 0 Hz (so now the samples are complex-valued), digitally filters each to 50
MHz bandwidth, decimates by 2, and then up- or down-converts the two channels
to center frequencies of +/-25 MHz (still complex). Finally both channels are added
so that -50 to 50 MHz data emerges from the DIF module in 16-bit “I”+16-bit “Q”
format at 100 MSPS.
Following the DIF output is a cascade of FPGA modules which can be pro-
grammed to perform a variety of functions. The favored strategy currently is as
shown in Figure 1.1: mitigation of radar pulses using APB, channelization into 100-
kHz bins using a 1K FFT, and integration to generate power spectra.
The APB is designed to detect and blank radar pulses, which typically are the
dominant source of external L-Band RFI below 1400 MHz. Radar pulses range from
2-400 µs in length and occur 1-75 ms apart [1]. To detect these pulses, the APB
maintains a running estimate of the mean and variance of the sample magnitudes.
Whenever a sample magnitude greater than a threshold number of standard devia-
tions from the mean is detected, the APB blanks (sets to zero) a block of samples
4
beginning from a predetermined period before the triggering sample, through and
hopefully including any multi-path components associated with the detected pulses.
APB operating parameters are adjustable and can be set by the user.
Following the APB is a length-1K complex FFT, which achieves approximately
98% duty cycle in performing the FFT computations (i.e., 98% of the data is FFTed,
and the rest is lost). A triangular window is applied before the FFT. Planned but
not yet implemented is a frequency-domain blanking (cross-time blanking as in Sec-
tion 3.6.1) module, which is similar in concept to the APB, except applied indepen-
dently to each frequency bin. The purpose of this module will be to exploit the
processing gain achieved through channelization to detect and excise weak, relatively
narrowband RFI. The FFT output is processed through a “spectral domain proces-
sor” (SDP) module which computes magnitude-squared for each frequency bin and
computes a linear power average over many FFT outputs. These results are passed
at a relatively low rate to a PC via a capture board. Total power can be computed
by summation of frequency bins within the digital hardware, or the same process can
be implemented within the PC for increased flexibility in monitoring RFI, selecting
subbands, and so on.
This radiometer has been used in both radio astronomy and remote sensing exper-
iments locally at the ElectroScience Laboratory [3] and at the Arecibo observatory
in Puerto Rico [2, 4]. The results qualitatively show the APB approach to be suc-
cessful for pulsed RFI mitigation, although a detailed performance assessment will be
discussed in this thesis.
5
In Chapter 2, a simulation study of the APB algorithm is discussed. The effects of
each controlling parameter and optimization of the overall process has been investi-
gated. Performance of the blanking algorithm is evaluated and presented in the same
chapter. Chapter 3 describes a deployment of the system for sky observation. This is
of interest in order to provide quantitative assessment of the system performance. A
discussion and simulation on post-process data enhancement including algorithms to
detect non-pulsed RFI, e.g. narrow-band RFI, are also reported. Finally, hardware
issues regarding implementation on field-programmable gate array (FPGA) devices
are summarized in Chapter 4.
6
CHAPTER 2
Examination of Asynchronous Pulse Blanking Technique
2.1 Principle of asynchronous pulse blanking
The APB algorithm is intentionally simple to keep implementation in hardware
feasible. As illustrated in Figure 2.1, a pulse-detection criterion is to declare RFI
detection whenever any high-power signal samples are observed. For each sample x in
the data stream, ||x||2 (the squared-magnitude of sample, i.e. the power) is computed
and compared to the threshold. This threshold is defined as βσ + m where σ and m
are the standard deviation and the mean, respectively. Both mean (m) and standard
deviation (σ) are estimated from previous data samples. The β is a parameter to
determine “aggressiveness” of detection process. Large β value may allow small RFI
to pass through while small β value will ensure that all RFI will be detected but
it might trigger some large noise peaks and treat them as RFI. Consequently, the β
parameter should be optimized to trade-off loss of noise power integration time versus
blanking performance. Section 2.4.2 discusses how to choose a proper value for this
parameter.
From a hardware perspective, computing the standard deviation is an expen-
sive process both in terms of implementation complexity and processing time. The
7
0 2 0 4 0 6 0 8 0 1 0 0 1 2 0 1 4 0 1 6 0 1 8 0 2 0 0T i m e ( � s e c )
P ower(li nearuni t s)T h r e s h o l d
N B L A N K
N S E P
N W A I TP u l s e R F I d e t e c t e d
B l a n k i n g R e g i o n
P u l s e R F I d e t e c t e d
T i m e I n d e xFigure 2.1: Simulated sampled signal. The blanker is triggered whenever the powerof the input data exceeds the threshold. It is possible that another pulse may bereceived while the blanker is occupied. Extra blankers are required to ensure that thewhole pulse can be completely removed.
8
variance (σ2) is rather preferable to deal with. Therefore, the detection criteria is
rearranged in equivalent form as:
(||x||2 −m)2 ≥ β2σ2 (2.1)
Because some portion of the pulse typically appears in the data stream before its
power is high enough to trigger the threshold (Figure 2.1), it is required in the pulse
removing process also to remove samples prior to a detection. To blank pre-detection
samples, a buffer (memory) is required in the system. The detection process is then
performed on the sample entering the buffer while the blanking operations and other
successive operations are performed on the samples exiting the buffer. A simple
first-in-first-out (FIFO) buffer suffices; and the length of the FIFO (FIFO LENGTH)
determines maximum number of samples that can be blanked before the detected
sample.
Even though a FIFO buffer allows all the samples in the buffer to be blanked if
the pulse is detected, this is not always necessary. Blanking the whole buffer may
removed desired noise information and should be avoided if possible. To control the
length of the pre-pulse blanking, the parameter Nwait is introduced. Once detection
is asserted, a process is initiated to wait for Nwait samples before the FIFO output
is set to zero. The number of pre-detection samples remaining in the buffer after
Nwait clock cycles have elapsed is then FIFO LENGTH - Nwait. The FIFO output is
then continually forced to zero for a specified number of clock cycles. The length of
this blanking “window” is determined by the final APB parameter, Nblank. The set
of parameters β, Nwait, Nblank, and FIFO LENGTH are the basic conceptual quantities
that control the APB process.
9
Figure 2.2 demonstrates the blanking process with the use of a FIFO. The pulse
is detected (at the FIFO input) at time t2. On the left panel, Nblank was set to 4
clock cycles; Nwait is not utilized (Nwait=0). The blank signal is therefore asserted
immediately after detection. As a result, 4 samples, starting from the one appearing
at the FIFO output at the time of detection, will be removed. In contrast, with
Nwait = 2 (the right panel), the blank signal has been delayed for 2 completed clock
cycles leaves only one pre-pulse sample is set to zero. Nblank of 2 is sufficient to
completely remove pulse contribution in this case.
Figure 2.3 illustrates simulated incoming signal power. A simulated pulse is also
shown in this figure. It can be clearly seen from the figure how the blanking region
is determined for given parameters. The combination FIFO LENGTH-Nwait controls
the size of the blanking region prior to detection. This value should be sufficiently
large to ensure that even largest pulses can be completely removed. The size of the
total blanking region is controlled by the parameter Nblank which obviously must be
large enough to ensure that any possible multi-path contribution of the pulses can be
successfully blanked.
2.2 Hardware Issues
Since the technique is intended to be implemented using FPGA, the algorithm
needs to be designed to exist within limited hardware resources.
FIFO LENGTH
Ideally, the FIFO should be made as large as possible in order to allow the largest
pre-detection period. However, hardware issues (i.e. size of the FPGA used) limit
the maximum FIFO size which can be included. For the IIP radiometer and for
10
P u l s e D e t e c t e dB l a n k t 0 t 1 t 2 t 3 t 4 t 5 t 6 t 7 t 0 t 1 t 2 t 3 t 4 t 5 t 6 t 7I n p u t F I F O O u t p u tP u l s e D e t e c t e d
B l a n k O u t p u t( 1 )B l a n k O u t p u t( 2 )B l a n k O u t p u t( 3 )B l a n k O u t p u t( 4 )
P u l s e D e t e c t e dW a i t( 1 )W a i t( 2 )
B l a n k O u t p u t( 1 )B l a n k O u t p u t( 2 )
P u l s e D e t e c t e dB l a n kS t r o n g p u l s et r i g g e r s t h e d e t e c t o r
t i m e t i m e
Figure 2.2: FIFO buffer and Nwait to control pre-pulse blanking region.
11
0 5 10 15 200
10
20
30
40
Pow
er
(dB
)
Sample Index (× 1000)
InputOutput
FIFO_LENGTH - Nwait
Nblank
Pulse Detected
Figure 2.3: Definition of Nblank and Nwait: note time is incresing from left to right
the remainder of the simulations performed in this thesis, FIFO LENGTH was set to
1 K due to size limitations of the FPGA components used in the prototype system.
This length of FIFO (1 K – representing 10.24 µsec of data given the 10 nsec sample
spacing) is shown to be sufficient for application at L-band (Section 2.4)
Running Filter
The APB algorithm requires a running estimate of the mean (mt) and variance
(σ2t ) of the incoming signal power. These estimates are computed as:
mt = µmean ×mt−1 + (1− µmean)× ||x||2 (2.2)
σ2t = µvar × σ2
t−1 + (1− µvar)× (||x||2 −mt)2 (2.3)
where mt−1 and σ2t−1 are the mean and variance generated on the previous clock
cycle, and µmean/var ≈ 1 are constants that control the response rate of these simple
averaging filters. In the APB hardware, the above computations are accomplished
by groups of fixed-point multipliers, adders, and delay registers. In addition, a state
machine controls these filters so that contributions from incoming post-detection data
12
that will be blanked are not included in computing mt and σ2t . The quantities (||x||2−
mt)2 and σ2
t available from equation (2.3) additionally make it convenient to perform
the pulse detection test as (||x||2 −mt)2 > β2σ2
t , because the square root operation
needed in computing σt from σ2t is then avoided. Further discussions of the threshold
will thus focus on β2 as opposed to β due to this fact.
Large Bit-width
Due to the squaring operation in computing ||x||2 (represented originally as 16 bit
I and 16 bit Q), a wide datapath of 32 bits is required to preserve accuracy for this
quantity. The estimate of σ2 further computes the square of ||x||2, requiring 64 bits.
The high dynamic range of the filter constants µ ≈ 1 and (1−µ) ≈ 0 further increases
the number of bits required in estimating the running mean and variance of ||x||2.
Use of 12 bits (µ = 0.999756) to capture both µ and (1 − µ) was found desirable
in setting an appropriate averaging length for the filter. The resulting arithmetic
operations grow to a maximum size of 76 bits, and severely limit the computation
speed of the detector due to this large size.
Processor Speed
Because of the large bit growth of the APB processor and speed limitations of the
FPGA components currently used, the detection operation can not be performed on
100% of the incoming data. The current implementation of the design can process
only one in four samples (25%) of the data for detection. A decimation controller
is introduced to sub-sample the input data stream to every 4 samples (40 nsec for
10 nsec sample spacing) for detection operations. However, since radar pulses are
typically longer than this sub-sampling period, they are still detectable even under this
13
limitation. This sub-sampling operation refers only to samples analyzed for detection,
the remainder of the system, including the blanking operation, still operates on 100%
of the incoming data. Note an average of four successive samples could also be used
in the decimation controller rather than the sub-sampling method implemented, and
would likely result in improved detection performance. However for simplicity in
implementation the sub-sampling method was utilized.
