2009:102
M A S T E R ' S T H E S I S
Modelling the electromagnetic phenomena associated with
atmospheric storms and simulation of the IME-HF
Analyser onboard the TARANIS satellite
Petr Váòa
Luleå University of Technology Master Thesis, Continuation Courses
Space Science and Technology Department of Space Science, Kiruna
Modelling the electromagnetic
phenomena associated with atmospheric storms and simulation of the IME-HF
Analyser onboard the TARANIS satellite
MASTER’S THESIS
Petr Váňa 2009
Supervised by:
ing. Zdeněk Kozáček, CSRC s.r.o. Brno Dr. Victoria Barabash, LTU Kiruna Dr. Christophe Peymirat, UPS Toulouse
Master professionnel techniques spatiales et instrumentation, Université Paul Sabatier Toulouse, France
Joint European Master in Space Science and Technology, Luleå University of Technology, Sweden
Abstract
The aim of this work is to study the transient electrical high frequency phenomena related to the thunderstorms. The mathematical model describing the trans-ionospheric pulse pairs (TIPPs) as a representant of these phenomena has been successfully developed. This knowledge is applied as a stimulus to the IME-HF (Instrument Mesure Electrique – voie Haute Fréquence) module of the TARANIS (Tool for the Analysis of RAdiation from lightNIng and Sprites) satellite. The TARANIS is a CNES (Centre National d'Études Spatiales) mission targeted to observe the magnetosphere-ionosphere-atmosphere coupling and the thunderstorm related phenomena from Earth’s orbit.
The IME-HF is an onboard module specialized for measurement of the electrical part of the electromagnetic spectra in the bandwidth from 100 kHz to 35 MHz. Both hardware and function of the IME-HF are described, along with the hardware implementation of the detection algorithm for the transient events like TIPPs. The VHDL (Very-high-speed integrated circuit Hardware Description Language) is used as a tool for the hardware implementation into the FPGA (Field-Programmable Gate Array) device. The primitive sine waves and then the mathematically modeled TIPPs are used to stimulate the IME-HF hardware and evaluate the developed algorithm.
Signals with and without the added artificial noise are used for stimulation. Only the noisiest signals are detected with an extended delay, otherwise the detection algorithm is working with the optimal performance. The results show, that the IME-HF with the implemented detection algorithm can sucessfully detect TIPPs with minimum initial pulse voltage 55 mV (the white noise level present is 10 mV) in 22 out of 25 cases. The future work that is discussed in the final chapter, will constist of implementation of the additional operational modes, telemetry and telecommand to the IME-HF.
Keywords: TARANIS, transient events, ionospheric dispersion, trans-ionospheric pulse pairs, IME-HF, VHDL, mathematical modelling, hardware stimulation.
Résumé
L'objectif de ce travail est d'étudier les phénomènes électriques transitoires en haute fréquence liés à l'orage. Le modèle mathématique décrivant les paires d'impulsions trans-ionosphériques (TIPPs) en tant que représentant de ces phénomènes est développé avec succès. Ces connaissances sont appliquées comme un stimulant pour le module IME-HF du satellite TARANIS. Le TARANIS est une mission du CNES ciblée pour observer les phénomènes liés aux orages à partir de l'orbite terrestre.
L'IME-HF est un module spécialisé pour mesurer une composante électrique de 100 kHz à 30 MHz. Le hardware et la fonction de l'IME-HF sont décrits, avec l’implémentation hardware de l'algorithme de détection pour les événements transitoires, comme TIPPs. Le VHDL est utilisé comme un outil pour l'implémentation hardware dans le dispositif FPGA. Les ondes sinusoïdales primitives, et le TIPPs modélisé mathématiquement, sont utilisés pour stimuler le hardware du IME-HF et pour évaluer l'algorithme développé.
Des signaux avec et sans le bruit artificiel ajouté sont utilisés pour la stimulation. Seuls les plus bruyants signaux sont détectés avec un délai prolongé, sinon l'algorithme de détection travaille avec des performances optimales. Les résultats montrent que l'IME-HF avec l'algorithme de détection mis en œuvre avec succès peuvent détecter TIPPs avec une tension de première impulsion de 55 mV (niveau de bruit blanc présent est 10 mV) dans 22 des 25 cas. Les travaux futurs sur le IME-HF sont abordés dans le chapitre final.
Mots-clés: TARANIS, événements transitoires, la dispersion ionosphérique, paires d'impulsions trans-ionosphériques, IME-HF, VHDL, modélisation mathématique, la stimulation du hardware.
Preface
This work was sponsored by a scholarship from the European Space Agency, Directorate of Human Space Flight and Exploration. I would like to thank to Dr. Piero Messina for this opportunity, which allowed me to fully concentrate on my thesis work, and which made the international travels much more bearable.
The thunderstorm knowledge is important for all the aircrafts or spacecrafts that fly through the planets atmosphere. The recent events, like the crash of the Airbus A330 flying as AF 447 on the 1st June 2009 are considered to be caused by an electromagnetic storm. And the storms in Earth’s atmosphere are very mild in comparison with thunderstorms observed for example by the Cassiny spacecraft on Saturn. For the travels inside atmospheres like this, the better knowledge of the thunderstorm phenomena would surely be beneficial.
