CONCOM: Control Over Noisy COmmunication Media

Research Fellow: Anatoly Khina
This project has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Składowska-Curie grant agreement No 708932.

Background

Two of the fundamental theories within the discipline of Electrical Engineering are those of Control and Communications (along with its mathematical foundation—information theory). These theories have been studied extensively by mathematicians and engineers throughout the 20th century, and have gone separate ways. The primary objective of control theory is to stabilize and control the behaviour of a given dynamical system in a desired fashion, by changing the system input according to the measured output of the system (“feedback”). In this theory, adapting according to the measured feedback with as small delay as possible is of grave importance, especially when the system is unstable in the absence of feedback. The theories of communication and information deal with conveying reliably data over noisy media (or “channels”). Information theory seeks to determine the maximal reliable-communication rates possible, disregarding and often undermining delay and complexity of computation. Communication theory attempts to approach the rates promised by information theory using “practical tools”. In past decades, control theory was mainly used in well-crafted closed engineering systems (e.g., car and aerospace industries). In the current technological era of ubiquitous wireless connectivity, the demand for control over (noisy) wireless media is ever growing, enabling numerous new possibilities, both teleoperation technologies in which an expert controls the input of the system and automated technologies where constantly collected data is analyzed and acts with minimal human intervention.

Another issue that is encountered more and more in practice is that of attacks on cyber-physical systems (CPS) (that comprise physical elements that are connected and controlled via communications links). A few examples include:

Goal

The grand goal of this research is to build a unified framework for communication and control (CONCOM) that allows to determine the fundamental limits of stabilizability and control performance of dynamical systems over noisy communication links as well as designing practical schemes and codes that approach these limits.

To achieve this goal we shall follow two approaches:

Separated CONCOM design: This approach follows the OSI model and separates the CONCOM design into two parts: A communications task that transforms the communications link to a noiseless finite-rate one (albeit with random delays), and a control task over the resulting noiseless finite-rate links. However, even if one conforms to this traditional separation paradigm, it turns out that standard compression (quantization) and error correcting techniques fail to stabilize unstable control systems. The underlying reason for this is that any inaccuracy in the controller's action, on account of the noisy link, is magnified at every subsequent time step due to the unstable nature of the system. Or in Aristotle's words: The least initial deviation from the truth is multiplied later a thousandfold.
Joint CONCOM design: Due to the inherent delay of separation-based solutions, they are prone to be suboptimal when applied to the real-time CONCOM settings (as well as very demanding in computation and memory resources). A more ambitious task is therefore to design a scheme that carries out the entire CONCOM task jointly. This has the potential to
- Achieve far better control performance and provide stronger control guarantees.
- Require far less computational and memory resources.
- Allow graceful improvement in performance with the decrease of the noise power, in contrast to separation-based schemes whose performance saturates due to their digital nature.

To protect such systems against cyberattacks, we would also like to construct authentication and security mechanisms on the physical side of the systems, to supplement the currently employed cybersecurity ones.

Potential Impact

Autonomous Vehicles

Photo taken from this NBC news article.

Check this MIT tech review about self-driving cars.

Autonomous driving and active safety are the two major objectives of the car industry today. As car accidents are one of the leading causes for death worldwide, designing smart cars that react to the changing environment via smart sensors ("feedback") to avoid accidents is a global top priority. Additionally, a fully automated driving system can also increase dramatically the efficiency of existing transportation infrastructure and reduce the ever growing need for new infrastructure.

Remote Surgery (and Other Procedures)

The remote-controlled Da Vinci robot surgeon performing surgery.

Photo taken from the following interesting BBC article.

Carrying out medical surgery remotely via a controlled robot has been successfully applied in the western world in recent years. These procedures require accurate and timely actions, which translate into high reliability, high controllability and low delay. Thus, present-day solutions use expensive infrastructure (fiber-optic connectivity and designated hardware) and are mainly used to enhance the capabilities of surgeons and to minimize the adverse effects of invasive procedures performed on patients that are located physically close to the physician providing the treatment. A unified CONCOM theory should allow the same over standard infrastructure (wireless media and off-the-shelf computers), and to much longer distances. This will make surgeries and other medical treatments accessible and widespread to all, allowing, among others, to facilitate organizations such as Médecins Sans Frontières treating patients in rural areas and developing countries remotely from the other side of the globe and allow to volunteer part time. Moreover, this will likely result in a large increase in volunteers that will be able to volunteer part-time without the need to relocate. Other (non-medical) procedures, e.g. cosmetic procedures, will enjoy similar growth in accessibility using this technology.

