Ground Station Activity Planning through a

Multi-Algorithm Optimisation Approach

Giuseppe Corrao∗, Roberta Falone†

∗Satellite Systems and Applications, † Mission Planning - Telespazio SpA∗Via Tiburtina, Rome, † Fucino Space Center, Ortucchio, Italy

Email: giuseppe.corrao@telespazio.com

roberta.falone@external.telespazio.com

Ennio Gambi, Susanna Spinsante

Department of Information Engineering

Marche Polytechnic University

Via Brecce Bianche 12, Ancona, I-60131, Italy

Email: {e.gambi, s.spinsante}@univpm.it

Abstract—Space missions management requires to cope witha number of different issues, among which the need of properlyhandling contacts among one or more ground antennae, and oneor more satellites belonging to a space mission. The manuscriptpresents a planning approach designed to support the automaticallocation of contact opportunities among a set of ground stationsand a set of satellites, with the aim of providing the mostefficient configurations, in terms of contact time maximisationand compliance to the requirements. The optimisation strategyherein discussed builds upon the integration of techniques basedon Genetic Algorithms, Graph Theory, and Linear Programming,in order to output the optimum, or sub-optimum, schedulingplan. Simulations show that the integration of such techniquesallows to generate conflict-free and high-performance solutionsto the scheduling problem.

I. INTRODUCTION

The management of space operations involves several issues

among which, in the context of Ground Segment activities,

the problem of managing contacts among one or more anten-

nae, with one or more space missions (each of them being

composed by one or more satellites) [1]. This condition often

leads to the temporal superimposition of two or more satellite

visibilities toward the same antenna: all but one visibilities

must be discarded, but the right solution is not always trivial

to find. The problem may be very complex, depending on

the different (and often contrasting) requirements issued by

the users, that, in this case, are represented by the satellite

operators. The first approaches to this problem were quite

trivial, i.e to build and set up a brand new serving antenna for

each new space mission, or to define single, mission-specific,

and not re-usable schedulers. The problem of automated

generation of tracking plans for ground antennas emerged with

the increased frequency in space missions development [2]–

[4]. Nowadays, one-shot solutions are no more sustainable:

while in the past space missions were limited in number, at

present we see an increasing number of Earth Observation and

Navigation missions that make the near space very crowded.

Manual approaches to planning issues are no longer acceptable

for a really effective management, especially in the case of a

network of Ground Stations that must support a constellation

of LEO (Low Earth Orbit) satellites [5].

In a typical schedule, each of the satellites that have access

to a certain Ground Station has a number of Data Downlink

�

�

�

�

�

�

�

�

�

�

�

����������������

������������������������

Fig. 1. Typical input activity schedule: some clashing services are evidenced

Opportunities (DDOs) assigned. Some of the contacts may be

conflict-free, while others may lead to conflicts, that require

proper actions to be avoided. A classical scenario is shown

in Fig. 1, where some of the time overlapping DDOs are

evidenced. In this context, the aim of a Ground Station

Planning System is to find the best arrangement of conflict-

free DDOs, to maximize performances, while accounting for

specific requirements.

This paper proposes a planning approach for the automatic

allocation of contacts between any collection of Ground

Stations, and any collection of satellites, in order to find

the most effective arrangement of conflict-free visibilities,

to maximize performances while accounting for given re-

quirements expressed through a suitable cost function. The

optimisation techniques herein presented make use of GAs

(Genetic Algorithms) [6], Graph Theory (GT) and Linear

Programming (LP) to get an optimal, or near-optimal, solution

to the scheduling problem: combining those strategies in a

single tool allows to determine an highly fitted and conflict-

free resource allocation plan. Despite GAs have been used for

solving scheduling problems in a variety of fields since long

time, due to their stochastic nature, related technical literature

about the application of such algorithms in the field of Ground

Station scheduling optimisation is not so rich. In [7], Li et.

al present a TT&C (Tracking Telemetering and Command)

task planning algorithm for multi-satellite, but it is based on

SDMA-CDMA (Space Division Multiple Access - Code Divi-

sion Multiple Access), i.e. quite classical multi-user allocation

approaches, to optimise the whole TT&C efficiency. In [8], the

978-1-4673-4688-7/12/$31.00 ©2012 IEEE

author presents a tool for ground support optimisation built

upon GA concepts; the solution herein proposed differs in

the adoption of a multi-objective strategy, and the application

of linear programming theory too. The paper is organized

as follows: Section II defines the basic requirements that

guide the development of the optimisation strategy; Section III

provides details about the different components of the planning

tools, and the design choices performed; Section IV discusses

simulations and results in a real scenario; finally, Section V

concludes the paper.

