I am interested in designing a robust controller for a small satellite with antenna, sloshing propellant and flexible panels on board. I am mostly interested in achieving arcsec precision pointing given such a scenario. Avoiding microvibration and jitter analysis from reaction wheels.

  • $\begingroup$ This is a super interesting question! What sensors do you imagine this using? $\endgroup$ Mar 13, 2019 at 17:28
  • $\begingroup$ Hi @Knudsen! Just the usual, nothing special. Sun sensors, magnetometers, IMU, maybe a star tracker. $\endgroup$
    – Watch This
    Mar 14, 2019 at 19:48
  • $\begingroup$ If you are dealing with sloshing you may need something with a higher update frequency. Have you looked into Model Reference Adaptive Control (MRAC) $\endgroup$ Apr 9, 2019 at 15:42
  • $\begingroup$ Hi @KnudsenNumber! Thank you for your reply. It still is a Nanosat and, unlike CryoCube-1, has non-cryo propellant on board. The elimination of sloshing is to mitigate the last bit of arcsec numbers, amongst others micro-vibration/jitter. IMUs on board should cover the high freq sensing. Traditionally folks from the space domain have been using H-inf for GEO/Telecom sats with flexible appendages. It gives perfect support for model uncertainty. Even demonstrated marcsec precision in simulations.... $\endgroup$
    – Watch This
    Apr 9, 2019 at 16:42
  • $\begingroup$ ....But lately for drag free attitude control of LISA mission (12DoF) people have been looking into QFT. Which is also a robust strategy. Was wondering what the differences are. I have looked into MRAC earlier for platooning of cars and velocity vector fields of UAVs, but not for arcsec flexible body pointing. What are your justifications for using MRAC for an arcsec flexible nanosat mission? Thanks for getting back at the topic ;) $\endgroup$
    – Watch This
    Apr 9, 2019 at 16:42

2 Answers 2


Among Quantitative Feedback Theory and H infinity, which one is preferred for robust control of satellites esp. small satellites? And why?

AFAIK Neither.

First, you should define what you mean by "small satellite", the definition can vary between 30kg to 300kg.

Second, in a small satellite, panels are generally mounted over faces rather than on deployable surfaces to avoid flexible mode disturbance, also flexible modes are usually specified to exist only in high frequencies, partially due to launcher restrictions. You can have guidance laws that prevent high torques from exciting the first mode, but that basically means limiting acceleration when needed.

Antennas in most small satellites are not steerable because this kind of mechanical part is prone to failure.

Many "small" satellites have no propulsion, so no problem with sloshing either. Some of the ones that do have propulsion also rely on fuel tanks with rolling diaphragm, so sloshing is severely attenuated as well.

Jitter is then mostly driven by reaction wheels, whose average speed in orbit requires very complex and mission specific analysis. However, jitter happens at high frequencies, above 100 Hz, while I'm yet to hear about any AOCS software running above 64Hz. Most of systems I know run at 10Hz or below. This means that no mater what control technique you are using, there is no way you can control jitter.

If you were to consider a large spacecraft such as GeoEye-1 then these techniques could be closer to making sense.

Finally, "arcsecond accuracy" is with respect to what? An inertial direction or maybe a location on Earth? If it is the latter, there are many other error contributors, including but not limited to time synchronization on board and Earth ephemeris prediction and computation.

So case in point, unless you are working with an academic problem, I doubt you gain much from robust control theories. You might as well use them, but when you find yourself in times of trouble, mother Mary will likely tell you to redesign the spacecraft rather than change your control algorithms.

  • $\begingroup$ This doesn't even attempt to answer which of QFT and H Infinity is better. Obviously you can design systems that are easier to control, but if you are stuck with something that is difficult then you have to deal with it. BTW pointing accuracy is never defined with respect to something. It is the ability to accurately point at any arbitrary point in space. $\endgroup$ Sep 16, 2019 at 20:51
  • $\begingroup$ @knudsen he does make a good point about you not having specified much about the satellite. It would be easier to answer if we knew if you spoke of a cubesat or something bigger than a human. $\endgroup$ Sep 17, 2019 at 12:52
  • $\begingroup$ @MagicOctopusUrn the OP said a nanosat, which is generally the same language used for CubeSats. If you want more information writing an answer saying the question is incomplete, makes no sense. The comment sections are for asking the OP questions and improving knowledge. $\endgroup$ Sep 18, 2019 at 1:13
  • $\begingroup$ Thank you for your comments! Well, imagine the case of NEA Scout. So a NanoSat with large solar sails and possibly underactuated actuators or lack of torque power availability. In such cases, robust control theory provides better confidence in your solutions. Sorry for asking too many questions on one post, viewing these problems separately before combining them is a better alternative. If you want to extend the capabilities of the current satellites, at times you might end up with a very unconventional design. A de-orbiting satellite with complex robotic arms/flexible nets/harpoon for e.g $\endgroup$
    – Watch This
    Feb 2, 2020 at 10:59
  • $\begingroup$ Consider a space tug/deorbiting satellite (25- 50 kg class), with some robotic arms/flexible nets/ harpoon ( something that gives rise to flexible modes), and half of the satellite mass is fuel because de-orbit. The other satellite has unknown inertia/modal properties and is tumbling around a single axis. And you wanna dock precisely, so good precision wrt body pointing. In such cases, robust control strategies like fixed-order H-inf or LPV will save you big time. $\endgroup$
    – Watch This
    Feb 2, 2020 at 11:08

The answer is both, and none at the same time. Control first, and the system later is the wrong way of looking at things. Model/system characterization/identification always comes first, before you synthesize a controller. It is the system characteristics that drive the control design process.

Some previous work on QFT and H-infinity controller comparison: doi.org/10.1115/1.2807067

More typical to the industry is the utilization of H-inf controllers. Why? My guess is that there is enough industrial support/heritage surrounding it, compared to QFT. Take it with a grain of salt though


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.