Neuroscience

For an explanation about this experiment (and more) you are invited to watch this talk by John Doyle.

Neuroscience is a field where many distributed CONCOM scenarios are encountered. The CONCOM framework has the potential to shed light on various brain functionalities and tradeoffs. One such illustrative example (demonstrated by Halle, above) is as follows. The maximal speed at which text can be read, say, from your smartphone, when moving it from side to side is lower than the maximal speed at which the same can be done when moving the head horizontally instead. The discrepancy between these two speeds is the result of different CONCOM mechanisms that are present in the visual system: head motions are tracked and compensated for by a fast low-resolution vestibule-ocular reflex; the cortex is responsible for the second slower high-resolution loop which accounts for the hand ("target") movement.

Ocean Exploration

Photo taken from the following Bold Business article.

For more details, check:

The Nippon Foundation—GEBCO Seabed 2030 Project webpage and this concept paper.
The UN Decade of Ocean Science for Sustainable Development webpage
The following paper which is part of “Oceannobs'19: An Ocean of Opportunity”.

Ocean exploration and seabed mapping: Although the ocean covers more than 70% of the globe, less than 10% of its seabed has been explored to a significant degree. Exploring the ocean depths—earth's last frontier, as described by UNESCO—is likely to reveal new sources for energy, medical drugs, and food, as well as facilitate our understanding of the marine biodiversity, and prediction of natural disasters and climate change. Ocean exploration and seabed mapping are carried through coordinated effort of swarms of autonomous underwater vehicles (AUVs). Since electromagnetic underwater communications is prone to severe attenuation, distributed control of these AUVs needs to be carried using acoustic (mechanical) waves, the propagation speed of which is merely (approx.) 1500 m/sec.

Nanobrobots & Nanomedicine

Photo taken from the following New Atlas article and is based on the following Nano letters publication.

For a review of recent advances in the design of nanobots, check this article.

Recent advances in biology, robotics, nanomaterials and nanoelectronic devices, bring us closer to fulfilling Hibbs's and Feynman's vision of nanomedicine or, as they put it to “swallow the surgeon”, namely, designing nanorobots that monitor, deliver drugs, operate and more [12]. Due to their size and the distributed nature of the tasks that they need to perform, swarms of nanorobots that work cooperatively need to be developed, where again the control tasks need to be communicated over fluids using acoustic waves.

Achievements

Separated CONCOM design

We constructed encoding and decoding techniques of error-correcting codes that are good for control (with high probability!) and implemented them in Python:

A. Khina, W. Halbawi, and B. Hassibi, “(Almost) practical tree codes,” Technical Report , Oct. 2017. [Paper]
Sequential stack decoder [Python3]
Sequential stack decoder [Python2]

We further constructed adaptive fixe-rate quantizers for linear LQG control over noiseless finite-rate links that are greedily optimal and implemented them in Python:

A. Khina, Y. Nakahira, Y. Su, H. Yıldız and B. Hassibi, “Algorithms for optimal control with fixed-rate feedback,” Technical Report , Sep. 2018. [1 col] [2 col] [@arXiv]
H. Yıldız, Y. Su, A. Khina, and B. Hassibi, “Event-triggered stochastic control via constrained quantization,” Data Compression Conference (DCC), Snowbird, UT, USA, p. 612, Mar. 2019. [Paper] [@IEEE Xplore]
A. Khina, Y. Nakahira, Y. Su, and B. Hassibi, “Algorithms for optimal control with fixed-rate feedback,” in IEEE Conference on Decision and Control (CDC), Melbourne, VIC, Australia, pp. 6015–6020, Dec. 2017. [Paper] [@IEEE Xplore] [Slides]
A. Khina, Y. Nakahira, and B. Hassibi, “LQG Control with fixed-rate limited feedback,” Information Theory and Applications (ITA) Workshop, La Jolla, CA, USA, Feb. 2017.
A. Khina, Y. Nakahira, and B. Hassibi, “Control with fixed-rate limited feedback,” Control with fixed-rate limited feedback,
Adaptive quantizer for control [Python3]

For variable-rate feedback with packet drops and acknowledgments (as in TCP/IP) with possible delays, we developed bounds and constructed schemes (that are optimal or close-to-optimal in certain cases):

A. Khina, V. Kostina, A. Khisti, and B. Hassibi, “Tracking and control of Gauss–Markov processes over packet-drop channels with acknowledgments,” IEEE Transactions on Control of Network Systems, vol. 6, No. 2, pp. 549–560, June 2019. [Paper] [@IEEE Xplore]
A. Khina, V. Kostina, A. Khisti, and B. Hassibi, Sequential coding of Gauss–Markov sources,” in IEEE Information Theory Workshop (ITW), Kaohsiung, Taiwan, pp. 529–533, Nov. 2017. [Paper] [@IEEE Xplore] [Slides]
A. Khina, V. Kostina, B. Hassibi, and A. Khisti, “Rate–cost tradeoffs over lossy channels,” Asilomar Conference on Signals, Systems, and Computers , Pacific Grova, CA, USA, Oct./Nov. 2017.