This work is a final stage of a two year Erasmus Mundus Master’s programme called “Joint European Master in Space Science and Technology” and known as the “SpaceMaster”. I feel very honoured to be part of this programme and at this point I would like to thank all the people responsible for creating this opportunity for students all over the world. Namely to Sven Molin and Victoria Barabash as the head of the SpaceMaster consortium at the Luleå University of Technology in Kiruna and to the course coordinator Mr. Chrisophe Peymirat at the Université Paul Sabatier in Toulouse.
This thesis would not be here without my supervisor ing. Zdeněk Kozáček from the company CSRC s.r.o. who created for me the opportunity to work on this very interesting subject and who provided me with the assistance and useful advices from the beginning to the end. Also I want to thank to Jaroslav Chum from the Institute of Atmospheric Physics AS CR for the consultations regarding the theoretical background of the transient phenomena.
I would like to thank to all my friends, classmates and colleagues within or outside the SpaceMaster programme for their personalities, advices and opinions, their just being there and their endurance during my bad days. And at last but not least, I would like to thank my parents, which supported me in so many ways that I cannot even imagine and I am very grateful for having such a wonderful family.
List of acronyms
ADC... Analog-to-Digital Converter CNES... Centre National d'Études Spatiales CSRC... Czech Space Research Centre EM ...Electromagnetic FPGA... Field-Programmable Gate Array IME-HF ... Instrument Mesure Electrique – voie Haute Fréquence HF-A...High Frequency Analyzer IAP AS ... Institue of Atmospheric Physics, Academy of Sciences IC (discharge) ... Intra-Cloud LF ...Low-Frequency LPC2E ... Laboratoire de Physique et Chimie de l'Environnement MAC ...Multiplier-Accumulator MF ... Medium-Frequency MEXIC ... Multi EXperiment Interface Controller RAM ... Random Access Memory TARANIS...Tool for the Analysis of RAdiation from lightNIng and Sprites TC/TM...Telecommand/Telemetry TGF ... Terrestrial Gamma-ray Flashes TLE... Transient Luminous Event TIPPs ... Trans-Ionospheric Pulse Pairs UV ...Ultra-Violet VHDL...VHSIC Hardware Description Language VHSIC ... Very-High-Speed Integrated Circuit VHF ...Very-High Frequency
Contents
1 Introduction ... 1
2 Storms in the medium and upper atmosphere ... 3
2.1 TARANIS satellite mission... 3
2.2 Ionosphere ... 5
2.3 Ionospheric dispersion... 6
3 IME-HF ... 11
3.1 Hardware description ... 12
3.2. Operational description ... 14
3.3 Detection algorithm ... 16
4 VHDL implementation... 17
5 Measurement, stimulation and evaluation... 21
5.1 Sine wave stimulation ... 21
5.2 TIPPs stimulation ... 25
6 Summary and discussion ... 29
7 References ... 30
1 Introduction
The man’s interest in thunderstorms dates back to the primitive ages, when it was this atmospheric effect that brought fire to mankind. With the fire it brought fear, then the invention of the lightning rod in the mid-18th century by P.Diviš and B.Franklin and finally a need to predict weather. Even now in the 21th century there are still phenomena associated with thunderstorms that still remain a mystery and that man is eager to discover.
The study of the thunderstorms today is done by a very wide range of instruments, from the ground stations and the airplane based measurement up to the Earth orbiting satellites. The first meteorological satellite Vanguard 2, launched in 1959, can be considered as a pioneer satellite that was partially tasked to observe the thunderstorms from space as a part of its weather observation mission.
However it was the discovery of sprites in the early 90’s that initiated the scientific interest in these previously unkown phenomena [15]. This resulted in launching the ALEXIS satellite in 1993 and FORTE satellite in 1997. These satellite missions focused on the thunderstorm related effects.
The planned TARANIS (Tool for the Analysis of RAdiation from lightNIng and Sprites) satellite mission, that is the subject of this thesis, should improve our understanding of the thunderstorms and its effects. From the broad spectra of the phenomena that will be studied by the TARANIS mission this thesis focuses on a high frequency transient electrical emissions that will be measured by the onboard instrument IME-HF (Instrument Mesure Electrique – voie Haute Fréquence). The instrument main parts are the bipolar HF antenna, the signal pre-amplifier and the HF analyzer (HF-A) controlled by the FPGA (Field- Programmable Gate Array). Only the HF-A part of the IME-HF will be discussed in this thesis.
This thesis was performed in the company CSRC s.r.o. (Czech Space Research Centre )which is responsible for the IME-HF instrument development and integration. Close cooperation was held with the IAP AS CR (Institute of Atmospheric Physics, Academy of Sciences of the Czech Republic) and with the LPC2E (Le Laboratoire de Physique et Chimie de l'Environnement), Orléans, France.
The TARANIS mission and storm phenomenon are discussed in Chapter 2. Particullary the ionospheric effects, particullary the ionospheric dispersion are explained in Sections 2.2 and 2.3. The physical model for simulating the so-called trans-ionospheric pulse pair (TIPP)
is analysed in Section 2.4. Chapter 3 is dedicated to the IME-HF module. The IME-HF hardware is discussed in Section 3.1; operational aspects of IME-HF are explained in Section 3.2 and the detection algorithm for transient events is presented in Section 3.3.
Implementation of this algorithm is explained in Chapter 4. The overall performance of the created algorithm is evaluated in Chapter 5. The evaluation of performance is based on simple sinewave stimulation in Section 5.1 and on stimulation with previously developed TIPPs signals, done in Section 5.2. Finally, Chapter 6 concludes the previously performed measurements and integrations along with a work to be done after finishing this thesis.