Control of Blanking Process
When a detection is obtained, a simple state machine is initiated to wait for
Nwait clock cycles, then blank the output of the FIFO for Nblank clock cycles. The
state machine controlling this process is called a “blanking timing register” (BTR).
Figure 2.1 demonstrates a case that a second RFI pulse is received while the BTR is
occupied for blanking process of the recently detected pulse. Note the second pulse
exceeds the detection threshold only within the period of the first blanking operation,
and does not exceed the threshold after the first blanking operation has completed.
If no second BTR were initiated, a portion of the second pulse would be allowed to
remain in the data. However a second blanking operation is initiated and continually
removes second pulse contributions.
Although the BTR is simple to implement in hardware, it still occupies a non-
negligible portion of the available FPGA logic. Therefore it is not possible to have an
unlimited number of BTR’s in the system. Because a BTR is unavailable to begin a
blanking process for Nwait +Nblank cycles after it has been “triggered”, the possibility
of having no BTR’s available to blank a detected pulse must be considered if only a
finite number of BTR’s exist.
14
A limitation in the number of BTR’s also raises another issue: a large pulsed
interferer presumably could trigger all available BTR’s on successive clock cycles, with
the resulting blanking windows overlapped almost completely at the FIFO output.
The ability of the APB to blank detected pulses would then be compromised for
approximately Nwait+Nblank clock cycles when the initial BTR would become available
again. This problem motivates the introduction of a parameter Nsep, defined to be
a number of clock cycles after a detection during which no further BTR’s will be
triggered. If M BTR’s are available, then setting
Nsep =1
M(Nwait + Nblank) (2.4)
will ensure that a BTR is always available for triggering once Nsep has elapsed. Al-
though Nsep values larger than the above value would also satisfy this requirement,
minimizing Nsep is desirable to ensure maximal pulse detection.
While introduction of Nsep manages the use of a finite number of BTR’s, it also
may compromise the ability of the APB to remove detected pulses, if a detection
occurs within Nsep clock cycles of an earlier detection. Although a new BTR process
will not be initiated for Nsep clock cycles, the APB algorithm can still be designed
so that all post-detection signal information within Nsep clock cycles will be blanked.
Referring to Figure 2.3, the post-detection blanking region contains Nblank + Nwait −
FIFO LENGTH samples for a fixed Nblank. Choosing
Nsep ≤ Nblank + Nwait − FIFO LENGTH (2.5)
will ensure that all signal information within Nsep samples after a detection will be
blanked. Figure 2.4 demonstrates a simulated case when this criterion is violated.
15
P u l s e d e t e c t e dT h r e s h o l d
B l a n k i n g R e g i o n( B T R # 1 )
N s e pP o s t & d e t e c t i o n R e g i o n
Figure 2.4: Simulated case for too small Nblank. For Nsep clock cycles after eachblanking process is initiated, new trigger would not be recognized. Small blankingregion would fail to suppress successive pulses received during Nsep period.
16
Combining equations (2.4) and (2.5) results in the condition
Nblank ≥M
M − 1FIFO LENGTH−Nwait (2.6)
to ensure that all samples exceeding the specified threshold will be blanked. Although
this equation appears to require that Nblank increase with FIFO LENGTH, the associated
Nwait would also likely increase with FIFO LENGTH so that Nblank would not necessarily
increase. The digital radiometer prototype at present contains 4 BTRs; this number
will be assumed in the simulations of Section 2.4. For this case with FIFO LENGTH =
1024, Nblank should be chosen ≥ 1366 − Nwait. Note that even though the choice
of equation (2.6) ensures that all samples exceeding the threshold will be blanked,
successive pulses occurring within the Nsep period of an initial pulse will still not have
a complete blanking window surrounding all their contributions (Figure 2.5).
17
P u l s e d e t e c t e dT h r e s h o l d
B l a n k i n g R e g i o n ( B T R # 1 )
N s e p
Figure 2.5: Though Nblank is sufficiency large to blank all samples during Nsep period,Eq. 2.6 can not guarantee that successive pulse occurs within Nsep period can becompletely suppressed (circled area).
18
2.3 L-band Interference Surveyor/Analyzer (LISA)
2.3.1 The Design
In order to perform a more complete assessment of the APB algorithm, it is
desirable to study performance in a complex and varying RFI environment. However,
experiments at L-band with the radiometer prototype to date have occurred only in
the local environment and at the Arecibo facility. Furthermore, it is desirable to
perform these assessments in a software rather than hardware environment, so that
quantitative information can be obtained as APB parameters are varied for a fixed
specific dataset. The dataset from the LISA sensor provides an excellent resource for
these studies.
LISA, completed in September 2002, was developed as a means to observe sources
of radio frequency interference (RFI) in the region of 1200-1700 MHz from an air-
borne platform [5, 6, 7]. LISA’s block diagram is shown in Figure 2.6. Physically,
LISA is comprised of a small antenna/ front-end unit (AFE) and an equipment rack.
The antenna unit consists of a nadir-facing cavity-backed planar spiral antenna with
an integrated custom RF front-end including filtering and calibration circuitry. The
antenna has a very broad pattern (approximately “cos θ”) and is reasonably well-
matched over the span of the observations reported here. The antenna unit is con-
nected to an equipment rack mounted in the aircraft cabin by a long and fairly lossy
coaxial cable. Although the cable loss degraded the sensitivity of the instrument,
the resulting gain profile was an important factor in preserving the linearity of the
system. While observing strong RFI waveforms, this consideration was paramount,
but comes at the expense of the system’s ability to detect weak RFI.
19
F r o n t � e n dA D CA D C IQ
S p e c t r u m A n a l y z e r2 5 6 KF I F O
B P F : 1 2 0 0 ' 1 8 0 0 M H zL o n g C o a x2 0 M S P S
L P F : 7 M H zD i r e c t ' C o n v e r s i o n R e c e i v e r 1 2 0 0 ' 1 7 0 0 M H z
Figure 2.6: Block diagram of an L-band Interference Surveyor/Analyzer (LISA)
Inside the equipment rack, the signal is delivered to a custom-designed coherent
sampling subsystem. This subsystem uses a direct-conversion receiver to tune (under
PC control) anywhere between 1200 MHz and 1700 MHz. “I” and “Q” signals at
baseband are low-pass filtered with ∼ 7 MHz cutoff and sampled at 20 MSPS, yielding
a digitized bandwidth of ∼ 14 MHz. The output samples are queued in a 16K-sample-
long first-in-first-out (FIFO), whose contents are acquired by means of the PC parallel
port. These 16K “captures” represent an 819.2 µsec sample of the received field
at the current center frequency, and require approximately 1 sec to be transferred
20
and recorded on the computer. During flight, the coherent sampling system was
successively tuned through center frequencies of 1250, 1264, ..., 1698 MHz; these
distinct bands are called channel 1, 2, etc. in what follows. Channels 1–14 and their
corresponding operating frequencies are tabulated in Table A.1, Appendix A, as a
reference.
For each channel, 5 819.2 µsec captures are performed before tuning to the next
center frequency. After completing a sweep of all channels, a backup operation was
performed to a separate storage component in the computer system. The time re-
quired for this operation caused a 10 to 15 minute delay between “sweeps”. The final
dataset provides a high temporal resolution but very low duty cycle observation of 4.1
msec captured per LISA channel per 12 to 15 minutes. This high resolution recorded
data allows the APB algorithm to be simulated and studied in softwares opposed to
hardware.
2.3.2 Deployment
For the datasets considered here, LISA was installed in NASA’s P-3 research
aircraft, which is based at the Wallops Flight Facility (WFF) located at Wallops Is-
land, VA. The LISA antenna unit was mounted in the tail radome. The loss due to
transmission through the radome is unknown and may be another factor degrading
sensitivity. The data used in the simulations to follow were collected during a single
flight on January 3, 2003 along an east-west track from Wallops Island, VA to Mon-
terey, CA at approximately 22,000 ft. These data obviously include observation in
a variety of RFI environments. Data from LISA channels 6-13 (1310 − 1428 MHz)
is used in the simulations; for each of these channels, 145 captures (5 captures × 29
21
sweeps) were available. LISA was also deployed in the “Wakasa Bay” remote sensing
campaign in Japan through Feb of 2003; data on the transit flights from Monterey to
Japan were also collected [7]. However, the RFI environments in these flights show
less variation than that of the January 3rd flight, and so these data are not considered
further in this simulation.
Figure 2.7 illustrates the navigation path of P-3 aircraft on January 3rd. During
this transit flight, the aircraft was in proximity of several Air Route Surveillance Radar
(ARSR) radar sites. Several interferers were observed at center frequencies ranging
from 1313 to 1349 MHz, and in many cases can be correlated with known air traffic
control sites. Figure 2.8 shows a sample of an interferer with multipath contributions
detected close to one of ARSR sites in Illinois (Figure 2.9). The frequency of this
pulse is 1340 MHz with a pulse width of 6 µsec. The power seen at the antenna
terminal is plotted in Figure 2.10 along with estimated power according to a free-
space propagation model. The plot shows some feature even though the model seems
to overestimate the power level. Considering rotation period of the radar suggests
that the signal is possibly received from the side-lobe of the radar’s antenna; this
explains the difference between the estimated power level and observed data.
Other interferers, whose pulse widths range from 2–6 µs, were detected when the
aircraft was passing other radar sites as well. Some of those pulses were received
without significant multipath effects while some of them show multipath contribu-
tions. Generally, this campaign provides useful datasets for RFI mitigation testing
purposes.
22
−120 −110 −100 −90 −80 −7035
40
Figure 2.7: Map of January 3rd transit fight from Wallops Island, VA to Monterey,CA, including known locations of ARSR systems (marked with ’×’). Edge markingsare latitude and longitude.
500 505 510 515 520 525 530−80
−60
−40
Pow
er (
dBm
)
Time(µs)
(a) Pulse length of approximately 6 µS is observed. The multipath contribution isalso received.
1325 1330 1335 1340
−100
−50
Frequency (MHz)
Pow
er (
dBm
)
(b) FFT reveals the frequency of the pulse is approximately 1340 MHz
Figure 2.8: Pulse interference at 1340 MHz during transit flight over Illinois.
23
Figure 2.9: Navigation path of LISA flight. The location of the aircraft (lat. 39.12◦
N, long. 90.87◦ W), marked with dot and circle, where interference in Figure 2.8 isreceived. The nearest ARSR site (lat. 38.70◦N, long. 90.39◦ W) is highlighted.
16:39:41 16:39:44 16:51:25 16:51:26 17:03:24 17:15:36−70
−60
−50
−40
−30
Pow
er (
dBm
)
Time(HH:MM:SS)
EstimatedMeasured
Figure 2.10: Estimated power (solid line) transmitted from the ARSR site in Fig-ure 2.9. Free-space model has been used assuming all power is received from themain-lobe of the radar’s antenna. The measured received power is also included(dotted line).
