Radar Warning Receiver (RWR) Time-Coincident

Pulse Data Extraction and Processing

Wesley A. Gee
Senior Member, IEEE
Scientific Research Corporation
Atlanta, GA USA
[email protected] or [email protected]

Abstract—This paper will present an approach for video conversion interface circuit is the extended-range
implementing an airborne radar warning receiver (RWR) detector logarithmic video amplifier (ERDLVA).
pulse data extraction (PDE) and processing technique
capable of handling multiple time-coincident radio-frequency This work presents an approach for accurately
(RF) pulse input threat emitter signals received in a dense interpreting incoming time-coincident pulse signal data for
threat radar signal environment. The pulse data extraction highly sensitive RWR systems in dense threat emitter
technique is specifically developed to accommodate the RF to environments. The concepts described use some well-
video response of current commercial-off- the-shelf (COTS) known real-time signal processing techniques optimized for
or militarized COTS (MOTS) extended-range detector specific ERDLVA noise and waveform performance
logarithmic video amplifiers (ERDLVAs), including characteristics along with simultaneous RWR multi-
extracting signal information in the presence of device noise. quadrant data processing. This combination of techniques
The approach uses a combination of time-domain filtering, facilitates identification of threat emitter locations in a non-
adjustable thresholds, and edge detection techniques ideal noisy environment. The approach enables accurate
combined with a four-quadrant data comparison to extract interpretation of time-coincident RF threat signals as a
individual pulse data from overlapped multi-source and standalone monopulse detection method, and can also be
uncorrelated noisy pulse inputs, and the processed data used to complement current deinterleaving and processing
output can be used as part of the presented PDE algorithm
methods.
and also may provide pre-processing for currently used pulse
deinterleaving and sorting methods. The developed technique Section II of the paper will discuss the motivation
has the potential to provide system response that is optimized behind this work. Section III will present a block diagram
for various threat environments via pre-deployment of a typical four quadrant RWR system and include
programming or continuous-time dynamic variations in the description of the key components included in RF to video
PDE. conversion circuits. Components of the system where
specific performance characteristics contribute to the pulse
Keywords - RWR, radar warning receiver, pulse processing,
data extraction processing methods will be identified and
signal processing, electronic countermeasures, ECM
explained. Section IV of the paper will show an example of
detected, partially coincident RF input pulses that might be
I. INTRODUCTION received and detected by a single quadrant of an RWR
With expected increasing range and accuracy of modern subsystem using a COTS ERDLVA device. Section V will
radar-guided ground-to-air and air-to-air weapons systems, provide a description of the PDE algorithm and an outline
there is a need for combat aircraft RWRs to receive, detect, of specific components of the algorithm including noise
and identify threat radar signals at greater distances via floor & minimum detectable level threshold setting, signal
increased receiver sensitivity. However, increased receiver multi-sampling variation, moving average filter window
sensitivity can also result in an overload of RF sensor point sizing, and d/dt or derivative threshold settings.
receive/detection circuits in dense threat environments Section VI presents simulated results for a low-probability
where a significant increase in time-coincident or pulse-on- coincident threat emitter signal group to demonstrate the
pulse (POP) signals may be present. In addition to effectiveness of the processing techniques for a difficult to
fundamental operational challenges, there is a demand for distinguish time-coincident RF signal group. A comparison
Department of Defense (DoD) RWR systems to utilize will also be presented to show the differences in raw video
COTS or MOTS circuit and system components to provide data vs. PDE-processed video signals for the algorithm
acceptable performance while minimizing hardware cost output. The data extraction technique will be presented
and development time. In RWR systems, a typical RF to operating on synthesized pulse data that includes Gaussian
random noise used to replicate non-ideal RWR system