2 Storms in the medium and upper atmosphere
2.1 TARANIS satellite mission
The name TARANIS comes from the old Celtic mythology, where Taranis was known as the god of thunder. In the case of the satellite mission, it stands for the Tool for the Analysis of RAdiations from lightNIng and Sprites [1]. The main goal of the TARANIS mission is to study the magnetosphere-atmosphere-ionosphere coupling via the transient processes.
The processes in question occur during the thunderstorms and they are also reffered to as
“Transient Lumionus Events” (TLEs). The most commonly observed emmision from ground, the so-called “sprite”, is a manifestation of electrical breakdown of the upper-atmosphere at 40-80km altitude [2]. The “blue jet” most resembles the classical lightning, except it is a discharge propagating upwards into the stratosphere from the cloud tops. Furthermore the
“elve” is a concentric ring of optical emmisions propagating horizontaly outwards at the bottom edge of the ionosphere. There is also documented the “gigantic jet” that is a discharge where a blue jet triggers a sprite, creating the electrical breakdown from the thunderstorm clouds directly up to the bottom of the ionosphere. These events are shown in Fig. 1.
Non-luminous events are also associated with the thunderstorm transient phenomena such as Terrestrial Gamma Ray flashes (TGFs), runaway electrons and Trans-Ionospheric Pulse Pairs (TIPPs). The TIPPs will be discussed in the section 2.3. Furthermore the TARANIS mission has been extended to incorporate the transient precipitations and accelerations of energy electrons, regardless of whether they are directly linked to TLEs. As to the time of writing this thesis the TARANIS mission has four main objectives [1]:
Detection and characterisation of TLEs
Detecting TGFs and studying generation mechanisms of TLEs and TGFs
Characterising runaway electrons that are accelerated upwards from atmosphere to the magnetosphere
Identifying the effects of TLEs on coupling between the ionosphere and the magnetosphere.
Figure 1: Overview of numerous transient events occuring in the atmosphere during the thunderstorms [2].
For achieving these objectives the satellite will carry the following payload elements (Fig. 2):
MCP, set of 2 cameras and 3 photometers (30 images/s, 512x512 pixels and luminance measurements of the different high-resolution spectral bands)
XGRE, set of 3 X and gamma-ray detectors (measurements of high-energy photons between 20 keV and 10 MeV and relativistic electrons between 1 MeV to 10 MeV)
IDEE, set of 2 electron detectors (measurements of high-energy electron spectrums between 70 KeV and 4 MeV)
IME-BF, LF antenna to measure electric fields between 0 and 1 MHz;
IME-HF, HF antenna to measure electric field between 100 kHz and 35 MHz
IMM, a triaxes search coil magnetometer to measure the alternative magnetic field:
- IMM-BF, 2 mono-band coils to measure the LF field;
- IMM-MF, 1 dual band coil to measure the LF and MF field;
Figure 2: The block schematic of the TARANIS payload. Each MEXIC box provides the electric power to the instruments along with the management of the payload modes and the interface with the Mass Storage and the platform computer. [4]
This work is focused on the IME-HF analyser, which will be further discussed in chapter 3.
2.2 Ionosphere
Ionosphere is a part of atmosphere, where UV and X-rays from the Sun are absorbed, creating a strongly ionized environment. The ionosphere can be further decomposed into different layers with various properties: D, E, F (F1, F2). The most significant layer in the terms of dispersion is the F layer, which during the day time can be separated into F1 and F2 layers.
The ionized environment of D and E layers is created by molecular ions. F1 layer can be seen as a transition between molecular and atomic ions. The atomic ions, most significantly O+, dominate in the F2 layer. The highest ionisation occurs in the F2 layer in the altitude approximately 250 km. The graph of ionisation level in different altitudes can be seen in Fig. 3.
Figure 3: Electron concentration as a function of altitude. The concentration also depends on the time of the day and solar activity (dashed line is solar minimum, full line is solar maximum). [Tascione, T.F., Introduction to the Space Environment, 2nd Ed]
2.3 Ionospheric dispersion
The aim of this chapter is to create a theoretical background for the TIPPs modelling. For studying of the dispersion in ionosphere, the cold plasma approximation is considered. The plasma can be regarded as cold, if the thermal movement can be neglected in the comparison with the movement caused by wave motion [5], [7]. The complete description and background theory for this topic can be found in [5] or [6].
The dispersion is a phenomenon when the wave propagation depends on wave frequency.
The wave number k is a function of the wave frequency ω, known as a dispersion relation.
The wave number k can be directly related to the refraction index n as
2 2 2 2
ω k
n = c . (1)
From [5], the dispersion relation for a wave in cold plasma can be written as
( ) ( )
( θ θ θ) θ θ
θ
2 2
2 2 2 2 4
2 2
2
cos sin
2
cos 4
sin cos
2 sin
P S
P D SP
RL SP
SP n RL
+
+
−
± +
= +
, (2)
while using the convention of Stix notation [6]:
(R L) D (R L)
S = + = −
2 , 1
2
1 , (3)
∑
∑ − = − +
−
=
i i
i
i i
i
Y R X
Y L X
1 1 1 ,
1 , (4)
ω ω ω
ω ci
i pi
i Y
X = , =
2 2
. (5)
The parameters used here are the plasma frequency ωpi, the cyclotron frequency ωci and the angle between the propagation vector and the geomagnetic field vector θ. The L and R terms were reduced using the collisionless plasma approximation, stating that the electron collision frequency ν is much smaller than the wave frequency of interest ω (0.1 MHz – 35 MHz).