We constructed a causal commmunication scheme with feebdack over noisy links that are provably good for control:

A. Lalitha, A. Khina, T. Javidi, and V. Kostina, “Real-time Binary Posterior Matching,” Information Theory and Applications (ITA) Workshop, La Jolla, CA, USA, Feb. 2019.
A. Lalitha, A. Khina, T. Javidi, and V. Kostina, “Real-time Binary Posterior Matching,” in IEEE International Symposium on Information Theory (ISIT), Paris, France, pp. 2239–2243, July 2019. [Paper]

Joint CONCOM design

We developed a joint CONCOM design for scalar LQG settings along with bounds on its performance and compared it to the separation-based scheme (with full implementations in Python!):

O. Lev and A. Khina, “Gauss–Markov source tracking with side information: Lower bounds,” in IEEE International Symposium on Information Theory and Applications (ISITA), submitted, Apr. 2020. [Paper] [Extended version] [Extended version@arXiv]
O. Lev and A. Khina, “Schemes for LQG control over Gaussian channels with Side Information,” in IEEE Conference on Decision and Control (CDC), submitted, Mar. 2020. [Paper] [@arXiv]
A. Khina, E. Riedel Gårding, G. M. Pettersson, V. Kostina, and B. Hassibi, “Control over Gaussian channels with and without source–channel separation,” IEEE Transactions on Automatic Control, vol. 64, No. 9, pp. 3690–3705, Sep 2019. [Paper] [@IEEE Xplore]
A. Khina, G. M, Pettersson, V. Kostina, and B. Hassibi, “Multi-rate control over AWGN channels via analog joint source—channel coding,” in IEEE Conference on Decision and Control (CDC), Las Vegas, NV, USA, pp. 5968–5973, July 2016. [Paper] [@IEEE Xplore] [Extended version] [Extended version@arXiv] [Slides]
A. Khina, G. M, Pettersson, V. Kostina, and B. Hassibi, “Control over Gaussian channels via non-linear analog mappings,” Southern California Control Workshop (SCCW) , Caltech, Pasadena, CA, USA, Apr. 2017.
A. Khina, G. M, Pettersson, V. Kostina, and B. Hassibi, Multi-rate control over AWGN channels via analog coding, in PARADISE Workshop, Caltech, Pasadena, CA, USA, Feb. 2017.
Control over AWGN channels code package [Python3]

Authenticatoin of Cyber-Physical Systems

We propose a new learning-limited paradigm for CPS here:

M. J. Khojasteh, A. Khina, M. Franceschetti and T. Javidi, “Authentication of cyber-physical systems under learning-based attacks,” CoRR, Jan. 2019. [@arXiv]
M. J. Khojasteh, A. Khina, M. Franceschetti and T. Javidi, “Learning-based Attacks in Cyber-Physical Systems: The Scalar Case,” IFAC Workshop on Distributed Estimation and Control in Networked Systems (NecSys), Vol. 52, no. 20, pp. 369–374, Sep. 2019. [Paper] [@ScienceDirect]

Publications

Journal Papers and Preprints

M. J. Khojasteh, A. Khina, M. Franceschetti and T. Javidi, “Authentication of cyber-physical systems under learning-based attacks,” CoRR, Jan. 2019. [@arXiv]
A. Khina, Y. Nakahira, Y. Su, H. Yıldız and B. Hassibi, “Algorithms for optimal control with fixed-rate feedback,” Technical Report , Sep. 2018. [1 col] [2 col] [@arXiv]
A. Khina, E. Riedel Gårding, G. M. Pettersson, V. Kostina, and B. Hassibi, “Control over Gaussian channels with and without source–channel separation,” IEEE Transactions on Automatic Control, vol. 64, No. 9, pp. 3690–3705, Sep 2019. [Paper] [@IEEE Xplore]
A. Khina, W. Halbawi, and B. Hassibi, “(Almost) practical tree codes,” Technical Report , Oct. 2017. [Paper]
A. Khina, V. Kostina, A. Khisti, and B. Hassibi, “Tracking and control of Gauss–Markov processes over packet-drop channels with acknowledgments,” IEEE Transactions on Control of Network Systems, vol. 6, No. 2, pp. 549–560, June 2019. [Paper] [@IEEE Xplore]