US Government work not protected by U.S. copyright 0752

RF/video performance characteristics. Finally, the paper A description of the RWR channel functional blocks is
will provide a summary along with prospective future provided below:
effort that can be undertaken to continue development and
refinement of the coincident pulse extraction technique. • Input RF Filter or Multiplexer: Provides filtering
and frequency division to accommodate reduction
II. MOTIVATION of total signals processed by each ERDLVA of a
multi-ERDLVA system.
The deployment of RWR processor algorithms
specifically tailored to function efficiently with current • RF Amp: RF gain optimized for utilized detector
COTS/MOTS ERDLVA devices provides the opportunity device.
for DoD system designers and maintainers to leverage off
• Detector: Detector circuits will generally linearly
of current industry innovation and competition in
add coincident RF input pulses (within
ERDLVA development in their effort to support or update
device/circuit dynamic range) which results in
legacy RWR systems at a lower cost and with a larger
compound multi-level video pulse signals and the
source of hardware suppliers. By using COTS ERDLVAs
requirement for the PDE technique.
and/or common RF to video logarithmic conversion
techniques, costs associated with development and • Special video processing: Included in custom non-
acquisition of customized hardware can be shifted to the COTS circuitry. Fixed specialized hardware
software/firmware development area. Moving development processing techniques fundamentally limit
to the area of software/firmware facilitates overall reduced reconfigurability of the RWR system due to custom
system costs along with flexibility in implementing hardware that may cost more and be more difficult
operational modifications through system reprogramming to manufacture and support.
as opposed to hardware redevelopment.
• Log Video Amp: Logarithmic video amplification
A significant amount of investigation and research and drive circuitry compatible with cabling and
exists in the area of pulse deinterleaving methods for input signals of the processor subsystem.
sorting and identifying multiple non-coincident pulse
signals in RWR systems [1]-[4]. These algorithms are • Sig Processor, Interface: Provides input
tailored to accommodate narrow, interleaved, mostly non- terminations to accept log video signals from
coincident pulses. In an RWR system with lower RF/video section.
sensitivity, the probability of time-coincident signals is • Sig Processor, analog-to-digital converter (ADC):
expected to be lower due to shorter detectable range and Provides sampling of time-coincident video pulses
smaller effective sensing area. With minimal signal prior to pulse data extraction being performed.
coincidence and narrow pulse-width signals, existing
deinterleaving algorithms can provide acceptable • Sig Processor, µP: Microprocessor executes pulse
functionality with minimal or no video pre-processing. processing algorithms including PDE techniques
However, in an extended range system with more sensitive for digitized video from ADC.
receivers in an environment of modern wide pulse and high
duty cycle signals (e.g. pulse Doppler) an increase in time- IV. TIME-COINCIDENT VIDEO SIGNALS
coincident signals is expected. Overlapping pulses detected An example of additive detected video output for time-
in this environment may create ambiguous ‘wildcards’ that coincident RF input pulses that might be received in a
must be discarded without signal pre-conditioning. single quadrant of an RWR subsystem and processed by a
COTS ERDLVA is shown in Fig. 2. The figure shows two
III. TYPICAL RWR SYSTEM ideal, noise-free pulses and the resultant additive signal (re-
A block diagram of one RWR system quadrant channel scaled for clarity). Also, the flat additive response levels
is shown in Fig. 1. One ERDLVA signal path is shown are based on assumption that carrier frequencies of Signal 1
(DLVA) to simplify. A four quadrant RWR system would and Signal 2 are sufficiently spaced to allow ERDLVA
consist of four copies of this subsystem, with each antenna video filtering circuits to remove any mixing products
positioned at 90° azimuth spacing on the host platform. resulting from the non-linear response of the ERDLVA
detector diode.

DLVA Signal
ANT Processor
RF Filter Detector
Special
Interface

Video ADC
RF Processing
Log Video µP
Amp Amp

RF/Video Subsystem Processor Subsystem

Figure 1. One quadrant of an RWR system.

US Government work not protected by U.S. copyright 0753

 4.0 8.0 RrF pulses from
Signal 1+2 threat emitters
3.5 7.0
Signal 2 RWR System
3.0 Signal 1 6.0
0.15uW = -38.24dBm 2.68V RWR RF/Video
Data conversion
2.5 5.0 Subsystem
circuits
ts ts & display output
l 2.0
o
v + 4.0 l
o
v
Unprocessed
1.5
2.50V -40dBm = 0.10uW
3.0 ERDLVA
id t Deinterleaving &
1.0 2.20V -43dBm = 0.05uW 2.0
threat-specific ID
Processor algorithms
0.5 1.0

subsystem
0.0 0.0 ADC & µP
0 2 4 6 8 10 12 14 Decision algorithm
time
& volt/power
Figure 2. Detected coincident pulse addition. calculations