The cyclotron frequency ωci can be estimated by relation [7]
3 2 0
1 sin 3 1
+
⋅ +
=
e ci
ci
R h ω λ
ω . (6)
There ωci0 (5.45·106 rad.s-1 for electron) is the cyclotron frequency on the equator and Re = 6378 km is radius of the Earth. As can be seen, the cyclotron frequency is dependent on the altitude h and geomagnetical latitude λ.
Plasma frequency ωpi is related to ion (electron) concentration ni (ne) as
0 2
ω ε
i i i
pi m
q
= n , (7)
where mi is the ion mass, qi is the ion charge and ε0 is the permittivity of free space (ε0 = 8.8542·10-12 F.m-1). Since the electron concentration ne changes with altitude (Fig. 3), the electron plasma frequency is also dependent on altitude. Using the plasma frequency at any given altitude then with eq. (6) and Fig. 3 it is possible to obtain plasma frequency at any required altitude in the ionosphere.
Finally, the one-species electron plasma is considered. The dispersion relation (2) further reduces to the form
( θ) 4 4θ ( )2 2 2θ
2 2 2
cos 1
4 sin 1 1
1 1
2 1 sin 1
Y X X Y
X Y n X
−
− +
− ±
−
−
= . (8)
The relation (8) is known as the reduced form of the Appleton-Hartree equation.
2.4 Trans-ionospheric pulse pairs
The trans-ionospheric pulse pairs or TIPPs were first observed by the Blackbeard experiment onboard the ALEXIS satellite in 1993 [3]. The TIPPs are twin broadband pulses with maximum frequency up to around 150 MHz, with an average duration of 10 µs and an average delay between the pulse and echo of 50 µs [9], [11]. The pulses are characterised by a dispersion of the signal, which is caused by the ionospheric passage (see 2.3). Thus, the TIPPs source lies in the lower part of the atmosphere. The example of TIPPs detected by the Blackbeard experiment is shown in Fig. 4.
Figure 4: Example of TIPPs. Picture is taken from [11].
In [11] was argued that TIPPs could originate from upward propagating lightning discharges or the origin of the TIPPs is considered to be an intra-cloud (IC) discharge [8].
There is also a dispute between the two possible origins of a second echo pulse. This is shown in Fig. 5. The first model assumes that the echo is merely a reflection of the first pulse from the ground. The second model states that the echo pulse is originating from the higher parts of the atmosphere several tens of kilometers above thunderstorm (high-altitude discharge).
Figure 5: Two possible models of the TIPPs origin. Figure is taken from [8].
Another observed phenomenon of the TIPPs is the bifurcation at lower frequencies as the result of mode splitting [11]. This is caused by the Earth’s magnetic field with the most significant effect of the altitude of F-layer which has the highest charged particle concentration. The separation is determined by the magnetic field strenght and the angle between the wave propagation vector and the geomagnetic field vector θ.
Consider the case θ = π/2 when the signal propagates through the ionosphere.In the first mode called ordinary “O” mode the electric field is perpendicular to the magnetic field and signal travels in the sense of their vector multiplication. This is the wavemode of plasma without the magetic field. The second mode travels as the extraordinary “X” mode and it exists only with the presence of the magnetic field.
Considering the most simple case, when θ = 0, the eq. (8) has two solutions Y R
nR X =
− +
=1 1
2 (9)
and
Y L
nL X =
− −
=1 1
2 . (10)
These solutions were caled R and L because they represent wavemodes with left-hand and right-hand polarisation [6].
For the simulation model the trans-ionospheric pulse is considered to be a broadband pulse originating from the thunderstorm cloud during the IC (intra-cloud )discharge with a ground
reflection creating the echo. The group velocity vg can be derived from the dispersion relations (9) and (10) using
k
vg r
r r
∂
=∂ω
. (11)
Then the time of arrival of the modes of the first pulse can be estimated as
( ) ( )
( ) dh
R c c R
dh h t v
t
h c
c p
free
h gR
free
R ⋅
+ + +
⋅ +
= +
= ∫ ∫ 2 2 2
2 2 2 2
2 1 1
ωω ω
ω ω ω
ω ω , (12)
( ) ( )
( ) dh
L c c L
dh h t v
t
h c
c p
free h gL
free
L ∫ ∫
− + −
⋅ +
= +
= 2
2 2
2 2 2 2
2 1 1
ωω ω
ω ω ω
ω ω . (13)
There the tfree (hfree) is time when the signal travels through non-dispersive environment that is considered equal for both modes. The integral part represents the time the signal travels through the ionosphere. The plasma frequency ωp and the cyclotron frequency ωc are functions of altitude.
The second pulse is delayed by ground reflection. The delay can be estimated as
c t 2hth
∆ = , (14)
where hth is the thunderstorm height.
The graphical representation of TIPPs created using equations (12), (13) and (14) can be seen in Fig. 6.
More complicated cases with arbitrary angle θ and multiple-species plasma were simulated numerically by MATLAB program written by J.Chum [personal reference]. These simulated signals were used for stimulating the IME HF-A later in Chapter 5.2.
Figure 6: Simulation of the trans-ionospheric pulse pairs dispersion given by equations (12), (13) and (14). Parameters: critical electron plasma frequency at the F2 layer fpeF2 = 7 MHz, geomagnetical latitude λ = 50°, IC discharge height hth = 7 km. Height of the satellite 700 km corresponds to hfree = 200 km (it is roughly assumed that the ionosphere starts at 100 km altitude and fades at 600 km altitude, only the regions with the high ion concetrations are considered). The birefringence (or mode splitting), represented by green line for the L mode and red line for the R mode, occurs due to the presence of the magnetic field. The blue lines are the ground echo.