24
2.4 Simulation and Performance Evaluation
The APB algorithm is examined in various aspects. First, an optimum choice of
β2, Nblank and Nwait parameters is investigated. Next a χ2-test [8] is utilized in order
to verify the success of the algorithm by evaluating statistical properties of the data
before and after pulse-blanking. Finally, errors introduced from the pulse-blanking
stage will be explored.
2.4.1 Simulating the hardware process
This section highlights some concerns about simulation of the radiometer’s digital
processing units. With a few exceptions, all aspects of hardware limitations have been
taken into account in the simulating software. For instance, the number of BTR’s was
limited to 4, and the FIFO LENGTH was set to 1024 according to the current hardware
prototype. The running average used a 12-bit bus-width for µ and (1−µ). However,
the efficiency of the FFT process (98% of data is FFTed), and the issue of a fixed-
width datapath (16 bits) was not considered; these issues do not significantly affect
the performance of the algorithm.
Another difference in the simulating software compared to the process imple-
mented in the prototype is the APB’s initialization stage. Figure 2.11 shows the
state machine of the APB’s initialization process implemented in the FPGA. After
a system global reset, the processor performs a “force-update” mode, using the first
262144 samples to compute an initial mean and average before starting the pulse-
detection cycle. Another simulation study shows that using 12-bit µ and (1−µ), and
a 24-bit datapath results in convergence (with 4% and 13% variation in mean and
variance, respectively) after 20000 snapshots. In this simulation, the first 5 captures,
25
providing (5 ×16384) / 4 = 20480 snapshots (assuming a decimation of 4) were used
in the the force-update stage. A longer initialization time was prohibited due to the
fact that the the noise environment varied every 5 captures depending on the location
of the aircraft. Furthermore, 5 captures may not be sufficient for the initialization
process since 20000 snapshots convergence assumes as RFI-free environment, whereas
the LISA first 5 captures might contain corrupted samples that could bias the statis-
tics of the data. The second stage of the simulated initialization is performed by
processing the same set of data with the pulse blanker turned on so that the effect of
RFI contributions on the running average computation was minimized. Figure 2.12
illustrates the result of these processes using the first 5 captures of LISA’s channel 7
data. R e s e t W a i t 2 6 2 1 4 4 c l o c k c y c l e sR e s e t A l l F o r c e u p d a t e m e a n / v a r i a n c e S t a r t p u l s e � d e t e c t o r/ p u l s e � b l a n k e r
Figure 2.11: Initialize stage of APB process
2.4.2 Choosing β2, Nblank, and Nwait
Parameters β2, Nblank, and Nwait affect the number of pulses that will be removed
in the blanking process. Nwait and Nblank directly control the size of blanking region.
26
(a) Mean (b) Variance
Figure 2.12: Initialization process for the simulation software. The process is slightlydifferent than that implemented in the actual hardware due to the amount of datathat is available. Several jumps in the first stage (force update) corresponds to thepresence of RFI. After 20480 samples (5 captures), the mean and variance converge.The second stage (blanker on) of the process repeats the mean/variance computationwith the blanker turned on to reduce the effect of pulsed RFI.
β2 controls the sensitivity of detection, and can also influence the amount of blanked
data.
β2 and a performance of pulse detection process
To investigate choice of β2, data from LISA channel 7 was examined to determine
the percent of samples exceeding a specified threshold determined by β. Because
the mean measured noise power can be expected to vary significantly throughout a
cross-US flight, the mean and variance estimates used to determine the threshold are
computed by the running average described in Section 2.1. This running average
used Nblank = 2048 and Nwait = 0 in removing pulse contributions from the mean
and variance computations, as described in Section 2.2. These Nblank and Nwait
27
result in fairly large blanking region; 1024 samples (representing 51.2 µS of data)
before and after detection point would be removed. This choice seems overestimate
to ensure that all pulses would be completely suppressed in blanking process so that
the mean/variance computation could be accurately determined.
Figure 2.13 illustrates the percent of samples out of the 145 16K captures that
exceed the threshold specified by β2 on the horizontal axis. The dashed line included
in the Figure represents the percent of exponentially distributed noise that would
exceed the same threshold. An exponential, rather than Gaussian, noise distribution
is more appropriate here due to the very short integration time of the incoming
power for this system. In this case, the presence of RFI in the dataset causes the
percent blanked to exceed that of the simulated noise for large β2 values; however
as β2 is reduced, the two curves become more identical because the detection of
noise dominates both processes. In this case, the turning point with the LISA data
appears to be in the range β2 ≈ 30 to 40. Further simulations with other LISA
channels showed β2 values ranging from 40 to 90 to be reasonable; smaller β2 values
clearly led to excessive blanking as should be expected.
Nblank/Nwait and a performance of pulse blanking process
A simulation investigating the effect of Nblank has also been performed. In this
study, two threshold levels are defined: a higher “detection threshold”(β2=90) used
for pulse detection as usual while Nblank is varied. Nwait was kept at “0” at this step.
The output of the blanker is then examined to determine the number of samples
remaining that exceed a lower “reference threshold”(β2=30 using the same mean and
variance computations as those for the detection threshold). This quantity is labeled
Pout in what follows. The total number of samples exceeding the reference threshold
28
10 20 30 40 50 60 70 80 900
0.5
1
1.5
2
2.5
3
3.5
4
4.5
β2
Pu
lse
s d
ete
cte
d (
%)
DataEstimation
Figure 2.13: Percentage of samples exceeding threshold for given β2. All 145 capturesof LISA’s channel 7 data are used.
in the original data is also determined, and labeled Pin. The ratio (Pin − Pout)/Pin
then provides information on the effectiveness of the higher threshold blanker with
a given Nblank at removing lower level pulse contributions. Figure 2.14 illustrates
these quantities for a single LISA capture. Note in some cases, samples exceeding
the higher threshold are not blanked in this dataset; this is because such “trigger”
samples may be missed by the subsampled detector if they are less than 4 samples
long.
Figure 2.15 plots (Pin − Pout)/Pin as a percentage (solid curve) from the entire
channel 7 dataset, with Nblank ≥ 1536 which clearly satisfies equation (2.6). The curve
shows only a modest variation with Nblank, and an Nblank value in the range 1536 to
2048 appears appropriate in this case. This can also be interpreted as indicating that
pulsed interference longer than 76.8µsec (i.e. 1536 samples) is not significant in this
29
data. Of course, this Nblank parameter will vary for RFI environments dominated
by different types of sources, but the cross-US flight considered here should be fairly
representative of other datasets. Results for smaller Nblank values may show more
sensitivity in Figure 2.15, but are not explored here due to the limitation of equation
(2.6). Data for smaller Nblank values could be generated by varying the FIFO LENGTH
and Nwait parameters, and will be explored in future work.
The saturation of the solid curve at a maximum value of approximately 80%
is caused by the presence of spurious internally generated interference in the LISA
dataset [7]. The dashed curve is computed by redefining Pin and Pout to included
only consecutive sets of 2 or more samples that exceed the reference threshold. In
this case, a removal of approximately 98% of the “low-level” pulse components is
achieved.
0 100 200 300 400 500 600 700 800 9005
10
15
20
25
Po
we
r (d
B)
Time (µs)
InputOutput
Detection Threshold (β2=90)
Reference Threshold (β2=30)
Figure 2.14: Detection Threshold (for detecting pulses) and Reference Threshold (forestimating number of pulses). Nwait=0, Nblank=1536 are used in the APB outputresults shown
30
2048 2560 3072 3584 409670
80
90
100
Bla
nke
d S
am
ple
s (
%)
Nblank
All pulses exceeding β2=30 threshold
Ignore pulses shorter than 100 ns long
Figure 2.15: (Pin − Pout)/Pin × 100%, where Pin and Pout are the number of sam-ples exceeding the “reference threshold” at input and output of the APB algorithm,respectively.
The possibility of increasing Nwait in order to reduce unnecessary blanking in
the pre-pulse region was investigated. The percentage of blanked samples for Nblank
less than 1024 samples was examined. Blanked samples evaluated from channel 13
data are shown in Figure 2.16. Recall that upon detection, the blanking region
begins at 1024 samples before detection point. Nblank samples after the beginning
of the blanking region would be set to zero. In Figure 2.16, it is clearly seen that
even Nblank of 768 samples is not sufficiency large to blank all corrupted sample.
Consequently, setting Nwait to 768 and Nblank to at least 597 (eq. 2.6) would be
sufficient to successfully remove pulsed RFI in this frequency range.
Figure 2.17 shows the percentage of blanked samples of channels 7–12. Similar
trends can be observed. However, unlike channel 13, channel 7 seems to contain much
longer pulses; the blanker does remove a portion of the pulse even with Nblank of 64.
Due to the somewhat small size of the FIFO (1024 length, representing 51.2 µsec of
LISA data), immediate blanking of the FIFO is preferable. For this reason, Nwait is
31
256 512 768 10240
20
40
60
Bla
nked
Sam
ples
(%
)
Nblank
Channel 13
Figure 2.16: Percentage of blanked samples ((Pin − Pout)/Pin × 100%) of LISA’schannel 13 data. The pulses shorter than 100 ns were ignored in this computation.
maintained to “0” to ensure that even such large pulses can still be removed. Future
work can explore a preferred value of Nwait for larger FIFO LENGTH parameter.
2.4.3 Output χ2 test
Now that means for choosing the APB parameters have been established, it is
of interest to quantify the quality of the output data. Because thermal noise power
should approach a Gaussian distribution when integrated sufficiently, the χ2 test
against a Gaussian distribution [8] can be utilized to evaluate if the output of the
blanker satisfies this expectation. However, the entire dataset for a specific channel
cannot be applied in this test, as the mean noise power in a channel will vary as differ-
ent locations are observed. The test was instead performed using sets of consecutive
5 16K captures, all of which were measured within 5 seconds. The χ2 statistic using
4 degrees of freedom was computed using data power integrated over 128 samples in
order to approach the Gaussian distribution.
32
256 512 768 10240
50
100
Bla
nked
Sam
ples
(%
)
Nblank
Channel 7
256 512 768 10240
20
40
60
Bla
nked
Sam
ples
(%
)
Nblank
Channel 8
256 512 768 10240
10
20
30
Bla
nked
Sam
ples
(%
)
Nblank
Channel 9
256 512 768 10240
10
20
30
Bla
nked
Sam
ples
(%
)
Nblank
Channel 10
256 512 768 10240
20
40
60
Bla
nked
Sam
ples
(%
)
Nblank
Channel 11
256 512 768 10240
20
40
60
Bla
nked
Sam
ples
(%
)
Nblank
Channel 12
Figure 2.17: Percentage of blanked samples of LISA’s channels 7–12.
33
Original data from one set of LISA channel 7 data is illustrated in the upper plot of
Figure 2.18 before and after the APB algorithm is applied with β2 = 40 and Nblank =
2048. The presence of interference in the original dataset results in a high χ2 value
of 262.53, indicating that the data are not likely to be from a Gaussian distribution.