II. DESIGN REQUIREMENTS

With the aim of designing a flexible and re-usable planning

tool, suitable for application to a number of different mission

contexts, the first requirement to comply with is the multi-

mission and multi-antenna property. The planning tool shall

be able to schedule the time visibilities of several satellites,

belonging to different missions, not only according to the

availability of the ground antenna placed in the same site

where the tool is beign applied, but also by taking into account

potential contact opportunities provided by other antennas,

displaced in different sites, within the same ground network.

The other requirements considered during the development

of the planning tool came from the real world experience,

and were collected at the Italian Fucino Space Centre [9].

The generation of a conflict-free scheduling plan is a basic

requirement: the supporting activities of a ground antenna

require that antennas working at a single frequency band can

track only a single satellite at a time. Different satellites relying

on the same antenna, at the same time, generate conflicts. As

a consequence, the contact table output by the planning tool

shall avoid conflicts. Two visibility opportunities are said to

be conflicting when the following condition holds:

(LoS + ST )i > AoSj (1)

where i and j are two indexes used to order in time different

services on the same antenna, and refer to the i-th and the j-th

visibility opportunity, respectively, with j > i; LoS is the time

instant at which the satellite signal is lost (Loss of Signal);

AoS is the initial satellite signal acquisition time (Acquisition

of Signal); finally, ST is the antenna Switch Time, sometimes

also called reconfiguration time.

Other operational requirements refer to: the possibility of

providing scheduling plans also for LEO (Low Earth Orbit)

satellites; the definition of mission-specific values for the

minimum time duration of the satellite - supporting ground

antenna contact; the minimum elevation over the horizon from

which the antenna will start to track the satellite; the possibility

of specifying the relative priority of each mission with respect

to the others supported by the same antenna (where lower

values correspond to higher priority missions); the definition,

for each antenna, of a cost figure that represents how much

better is it to select that antenna with respect to others (in the

event it is possible to replace a contact opportunity provided

by a third part antenna with a contact opportunity provided by

a owned antenna). Besides all the previous requirements, the

�

�

�

�

�

�

�

����������

��������

�� � � �����

�� � � �����

�� � � �����

�� � � �����

�� � � �����

�����

Fig. 2. Phenotype to genotype mapping process

planning tool should also enable the definition of the maximum

time interval for which a satellite may remain un-tracked by

any of the antennas in a network; the minimum number of

daily contacts, for each satellite of a mission; the definition

of the antenna ST; the configuration of possible unavailability

time intervals for a given antenna; the possibility to force a

contact in case of contingency, and the necessary reallocation

of resources among the previously planned contacts.

III. THE PLANNING TOOL

The first step in the design of the planning tool consists in

obtaining the complete, and start-time ordered, list of all the

geometric visibility opportunities, computed by means of an

adequately accurate orbit calculator. The geometric visibility

of a satellite, with respect to the supporting antenna in charge

of it, joint the Switch Time of the antenna itself, represents a

so-called service.

A. Application of Graph Theory

In the framework of the proposed optimisation strategy,

the GT is used to set-up the initial conditions for the GA

execution, by resorting to the concept of Adjacency Matrix

(AM).

The GA approach requires to establish a phenotype to geno-

type mapping. In our application, the phenotype is represented

by the list of all the services; it has to be mapped into a

chromosome, the length of which equals the cardinality of the

service list. The i−th service is mapped into the i−th gene of

the chromosome. Genes may assume only binary values (0, 1),depending on the corresponding service being scheduled or not

by the algorithm. The mapping process is depicted in Fig. 2.

The AM allows to randomly generate sets of conflict-free

chromosomes, recognize and classify conflicts (by using the

AM, a list of conflicts is generated, and relevant property

classification performed), and to define a so-called schema.

Schemas are templates that identify a subset of strings with

similarities at certain string positions. They allow the gen-

eration of conflict-free random populations, that drive the

evolutionary process until a successful end. A schema may be

thought of as what identifies a subset of population elements

with given similarities.