3 IME-HF
The studies of the lightning radiofrequency emissions in the VHF range performed from the Earth’s orbit already have a short history. First studies were done by the Blackbeard experiment onboard the ALEXIS satellite, launched in 1993. The frequency measurement range of the Blackbeard experiment was 25 to 100 MHz. This effort was followed by the FORTE satellite in 1997. In this experiment, the vast majority of information was obtained in the range of 26-48 MHz (low-band) with simultaneous readings either at 118-140 MHz or at low-band but with alternate polarisation [3]. It can be said that the IME-HF instrument is filling the gap in logical continuation of studying the electric part of the VHF electromagnetic field associated with the thunderstorm events.
3.1 Hardware description
The IME-HF (Instrument Mesure Electrique – voie Haute Fréquence) instrument consists of a double-wire Hertz dipole (the sensor) and a HF-A analyser circuit. The sensor intended to measure one component of the electric field from 100 kHz to 30 MHz along the axis of the two aligned wire antennas. This allows to measure events like TIPPs and contrary to ALEXIS or FORTE, the frequency range contains the plasma frequency of the ionosphere. Each antenna is equipped with a preamplifier. The antenna is shown on Fig. 6 and its connection to the HF-A is shown on Fig. 7.
Figure 6: The IME-HF sensor antenna. Image was taken from [13].
Figure 7: Connection of the antenna to the HF-A analyser [13].
Figure 8: IME-HF block diagram [13].
The analyser is mainly dedicated to the data processing of the HF signal with 12 bit depth on the sampling frequency of 80MHz [4]. The schematic of the HF-A is presented in Fig. 8.
Initialy the IME HF-A performs a diffference between the potentials measured by antennas.
This difference is fed through the 10-th order bandpass anti-aliasing input filter. This signal is then processed in two ways.
Firstly there is the “fast” serial A/D converter (fADC) with the sampling frequency of 80 MHz and 14 bit depth. Only 12 bits are transmitted further and processed. There is a gain control to select the desired 12 bit range. The 12 bit data from the fADC are stored in the circular RAM buffer for further processing and event detection by the FPGA.
Secondly there is the “slow” signal path. The signal is divided by 9-th order bandpass filters into 12 channels across the whole frequency range. The centre frequencies of these bandpass filters are 1.5, 4.5, 7.5, 10.5, 13.5, 16.5, 19.5, 22.5, 25.5, 28.5, 31.5 and 34.5 MHz. Each channel further contains its own amplifier and a 12 bit “slow” serial A/D converter (sADC) with the 12 µs time interval between samples. Data from the 12 channels are fed to the FPGA and processed there.
The outside world communication of the IME-HF is provided by the MEXIC interface connected to the FPGA. Beside the telecommand/telemetry (TC/TM) the FPGA is responsible for data formating, event detection and driver for the other hardware peripherals.
3.2. Operational description
According to [13] the IME HF-A has several working modes: Off mode, Standby mode, Test mode, Servicing mode and Science mode. The normal operation with data acquisition and TC/TM is provided by Science mode. This mode is the topic of this chapter.
In the Science mode, the HF-A operation is further divided into 2 configurations: survey mode and event mode. The input data that are processed in these two configurations are:
- averaged values of the 12 signals obtained from the bandpass filter banks, sampled at 12 µs
- waveform sampled at 80 MHz with selected 12 bit depth
The data are stored in 2048 bytes blocks with single sample stored in 12 bits. Effectively there are 1365 samples in every block.
Survey mode
In survey mode, the data are transmitted to MEXIC by a package of 36 blocks containing both the measured data and the housekeeping information in the block headers. The telemetry rate of raw data without the headers is 40 kbits/s. That means the transmission time of one package is 14.7456 s. There are three different types of data: SP, SK and SD. They are transmitted to MEXIC by couples SK-SD (default), SD-SP or SK-SP. Each data type is forming one superblock (18 blocks) and the telemetry package contains two superblocks, each with different data type.
The data types are defined as follows:
SP: 1024 averaged signal samples from 12 bandpass filter banks on continuous mode. The time, over which is the samples are averaged, is 12.288 ms, thus 1200 averages fit in the single transmission interval (14.7456 s). Since 2034 averaged samples from each channel can fit into a superblock, there is a space for partial overlap. In this data type, 1800 samples per superblock are used. The slide between the subsequent sample intervals used for calculation of the averages is 682.6667 samples. In the implementation it means slide 682 samples, then 2x 683, then again 682, 2x 683 and so on.
SK: Snapshots of short waveform from the fast ADC. Each block contains one snapshot. 18 snapshots, forming one superblock, are equidistantly distributed inside the 14.7456 transmission interval. Duration of one snapshot is 17.0625 µs and the time interval between subequent snapshots is 0.8192 s.
SD: Longer waveform from the fast ADC, that fits into one superblock. The time of capture of this waveform is selected by the signal VCIL_S from the detection algorithm described in section 3.3. The length of the waveform sampled at 80MHz with 12 bit depth that can fit into one superblock is 307.125 µs. The waveform is stored in the circular buffer thus a time offset can be set. That means the time interval of the selected waveform can be shifted with respect to the time of capture. Time offset is set by a command.