The lower plot illustrates the χ2 statistic of the non-blanked data after blanking
with β2 = 40 versus Nblank, and shows a greatly reduced value compared to the pre-
blanking case. Critical values based on α = 1% and 10% (the probability of incorrectly
classifying a true Gaussian distribution as non-Gaussian) are also illustrated in the
figure. Clearly for this example, the APB output data is much more Gaussian than
the input data, particularly for Nblank exceeding 1366. The poor performance for
Nblank = 1024 is not surprising, since this case does not satisfy equation (2.6), allowing
the possibility that some detected samples remain unblanked.
Simulations from other LISA data subsets show similar results, with a few excep-
tions. In particular, LISA channel 6 (1310 MHz- 1330 MHz) sometimes contains very
strong interference from multiple aviation radars, and a large value of χ2 remains
even after blanking with Nblank up to 4096. The limitation of the fixed FIFO LENGTH
parameter is an issue here, and future work will examine if increasing this parameter
can improve these problematic datasets. However it should certainly be expected that
there are cases with exceptional RFI corruption for which the APB algorithm cannot
retrieve the original noise power.
34
0 20 40 60 80 1000
10
20
30
40
Pow
er
(dB
)
Sample Index ( ×1000)
Input (χ2 = 262.53)
Output (Nblank
= 2048)
1024 2048 3072 40960
5
10
15
20
χ2
Nblank
Critical Value(α=1%)
Critical Value(α=10%)
5 × 16K-sample
Figure 2.18: Top: Noise power from 5 captures sucessively taken in the vicinity ofWheeling, IL (approx. 39.87◦ N lat, 80.67◦ W long) at 22,000 ft. LISA is tuned tochannel 7 (1324 - 1344 MHz). Several examples of pulsed RFI are observed. Bottom:χ2 of APB output as Nblank is varied (β2=40, Nwait=0).
35
2.4.4 Effect of blanking on integrated spectra
The ideal pulse-blanking algorithm would remove only RFI information, without
changing properties (particularly the mean power level) of the remaining noise infor-
mation. One questionable issue of the APB algorithm is the impact of forcing data to
zero when pulses are detected. This introduces discontinuities into the signal which
may lead to undesired effects on the final output, as well as calibration uncertainties.
Note after the APB operation an FFT is performed in the interference suppressing
radiometer of Section 1.1; clearly the impact of blanked samples on the FFT output
should be investigated.
To examine these effects, APB outputs were processed through FFT and integra-
tion operations. Each 16K LISA capture after blanking was first separated into 32
512 sample “frames” (i.e. a 512 point FFT operation is used). Prior to the FFT, each
frame can be categorized as either BLANK (contains no non-zero samples), NO BLANK
(contains no blanked samples), or PARTIAL BLANK (some samples are blanked), as
shown in Figure 2.19. The FFT is performed on each frame, the power computed in
each FFT bin, and all results in each FFT bin are averaged.
It is clear that the only effect of the BLANK category is to decrease the noise power
level of the final average. It is trivial to correct for this effect simply by counting
the total number of frames and the number of BLANK frames. However, the effect of
the PARTIAL BLANK frames, which contain discontinuities, is more complex. An FFT
operation on such a frame clearly will produce a distorted spectrum and a reduced
noise power level, with the degree of distortion and power reduction related to the
number of blanked samples within the frame. Clearly, narrow band noise sources may
experience some distortion in this process; however, noise sources with bandwidths
36
200 250 300 350 4000
10
20
30
40
Po
we
r (d
B)
Time (µs)
InputOutput
BLANK PARTIAL BLANK
NO BLANK
Figure 2.19: Definition of BLANK, PARTIAL BLANK and NO BLANK frames
larger than a few MHz, which are the subject of this work, are not considerably
affected. An example PARTIAL BLANK spectrum is compared to the corresponding
NO BLANK average spectrum in Figure 2.20. The reduction in power level is clearly
visible, along with a moderate distortion in the overall shape of the spectrum. Note
the large power levels observed near the center of this spectrum are due to an internal
DC component of the measured power, not due to external interference.
Various means for coping with the PARTIAL BLANK issue can be conceived. A
simple strategy (called method 1) is to eliminate such frames from the averaging
operation; however this approach may also eliminate a large fraction of the incoming
data in high RFI environments. A second approach to retain these frames while
correcting for the power level reduction is called “method 2: instantaneous scaling.”
By Parseval’s theorem, the effect of blanking on total average power of the frame can
be corrected simply by increasing the power of the computed spectrum by N/Nrem,
where Nrem is the number of non-blanked samples in the frame. This correction is
applied to the power level of each frame before including the frame in the average
37
computation. A final approach to is to included all (unscaled) frames in the spectral
average operation, and to maintain a separate count of the total number of non-
blanked time-domain samples included in the average, labeled Ntot, 6=0. Only the final
average power is scaled by Ntot/Ntot, 6=0, where Ntot is the total number of time domain
samples that make up the average operation. This approach is termed “method 3:
slow scaling”.
The upper plot of Figure 2.21 compares average spectra from a single Channel 6
LISA 16K capture before any blanking operations with the averages obtained from
methods 1 and 2. APB parameters β2 = 90 and Nblank = 2048 were used in the
APB algorithm. Both blanking methods are seen to be highly effective in removing
contributions from a pulsed interferer centered at approximately 1315 MHz in this
example. Results from Methods 1 and 2 are also observed to be very similar, demon-
strating that the instantaneous scaling approach is correcting for the reduced noise
power level due to blanking effects. To highlight the differences between methods 1
and 2, the lower plot of the figure illustrates the difference (subtracted decibel values)
between the method 2 (instanteneous scaling) and 1 (NO BLANK only) average spectra.
Differences are generally within 1.5 dB in all cases, and appear noiselike, indicating
that further averaging would likely make these differences less significant.
The difference between averaged spectra for methods 3 (slow scaled) and 1 (NO
BLANK only) is also illustrated in the lower plot; errors from method 3 are observed
to be somewhat smaller than those from method 2 on average. Clearly method 3 is
a simpler operation than that of method 2 (favorable for hardware implementation)
since corrections are required at a much slower rate. In addition, method 3 should
be preferable to method 2 because method 2 allows PARTIAL BLANK frames with a
38
great deal of blanking to be weighted equally in terms of the power averaging compu-
tation. However, these frames also have the largest degree of spectral distortion, so
reducing their weight should be advantageous. Figure 2.22 illustrates the mean error
for methods 2 and 3 (compared to method 1) for LISA’s channels 7 through 13 when
averaging results over the entire dataset and the entire frequency spectrum. The
mean error is of interest because it indicates the degree to which the average power
level is not being corrected properly. Results clearly show the method 3 error (slow
scaling) generally to be smaller than that of method 2. A hardware implementation
of method 3 is described in Section 4.3
1310 1312 1314 1316 1318 1320 1322 1324 1326 1328 13300
10
20
30
40
Po
we
r (d
B)
Frequency (MHz)
NO BLANK (Averaged)
PARTIAL BLANK (Single)
Figure 2.20: Frequency spectrum of an example PARTIAL BLANK frame, compared tothe entire NO BLANK average for the corresponding capture
39
1310 1312 1314 1316 1318 1320 1322 1324 1326 1328 133015
20
25
30
35
40
Po
we
r (d
B)
Frequency (MHz)
1310 1312 1314 1316 1318 1320 1322 1324 1326 1328 13301
0. 5
0
0.5
1
1.5
Po
we
r (d
B)
Frequency (MHz)
OUTPUT(Inst. Scaled) - NO BLANK
OUTPUT(Slow Scaled)- NO BLANK
Input
NO BLANK
Output(Inst. scaled)
Figure 2.21: Impact of PARTIAL BLANK on the final output. Top: Method 2: Instan-taneous Scaled averages compared to method 1. Bottom: The error of methods 2(inst. scaled) and 3 (slow scaled)
40
7 8 9 10 11 12 130.1
0.05
0
0.05
0.1
Channel Index
Mean E
rror
(dB
) Slow ScalingInst. Scaling
Figure 2.22: Mean error in total power estimate from methods 2 and 3 in LISAchannels 7-13
2.4.5 Frequency domain blanking
Although the digital radiometer prototype does not implement RFI mitigation
strategies in hardware after the FFT operation at present, it is of interest to simulate
the expected performance of such approaches. The “channelization” of the FFT
should allow an improved signal-to-noise ratio in detecting pulsed interference within
a single FFT bin. A hardware blanking algorithm could conceivably operate on each
FFT bin in real time by using a strategy identical to that of the APB processor. Such
an approach would allow lower level, rapidly pulsed RFI to be removed if missed by
the original APB.
This algorithm was simulated using the LISA data of 145 captures in channel 6.
After passing each capture through the time-domain APB algorithm with β2 = 40,
and Nblank = 4028, each 16K capture was split into 32 512-point frames. An FFT
operation was then applied to each frame, resulting in a total of 32 × 145 = 4640
temporal samples for each FFT bin. For each bin, a second APB algorithm with
β2 = 90, Nblank = 4, and FIFO LENGTH − Nwait = 2 was then applied to these 4640
samples, with the mean and variance computed by the averaging filter process used
41
1310 1315 1320 1325 1330
10
20
30
40
50
Frequency (MHz)
Po
we
r (d
B)
Input (Max Hold)
Input (Average)
1310 1315 1320 1325 1330
10
20
30
40
50
Frequency (MHz)
Po
we
r (d
B)
Output (Max Hold)
Output (Average)
Figure 2.23: Frequency-domain APB algorithm results.
in the original APB algorithm. Figure 2.23 illustrates the average spectra before and
after the frequency domain blanking operation, and shows a slight change in results
near the center of the spectrum. The effect of the blanker is more obvious in the
“max-hold” spectra also illustrated in the plot; “max-hold” refers to the maximum
value of the 4640 temporal samples. Clearly a significant degree of RFI is included
in this dataset near the center frequency 1320 MHz; the frequency domain blanking
operation reduces this interference so that corruption of the average spectrum is less
significant.
2.4.6 χ2-filter: alternative pulse blanking technique
An alternative approach, “χ2-blanking” can also be considered if blanking at a
slower temporal rate is deemed acceptable. In this method, the χ2-test is performed
on data from each bin. If the χ2-value exceeds a specified critical value, samples
in the dataset exceeding a power threshold are removed and χ2 then re-evaluated.
42
This iteration is repeated until the distribution satisfies the χ2-test for a specified
critical value. The requirement for a slower temporal rate here is due to the com-
plexity and iterative nature of this algorithm, which is not suited for integration in
hardware. However the algorithm could be applied to already-integrated data as a
post-processing step in software.
2.4.7 Conclusions
Results of the study show the APB approach generally to be effective in reducing
corruption from temporally localized RFI. The simulations performed on the LISA
dataset, while not completely general, arguably should be representative of a wide
range of RFI environments. For these simulations, use of β2 values ranging from 40
to 90 appears to effective, along with Nblank > 1536 (≈ 76.8µsec blanking window)
and Nwait = 0 (immediate blanking). Use of “slow-scaling” of the output power after
blanking was found preferable for correcting errors in the estimate of mean power due
to blanking, and averaged spectra after blanking showed only a modest distortion of
the underlying noise power. Further studies will continue to explore a preferred value
of Nwait for larger FIFO LENGTH parameter, as well as studies with other data sets. In
particular, an airborne version of the digital receiver prototype has been developed
for RFI observations at C-band [9]. Results from flights of this system will provide a
substantial set of RFI observations for further evaluating APB performance in varying
RFI environments.