As seen in the figure, ‘Signal 1’ (Sig1) and ‘Signal 2’ Baseline

threshold Moving avg
(Sig2) video voltage levels do not add linearly. However, window adjust
RF power adds linearly at the detector and the video output
voltage is determined by the sum of the coincident RF ‘n’ sample d/dt threshold
energy signals. The detected video voltages for Sig1, Sig2, adjust adjust
or ‘Signal 1+2’ (Sig12) are given by:
RWR Hardware
Sig *volts = [( Sig * − Psens ) dBm ] × v / dB + vmin (1) PDE algorithms
Other sorting algorithms
where v/dB is the log slope or linear video voltage
Figure 3. Single RWR Quadrant and PDE algorithm flowchart.
output transfer function of the ERDLVA as a function of
RF input power in dBm and vmin is the lowest output Specific PDE algorithm components are utilized to
voltage that also corresponds to ERDLVA sensitivity, Psens. condition and extract data from coincident signals as
The additive RF input power of the coincident areas, or described below:
overlap, of the two RF pulses is given by:
• Baseline noise floor & minimum detectable level
( Sig12) dBm = 10 log[1000 × (Sig1 + Sig 2 )Watts ] (2) threshold setting: Level determined by combination
of lowest ERDLVA video voltage out and overall
The example signals and calculations shown in Fig. 2 expected device noise floor. Could be dynamically
assume v/dB = 0.10 volts and ERDLVA sensitivity, Psens, is modified to account for detected errors for the long
equal to -50dBm with a corresponding minimum video range, i.e. low video voltage, signals.
voltage, vmin, of 1.5 volts. To obtain accurate voltage for a
later-arriving and coincident pulse, the calculated power of • Signal multi-sampling: Sampling a time window of
the preceding pulse(s) will be used for ‘normalization’, repetitive signals ‘n’ times will provide a 1/sqrt(n)
with video voltage of the second arriving pulse given by: improvement in noise superimposed on the
detected pulsed RF signals included in that
( Sig 2) volts = (2) window. Only useful if monitoring a consistent
pattern of distant fixed signals that may exhibit
(10 log[1000 × ( Sig12 − Sig1)Watts ] − Psens ) × v / dB + vmin
periodicity in overlapping pulses. Not used for
random signals where monopulse interpretation of
V. PULSE DATA EXTRACTION aperiodic coincident signals is required.
The PDE algorithm includes continuous-time signal • Moving average filter window sizing and offset: A
conditioning, pulse edge detection, and decision algorithm simple, effective filter for time-domain processing
for valid pulse edge identification. The PDE obtains of pulsed signals, the moving average filter reduces
magnitude and timing information from noisy coincident random noise while retaining a sharp step response
detected video pulse voltages in each system quadrant by [5]. This response is optimal for the PDE approach
setting a baseline lowest acceptable voltage, averaging to of extracting pulse edges in near real-time to
reduce signal noise, extracting pulse edge information, then extract time and magnitude information. Offset or
identifying magnitude and time for valid pulse edges. A delayed level reading, taking max of several data
flowchart showing one RWR quadrant with PDE and points, is used to ensure voltage read is sufficiently
decision algorithms is illustrated in Fig. 3. ‘inside’ the pulse and not rise-time averaged.

US Government work not protected by U.S. copyright 0754

 • Derivative (d/dt) threshold settings: d/dt threshold
selects the voltage delta from consecutive data
samples to separate actual pulse edges from
random noise spikes. This level is adjusted by
3 4
weighing expected ERDLVA minimum voltage,
RF/video log transfer curve, and expected video 1
signal noise levels.
• Voltage/power calculations and decision algorithm:
Magnitudes of processed waveforms are compared
to determine threat headings. Rising edges used to
determine pulse start and determine if subsequent 2
pulses need to be ‘normalized’ due to pre-existing
pulse and using data from all four system
quadrants. Note that a required element of the PDE
is simultaneous comparison of input from all
quadrants to extract two valid quadrant voltages
(i.e. associated with one threat). Figure 5. Signal position and relative range used for example.

VI. PDE ALGORITHM EXAMPLE Fig. 4 qualitatively illustrates the high level of
A multiple pulse signal threat environment is simulated probability for varying levels of time-coincidence for the
to demonstrate operation and effectiveness of the PDE four presented signals, with the time from 0-15us
algorithm. Fig. 4 shows relative power levels for the exhibiting the greatest percentage of overlap in this
simulation (unitless) of the four signals for a randomly example. A quantitative probability of signal overlap can
selected 100us time sample while Fig. 5 shows relative be computed using ‘window functions’ in a similar fashion
azimuthal threat positions. The example uses a low- that probability of intercept (POI) might be calculated for
probability scenario consisting of four uncorrelated uncorrelated scanning antennas or sweeping receivers.
coincident signals with various pulse widths and PRIs, with Closed-form solutions for various parameters including
a 15us interval highlighted where close and coincident probability of intercept for any two signals in a given time
monopulse information is difficult to extract. Table I and mean period between signal coincidence can be found
provides signal IDs, absolute power incident to aircraft (if in [6] but will not be presented in detail in this work as the
emitter positioned at antenna boresight), pulse width, and focus will be on the extraction of information from the
PRI. multi-signal coincidence condition expected to exist.