Event mode
Data storage for this mode is initiated by the signal VCIL_UD from the detection algorithm described in section 3.3. The available memory for the event mode is 128Mbits. The space of this memory is distributed between the space dedicated to the fADC waveforms, sampled at 80MS/s with 12-bit depth (90%) and to the sADC unaveraged bandpass filter bank data, sampled at 83,33 kHz (12 µs) (10%). The graphical representation of the memory distribution can be seen in Fig. 9. Data from the first event are stored to the whole memory range. Each new VCIL_UD event overwrites the upper part of the memory with the new data, while leaving the lower part with the data from the previous events. Only 3 waveforms can be stored, while maximal number of events registered can be 24.
Figure 9: Allocation of the 128 Mbit memory in the event mode. Maximal number of stored events NA is 24. Only waveforms from the first 3 events are stored. Each new incoming event
overwrites part of the memory with new data, thus reducing the stored length of the previous event.
3.3 Detection algorithm
The implementation of the detection algorithm to the FPGA is one of the main practical impacts of this thesis. The detection algorithm is sensitive to signal variation; it works with the data input from the 12 channel bandpass filter banks with the sampling period Tsa = 12 µs.
According to [4] and [13], the detection runs in the following steps:
1) For each frequency channel i, a maximum of amplitude mi is searched on time interval N*Tsa. The number N is set by a command. A default value is N = 16. Thus, e.g., for N = 16, the maximum is searched in the time interval of 192 µs. The time interval in which the maxima are searched is slid by N/2 in the next step.
2) Mean values ai for each frequency channel are calculated over time M* Tsa, where M = 2h. The h is an integer set by the command (default h = 20, which makes the time interval of the mean value update 220* Tsa ~12.58 s).
3) A rank R is attributed to each processed interval using formula
( )
∑=
−
= 12
1 i
i i
i m a
w
R , (15)
where wi is the weight of i-th frequency channel. The weights wi are set by the command.
If wi = 0, then the i-th frequency channel is not used in detection. It is supposed that channels with the higher (thus more important and less noisy) frequency have higher weights.
4) A maximum of R is searched during the transmission of the last data package (36 blocks, 14.7456 s). If a new maximum is found a signal VCIL-S is generated and the 307.125 µs time interval centered on TS - TSOFF is copied from the circular buffer to the selected data memory. TS is the time of capture when the signal VCIL-S was generated and TSOFF is an offset set by the command.
5) If the rank R exceeds a threshold value P, R > P, then a signal VCIL-UD is generated.
The threshold value P is set by a command. The signal VCIL-UD is used for triggering the event mode. If P is set to a value which is larger than a maximum possible value of R, then the signal VCIL-UD will never be generated. The signal VCIL-UD is transferred via MEXIC to the satellite.
4 VHDL implementation
This chapter covers one of the practical impacts of the thesis, namely the design and implementation of the detection algorithm to the Xilinx® FPGA on the HF-A analyser. The FPGA (Field-Programmable Gate Array) used in the HF-A engineering model is a Xilinx® Virtex® 4 platform, model XC4VFX40. The further engineering and flight versions of the IME-HF will have space qualified modification Virtex® 4QV. The detection module is a part of a larger module that takes care of TC/TM, memory operations, ADCs control and data formatting.
The VHDL (Very-high-speed integrated circuit Hardware Description Language) was chosen as a tool for the algorithm implementation. Since VHDL is used to describe hardware, it should not be seen strictly as a programming language. The detection algorithm was divided into logical blocks, which resembles the paragraphs in section 3.3 and these blocks were individually implemented and then interconnected. The FPGA was selected for this task due to the fact, that the hardware described in the VHDL is working in parallel way, additionaly there is no space-qualified microprocessor fast enough to provide data processing on the required 80MHz frequency.
The detection algorithm is driven by bus clock working on 100 MHz. Every action takes place on a rising edge of the clock to avoid race conditions. The block schematic of the detection module and its surroundings can be seen in Fig. 10. The detection module itself, called detect_mod, is wrapped in the sadc_detect module that acts as the interface to the sADC signal and to the outside variables N, P, M and wi. On the input, the sadc module processes the sADC signal from 12 bandpass filter channels. The output from sadc that is used by the detect_mod is represented by these two interfaces:
- The unmodified signal from each channel with 12 µs interval, represented by a 12-bit data bus s_data, 4-bit address bus s_addr and a signal s_ready.
- The averaged signal from each channel with 12.288 ms interval. Averages are calculated from 1024 unmodified signal samples. This interface is represented by a 12-bit data bus sp_data, 4-bit address bus sp_addr and a signal sp_ready.
Whenever there are new data present, the sadc module successively puts the new data on the sp_data (sp_data) bus and addresses one of the 12 periphery channels max_mod (avg_mod) with s_addr (sp_addr). Peripheries are alerted on the new data by driving the signal s_ready (sp_ready) high for one clock period.
Figure 10: Block schematics of the VHDL implementation of the detection algorithm. The interface to the A/D converter is the data input. Additional input parameters are N, P, M. The w parameter is a data bus, w_sel is corresponding address selection. Output from the whole unit are the VCIL_S and VCIL_UD signals discussed above in Chapter 3.3. S_running and UD_running inputs serve as gates for VCIL_S and VCIL_UD generation.