Let us assume that the list of services provided by the orbit

calculator includes N services. An N×N AM is set up, so that

�

�

�

�

�

�

�

�����

������

�����

�����

������

������

Fig. 3. Scheduling diagram with clashing services and corresponding AM

each i-th row of the matrix represents the clashing conditions

of the i-th service with respect to all the other services in the

list. As a consequence, the matrix will exhibit the following

properties:

ai,j = 1, for i = j

ai,j = 0, for i < j

and

ai,j = 0, for i > j

if the i− th and j − th services are clashing ones,

ai,j = 1, for i > j

otherwise.

Fig. 3 shows a sample AM corresponding to the clashing

services reported in the scheduling diagram.

At this point, it is possible to introduce a schema, that is

generated by generating strings of values of length N (the

number of services in the list). A one-to-one relationship is

established between the elements in a schema and the genes

in a chromosome. The possible values in a schema are:

0 : the corresponding service is not selected;

1 : the corresponding service selection is forced, irrespec-

tive to subsequent outcomes provided by the algorithm;

-1 : services corresponding to these positions will be sub-

jected to true optimisation, as they are not constrained to

pre-defined decisions (don’t care).

By the definition of a schema, the optimisation search space

is reduced from 2N to 2N−k, being k the number of positions

anchored by the schema.

B. Application of Genetic Algorithm

GA is an optimisation-and-search technique based on the

principles of genetics and natural selection: a set of candidate

solutions (the current population) are allowed to evolve under

determined natural selection rules, to a state that maximizes

the associated fitness, i.e. minimizes a given cost function.

The search space is iteratively explored, so that the candidate

solutions are evaluated, and then used to generate new ele-

ments in the population (i.e. the next generation). The process,

that may include selection, mating and offspring’s generation,

and mutation, comes to an end due to predefined stopping

conditions.

1) Generation of initial random population: The generation

of the initial random population requires the chromosomes

belonging to it to be conflict-free. In order to comply with this

constraint, the AM is again invoked: an index k ∈ [0, n− 1]is almost randomly chosen (greater probability is assigned to

lower index values, in order to keep limited the number of 0’s

in the initial chromosomes, i.e. the number of lost services).

Then, genes in 0 to k − 1 positions are set to 0, whereas for

positions k to n, the AM is analysed, in order to find possible

non conflicting services. Each chromosome generated by the

process is then compared to the schema previously defined:

values in the randomly generated chromosome, corresponding

to positions where −1’s are present in the schema, are left

unaltered. Chromosome values in positions corresponding to

a 0 or a 1 in the schema are constrained to the values defined

by the schema itself.

2) Fitness definition and evaluation: In a GA approach,

each possible solution, i.e. each chromosome, is evaluated with

respect to a fitness function, based on targets and constraints

the solution has to comply with. When a solution does not

comply with targets and constraints, it is assigned a penalty,

that increases the value of the requested fitness associated to

that solution. In the present work, the fitness function is a

multi-objective one, that accounts for the requirements on:

supporting antenna cost (AC), satellite priority (SP), maximum

time gap between two consecutive contact opportunities (TG),

and minimum number of passes in a given time interval (NP).

The Global Fitness (GF) function is consequently defined as

a weighted polynomial:

GF = wC ·AC + wP · SP + wG · TG+ wN ·NP (2)

where each single fitness function has been normalized. The

relative values assigned to the weighting coefficients (wC , wP ,

wG, and wN ) may vary, depending on the optimisation target;

in our study we considered wC = wP = 1, and wG = wN =10. In general, there are as many objectives as the number of

mission requirements.

3) Verification of the stopping conditions: Before moving

to the generation of subsequent populations, it is necessary to

verify if the stopping conditions are satisfied or not. In the

framework of the proposed planning tool, three stopping con-

ditions are defined: 1) the number of populations generated by

the iterative execution of the algorithm reaches the maximum

value fixed at run time; 2) the average fitness value of the

current generation reaches the lower limit value at run time;

3) the maximum variation of the average fitness value is lower

than the value defined at run time, for at least three successive

iterations. If none of these conditions is verified, the iterative

execution of the selection - reproduction - evaluation processes

gets started.