TABLE I. THREAT SIGNAL MAGNITUDES AND PULSE TIMING

The simulation uses 40mV RMS random noise on the
video signals, nominal for typical COTS ERDLVAs and
Signal Incident Simulated Heading Pulse width logarithmic amplifiers at example power levels. The
PRI (us)
ID Power (dBm) (degrees) (us) ERDLVA model has a video output minimum voltage of
1 -48 298 1.5 25
+0.6 volts with an RF/video log transfer curve of
70mV/dB. A simulation sample rate of 25nS is used,
2 -48 118 8 26 corresponding to a 40MHz RWR system ADC. Ideal
3 -49 335 1 30
Gaussian antenna response with 80° beamwidths are used
to replicate response of typical RWR system cavity-backed
4 -46 347 3 75 spiral antennas. The PDE algorithm variables in this
example are set to:
• Baseline threshold = 0.55V
1
• Multi-sampling, ‘n’ = 1 (monopulse detection)
15us time window used for
monopulse detection simulation 2
• Moving pts avg/offset = 2/5
Relative on‐boresight power

3
4 • d/dt threshold = 0.14V

PDE-processed detected waveforms are shown in Fig. 6

and a graphical representation of the corresponding four-
quadrant PDE-processed video output signals for the
example is provided in Fig. 7. The data in Fig. 7 consists of
pulse edge information extracted from examination of the
0 10 20 30 40 50 60 70 80 90 100 processed video signals with start and stop information for
t (us) the extracted pulses notated below the figure.
Figure 4. Signal relative power and timing used for example.

US Government work not protected by U.S. copyright 0755

Sample algorithm name ID formats can be interpreted from 315 deg PDE algorithm output
1.5
the names: ‘P1W_start’ is start of the first pulse which is 1.0
located to the west of the platform, i.e. with the two

Vout (v)
0.5
maximum detected voltages at the 225º and 315º quadrant 0.0
receivers. Part of the PDE decision process is to determine -0.5
when a positive edge is processed as an individual input -1.0
t (us)
pulse or as a summation of separate, coincident pulses. This -1.5
0.0 5.0 10.0 15.0
is achieved by ‘holding’ the voltage/power signal in a
particular quadrant at an initial ‘start’ signal until a 1.5
45 deg PDE algorithm output
corresponding ‘stop’ is identified. For the signals in Fig. 6 1.0
this is illustrated by ‘P1E_start’ edge identification in the 0.5

Vout (v)
45° quadrant followed by ‘P1N_start’ edge. Since the 0.0
‘P1E_stop’ edge has not yet been detected, equations (1) -0.5

thru (3) (or applicable derivations) would be used to -1.0

t (us)
calculate the power for ‘P1N_start’ based on the absolute -1.5
0.0 5.0 10.0 15.0
voltage/power and the previously observed ‘P1E_start’
voltage/power. The ‘stop’ edges are not used to extract 1.5
225 deg PDE algorithm output

magnitude data, they merely ‘clear’ any stored 1.0

voltage/power values from associated quadrants. However,

Vout (v)
0.5
0.0
the stop edges are also used to provide end of pulse info -0.5
and pulse width, up to the limitations of the ERDLVA -1.0
t (us)
RF/logarithmic amplifiers in the system. -1.5

Also significant to note, since direction finding relies on a 0.0 5.0 10.0 15.0

comparison of the relative power and detected voltages in 135 deg PDE algorithm output
1.5
two adjacent quadrants with the two strongest detected 1.0
signals at that particular time, using just the initial edge 0.5
Vout(v)

magnitude is sufficient to extract accurate data for pulse 0.0

shapes that may not be perfectly flat or symmetrical, as any -0.5

non-symmetry will be seen to relatively track in both -1.0

t (us)
-1.5
detecting quadrants. 0.0 5.0 10.0 15.0
Table II shows simulation results for signal detection
accuracy. With PDE parameter settings for this simulation
and several runs with random noise variation, relative video
voltage errors of less than 10% were observed as compared
to a noise-free, ideal signal. Voltage variations in the
simulated system translate to maximum azimuth direction
Figure 7. Threat signal position/mag and four-quadrant PDE-processed
error ranges below 8.0° compared to an ideal signal, video output.
comparing well with theoretical errors expected for similar
noise figures and other system parameters fixed [7]. The importance of pre-processing is illustrated in the 45°
quadrant waveform data. Fig. 8 shows the 45° waveform
while Fig. 9 shows associated magnitude data outputs, with
comparison of processed vs. unprocessed signals. The
excessive false pulse detection signals in Fig. 9(a) illustrate
the increase in noise-induced errors without waveform
processing. When processing is applied (and all quadrants
simultaneously compared), useful signal information from
the noisy waveform can be extracted as shown in Fig. 9(b).