The module detect_mod is the implementation of the detection algorithm itself and its structure is further divided into these modules:
max_mod: The 12 modules max_mod are designed to perform step 1) of the detection algorithm in the Section 3.3. All 12 modules are connected with the tri-state data bus s_data and address bus s_addr to the sadc module. This bus provides the input data for calculating the maxima. When the inside register is filled with N/2 samples of signal data, the process changes to the state, when this register is searched for its maximum value. There are two variables to store maximum inside the max_mod. In the search state, the previous maximum is stored and the new maximum is chosen from the N/2 samples stored in register. As soon as the new maximum is ready, the tri-state signal calc_upd is driven low to alert the calcR_mod that the new data is present. Each operation in search state occurs in one clock cycle, due to the marginal and speed limits. Thus when e.g. N/2 = 8, then the maximum is updated after 8 cycles of maximum search. When the particular max_mod is addressed by ch_sel, the process compares the old and new maximum, each representing subsequent N/2 interval, and the highest value is put to the tri-state bus max_data, that was previously kept in the high impedance state.
avg_mod: These 12 modules are designed to perform step 2) of the detection algorithm in the section 3.3. The busses sp_data and sp_addr are the data inputs, providing the average of 1024 samples from the sadc module. When the particular avg_mod is addresed by sp_addr, the data on the sp_data bus are added to the existing sum and the sample counter is incremented. If the sample counter is equal to 2M-1024 = 2M-10, then the adding stops and the sum is divided by the total sample count 2M-10, providing the required average. The division can be performed easily, since the divider is a power of two, thus a bit shift by M-10 is all that is required. After the division, the sum and the sample counter is zeroed and the tri-state signal calc_upd is driven low to alert calcR_mod on the actualized data.
calcR_mod: This module performs the calculation of the rank R according to step 3) in section 3.3. To do that, the equation (15) is implemented into the hardware. Due to the speed limitations, each single calculation step is performed in one clock cycle. For performing the multiplication, the advantage of Virtex® 4 DSP48 slice is exploited, and the multiplier-accumulator (MAC) from IP Core is used. This is represented in the VHDL code by the multac component. The main process in
calcR_mod is a state machine with rs and ch as state variables.
There the variable rs has two states:
sel – state where next data channel is selected and the difference max_data – avg_data in eq. (15) for the previous data channel is calculated. Then the state of rs is changed to c1.
c1 – state in which the rest of the calculation with the current data channel takes place. This is performed by updating the MAC inputs and driving its clock enable signal cex high. The inputs to the MAC are the difference max_data – avg_data calculated in the sel state and the weight wi_data which is supplied by an external bus of the same name. Then the state of rs is changed to sel.
The state variable ch represents the pseudo-channel number from 1 to 16. There are only 12 real channels; rest of the states is used as calculation cycles.
Initially after reset, the process waits for the tri-state signal calc_upd to be driven low. When that happens, the state of the machine is rs = sel and ch = 1 and the MAC is reset. The state variable rs switches between sel and c1 states, while the state variable ch increments up to 13. One clock cycle is required to update the data on the max_data, avg_data and wi_data buses. In the first run with rs in the sel state, no difference is calculated and channel 1 is selected by address bus ch_sel. In the second run with rs in the sel state, the updated max_data and avg_data buses are used for calculation for the channel 1 and at the same time channel 2 is selected. In the last run (ch = 13), only the difference is calculated from the already selected channel 12. The states ch = 14 and ch = 15 are used to wait for the MAC pipeline to complete the calculation. In the final state ch = 16, the calcR_mod output R is updated with the calculated rate R.
Then the process waits for the tri-state signal calc_upd to be driven low again.
This process algorithm was proven to be the so far best compromise between the speed, the bus clock load and the FPGA resources usage. It takes 30 clock cycles (300 ns) from detecting the calc_upd in low state to updating the output R.
detect_mod – implementation:
This is implementation of the detection module itself. It is responsible for steps 4) and 5) of the detection algorithm described in the section 3.3 and for generating the output signals VCIL_S and VCIL_UD. In each clock cycle without reset, the value of R from the calcR_mod is compared either to the value
P or the internal value R_tmp. If it is higher than either of these values, the signals VCIL_UD and VCIL_S are generated respectively. The inputs S_running and UD_running are used to block the generation, if the survey mode or the event mode are still processing the data from the previous upset.
The internal value R_tmp is updated each time the value of R is higher. Next R is compared to this new value of R_tmp and so on. R_tmp is cleared after every package transmission interval, signalled by the R_clear clock.
5 Measurement, stimulation and evaluation
This chapter will cover the hardware measurements on the IME HF-A module by stimulating its input mostly by signals characterised in Chapter 2 and by evaluating the function of the implemented detection algorithm and an analog signal path inside the HF-A.
5.1 Sine wave stimulation
The initial measurement that was performed on the IME-HF was the detection of the burst of the simple sine waves. The sine wave frequency was successively set according to the centre frequencies of the bandpass filters mentioned in Section 3.1. Each channel was tested separately on its detection sensitivity to the signal level (sine wave amplitude). Every measurement consisted of 10 bursts with the given frequency and amplitude. The measured parameter was the number of succesfully detected events out of 10 possible. The necessary parameters for the detection module in the FPGA had values set to: N = 8, M = 20 and
(∀i∈ 1,12 )(wi =1). (16)
Since the R updates every N/2 signal samples, the detection was considered succesful if the signal VCIL_S was generated within the (N/2+1)*12 µs = 60 µs interval after the burst initiation. Example of this burst can be seen in Fig. 11.
