Power-aware improvement in signal detection. Page: 4 of 15
This article is part of the collection entitled: Office of Scientific & Technical Information Technical Reports and was provided to Digital Library by the UNT Libraries Government Documents Department.
The following text was automatically extracted from the image on this page using optical character recognition software:
Four algorithms are available to operate on the data in order to
yield an estimate of the TEC value. The four algorithms will be
referred to as: 1) least-mean-squares (LMS), 2) maximum-
likelihood (ML), 3) software trigger box (ST), and 4) match-
filter bank (MF). Using only the time-frequency data pairs
provided by the channels of the analog trigger box, the first two
algorithms use curve fitting techniques. While the LMS is a
deterministic algorithm, the ML is force to be deterministic by
only allowing for 20 iterations to be performed . The
remaining two algorithms make use of the 2048-sample, 12-bit,
digitized waveform data.
Using digitized waveform data, the software trigger box
algorithm utilizes frequency domain processing to provide an
estimate of TEC. The software trigger box algorithms first
transforms the time-domain data into a non-overlap spectrogram
composed of boxcar-windowed, 32-sample FFTs. Upon
determining the associated time indexes for the maximums of
each of the seventeen non-negative frequency bins, a maximum
likelihood algorithm is performed on the data pairs constructed
from the bin-maximum times and the center frequencies of the
The match-filter bank algorithm also utilizes frequency-domain
processing. By generating simulated exemplar time-domain
waveforms of different TEC and transforming them into the
frequency domain, a bank of match filter can be constructed that
spans the space of possible TEC values. A correlation peak is
rendered by performing a fast correlation algorithm on the
waveform data and a TEC-specific filter. Exploration of the
match filter bank for the greatest correlation-peak value is done
with a "focus-in" decision tree so that only ten fast correlations
are performed to yield an estimate of TEC. However, since the
value of the winning peak is not quantized or hard constrained
like the TEC estimate, this peak value will be used for post-
processing detection work in this paper for the match-filter bank
2.4 Power Measurements
Power usage measurements for the four algorithms were
obtained through experiments conducted on a 266-MHz
PowerPC 750 microprocessor running the VxWorksTM
operating system. Both time-to-execute values and power usage
estimates (RMS and peak current) were determined for the
PowerPC 750. The time-to-execute values are average values
over a test set of 21 trials cycled 20 to 100 times. Each trial
used synthetically generated data that simulated a chirp-signal
event being received by a space-base receiver system containing
an analog trigger box and a waveform digitizer.
Table 1. Power Measurements for PowerPC 750
Algorithm Current Execution Energy
(amps-peak) Time (Joules)
Least Mean 2.06 3.4 ps 18.7e-6
Maximum 2.06 183 ps 1.02e-3
Software 2.18 8.34 ms 47.3e-3
Match Filter 2.04 470 ms 2.35
Power usage for the PowerPC 750 executing the benchmaking
code is presented in Table 1. The Jet Propulsion Laboratory
(JPL) power-aware testbed consists of a Wind River PPC750
266-MHz processor board that is running VxWorks 5.4.2. The
processor operates at a constant 2.67V and current consumption
is measured with a Tektronix TDS 7104 Digital Phosphor
Oscilloscope. Current is sampled with the Tektronix TCP202
probe that is wired to the board. Software compilation is done
with a VxWorks Tornado 2.0.2 programming tools which uses
the GNU C compiler.
The software is compiled and downloaded to the testbed with
the Tornado target server shell. The programs are run until an
"average" current signal snapshot is taken with the oscilloscope.
The "average" signal is determined manually by watching the
current response during several program runs. The snapshot is
taken when the current response produces a fairly consistent
signal and consistent measurement value.
2.5 Post-Processing Detection
The output of the four algorithms can be compared against
unique thresholds to determine if a false alarm has been
generated by the analog trigger box. Figure 4 shows the concept
expressed in terms of its effect on the ROC curve. Of course,
lost detections can not be corrected for in post processing since
the analog trigger box cues the collection of data and the
execution of the algorithms. Unique thresholds for each
algorithm are needed since the algorithms arrive at their results
differently and to the point of this paper, require different
amounts of power to obtain those results.
Receiver Operating Characteristic: 3 dB SNR
o.se + double threshold-
10 10 10 f 1 l 10a
Probability of False Alarm
Figure 4. Post-Processing Effect on ROC Curve
3. DATA CREATION
Using computer simulation and Monte Carlo experimentation,
the data needed to derive post-processing thresholds and an
estimate of performance of the thresholds can be created.
Parameters needed for the receiver simulation include the
signal-to-noise ratio (SNR) and analog trigger box thresholds,
Here’s what’s next.
This article can be searched. Note: Results may vary based on the legibility of text within the document.
Tools / Downloads
Get a copy of this page or view the extracted text.
Citing and Sharing
Basic information for referencing this web page. We also provide extended guidance on usage rights, references, copying or embedding.
Reference the current page of this Article.
Briles, S. D. (Scott D.); Shriver, P. M. (Patrick M.); Gokhale, M. (Maya) & Harikumar, J. (Jayashree). Power-aware improvement in signal detection., article, January 1, 2003; United States. (https://digital.library.unt.edu/ark:/67531/metadc934100/m1/4/: accessed April 22, 2019), University of North Texas Libraries, Digital Library, https://digital.library.unt.edu; crediting UNT Libraries Government Documents Department.