4) Selection: The selection process aims at selecting pop-

ulation elements that will be included in the Mating Pool, to

enable the reproduction phase. Two techniques are available

�

�

�

�

�

�

�

����������

�����

��

���

Fig. 4. Limited overlap among clashing services

for the selection process: trunking and tournement selection.

The former is a deterministic strategy that allows to select the

best k elements in a population, according to their GF values

(lower values correspond to better solutions); the latter puts k

elements, randomly selected from the current population, in a

competition. The winner element enters the Mating Pool. The

GF values are used for selection also in the second case.

5) Reproduction: In the present context, the reproduction

step is implemented by means of the following operations:

1) single point crossover: irrespective of the point at which

a chromosome is broken and re-combined to the other parent

chromosome, the children chromosomes will verify the condi-

tion of representing conflict-free solutions (specific checks are

provided to guarantee this property); 2) mutation: a specified

amount of the don’t care genes may be mutated, always

keeping the chromosomes as conflict-free; 3) elitism: the best

chromosome in the current population is chosen to replace

the worst element of the next generation; by this way, an

improving evolution is ensured, and subsequent populations

will never be worse than previous ones.

Once the new population has been generated, the iterative

process starts again, from the fitness definition and evaluation

step.

C. Application of Linear Programming

The GA poses some limitations to the solution of the

optimisation problem, that may be overcome by means of a

subsequent LP step. As a matter of fact, due to the mapping

strategy applied, the GA may only select one out of two

conflicting services, without taking into account the amount

of their overlap. The GA is not able to select the partial

scheduling of one or both the passes, even when they feature

a very small overlap, as shown in Fig. 4.

Let us define a conflict set as the set of services that

are clashing with the service considered; a conflict domain

is an uninterrupted sequence of clashing services, so that

each service in the sequence is clashing at least with another

service, in the same sequence. Let real variables x1, x2, . . . , xn

represent the contact times between n satellites and the same

antenna. The application of LP is possible if the variables

belong to the same conflict domain. We aim at maximising the

objective function z = f(x1, x2, . . . , xn), in order to maximise

the antenna contact times, i.e. we have to solve for max

z = c1x1+c2x2+. . .+cnxn, where the constraining conditions

�

�

�

�����

����������

���

���

���� ����������� ����

���� ����������� ����

���� ���������� ���� ����������

����

����

����

����

�����

Fig. 5. a) Constraints definition, b) Service time duration

on each service xi are generated at run-time, according to

the configuration of the conflict domain. For the example

shown in Fig.s 5 a) and b), we have to solve for max

z = c1x1 + c2x2 + c3x3 + c4x4, where {x1 . . . x4} represent

the contact times allocated to satellites {1 . . . 4} respectively,

and the following constraints hold:

x1 + x2 ≤ SStopT2 − SStartT1

x1 + x2 + x3 ≤ SSTopT3 − SStartT1

x1 + x2 + x4 ≤ SStopT4 − SStartT1

x2 + x3 ≤ SStopT3 − SStartT2

x2 + x4 ≤ SStopT4 − SStartT2

x3 + x4 ≤ SStopT4 − SStartT3

further: x1 ≤ ∆t1, x2 ≤ ∆t2, x3 ≤ ∆t3, x4 ≤ ∆t4.

SStartTn and SStopTn represent the service start time and

stop time, respectively; ∆ti is the i − th service duration.

By resorting to the application of the simplex algorithm it is

possible to solve also very complicated clashing conditions.

D. Setting the weighting coefficients of the multi-objective

fitness function

The weighting coefficients applied to maximise the multi-

objective fitness function are provided by the GA to the LP

process. Such coefficients are run-time computed on the basis

of the whole global planning period, despite the local scale of

the LP optimisation. The LP process deals with the solution

proposed by the GA step: if no conflicts are present, the

LP process accepts the GA output as a valid one. On the

other hand, if the GA outputs a conflicting solution, the LP

process sets multiplicative factors to increment the weighting

coefficients of the services that have been scheduled by the

GA step, within the clashing event. According to this strategy,

even if the inclusion of such services into the final schedule

is not ensured, such services will be somehow ”favoured” by

the allocation of higher weighting coefficients, during the LP

optimisation process. A further recursive optimisation routine

is executed, if the time assigned by the LP process to a

given contact is less than the minimum one set during the

�

�

�

�

�

����������������

������������������������

Fig. 6. Short portion of the schedule generated by the planning tool on theMatera 2 antenna

planning tool configuration. In such a case, the whole service

is discarded from the conflict domain, and the LP optimisation

is repeated on the reduced domain.