Figure 11: Example of the sine wave burst. The signal was burst of 5000 cycles of sine wave with frequency 1.5 MHz and peak-to-peak voltage 1.2 V. The interval between bursts was set to 2 s, which was considered enough to establish zeroed value R_tmp to the noise level of the zero signal. The VCIL_S signal was detected 58 µs after the burst initiation.
The noise present in the signal was the background noise from the used generator, oscilloscope and internal noise of the HF-A. It can be characterised as a white noise with no significant peaks in the observed spectra (1 MHz to 35 MHz) on any frequency. The noise level was 8 mV peak-to-peak. The results of the measurement for each channel are shown in table in Fig. 12. Graphical representation of these data can be seen in Fig. 13.
Figure 12: Detection success rate. It is obvious, that the most sensitive are the channels 5 to 11. The channel 1 was not even able to detect 2 out of 10 sine burst events with the 400 mV amplitude.
Figure 13: Graphical representation of the data in Fig. 12.
The average detection delay for the 1.5 MHz signal is displayed in Fig. 14. Here it can be seen that in the two cases the detection occurred in the second update of the R value (later than 60 µs).
Figure 14: Detection delay and its average for the 1.5 MHz sine burst signal with amplitude 0.6 V.
The different sensitivity of the each channel is partly a characteristic property of the IME-HF instrument. It is an advantage that the channels 7 to 9 are the most sensitive, because the detected events like TIPPs start with higher frequency signal. There is also more parasite noise present in the lower frequency bands. It is considered for the detection algorithm to have the weights wi set to favour the higher frequency channels. Example of this setting will be shown in Section 5.2.
Variation in sensitivity is also a result of the bandpass filter banks characteristics shown in Fig. 15. Since the filters are not ideal, they do not completely attenuate the signal with undesired frequencies. Thus thanks to the overlapping; the detection algorithm with the weights set according to (16) is more sensitive to the frequencies from the middle range. The decrease of sensitivity of the channel 12 is caused by the drop of throughput of the input anti- aliasing filter to the HF-A for the highest frequencies.
Figure 15: The bandpass filter banks characteristics. These are the 9-th order analog filters.
Input signal level is -30dBm with 10dB amplification. Black line is the characteristics of the input anti-aliasing filter and it is not in scale to the rest of the lines. Figure is courtesy of J.Chum (IAP AS CR).
5.2 TIPPs stimulation
Testing of the HF-A continued with stimulating the device with the TIPPs signals discussed in Chapter 2. The signal was generated by MATLAB and then uploaded to the function generator as an arbitrary waveform. The sample of this signal can be seen in Fig. 16 and its spectral characteristics can be seen in Fig. 17.
Figure 16: Example of a TIPPs signal. The parameters for generation were: θ = 30.8°, fpeF2 = 7 MHz, hsat = 700 km, λ = 40°, hth = 7 km and a multi-species plasma with the relative ion concentrations: rH = 0.9, rHe = 0.09, rO = 0.01. (See section 2.4 for description).
Figure 17: Spectrogram of the TIPPs from the Fig. 16.
These samples in Fig. 16 and Fig. 17 are the most ideal cases with no noise present. Since the real operation conditions are subject to a very strong interference, the HF-A was also tested with a signal with added arbitrary noise. The simulated noise contributions were: white noise
with signal-to-noise ratio -10 dB related to the first TIP, broadband noise with the centre frequency at 12.5 MHz, bandwidth 9 MHz and SNR -10dB and finally a transmitter at 20 MHz with relative signal strength -6.9 dB. The resulting waveform is shown in Fig. 18. The spectrogram for this waveform is then shown in Fig. 19.
Figure 18: The TIPPs signal with noise contributions. The signal parameters are the same as for the signal in Fig. 16.
Figure 19: A spectrogram for the signal in Fig. 16. All the three kinds of noises: white noise, broadband noise at 12.5 MHz with 9 MHz width and a parasite transmitter at 20 MHz can be easily identified.
The simulated signals were fed to the HF-A input and the observed values were: the time of the event detection (te), background R value (R0) and R value at the time of the signal detection (Rsignal). The signal level was set to 1 V peak-to-peak voltage of the first TIP in all cases. The necessary parameters for the detection module in the FPGA had values set to: N = 8, M = 20 and w = [0, 0, 2, 5, 8, 10, 10, 10, 10, 10, 10, 10]. The partial results of this measurement are shown in Fig. 20.
fpeF2 = 4 MHz, no noise
te [µs] 40.0 16.8 59.2 42.8 54.2
Rsignal [-] 479480 450057 445745 474580 452194
fpeF2 = 4 MHz, added noise
te [µs] 100.4 44.0 133.2 23.6 36.0
Rsignal [-] 456977 472796 457026 455607 456865
fpeF2 = 7 MHz, no noise
te [µs] 33.8 53.8 35.0 40.2 21.0
Rsignal [-] 452896 465823 456607 457230 468689
fpeF2 = 7 MHz, added noise
te [µs] 28.4 33.2 24.4 15.6 21.6
Rsignal [-] 475949 476877 478910 477832 474777
363075 363240 339423 346164 351363 R0, no noise
present[-] 356195 355804 352410 340548 352928
Figure 20: The result of the TIPPs stimulation. The average R0 value with no added noise and no signal present is 352115 ± 7783. The R0 value with added noise was very hard to measure, since the generated pulse width is in the order of hundrets of microseconds and the communication speed with the HF-A was only 9600 bauds. It can be seen that the case of fpeF2
= 4 MHz with arbitrary noise present had two events detected later than 60 µs.