TABLE II. THREAT SIGNAL MAGNITUDES AND PULSE TIMING

Incident Start Pulse Simulated Error
Algorithm
Signal Power delay width Heading range
ID
(dBm) (us) (us) (degrees) (degrees)
1 P1W -48 2 1.5 298 -0.4 to 7.5

2 P1E -48 4 8 118 0.5 to 7.1

3 P1N -49 5 1 335 -1.6 to 6.3

Figure 6. PDE-processed video waveforms. 4 P2N -46 7 3 347 1.6 to 6.3

US Government work not protected by U.S. copyright 0756

 VII. SUMMARY AND PROSPECTIVE FUTURE EFFORT
45 deg video out
3.0
The development and simulation of an RWR pulse data
2.5 extraction technique for time-coincident threat signals has
been presented. The technique uses a unique combination
of well-known time-domain processing techniques with
video out (v)

2.0
simultaneous RWR multi-quadrant processing to enable
1.5
interpretation of time-coincident RF threat signals. The
1.0 approach was shown through simulations to provide an
accurate method of separating time-coincident detected RF
0.5 signals with video outputs typical of commonly available
COTS ERDLVAs.
0.0
0.0 5.0
t (us)
10.0 15.0 In this paper the presentation is limited to specific,
static cases to allow for ‘discrete’ explanation of the
(a) concepts. It is understood that in different threat
environments with highly dynamic threat/platform relative
45 deg video out spatial positioning an infinite number of signal
3.0
combinations can be encountered. This necessitates a more
2.5 exhaustive examination of the presented research to refine
and optimize the concepts for specific operational
2.0 environments and tailor the approach for complementary
video out (v)

operation with other existing, proven algorithms. An

1.5
expansion of the probability of coincidence for multiple
1.0
signals of various pulse widths and PRIs to understand the
effectiveness of the PDE algorithm for various threat
0.5 environments is also warranted. Other continuation or
future research expansion is expected to include addition of
0.0 a PDE code module that would dynamically optimize PDE
0.0 5.0 10.0 15.0
t (us) variables for general and specific threat environments as
well as hardware development concepts for novel RWR
(b) subsystem feedback and inter-communication. To refine
Figure 8. Single quadrant waveform: (a) no PDE processing applied (b)
results accuracy for practical systems more real-world
PDE processing applied. variables should be introduced including non-ideal antenna
response (inherent and for specific mounting
45 deg PDE & decision algorithm output
configurations) and system noise variation as a function of
1.5 operating temperatures.
1.0
0.5 REFERENCES
0.0
[1] Y. Kuang, Q. Shi, Q. Chen, L. Yun, K. Long, “A Novel SDIF-
-0.5 Based PRI Estimation Approach to Deinterleave Repetitive Pulse
-1.0 Sequences”,WSEAS Transactions on Mathematics, pp. 260-265,
t (us) Vol. 4, 2005.
-1.5
0.0 5.0 10.0 15.0 [2] R.J. Orsi, J.B Moore, R.E. Mahony, "Spectrum estimation of
interleaved pulse trains", IEEE Transactions on Signal Processing,
pp. 1646 - 1653, Vol. 47, Jun 1999.
(a)
[3] W. Jun, L. Peng, Y. Dong, L. Wei, Y. Xinyu, “A novel
deinterleaving algorithm of radar pulse signal based on DSP”, IEEE
1.5
45 deg PDE & decision algorithm output International Symposium on Industrial Electronics, pp. 1899-1903,
Jul 2009.
1.0
[4] T.L. Conroy, J.B. Moore, "On the estimation of interleaved pulse
0.5
train phases", IEEE Transactions on Signal Processing, pp. 3420 -
0.0 3425, Vol. 48, Dec 2000.
-0.5 [5] S. W. Smith, “Moving Average Filters,” The Scientist & Engineer's
-1.0 Guide to Digital Signal Processing, Ch. 15, California Technical
t (us) Pub, 1997.
-1.5
0.0 5.0 10.0 15.0
[6] A. G. Self, “Probability of Intercept“, ELINT: The Interception and
Analysis of Radar Signals, pp. 97-114, Artech House, 2006.
(b) [7] S. E. Lipsky, “Noise Accuracy of Amplitude Comparison DF
Systems“, Microwave Passive Direction Finding, pp. 248-252,
Figure 9. Data at decision algorithm output: (a) no PDE processing Scitech, 2004.
applied (b) PDE processing applied.

US Government work not protected by U.S. copyright 0757