43
CHAPTER 3
Sky observation with digital radiometer
3.1 Introduction
The previous chapter describes the pulse blanking process implemented in a digital
radiometer. Clearly, the blanking process can reduce pulsed RFI without significant
error. This chapter mainly concerns deployment of the radiometer for long-term sky
observations to quantitatively verify the functionality of the system as well as the
performance of the blanking algorithm.
3.2 Experimental Setup
The details of the IIP digital back-end were described in Section 1.1. The antenna
used for this experiment is a 10ft-diameter parabolic dish antenna illuminated by a
dual polarized cylindrical feedhorn. A picture of this antenna and the schematic
diagram of an antenna front-end (AFE) is shown in Figure 3.1. The AFE system
is attached to the feedhorn, and includes a switch for measurement of four distinct
ports: internal terminator and noise generator sources as well as the antenna in
44
either horizontal or vertical polarization. The AFE also includes filtering and low-
noise amplification stages. Tests showed the antenna ports to be reasonably well-
matched over the band of interest, 1325-1425 MHz, and additional low-loss isolators
are included on the antenna ports to further reduce any reflections. All front-end
components are installed within a temperature-controlled and temperature-monitored
enclosure.
Because the temperature of the internal terminator is monitored, its noise power
can be estimated. The noise diode is used as a noise source with estimated excess-
noise-ratio (ENR) of 27.3 dB. The process to obtain the ENR of the noise diode is
described in the next section.
3.3 Noise diode calibration
The excess noise ratio (ENR) of the noise diode is estimated. The calibration
involves the relation between the noise temperature and the noise power of objects
at known physical temperature.
The noise power, P , emitted by the object having physical temperature, T , is
described by the relation.
P = kTB (3.1)
where P is the noise power (in watts), T is the physical temperature (in Kelvins) of
the object, B is the bandwidth (in Hertz) of observation, and k is the Boltzmann
constant = 1.3807× 10−23m2kgs−2K−1. The relation between noise power generated
45
� � � � � � � � � � � � � � � � � � � � � � �� � � � � � � � � � � � � �� � � � � �� � � ! � " � # $ % & � ' ( ) *+ , - . / � " 0 �� ! � + $ %1 2 3 4 2 5 6 7 8 5 2 3 9 : ; 7 9 5 2 < = 1 2 3 4 2 5 6 7 8 5 2 > 9 : 7 5 9 ? ? 2 <
@ " ( A B 0 � C D E EFigure 3.1: Top: 10ft-diameter parabolic reflector. Bottom: Schematic diagram ofantenna front-end (AFE)
46
by each reference load and corresponding physical temperature can be modeled as:
Ph = aTh + b (3.2)
Pn = aTn + b (3.3)
Pc = aTc + b (3.4)
where Ph, Pn, Pc is the noise power (in watts) corresponding to the noise temperature
(in Kelvins) Th, Tn, and Tc. Subscripts “h”, “n” and “c” refer to Hot Load, Noise
Diode, and Cold Load respectively. The coefficient a is proportional to the total gain
of the system while the coefficient “b” accounts for additive noise. The noise power
Ph, Pn, and Pc are measured. The temperature Th and Tc are known. Therefore the
calibration coefficients a, b and the noise temperature Tn of the noise diode can be
determined.
Excess Noise Ratio (ENR) is defined as
ENR = Tn−To
To(3.5)
where To is the room temperature (290 K is assumed). To accurately estimate ENR,
the noise temperature needs to be measured precisely. However, an error due to
the cable loss between switch ports and the loads is unavoidable. The following
experimental setup is designed to account for this error.
3.3.1 Experimental setup for noise-diode calibration
Figure 3.2 illustrates the experimental setup. A matched load “Term-1 (Hot)”
and the noise diode are temperature-controlled at 45 ◦C. The load “Term-2 (Cold)”
was emerged into liquid nitrogen (LN2) whose physical temperature is 77 Kelvin.
The cable connecting switch ports to Term-1 and the Noise Diode are fairly short
47
semi-rigid cables whose loss (L1 and L2) are relatively small and can be neglected.
The cable from Term-2 is 1 ft of semi-rigid type and 4.5 ft of RG-58. Tho total loss
is measured to be L3 + L4 = 0.17 + 1.31 = 1.48 dB. Extra attenuators Lx, are to
be inserted between the semi-rigid cable from the switch port and the RG-58 cable
connected to the Term-2 (cold load). The function of Lx attenuators is to provide a
known loss.
N o i s e D i o d eT e r m 1 ( H o t ) L 1L 2L 3T e r m 2( C o l d )L N 2
L xL 4
Figure 3.2: Experimental Setup for noise-diode calibration
Five Lx values were tested. For each Lx value, the ENR of the noise-diode is
estimated. Figure 3.3 shows the relation between Lx and estimated ENR. The total
loss of Term-2’s cable is L3 + L4 + Lx = 1.48dB + Lx. The plot in Figure 3.3 can
be used to estimate the case for Lx = −1.48dB so that its value on vertical axis
48
represents a correct ENR at the switch port. Following this procedure yielded an
ENR of ∼ 27.3 dB, averaged over 100 MHz bandwidth (1325–1425 MHz)
0 1 2 3 1028
29
30
31
32
33
34
35
Attenuator Lx (dB)
Est
imat
ed E
NR
of N
oise
dio
de (
dB)
Figure 3.3: Estimated ENR vs Lx pad
3.4 Experiments
Although previous observations of a water pool [3] have indicated successful op-
eration of the APB system, difficulties in calibrating these measurements resulted
due to the large calibration target sizes required for far-field observation of ground
49
targets. An alternate experiment was initiated involving sky observations, since it
is expected that the sky provides a slowly varying brightness comprised of known
cosmic background, atmospheric, and astronomical source contributions [10, 11]. If
a high degree of stability of the LISR system can be demonstrated, sky observations
over long time periods should show only a slow evolution as various astronomical
sources enter the antenna pattern, while more rapid brightness variations would indi-
cate RFI effects. Long term sky observations also allow the possibility of calibration,
given that the brightnesses of astronomical targets are reasonably well known. The
sun and moon also provide opportunities for use as calibration targets. Overall the
goal of the campaign is to demonstrate reduction of RFI effects, including calibrated
information on the number of Kelvins of RFI reduction achieved.
At present, calibration using astronomical sources has yet to be achieved. For
this reason, data will be internally calibrated based on the known brightnesses of
the internal noise diode and terminator standards. Consequently, calibrated data
still include antenna and/or other front-end loss effects prior to the front-end switch.
Results are presented in terms of sky brightness temperature integrated over 5.3
seconds, so that high spectral resolution information (100 kHz channels) is available
in the band 1325-1425 MHz.
3.5 Results
Figure 3.4 illustrates the obtained noise temperature of the sky over 6 hours of
observations (00:00 – 6.00am, local time). The top plot shows the time history of
noise temperature over 100 MHz bandwidth. The color scale represents 50 Kelvin to
190 Kelvin, so that various types of RFI can be easily observed; results staying within
50
these limits over several hours indicate a highly stable system. Strong interferers are
observed near 1330 MHz, 1400 MHz, 1403 MHz, and other frequencies. The 1330 MHz
source is known to be a local air-traffic control radar, and generates the strong pulsed
RFI observed. The bottom plot in Figure 3.4 shows the average noise temperature
versus time. A 3 Kelvin oscillation due to the radar at 1330 MHz can be noticed from
this plot. The increase in power observed seen at 1:30am corresponds to observation of
the galactic plane. The average noise temperature of at ∼107 Kelvin is contributed
from several factors including component loss and ground contribution, which can
not be eliminated for internal calibration. The narrow-band continuous RFI is also
contribute to this measured value. The latter can be suppressed by cross-frequency
blanking technique described in Section 3.6.2.
Figure 3.5 provides similar plots from the same observation period, but results are
illustrated with the blanker (APB) turned on. A dramatic reduction in the 1330 MHz
source is observed, as should be expected given the pulsed nature of radar emissions.
Other non-pulsed RFI sources are not significantly affected. The average noise power
plot also confirms that the 3-Kelvin oscillation seen in the previous plot is solely
contributed by the radar at 1330 MHz.
3.6 Post-processing mitigation technique
The APB is clearly useful for detecting and removing pulsed RFI in real-time.
However, non-pulsed RFI is of interest as well. The RFI mitigation in this stage are
done in post-processing. Though the real-time mitigation is possible, given the com-
plexity of the algorithm to be implemented, post-processing by software is favorable
at present.
51
Figure 3.4: APB-OFF: Top–Spectrogram, color scale represents sky brightness tem-perature in Kelvin. Bottom–Average brightness temperature VS Time. Without RFImitigation some of data are corrupted. Pulsed RFI at 1330 MHz is expected andresponsible for 3-Kelvin oscillation in the average temperature plot.
52
T e m p o r a l n o n p u l s e d R F I
Figure 3.5: APB-ON: Spectrogram and average noise temperature plot observed withthe blanker turned on.
53
3.6.1 Cross-time blanking
In Figure 3.5, the contribution from temporal non-pulsed RFI can be seen at
1380 MHz. A strategy identical to APB – termed “Cross-time blanking” – can be
applied on time history of each frequency channel. Each of these time histories will be
separated into smaller time-frames and the APB will be applied on each of these time-
frames. Smaller time-frames would preserve slowly varying information, for example
the contribution from the galactic plane centered at 1:30 am. Using larger time-
frames would treat this slow-varying noise power contribution as RFI and remove
them. Clearly, the optimal length of these frames needs to be studied. For this
simulation, frame length of 10 minutes was used, given assumption that there is no
RFI with such a slow varying power.
Figure 3.6 shows the output of cross-time blanking process. The RFI at 0:30 am
and 5:00 am are clearly suppressed.
Figure 3.6: APB-ON with cross-time blanking applied
54
Frequency (MHz) Error (Kelvin) Frequency (MHz) Error (Kelvin)1400 9.59 1336 0.191403 5.30 1353 0.181333 0.49 1347 0.101350 0.47
Table 3.1: Contribution of narrow-band RFI to the average noise power (ordered bycontributed error)
3.6.2 Cross-frequency blanking
Though the average brightness plot in Figure 3.6 is considerably improved com-
pared to the data without any RFI mitigation applied (Figure 3.4), the contribu-
tion of narrow-band RFI can not be neglected. In fact, some of these interference
greatly contribute to the average noise power. Table 3.1 tabulates the contributions
of narrow-band RFI at some specific channels.
Cross-frequency blanking involves comparisons of results in different frequency
bins, in order to detect and remove more time continuous RFI sources. Again the
long time scales expected for sources in this process did not motivate immediate
consideration for implementation in hardware. A simple software based algorithm
has been developed. Three different approaches to detect this type of RFI have been
proposed. The first method (Xfreq-I) uses a simple threshold based on average
noise temperature over the 100 MHz band (Figure 3.7). However, without external
calibration, the instrument gain pattern over frequency limits the sensitivity of this
algorithm. As a result, weak RFI can not be detected.
