To achieve this goal I have followed two approaches:
* A control–communications separation paradigm (depicted schematically in Image 2) that relies on anytime-reliable codes and adaptive quantization techniques that are matched to the control setup. The quantization techniques can be further subdivided into variable- and fixed-rate quantization. After getting familiar with the relevant literature on control and low-delay communications, I was able to enhance the theory and performance of all these ingredients:
- Anytime-reliable tree codes over binary-input output-symmetric channels with low-complexity encoding and decoding [with Dr. Wael Hablawi and Prof. Babak Hassibi from Caltech].
- Anytime-reliable tree codes over binary-input output-symmetric channels with instantaneous feedback over the binary symmetric (bit-flip) channel [With Dr. Anusha Lalitha and Prof. Tara Javidi from UC San Diego, and Prof. Victoria Kostina from Caltech].
- Greedy algorithm fixed-rate quantization for control for disturbances with log-concage distributions (Gaussian included, namely linear-quadratic Gaussian control) [with Dr. Yorie Nakahira, Mr. Yu Su, and Prof. Babak Hassibi, all from Caltech].
- Event-triggered control: As in the previous setting but when the encoder may decide not to send any message at a particular time [with Mr. Yu Su, Mr. Hikmet Yıldız, and Prof. Babak Hassibi, all from Caltech].
- Bounds on the performance and an achievable variable-rate scheme for tracking and control over rate-limited links in the presence of packet erasures and acknowledgment signals (instantaneous or delayed by one time slot) [with Prof. Victoria Kostina and Prof. Babak Hassibi from Caltech, and Prof. Ashish Khisti from the University of Toronto].
- Bounds on the performance of control over rate-limited channels in the presence of side-information at the controller [with Mr. Omri Lev from Tel Aviv University].
- Bounds on the performance without control-objectives knowledge at the sensor [with Mr. Omri Lev from Tel Aviv University].
* A joint control–communications paradigm that makes use of low-delay joint source–channel coding techniques. Finding an optimal solution for the latter remains an open problem although good (suboptimal) solutions exist (see, e.g. Image 3). I was able to develop a principled way of applying such techniques to control and translate their communication cost (distortion) performance to those of control cost of the entire system. This work has been carried with collaboration with Prof. Babak Hassibi (Caltech), Prof. Victoria Kostina (Caltech), Mr. Gustav M. Pettersson (KTH), and Mr. Elias Riedel Gårding (KTH). I was further been able to design non-linear schemes that make judicious use of side information at the controller and/or without control-objectives knowledge at the sensor's end. This latter part has been carried with Mr. Omri Lev at Tel Aviv University.
For a comparison of the quadratic control costs achievable using the joint and separation-based CONCOM schemes (without side information and with control-objectives knowledge at the sensor) see Image 4.
Finally, while working on this project, I have been exposed to concerns coming from both academia and the industry regarding the vulanerability of CONCOM system manifested by recent attacks on several infrastructure facilities: the attack on the sewage system in Maroochy Shire, Australia; the Ukraine power grid cyber-attack; the German steel mill cyber-attack; the Davis-Besse nuclear power plant attack in Ohio, USA; and the Iranian uraniu-enrichment facility attack via the Stuxnet malware. To secure cyber-physical systems from such attacks, I have offered along with Dr. Mohammad Javad Khojasteh, Prof. Massimo Franceschetti, and Prof. Tara Javidi ― my colleagues from UC San Diego ― a new way to authenticate systems by relying on the learning limitations of an attacker that tries to learn the system parameters in order to successfully run a sophisticated attack.