Conference Papers

O. Lev and A. Khina, “Gauss–Markov source tracking with side information: Lower bounds,” in IEEE International Symposium on Information Theory and Applications (ISITA), submitted, Apr. 2020. [Paper] [Extended version] [Extended version@arXiv]
O. Lev and A. Khina, “Schemes for LQG control over Gaussian channels with Side Information,” in IEEE Conference on Decision and Control (CDC), submitted, Mar. 2020. [Paper] [@arXiv]
M. J. Khojasteh, A. Khina, M. Franceschetti and T. Javidi, “Learning-based Attacks in Cyber-Physical Systems: The Scalar Case,” IFAC Workshop on Distributed Estimation and Control in Networked Systems (NecSys), Vol. 52, no. 20, pp. 369–374, Sep. 2019. [Paper] [@ScienceDirect]
A. Lalitha, A. Khina, T. Javidi, and V. Kostina, “Real-time Binary Posterior Matching,” in IEEE International Symposium on Information Theory (ISIT), Paris, France, pp. 2239–2243, July 2019. [Paper]
A. Khina, Y. Nakahira, Y. Su, H. Yıldız and B. Hassibi, “Algorithms for optimal control with fixed-rate feedback,” Technical Report , Sep. 2018. [1 col] [2 col] [@arXiv]
A. Khina, Y. Nakahira, Y. Su, and B. Hassibi, “Algorithms for optimal control with fixed-rate feedback,” in IEEE Conference on Decision and Control (CDC), Melbourne, VIC, Australia, pp. 6015–6020, Dec. 2017. [Paper] [@IEEE Xplore] [Slides]
A. Khina, V. Kostina, A. Khisti, and B. Hassibi, Sequential coding of Gauss–Markov sources,” in IEEE Information Theory Workshop (ITW), Kaohsiung, Taiwan, pp. 529–533, Nov. 2017. [Paper] [@IEEE Xplore] [Slides]
A. Khina, G. M, Pettersson, V. Kostina, and B. Hassibi, “Multi-rate control over AWGN channels via analog joint source—channel coding,” in IEEE Conference on Decision and Control (CDC), Las Vegas, NV, USA, pp. 5968–5973, July 2016. [Paper] [@IEEE Xplore] [Extended version] [Extended version@arXiv] [Slides]

Conference Talks/Posters (without paper)

A. Lalitha, A. Khina, T. Javidi, and V. Kostina, “Real-time Binary Posterior Matching,” Information Theory and Applications (ITA) Workshop, La Jolla, CA, USA, Feb. 2019.
A. Khina, “Control over Noisy Communication Media,” Meet the Faculty Candidate Poster Session, IEEE Conference on Decision and Control (CDC), Melbourne, VIC, Australia, Dec. 2017.
A. Khina, V. Kostina, B. Hassibi, and A. Khisti, “Rate–cost tradeoffs over lossy channels,” Asilomar Conference on Signals, Systems, and Computers , Pacific Grova, CA, USA, Oct./Nov. 2017.
A. Khina, G. M, Pettersson, V. Kostina, and B. Hassibi, “Control over Gaussian channels via non-linear analog mappings,” Southern California Control Workshop (SCCW) , Caltech, Pasadena, CA, USA, Apr. 2017.
A. Khina, G. M, Pettersson, V. Kostina, and B. Hassibi, Multi-rate control over AWGN channels via analog coding, in PARADISE Workshop, Caltech, Pasadena, CA, USA, Feb. 2017.
A. Khina, Y. Nakahira, and B. Hassibi, “LQG Control with fixed-rate limited feedback,” Information Theory and Applications (ITA) Workshop, La Jolla, CA, USA, Feb. 2017.
A. Khina, Y. Nakahira, and B. Hassibi, “Control with fixed-rate limited feedback,” Control with fixed-rate limited feedback,
- Sequential stack decoder [Python]
IEEE ICSEE, Eilat, Israel, Nov. 2016.
A. Khina, G. M, Pettersson, V. Kostina, and B. Hassibi, “Multi-rate control over AWGN channels via analog joint source–channel coding,” IEEE ICSEE, Eilat, Israel, Nov. 2016.

Other Talks

“Control over Noisy Communication Media.” Talk given in a few universities during 2017.

Software

Control over AWGN channels code package [Python3]
Sequential stack decoder [Python2]