55
1340 1360 1380 1400 1420
80
100
120
140
160
180
200
Noi
se T
empe
ratu
re (
Kel
vin)
Frequency (MHz)
Data7σ−threshold3σ−thresholdRFI
Figure 3.7: Xfreq-I cross-frequency RFI detection algorithm. A simple threshold isderived from the average noise temperature over the 100 MHz spectrum. Th possiblelowest threshold (dashed line) is limited by the gain pattern of the instrument.
56
An alternative approach (Xfreq-II) is to utilize thresholding the derivative of
the data in frequency. This method is expected to remove the gain pattern ver-
sus frequency (Figure 3.8(a)). This is important because if such an algorithm were
implemented in hardware, it is unlikely that calibrated data would be utilized in
the algorithm. Xfreq-II method provides a means to improve detection sensitivity
comparing to the first approach; however, any ripple in the gain pattern also intro-
duces non-negligible derivative. This becomes the limitation of the derivative-based
thresholding approach.
1320 1340 1360 1380 1400 1420 14400
10
20
30
40
50
60
70
80
Der
ivat
ive
of N
oise
Tem
pera
ture
Frequency (MHz)
Data1σ−thresholdRFI
(a) Derivative of the data in frequency
1340 1360 1380 1400 1420
80
100
120
140
160
180
200
Noi
se T
empe
ratu
re (
Kel
vin)
Frequency (MHz)
DataRFI
(b) Detected RFI (marked with ’×’)
Figure 3.8: Xfreq-II cross-frquency RFI detection algorithm. The threshold is de-rived from the derivative of the data. This allows weak RFI to be detected.
Though not hardware-friendly, another algorithm designed to be insensitive to the
instrument gain pattern versus frequency has been investigated. For this approach
(Xfreq-III), the gain pattern is first estimated and then used to define a frequency
varying threshold. The gain pattern is estimated by interpolating local minima of the
57
frequency response curve as illustrated in Figure 3.9(a). The algorithm developed was
also enhanced to incorporate information from the cross-time post-FFT algorithm in
order to improve detection of corrupted FFT bins.
1330 1331 1332 1333 1334 133570
80
90
100
110
120
130
140
150
Frequency (MHz)
Noi
se T
empe
ratu
re (
Kel
vin)
DataLocal minimaEstimated uncalibrated gain pattern
(a) Gain pattern estimation
1340 1360 1380 1400 1420
80
100
120
140
160
180
200
Noi
se T
empe
ratu
re (
Kel
vin)
Frequency (MHz)
DataThresholdRFI
(b) Detected RFI (marked with ’×’)
Figure 3.9: Xfreq-III cross-frequency RFI detection algorithm. Gain pattern is firstestimated by interpolating local minima of the frequency spectrum (Left panel). Thisallows the contribution of uncalibrated instrument gain pattern to be considered inthe detection process.
Figure 3.10 shows the final results after incorporating all RFI mitigation tech-
niques: real-time APB, cross-time, and cross-frequency blanking. Clearly, the cross-
frequency blanking helps to remove biasing due to narrow-band interference, and
results in several Kelvins improvement in the average noise temperature estimation.
Note that remaining variation across the frequency is a result of the gain pattern of
the instrument before the calibration point. Calibration using external sources as
reference loads are required to eliminate this effect.
58
Calibration corrections for cross-frequency blanking again are simple to imple-
ment, involving simply scaling by the number of frequency channels retained relative
to the original number of channels. Note at L-band that high spectral resolution is
useful in removing contributions from hydrogen line emissions at 1413 MHz, as these
narrowband contributions can influence the accuracy of sea salinity retrievals from
space [11].
Figure 3.10: APB-ON with cross-time and cross-frequency blanking applied.
3.6.3 Temperature control and stability of the system
The average noise temperature plot in Figure 3.10 demonstrates very stable system
for at least over 6 hrs observation period. However, small variation, ∼ 1 Kelvin, can
be noticed. This error is a result of temperature drift of the analog down-converter
back-end. The additional hardware required for temperature monitoring/controlling
are disscussed in Section 4.4. Installation, calibration and possible improvement are
also discussed in the Sectin 4.4 as well.
59
3.6.4 Narrow-band interferences at 1400 and 1403 MHz
Among different narrow-band interference observed from the sky measurement
(Table 3.1), the ones at 1400 and 1403 MHz are of most interest. Both of them are
relatively strong, narrow, fairly constant over time and appear inside the protected
spectrum (1400-1427 MHz) Several brief experiments were done to investigate some
properties of these sources.
1400 MHz RFI
A simple survey was done over the roof-top of the ElectroScience laboratory.
The experimental setup consists of an L-band square horn antenna and a spectrum
analyzer. The antenna has been focused to different directions, approximately, North,
West, South, East and straight up to the zenith. Snapshots when the antenna was
pointing East and South are shown in Figure 3.11. No RFI at this frequency was
detected from pointing toward South as well as in other directions. This suggests the
1400 RFI could be partly contributed from external sources.
A possibility that this source is a harmonic of other signals was investigated.
The source at 700 MHz can be observed by spectrum analyzer located at ESL. The
multi-window tracking feature of the spectrum analyzer over an extended time period
reveals that the 1400 MHz source is not likely to be a second harmonic of this 700 MHz
source. The external source of RFI at this frequency remains unknown; however, it
was found that the radiation from radiometer back-end contributes to this RFI as well.
In fact, a tone at 700 MHz as well as 1400 MHz can be observed by the spectrum
analyzer. Figure 3.12 shows the frequency spectrum of 50 kHz span around the
center frequencies at 700 MHz and 1400 MHz. One feature of the spectrum analyzer
60
M a x .A v g .M i n .
(a) East
M a x .A v g .M i n .
(b) South
Figure 3.11: A survey of 1400 MHz RFI. The antenna was focused to different direc-tions. Strong RFI is found from the East while there was no other RFI received fromother directions.
allows the spectrum at two distinct frequencies to be observed simultaneously. An
antenna was pointing toward the radiometer back-end rack. With the radiometer
turned off, Figure 3.12(a) shows a source at 700 MHz while no 1400 MHz source
had been observed. With the radiometer turned on, Figure 3.12(b) illustrates strong
contribution at 700 MHz as well as its harmonic at 1400 MHz.
1403 MHz RFI
Though the major sources of these RFI are still undetermined, most of them
are likely to be from radiation from local equipment including the radiometer itself.
The survey at ESL’s roof does not detect RFI at this frequency from any direction.
This reduces the possibility of external contributions. However, the strong 1403 MHz
61
5 0 k H z 5 0 k H z(a) With radiometer turned off
5 0 k H z 5 0 k H z(b) With radiometer turned on
Figure 3.12: Multi-window plot show the spectrum centered around 700 MHz and1400 MHz measure at the same time. The antenna is directed into the radiometerback-end rack.
62
RFI are found to be radiated from the radiometer back-end and spectrum analyzer
themselves (Figure 3.13).
3.6.5 Conclusions
Current results indicate that the LISR system is providing stable sky observa-
tions, although a calibration using external targets will be required to reduce the
gain pattern versus frequency. The studies show the APB to be effective at reduc-
ing the influence of a local air traffic control radar, as well as the advantage of high
spectral resolution in removing narrowband temporal/continuous RFI sources. Both
of these suppression techniques are possible due to the use of a digital receiver, as
the rapid temporal processing and large number of FFT bins is only practical with
such a system, and the real-time processing achieved allows a manageable data rate
to be retained ultimately. Future works include long-term observations with exter-
nal calibration and improving the stability of the analog down-converter with more
sophisticated temperature control.
63
1 0 0 k H z(a) Antenna pointed toward spectrum analyzer(0◦ relative)
1 0 0 k H z(b) Pointed away (90◦ relative)
1 0 0 k H z(c) Pointed away (180◦ relative)
1 0 0 k H z(d) Pointed away (270◦ relative)
Figure 3.13: 1403 MHz radiation from spectrum analyzer. Observation at 1403 MHzwith a spectrum analyzer also found the RFI leaked from the instrument itself
64
CHAPTER 4
Hardware issues for implementation in FPGA Components
The first prototype of the L-band Interference Suppressing Radiometer-1 (LISR1)
was completed by late September 2002. This compilation of the radiometer included
only one ADC and therefore sampled only a 50 MHz bandwidth, but retained the
full DIF, APB, FFT, and SDP operations with a few exceptions. The FFT operated
at 14% duty cycle and APB processor has only one BTR implemented. A second
prototype, LISR2, was developed to remove the limitations of LISR1. A higher-
density series of FPGA components were used (the Altera “Stratix” line). This
allows 6 parallel FFT processors to be included so that 98% duty cycle computations
resulted; four BTRs are included in this version. A photo of LISR2 digital backend
is shown in Figure 4.1. Both these prototypes were completed prior to the efforts of
this thesis.
For both prototypes, the digital processing units (DIF, APB, FFT and SDP) are
programmed into 3 cascaded cards, each of which are controlled by a computer via
ethernet. Having digital processing units on separate cards makes implementation
of the scaling process more difficult. However instantaneous scaling could be im-
plemented in a single FPGA receiver if a higher-density FPGA were provided. As
mentioned in Section 2.4.4, instantaneous scaling at a high rate should be avoided.
65
Figure 4.1: LISR2 digital backend; the vertical cascade of three circuit boards nearthe left hand side contains the dual ADC sections (upper and lower boards) and thedigital channel combination and filtering (DIF) section (center board). The APBsection for removing temporal pulses is also implemented on the center board. Fol-lowing the vertical cascade to the right is the FFT processor, then the SDP section forpower computation and integration operations. Finally a “capture card” provides theinterface to the PC. Microcontrollers are also included on each card (the smaller at-tached circuit boards with ethernet cables) to enable PC setting of FPGA parametersthrough an ethernet interface.
Slow scaling, in contrast, simplifies hardware design and requirements. This process
requires information on the number of samples set to zero by the blanking stage,
which is carried over from the APB unit. Accordingly, a final prototype, LISR3, was
developed to resolve these communication issues. All the DSP units are now inte-
grated into a single FPGA. The large size of the program on this FPGA raises many
concerns on simulation and programming, including clock management and low-level
signal routing. Clock distribution for a Stratix FPGA is reviewed in Section 4.1, while
Section 4.2 deals with the signal-routing issue in detail.
66
4.1 Clock management for LISR3’s FPGA
Figure 4.2 illustrates the LISR3 digital back-end. The clock-generator (upper-left
corner of the figure) provides a 200 MHz clock for the ADC cards. The 10-bit data
sampled by the ADC device (Analog Device’s AD9410) are synchronized to the 100
MHz clock generated by the ADC itself. The ADC card also includes one PLD –
programmable logic device – to allow the AD9410 to be synchronized to an external
clock if needed. However, this function is not utilized in the LISR3 design, and this
PLD simply serves as a data and clock output-buffer. Even though registered through
the PLD, noticeable signal degradation can be observed in both the clock and data
lines (Figure 4.3). Figure 4.3 also points out that a distorted clock signal could affect
the duty cycle of the clock pulse seen by the following device. The measurement
was performed with the Stratix card attached (i.e. normal operating setup as in
Figure 4.2) excepted only DIF circuits were loaded into the Stratix processor. The
problem is expected to be even more critical if the entire processor (DIF, APB, FFT
and SDP), introducing much higher loading, is programmed.