IV. SIMULATIONS AND RESULTS

To test the effectiveness of the proposed approach, a soft-

ware tool has been developed from scratch, by the design of

a Java class library that implements the planning engine and

exposes a Graphical User Interface (GUI) for process manage-

ment. First, a real case scenario has been evaluated, referred

to the National Multimission Center - Matera Ground Station

(Italy). Then, a highly conflicting condition is simulated for

the BTS-2 antenna at Fucino Space Center (Italy).

A. Multimission Scenario at Matera Ground Station

Table I shows a subset of the Matera operational context;

for the purpose of testing the proposed planning optimisation

tool, let us focus on the Matera 2 antenna, that may experience

a number of potentially clashing services. By applying the

planning tool on a schedule of one week duration, a list

of services is obtained. Some of the services are totally or

partially scheduled, others are not scheduled at all. Fig. 6

shows a very short portion of the schedule obtained.

After the schedule has been generated, the planning op-

timisation tool is applied. Let us analyse the behavior of the

planning optimisation tool with respect to the clashing services

Oceansat 2 and Envisat. The conflict is solved by allocating

the Envisat service, and totally excluding the Oceansat 2

one, at a parity of their priority. The GA step selects the

Envisat pass as the winning one, based on the GA parameters

configured before running the planning tool, as reported in

Table II. The winning service is then passed to the LP step

for the subsequent optimisation. The remaining contact time,

that could be allocated to the Oceansat 2 service, is actually

too short and does not comply with the requirement on the

minimum contact time of 6 minutes, that’s why the Oceansat

2 service is not scheduled at all.

Terra and Envisat services are also clashing on the Matera

2 antenna. In this case, the Envisat service has higher priority

than the Terra one, and their overlap time interval is very

short. The planning optimisation tool assigns the resource to

the Envisat service (based on its priority). However, because

the remaining contact time is enough for the Terra service (i.e.

it is greater than 6 minutes), the latter is partially scheduled.

This result is enabled by the application of the LP step, that

avoids service denial when it is possible to save enough contact

time from an original clashing condition.

TABLE IMATERA MULTIMISSION CENTER RESOURCES ALLOCATION (SUBSET)

Antenna Mission Satellite Priority

Matera 2 Aqua Aqua 3

Oceansat 2 Oceansat 2 1

Envisat Envisat 1

SAC-D SAC-D 4

Terra Terra 2

Kiruna 1 Envisat Envisat 1

Cosmo Skymed Cosmo Skymed 1 1

Cosmo Skymed Cosmo Skymed 2 1

Cosmo Skymed Cosmo Skymed 3 1

Cosmo Skymed Cosmo Skymed 4 1

Svalbard 3 Envisat Envisat 1

TABLE IIGA AND LP PARAMETERS CONFIGURATION IN Matera 2 ANTENNA

PLANNING OPTIMISATION

Parameter Configuration

Selection Mode Truncation

Cut off value: 50

Stopping Conditions Max # generations: 100

Min Fitness Threshold: 1.0E − 4

Max Fitness Variation: 1.0E − 4

Sorting Mode Global Fitness

Global Fitness wC = 1, wP = 1

Polynomial Weights wG = 10

Population Size # chromosomes: 4000

Contact Period Minimum Time: 6 min

B. Simulated conflicting condition at Fucino Space Center

In a second set of simulations, we test the performance

of the proposed optimised planner in a highly conflicting

scenario. The BTS-2 antenna at the Fucino Space Center is

at present devoted to support the GIOVE-B satellite only, but

it is expected that in the near future the same antenna will

also support the PRISMA mission. As a consequence, it is of

interest to simulate what would happen in case a number of

different clashing services need to be supported by BTS-2.

