Link to research description in LUCRIS

I  have a broad interest ranging from pure mathematics to industrial applications of control. The main theme during my career has been modeling and control of uncertain systems, i.e. systems where the dynamics is unknown and you might have noisy unreliable data. I have a PhD in control (Lund 1992) and have been a professor in Automatic Control since 1999. During the years 2001-2010 I worked at Ericsson with mobile phone systems. From this period I have an interest in combining control and communication problems. From 2010 I am working full time as (ELLIIT) professor in Automatic Control at Lund University.

Since 2016 I have been engaged in the startup and manegement of the WASP-AS research school. 

A short CV is available here

Some day I should really update this page :-)

I am part of

• ELLIIT, a network organization for Information and Communication Technology (ICT) research at Linköping, Lund, Halmstad and Blekinge 
• LCCC, the Lund center for Control of Complex Engineering Systems

The following is a list of main present and past research

Researchers:Bo Bernhardsson, Anders J Johansson (Dept. of Electrical and Information Technology), Rolf Johansson, Olof Troeng

Funding:European Spallation Source

The European Spallation Source will be a major user facility at which researchers from academia and industry will investigate scientific questions using neutron beams.

Neutron methods provide insights about the molecular building blocks of matter not available by other means. Applications include research in life science, soft condensed matter physics, chemistry of materials, fundamental particle physics and engineering materials. More info here.

RF Field Control

The neutrons are produced by colliding high-speed protons with a rotating tungsten target. The protons are accelerated by oscillating electro-magnetic fields in more than 150 radio-frequency cavities along the 482 meter long linear accelerator. In order to avoid defocusing of the beam (which leads to activation of the accelerator structure), it is important that the amplitude and phase of the field in every cavity are kept very close to their nominal values. In order to achieve this individual feedback loops with sampling frequencies of 10 MHz will be used to regulate the fields.

The Department of Automatic Control is involved in modeling and controller design for the RF system through this group at the Department of Electrical and Information Technology.

Schematic description of the control problem for a single cavity

------------------------------------------------------------------------

with Anders Mannesson (AC), Fredrik Tufvesson (EIT), Atif Yaqoob (EIT)

The idea is to combine information from gyros, accelerometers and compass sensors with radio channel estimation and to determine the fundamental properties of such schemes. The hope is to enable two technology advancements:

• a new method for significantly reduced drift in low cost navigation systems
• improved radio channel estimation for moving terminals

Initial research show promising result as presented in the licentiate thesis by Anders Mannesson and in this presentation

The work is based on angle of arrival estimation using antenna arrays which is a well studied problem with many different algorithms resolving the individual rays impinging on the array. However, less attention has been given to so called virtual array antennas where only one receiver element is used. By tracking the movement of the element, an array with properties similar to a stationary array with multiple elements is formed. By combining the IMU and the radio channel information, a map of the local radio environment can be obtained, such as in this picture illustrating the complex valued radio channel for three impinging planar radiowaves in a noise-free situation

Tightly coupled nonlinear state estimation algorithms between IMU signals and radio channel estimates are used to simultaneously estimate this map and obtain improved pose estimates.

The major challenge is to battle the drift in sensors and radio crystal oscillators. We work with both real-world measurements and simulations to evaluate performance. Initial experiments show promising results. The following figure shows performance (simulations left, real experiments right) with realistic radio and IMU imperfections. There is a dramatic improvement by including radio channel information compared to using dead reckoning, especially for movements longer than 10 seconds.

Our future research now focuses on improving radio channel estimation and prediction by adding IMU information and using motion models.

The goal is to model and control large loudspeakers, so called marine vibrators, that are used to generate acoustic underwater signals. 

Performance requirements on out-of-band spectrum of the acoustic signal are achieved by careful characterisation of the dynamical behavior of the vibrators and use of frequenyc domain Iterative Learning Control (ILC). This has been found  to successfully reduce the impact of nonlinearities such as friction and backlash. A 40 dB suppression of out-of-band harmonics has been acheived in experiments (in air).

The project is funded by the Norwegian company PGS.

with Erik Johannesson, Ather Gattami, Anders Rantzer, Andrey Gulchak

Classical control theory assumes perfect communication, without limitations, between different parts of the control system and the process. The trend towards large-scale distributed systems makes it interesting to study the interplay between communication and control such as fundamental limitations of control performance arising from communication constraints.

In Erik Johanesson's PhD thesis several such situations are studied. We model the communication constraints as an signal to noise ratio (SNR) constraint for the communication channel. Such constraints give an incentive to perform coding and decoding of the transmitted signal in addition to the usual filtering and computation of control signals.

The problem of designing the controller, coder and decoder simultaneously is found to be solvable by convex optimization and spectral factorization giving efficient methods for their calculation.

The situation in the following figure with an SNR constraint on the channel is studied in

• E. Johannesson, A. Rantzer, B. Bernhardsson and A. Ghulchak, "Encoder and Decoder Design for Signal Estimation," in Proc. American Control Conference, Baltimore, USA, June 2010.
• E. Johannesson, A. Ghulchak, A. Rantzer and B. Bernhardsson, "MIMO Encoder and Decoder Design for Signal Estimation," in Proc. 19th Inter- national Symposium on Mathematical Theory of Networks and Systems, Budapest, Hungary, July 2010.
• E. Johannesson, "Signal Estimation over Channels with SNR Constraints and Feedback," in Proc. 18th IFAC World Congress, Milano, Italy, August 2011.

The main result is that if encoders and decoders are restricted to linear filters the optimal ones can be found by solving an extension of the Wiener filter equations where R is in L_infinity and X in H_2. The first term is the same as in the classical Wiener filter design and the new last term, where sigma is the SNR constraint on the channel, describes the added cost of imperfect communciation.

Problems concerning feedback design is also studied as described in the following figure. Here the communication channel is modeled as having an SNR constraint with the added possibility of using also a perfect communication feedback channel (through the filter B). The joint design of coder C, decoder D, and feedback filter B is solved through convex optimization.

• E. Johannesson, A. Rantzer and B. Bernhardsson, "Optimal linear control for channels with signal-to-noise ratio constraints," in Proc. American Control Conference, San Francisco, CA, USA., June 2011.
• E. Johannesson, A. Rantzer and B. Bernhardsson, "A Framework for Linear Control over Channels with Signal-to-Noise Ratio Constraints," 9th IEEE International Conference on Control & Automation, Santiago, Chile, December 2011.

Another line of research focuses on the impact of feedback for  communication under a delay constraint. In

• B. Bernhardsson, "Using a Gaussian Channel Twice", IEEE International Symposium on Information Theory, Istanbul, 2013

I study the energy needed to transmit one bit of information over a Gaussian channel achieving a certain bit error rate. The setup in the following figure is used, with a constraint on the average expected energy of the sequence u=(u_1,...,u_N).

In the paper only the case N=2 is treated but similar techniques can be used to derive the result in the following figure for any N using dynamic programming (the needed work grows linearly with N).

The red curve for N=infinity is the well known Shannon limit.