To prevent transmission errors, including loss of synchronization and bit-errors,
the clock signals are regenerated and all data are registered immediately after inputs
of the Stratix FPGA.
The data from each ADC is first re-synchronized to its own clock, then syn-
chronized with the “master” clock derived from one of the two ADC cards. Since
the properties of both clocks are identical, the choice of master clock is decided
from the signal-routing and layout perspective (Section 4.2). The master clock is a
100-MHz ADC clock, regenerated by Stratix’s Enhanced-PLL (dedicated phase-lock
loop circuits available in Stratix devices) immediately after the input pin. Stratix’s
67
Figure 4.2: LISR3 digital backend; underneath a heat-sink on the center board isthe Stratix FPGA which contains the entire processor. The boards on the left andright of the Stratix board are the ADC cards. The system clock used to run theStratix processor is also derived from one of these ADC cards. Above the Stratixcard is a “capture card” providing a 256K-FIFO buffer and the interface to the PC.Only a single Microcontroller card (the smaller card attached to the bottom of theStratix board) is needed in this prototype. Except for the Stratix card, the otherparts of the hardware are the same as in LISR1 and LISR2. The clock-generator andpower-supply (top-left and bottom-right respectively) are also the same as in previousprototype.
68
0 10 20 30 40 50
0
2
4
Time (ns)
Vol
tage
(V
) Clock (Stratix)Clock (A/D Card)Min. HighMax. Low
0 10 20 30 40 50
LowHigh
LowHigh
Time (ns)
(a) Clock
0 10 20 30 40 50
0
2
4
Time (ns)
Vol
tage
(V
) Data (Stratix)Data (A/D Card)Min. HighMax. Low
0 10 20 30 40 50
LowHigh
LowHigh
Time (ns)
(b) Data
Figure 4.3: Clock and data signal at the output of ADC cards (dashed line) beforecrossing the SCSI-3 connector to the Stratix card (solid line). The minimum high(1.7 V) and maximum low (0.7 V) voltage level according to LVTTL/LVCMOS re-quirement are also shown as a reference. The lower plots demonstrate the expectedlogic level to be recognized by the Stratix FPGA according to these reference levels.
Enhanced-PLL is capable of driving up to 6 clock networks. For the LISR3 design,
only two of the PLL output are enabled, supporting two global clock networks. A
single clock network indeed suffices; however, in this design, the APB and Microcon-
troller interface are driven by separate clock networks for design convenience.
In order to synchronize the data to the master clock, a simple dual-clock FIFO is
utilized. Quartus R©II design software provides a tool to implement single-clock/dual-
clock FIFO’s. Figure 4.4(a) illustrates data and control signals configured for this
purpose. wrreq (write-request) and rdreq (read-request) provides a means to control
the FIFO writing and reading processes independently. wrclk (write-clock) and rdclk
(read-clock) allows the FIFO to be written and read from separate clock domains.
The state-machine in Figure 4.4(b) controls the initializing process of each dual-clock
69
FIFO to ensure that the FIFO is approximately half-full before the first reading cycle
is allowed. Figure 4.5 illustrates the whole clock regeneration scheme.d a t a [ 1 5 . . 0 ]w r r e qw r c l kr d c l kr d r e q q [ 1 5 . . 0 ](a) Configuration
0 01 01 1 w a i t 1 2 c l o c k � c y c l e sw a i t 4 c l o c k � c y c l e s0 0 � I n i t i a l i z a t i o n1 0 � A l l o w w r i t i n g ( w r e n a = 1 )1 1 � A l l o w w r i t i n g ( r d e n a = 1 )
(b) State machine control initialization process
Figure 4.4: Dual-clock FIFO.
4.2 Floorplan and Signal-Routing
Figure 4.6 shows the layout of the latest Stratix FPGA compilation. The yellow-
shaded area represents allocated logic elements and dedicated circuits. Even though
only half of the resources are occupied, the resource allocation and signal-routing
need to be carefully determined to meet timing requirement for operation at 100
MHz. Figure 4.7 shows the floorplan of the FPGA. The labeled area shows the
LogicLockTMregion, a set of constraints used by Quartus R©II design software in order
to lock down design into specified regions in the device’s floorplan. The figure also
70
d a t aw r r e qw r c l kr d c l kr d r e q qM a s t e r C l o c kP L LA D C � c a r d( U p p e r b a n d )1 0 + b i t d a t a1 0 0 M H z c l o c k D u a l � c l o c k F I F O ( U p p e r )
S y n c . d a t a ( U p p e r )d a t aw r r e qw r c l kr d c l kr d r e q qM a s t e r C l o c kP L L
A D C � c a r d( L o w e r b a n d )1 0 + b i t d a t a1 0 0 M H z c l o c k D u a l � c l o c k F I F O ( L o w e r )S y n c . d a t a ( L o w e r )
Figure 4.5: Clock regeneration and synchronization scheme for Stratix FPGA
shows the interfaces to the other parts of the system. Limited by the physical ori-
entation (Figure 4.2), the data and clock from ADC cards are fed from the top and
bottom sides of FPGA. The clock signal from the lower part of the chip is designated
as a master clock since it can access the global clock network more effectively. The
nearest PLL to the clock input from the upper part of the chip is much further away
(also far from the clock output pin) and would cause additional propagation delay in
the clock line.
The “Rabbit”’s Microcontroller is linked to the Stratix FPGA via I/O banks on
the left of the floorplan while I/O banks on the right are allocated for data and clock
outputs. Therefore, in order to minimize routing expenses, the data flow should be
from the upper, lower and left of the FPGA to the output pins on the right of the
71
Figure 4.6: Final layout of Stratix FPGA. Yellow-shaded area shows occupied re-sources.
72
D I F( L o w e r )
D I F( U p p e r )F F TD at aFIFO I nt egrat or
A / D c a r d ( U p p e r b a n d )
O ut put
A / D c a r d ( L o w e r b a n d )
C omput er C ont rolP ower C omput ati on O ut put R egi st ers
Figure 4.7: Floorplan of the Stratix FPGA. All marked/labeled areas are assigned asLogic-Lock region. During “fitting” process, the simulator will attempt to allocateresources according to these given criteria. The connections to the rest of the systemare also shown.
73
floorplan. The layout seems to follow this flow with the exception of FFT module.
The FFT is placed on the rightmost of the FPGA next to the output pins. The choice
of FFT location is bounded by availability of DSP blocks. Stratix provides several
dedicated circuitry – memory blocks, DSP blocks and dedicated clocks – which are
assigned to specific locations. Though it is not required to confine each module into
the same block as those dedicated circuits, it is preferable to minimize the routing
resources needed.
Figure 4.8 shows the location of two columns of DSP blocks, used extensively
by the FFT and APB as a built-in multiplier circuit. Each Stratix device has 14
DSP blocks, 7 blocks per column. One DSP block can implement up to either eight
full-precision 9 × 9-bit multipliers, four full-precision 18 × 18-bit multipliers, or one
full-precision 36 × 36-bit multiplier with add or subtract features. One FFT processor
requires 1 block of DSP, therefore one entire DSP column must be assigned to the
FFT processors. The FFT module also places heavy demands in other resources
(Figure 4.6); it is not possible to place this module close to the left column of the
DSP where non-negligible portion of those resources have to be allocated for the
Microcontroller interface module as well.
FFT placement eventually constrains the possible location assignment for the rest
of the circuits. APB also requires DSP blocks for multiplication processes; the APB
has to be placed around the other column of the DSP resources. DIFs must be
close to the input pins to minimize routing delays. The APB also requires numerous
M512-RAM blocks (Figure 4.9). Hence, the top-left and bottom-left regions of the
floorplan become the perfect location for them. Finally, the location o the SDP (power
computation and integrator) as well as the output registers are fixed, forcing the rest
74
P L L
M 4 K R A M
D S PFigure 4.8: DSPs and M4K-RAM locations in Stratix FPGA. The location of PLLused to generate the “master clock” for the entire chip is also shown.
75
M 5 1 2 R A M
M � R A M ( n o t u s e d )Figure 4.9: M512-RAM and M-RAM locations in Stratix FPGA
76
of the data flow to reside along the center of the chip. The final design does not
utilize any M-RAM blocks. An alternative FPGA model that has less or no M-RAM
blocks can be proposed for future work to trade the unused portion of floorplan for
other useful resources.
77
4.3 Implementation of slow scaling process
The scaling process is accomplished by a combination of hardware and software
operations. The FPGA counts the number of blanked samples, then attaches this
number to the data stream. The controlling software extracts that information from
the received data block, computes the scaling factor and then finally corrects the
data.
The most important issue of the blanking counter design is to ensure that the
number attached to each data block represents the correct number of zeroed samples
within that block. Figure 4.10 shows the block diagram of the counter module and
associated units. The “Delay Register” is inserted between the blanking counter and
the 1-bit blank signal generated by the APB. The number of clock cycles delayed by
this register must be exactly the same as the total delay for data to go through the
FFT, power computation, and integration processes. Circuit analysis is required to
provide the number of delay cycles needed.
S D PT oC a p t u r e c a r dA P B
F F TB l a n k i n gC o u n t e r
P o w e rC o m p u t a t i o n" B l a n k " o r " D o n o t b l a n k "D e t e c t o rB l a n k e rF r o mD I F s D e l a yR e g i s t e r s I n t e g r a t o r
Figure 4.10: Detailed block diagram of APB and SDP.
78
Figure 4.11 illustrates test results. The plot show average power of integrated
spectra in blanker-off mode (solid line). The pulsing behavior is known to be pulsed-
RFI from the local radar. The blanker-on mode removes these pulses but also drops
the power level down several dB. The blanker-on with scaling process corrects for this
error and moves the power level back to the correct value, the average noise-power
without RFI.
B l a n k e r O F FB l a n e r O N ( n o s c a l i n g )B l a n k e r O N ( w i t h s c a l i n g )4 5 . 4 84 5 . 4 64 5 . 4 44 5 . 4 24 5 . 4 04 4 . 3 84 4 . 3 64 4 . 3 44 4 . 3 24 4 . 3 0 1 0 2 4 2 0 4 8 3 0 7 2 4 0 9 6Figure 4.11: Slow-scaling test results
4.4 Thermal Control for Analog Down-converter
A thermal control and measurement system was developed for the front-end unit,
in order to enhance system gain stability as well as to provide accurate control of
internal calibration load noise powers. The down-converter unit was initially designed
to operate without any thermal control; however, the results from early phases of
79
sky observation have revealed some signs of system gain stability problems. Tests
confirmed that the thermal control system for the front-end performs adequately
at keeping the front-end average temperature stable to within approximately 0.2 K
over long periods of time. The performance of the digital back-end should be only
minimally affected by temperature drift. Hence, issue of gain stability of the analog
down-converter, and its need for thermal control have been reviewed.
Figure 4.12 shows the temperature measured at the enclosure of the down-converter.