The switch time of the antenna is 300 s, its related cost

parameter is wC = 1, and we assume the antenna is the

only active one. We also assume that there are eight satellites

requesting support from BTS-2 (the same of Table I, with

the exception of Envisat). The planning tool is applied on a

schedule of 5 days, and a list of services is generated, some of

which are scheduled or partially scheduled, while others are

not scheduled at all. The generated list of services is shown

in Fig. 7. Following the schedule generation, the optimisation

process is applied. The parameters requested by the GA are

set according to Table II, with the exception of the maximum

number of generations that is limited to 50. It is also requested

that all the non-conflicting passes are included in the schedule.

Looking at clashing satellites Cosmo Skymed 1 and SAC-D,

thanks to the different priorities they feature, the optimisation

�

�

�

�

Fig. 7. The 5-days schedule generated by the planning tool on the BTS-2 antenna, for the eight clashing services simulated

Fig. 8. Optimised 5-days Gantt schedule on the BTS-2 antenna

tool is able to allocate all the passes available to Cosmo

Skymed 1. For the SAC-D satellite, some of the passes

result to be partially scheduled: this is obtained thanks to

the LP optimisation step, that is able to recover some contact

opportunities (that otherwise would have been denied), when

their remaining time is longer than the minimum contact time

requested of 6 minutes. The final schedule shows how all the

requirements have been satisfied, thanks to the optimisation

tool. No clashing services are now present, as confirmed by

the Gantt diagram in Fig. 8.

V. CONCLUSION

The manuscript presented a multi-algorithm planning ap-

proach to support the automatic allocation of contact oppor-

tunities among a set of ground stations and a set of satellites,

with the aim of providing the most efficient configurations,

in terms of contact time maximisation and compliance to

the requirements. The results obtained in real and simulated

operational contexts are strongly encouraging. The proposed

algorithm assesses complex situations in a very rapid manner,

and gives an optimal (or near optimal) solution, by supplying a

strongly fitted schedule that provides maximisation of revenues

(when the satellite contacts are sold to third parties), optimal

handling of antenna duty cycle, and assessment of Ground

Station location (during the design phase). The planning capa-

bility herein described is of great interest either for operational

and design purposes, and it is expected to have a significant

impact on practical adoption.

REFERENCES

[1] S. Damiani, H. Dreihahn, J. Noll, M. Niezette, and G. P. Calzo-lari, ”A Planning and Scheduling System to Allocate ESA GroundStation Network Services,” Proc. of The International Conference onAutomated Planning and Scheduling, Providence, Rhode Island, USA,September 22nd - 26th, 2007, available at http://abotea.rsise.anu.edu.au/satellite-events-icaps07/demos/4/ddnnc.pdf.

[2] Chien, S., Lam, R., Quoc V., ”Resource scheduling for a network ofcommunications antennas,” Proc. of the IEEE Aerospace Conference, BigSky (MT, USA), 1 - 8 Feb. 1997, Vol. 1, pp. 361 - 373.

[3] Chien, S., Govindjee, A., Estlin, T., Wang, X., Hill, R., ”Automatedgeneration of tracking plans for a network of communications antennas,”Proc. of the IEEE Aerospace Conference, Big Sky (MT, USA), 1 - 8 Feb.1997, Vol. 1, pp. 343 - 359.

[4] Whitley, L.D., Howe, A.E., Barbulescu, L., Watson, J.P., ”SchedulingSpace Ground Communications for the Air Force Satellite ControlNetwork,” Journal of Scheduling, Vol.7, pp. 7 - 34, 2004.

[5] C.H. Lee, K.-M. Cheung, ”Mixed integer programming & heuristicscheduling for space communication networks,” Proc. of IEEE 2012Aerospace Conference, 2012 IEEE, pp. 1 - 10.

[6] J. H. Holland, (1975), Adaptation in Natural and Artificial Systems,University of Michigan Press, Ann Arbor.

[7] J. Li, J. Bai, J.-P. Liu, P.-J. Yu, ”The TT&C Task Planning AlgorithmBased on Multi-satellite,” Proc. of 2010 International Conference onArtificial Intelligence and Computational Intelligence (AICI), 23rd-24th

Oct. 2010, Vol. 3, pp. 556 - 559.[8] T.P. Bagchi, ”Near Optimal Ground Support in Multi-Spacecraft Mis-

sions: A GA Model and its Results,” IEEE Transactions on Aerospaceand Electronic Systems, Vol. 45, Issue. 3, 2009, pp. 950 - 964.

[9] See http://www.telespazio.it/.