Background

Motivation

Satellites must operate in a hostile environment while successfully performing a variety of designed tasks. In orbit, satellites experience varying power conditions and extreme temperature ranges during periods of isolation from ground staff supervision. A crucial system to the spacecraft survivability and functionality is the power subsystem; with adequate power supply and distribution through all subsystems, the satellite can perform the mission tasks, however, without the proper power supply and distribution, the satellite mission will be in jeopardy. Additionally, on-board power is a finite, although dynamic, resource that is conventionally managed in a strict manner, limiting performance abilities. Increased power availability and better power management methods can lead to improved satellite capability.

We aim to increase the capability of satellites with power aware ideas and technologies through the use of smarter power management methods. In particular, we have focused on the area of remote sensing satellites, with the objective to increase the amount of on-board processing of remotely sensed data. The perceived benefits of this research is fourfold:

FORTÉ Mission Summary

For the first phase of this research, we have chosen to target a specific application similar to the Fast On-Orbit Recording of Transient Events (FORTÉ) satellite mission. FORTÉ is a Department of Energy (DoE) funded project that is a joint venture between Los Alamos and Sandia National Laboratories to study the correlation between optical and radio frequency (RF) emissions from lightning events in the Earth's atmosphere. The satellite has been a continued success since the launch in August of 1997. The resulting data has provided scientists with an understanding and insight into global lightning climatology, lightning physics, ionosphere dispersion effects on RF signals, and can be used in future space missions to aide severe storm monitoring.

The experiment payload consists of a suite of instruments sensitive to the optical and wide-band Very High Frequency (VHF) regimes of lightning signals. Data gathered by the instruments are sent through hardware devices which "decide" whether or not a valid signal was received. If specified threshold levels are exceeded, a triggering event occurs, and the data are recorded. An on-board storage device keeps the data until the next available downlink.

For more information on this satellite, please see the FORTÉ project homepage.

Remote Sensing Detection Problem

The objective of utilizing a satellite-based, remote-sensing system is to obtain pertinent information from orbit that is typically obscured in a dynamic physical environment. The physical environment masks desired signals with undesirable anthropogenic signals, natural phenomenon, and atmospheric distortion. Additionally, the desired signals may have varying, or random, event rates. Depending upon the application, the targeted signals may be extremely difficult to select from the "background" of unwanted signals. Discrimination between the desired and undesirable signals can occur at either or both of two levels: on-board the spacecraft and on the ground.

In order to focus on the desired signals at the first level, the remote-sensing system can perform event detection triggering, i.e., initiating the recording of incoming data (and/or activating other hardware/software processes) when distinguishing characteristics of the desired signal are observed in the incoming data. Characteristics of the target signal can include frequencies, amplitudes, or any other accessible information, and the defining properties of the desired signal are typically determined with hardware signal filtering. When the selected properties of the incoming signal are observed, a trigger occurs, activating processes to store and/or analyze the data.

In an ideal system, the trigger would occur only when a desired signal is received, however, in reality, the system may still trigger on an undesirable signal. Factors contributing to the false triggers can include "background" signals with properties similar to the desired signal, signal environment noise, unforseen or unexpected events also with properties similar to the desired signal, as well as instrument precision limitations. Thus, data downlinked to the ground can still contain unwanted information or events. The probability of having false triggers, or false alarms, can be reduced through the use of further post-trigger, signal processing techniques. If the spacecraft instrument design is flexible enough, instrument settings can be readjusted after further signal analysis with the aim of reducing the false alarms.

The associated on-board hardware/software system must perform the signal discrimination with constrained power budget limits. There is a direct relationship between the amount of on-board processing and the power consumed: the more processing that takes place on-board, the more power that will be consumed. Processing capability is therefore limited by power consumption since power on a satellite is a valuable, finite resource.

By performing post-trigger processing on the ground, the second level of signal discrimination alleviates the burden of expensive power consumption caused by increased processing. Instead, raw data are downlinked to the groundstation, further analyzed by project staff, and new instrument settings are determined based on the analysis. The new settings are then transmitted to the spacecraft during the next scheduled command uplink. Although this method reduces power demands on the spacecraft bus, the following drawbacks are also introduced:

time delays

Time delays are associated with transmission times and the time

taken for on-ground analyses. The satellite is not always in

view of a ground station where data can be downlinked and

commands uplinked; these transmission times occur only during

specific regions of the orbit (depending upon the spacecraft

orbit and ground station locations). Additionally, there is the time

delay associated with ground staff analyses, during which, the

satellite may travel out of communication range before an uplink

can be performed.

poorer data quality

Less on-board processing increases the chance of accepting more

false alarms. Undesired events uses up the limited storage space

for the desired event data, therefore, good quality data could be

lost. Project staff must then spend the time to also sift through this

mix of desired and unwanted data. Higher quality data would

reduce man-hours spent on selecting the good data from the bad

and allow for more time to be spent on the science, i.e., analyzing

the interesting signals.

larger data quantity

Additionally, pertinent information from the desired signals

cannot be extracted to reduce the size of the downlinked data

sets. The end-user must sift through larger data sets to obtain

the desired data. Additionally, more power will be required for

data downlink transmissions, power that could otherwise be used

for more processing. This could result in an undesired, but

required, off-line of the spacecraft processing during transmission

and reset after transmission.

By utilizing a system which is aware of component power consumption, a smarter power management approach can be applied to solve these associated remote-sensing detection problems, as well as improving satellite functionality.

Parallel Signal Processing Solution

The power aware method we've developed involves using parallel processing techniques coupled with signal processing (SP) algorithms. The SP algorithms can be run independently on multiple processors to analyze different or the same data sets. The idea is to scale power consumption by the choice of the algorithm to run, number of active processors, and the processor clock speed. The algorithms are used to determine specified parameters of the received signal. Each algorithm is chosen on the basis of the algorithm accuracy, power consumed, and time taken to execute; the more accurate the algorithm, the more energy that will be spent. The signal processing routines we are utilizing consist of the following:

The Power Aware Multiprocessor Architecture (PAMA) hardware board developed by the Information Sciences Institute is ideally suited for this task. PAMA consists of 4 processors connected by a programmable FPGA interconnect. The processors run the Linux operating system, Linux MPI-like communications between the processors, and a library that allows power level queries in addition to processor mode, clock frequency, and voltage settings. For more information on the PAMA hardware, please see the USC/ISI PAMA website.

In order to not confuse parallel programming concepts, note that we are not developing parallel algorithms, i.e., the algorithms themselves are sequential, but can be run simultaneously on the 4 processors.