The plot shows a strong correlation with ambient temperature as expected. Fig-
ure 4.13 plots the power observed at switch ports of the front-end for 17 hrs. System
gain variations, appear as slow fluctuations in power level that can be noticed in
every switch port. Finally, Figure 4.14 compares the power plot in Figure 4.13 and
temperature plot in Figure 4.12; a strong correlation can be noticed.
18:00 21:20 00:40 04:00 07:2020
25
30
35
Time (HH:MM)
Tem
pera
ture
(° C)
Down−ConverterAmbient
Figure 4.12: Temperature of down-converter compared to the ambient temperature.
80
20:42 23:23 02:03 04:44 07:2535
40
45
50
55
Pow
er (
dB)
Time (HH:MM)
Noise DiodeTerminatorAntenna (H−Pol)
Figure 4.13: Power at distict ports of the front-end switch: shown are Noise Diode,Terminator and Antenna (H-Pol). The front-end temperature is set to 45 ◦C
81
18:10 18:51 19:31 20:11 20:51 21:3236
37
38
Rel
. Pow
er @
ant
enna
por
t (dB
)
Time (HH:MM)18:10 18:51 19:31 20:11 20:51 21:32
25
30
35
Dow
n−C
onve
rter
Tem
p. (°
C)
Figure 4.14: Comparison of the power seen at antenna port and the tempearuture ofdown-converter.
82
To stabilize the down-converter inside temperature, a commercial thermal control
unit was purchased for use with a resistive-heater mounted on the enclosure’s inside
wall. The controller can be operated at 120 VAC. Its solid state relay output pro-
vides a current drive up to 15 amperes; the resistive-heater is chosen as a heating
component. The controller can be configured via the RS232 communication port of
computer. A simple software was also developed to monitor and record the tem-
perature during the measurement. The controller can be set for different modes of
operations, including on-off and PID control modes. On-off mode with adjustable hys-
teresis provides a simple solution for thermal control by switching on the output relay
if the input temperature is below the set point, or switching off the output otherwise.
In this mode, a “Dead-band control” parameter controls the minimum temperature
offset that can trigger the output switch. PID modes provide more control of the
heating process with the expense of algorithm complexity and a complicated calibra-
tion. Controller tuning parameters include “proportional bandwidth (P)”, “integral
gain (I)” and “derivative rate (D)” allowing the controller can be configured for P,
PI, PD or PID modes.
For the sky-observations in Chapter 3, the thermal control system was set in P
mode. Thermal insulators were added on every side of the enclosure, except the front,
to enhance the heating efficiency. The front panel of the enclosure is uncovered to
provide thermal ventilation if necessary (Figure 4.15(a)). The temperature set point
is at 45◦, a few degrees higher than the average temperature of the enclosure without
heating. The heater was fixed to one of the side wall of the enclosure (Figure 4.15(b)).
Two temperature probes were placed in the enclosure. The “Control” probe – used
to control heating process – was placed on the heater to ensure the most sensitive
83
D o w n � c o n v e r t e rT h e r m a l i n s u l a t o r
(a) Thermal insulation plan for a down-converter
(b) Top view of a down-converter enclosure: the lo-cation of control probe/heater (round marker) andmonitor probe (squared marker) are shown. Di-mension are in inches.
Figure 4.15: Thermal insulation plan and location of the temperature probes
84
control. A “Monitor” probe was located at the opposite corner of the enclosure
to provide rough information on temperature gradient on the heating surface. The
temperature history of down-converter for sky-observations is shown in Figure 4.16.
Figure 4.16(a) shows the success of thermal control process. Although Figure 4.16(b)
highlights the issue of temperature gradients, the results in Chapter 3 show adequate
performance of this temperature control system. Further improvement may includes
operating in PI mode to remove the temperature offset (from the set point at 45
◦C) and to further improve temperature stability. Reference [12] suggests a more
sophisticated temperature control technique that can be applied to this system as
well.
4.4.1 Conclusions
An integration of digital processing units into a single Stratix FPGA was com-
pleted. Several concerns need to be carefully considered, including clock distribution,
resources allocation, and signal-routing in physical level. Clock regeneration and re-
synchronization are required to ensure the quality of the master-clock signal, and to
synchronize the data stream from two ADCs. Routing delays must be minimal to
meet the timing requirement at 100 MHz operation. Blanking counter has been em-
bedded in FPGA to provide the number of blanked samples for slow-scaling process
(currently performed in software). An issue of thermal control has been addressed to
improve the system stability for long-term measurement. Current solution employs
PID controller operating in P mode. Thermal insulator was added to help improving
thermal stability. As a result, the temperature of down-converter enclosure appears
to be stable within 0.2 ◦C (at control point) for 6 hours of observation; however, the
85
0:04 1:32 3:01 4:29 5:5847.1
47.2
47.3
47.4
Time (HH:MM)
Tem
p. (°
C)
(a) “Control” probe
0:04 1:32 3:01 4:29 5:5843.5
44
44.5
Time (HH:MM)
Tem
p. (°
C)
(b) “Monitor” probe
Figure 4.16: Temperature of down-converter enclosure for sky-observation in Chap-ter 3. The temperature measured from “Control” probe is used as an input forcontrolling feed-back loop.
86
temperature gradients on the controlled enclosure can be noticed. The future works
will improve the thermal-control plan to reduce the temperature gradients. Further
system refinement is also of interest. The possible areas include integrating the digital
back-end unit onto a single circuit board to reduce signal degradation. Implemen-
tation of the entire slow-scaling process and cross-frequency blanking process in the
FPGA would be useful once higher-density FPGA is available.
87
CHAPTER 5
Conclusions
Performance of the APB algorithm has been studied with the simulation software.
With a few exceptions, the simulation software was designed to be identical to the
actual processes implemented in hardware. APB initialization stage is one of the
processes that cannot be identically realized in software. APB initialization requires
a large amount of data to be processed. However, the results show the simulating
process to have comparable performance comparing to the actual hardware. Results
of the study show the APB approach generally to be effective in reducing corruption
from temporally localized RFI. The simulations performed on the LISA dataset, while
not completely general, arguably should be representative of a wide range of RFI
environments. For these simulations, use of β2 values ranging from 40 to 90 appears
to be effective, along with Nblank > 1536 (≈ 76.8µsec blanking window). Use of “slow-
scaling” of the output power after blanking was found preferable for correcting errors
in the estimate of mean power due to blanking, and averaged spectra after blanking
showed only a modest distortion of the underlying noise power. APB-like algorithm
applied on a single FFT bin, later termed as the “cross-time” blanking technique, has
been studied with the same dataset.
88
Results from sky measurements indicate that the LISR system is providing stable
sky observations. The studies show the APB to be effective at reducing the influence
of a local air traffic control radar, as well as the advantage of high spectral resolution in
removing narrowband continuous RFI sources. Both of these suppression techniques
are possible due to the use of a digital receiver, as the rapid temporal processing and
large number of FFT bins is only practical with such a system, and the real-time
processing achieved allows a manageable data rate to be retained ultimately. Several
approaches for non-pulsed RFI blanking have been proposed. The advantages and
disadvantages of each method were highlighted. Future works will investigate the
performance and limitations of each approach in further detail.
LISR3, the new prototype that has all digital signal processing units embedded in
a single FPGA device, was completed. Issues of clock distribution and signal-routing
have been addressed to ensure that the timing-requirement of 100 MHz operation
is achieved. Temperature stability of the analog down-converter section has been
improved. The feasibility of incorporating the cross-time and cross-frequency blanking
techniques into a hardware will be investigated in future studies.
The major contribution of this thesis is to quantitatively verify the performance of
the APB technique from software simulations and actual measurements. The results
also show the gains that can be achieved through simple temporal and spectral RFI
removal techniques. It is certain that incorporation of digital receiver technologies into
passive remote sensing systems will increase in the future, as such technologies can
dramatically improve the capabilities of such systems in mitigating RFI. This study
provides the support that APB is a good candidate for such systems. In addition,
digital design also allows the size of prototype/product to be more compact. The
89
current prototype digital receiver combines all the RFI processing stages onto a single
FPGA. Market research also shows that rad-tolerant FPGA components of similar
size and performance are already available for space operations.
90
APPENDIX A
LISA’s frequency channels
Channel Frequency range (MHz) Channel Frequency range (MHz)1 1240 – 1260 8 1338 – 13582 1254 – 1274 9 1352 – 13723 1268 – 1288 10 1366 – 13864 1282 – 1302 11 1380 – 14005 1296 – 1316 12 1394 – 14146 1310 – 1330 13 1408 – 14287 1324 – 1344 14 1422 – 1442
Table A.1: Frequency range of LISA’s channel 1 to 14
91
BIBLIOGRAPHY
[1] S. W. Ellingson, “Characterization of some L-band signals visible at Arecibo,”tech. rep., The Ohio State University ElectroScience Laboratory, Febuary 2003.
[2] S. W. Ellingson and G. A. Hampson, “RFI and asynchronous pulse blankingin the 1230-1375 mhz band at Arecibo,” tech. rep., The Ohio State UniversityElectroScience Laboratory, Feb 2003.
[3] G. A. Hampson, J. T. Johnson, and S. W. Ellingson, “Design and demonstra-tion of an interference suppressing microwave radiometer,” in Proc. 2004 IEEEAerospace Conference, 2004.
[4] S. W. Ellingson and G. A. Hampson, “Mitigation of radar interference in L-bandradio astronomy,” ApJS, vol. 147 (1), pp. 167–176, 2003.
[5] S. W. Ellingson, G. A. Hampson, and J. T. Johnson, “Characterization of L-band RFI and implications for mitigation techniques,” in Proc. 2003 IGARSS,Toulouse, July 2003.
[6] S. W. Ellingson and J. T. Johnson, “Airborne RFI measurements over the Mid-Atlantic coast using LISA,” tech. rep., The Ohio State University ElectroScienceLaboratory, January 2003.
[7] J. T. Johnson and S. W. Ellingson, “Airborne RFI measurements using LISAduring transit to and in the wakasa bay campaign,” tech. rep., The Ohio StateUniversity ElectroScience Laboratory, July 2003.
[8] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, NumericalRecipes in C. New York: Cambridge University, 1992.
[9] J. T. Johnson, A. J. Gasiewski, G. A. Hampson, S. W. Ellingson, R. Krishna-machari, and M. Klein, “Airborne radio frequency interference studies at C-bandusing a digital receiver,” in Proc. 2004 IGARSS, vol. 3, pp. 1687–1690, Anchor-age, Alaska.
[10] J. Y. Delahaye, P. Gole, and P. Waldteufel, “Calibration error of L-band skylooking ground-based radiometers,” Radio Science, vol. 37, pp. 11/1–11/11, 2002.
92
[11] D. M. LeVine and S. Abraham, “Galactic noise and passive microwave remotesensing from space at L-band,” IEEE Trans. Geosc. Rem. Sens., vol. 42, pp. 119–129, 2004.
[12] R. Villarino, J. Fernandez, A. Camps, M. Vall-llossera, I. Corbella, N. Duffo, andF. Torres, “Design and test of the L-band AUtomatic RAdiometer (LAURA)temperature control,” in Proc. 2005 IGARSS, Seoul, Korea, 2005.
93