Design and implementation of DSP-based magnetic control system for capsule endoscope

Abstract

Early detection methods are key to reducing morbidity rates from digestive tract cancer which is currently one of the fastest growing cancers in the World. Capsule endoscopes (CEs) are a new technology that can be used to improve early detection of the gastrointestinal (GI) tract disorder. The device integrates the technologies such as image processing, optoelectronic engineering, information communication, and biomedical engineering. The capsule is the size and shape of a pill and contains an optoelectronic camera, antenna, transmitter, battery and optoelectronic illuminating light emitting diodes (LEDs). The small size of these devices enables them to offer many advantages over conventional endoscopes such as accessibility to the entire intestine and minimising the risk of perforation, particularly for patients with difficult anatomy (e.g. post-operative scar tissue). Currently used devices are passive and can only follow the natural transit of the intestines, and hence there is considerable interest in methods of controlled actuation for these devices. In this thesis, a novel actuation system based on magnetic levitation is designed, developed and implemented, utilizing a small permanent magnet embedded within the capsule and an arrangement of digitally controlled electromagnets outside the body. The proposed approach is that the magnet can be moved and oriented by DC magnetic force and torque produced by coils placed outside of the human body, with a suitable position feedback sensor enabling closed-loop control. Theoretical analyses of the proposed actuation system are presented which model the magnetic field, force and torque exerted by electromagnetic coil on the embedded magnet. Based on the distribution of the magnetic field, an optimal geometry for the coils is proposed in order to achieve a levitation distance which is realistic for the inspection of the GI tract. Two types of systems are investigated in the thesis, namely single-input single-output (SISO) and multi-input multi-output (MIMO), and the dynamics of these systems are modelled in state space form and hence linear controllers are designed for capsule actuation. The controllers are simulated using Matlab/ Simulink tools to realize the mathematical analysis of the system, and then implemented digitally in real-time using Texas Instruments (TI) TMS320F2812 Digital Signal Processor (DSP) to validate the proposed actuation system. In the SISO system, a linear one degree of freedom (1DOF) proportionalintegral- derivative (PID) controller is designed to move the inserted magnet in the vertical dimension within an area around the operating point and to maintain it at a desired position. A realistic simulation model is designed and implemented to evaluate the proposed controller. Simulation results have shown that the controller is able to successfully hold the embedded magnet in the desired position. For practical validation, the PID controller is implemented in real-time on the DSP system, where pulse width modulation (PWM) is generated to control the coil current, and Hall effect sensors are used for position feedback. Experimental results are obtained under step and square wave input demand. In the proposed system, high frequency noise on the position sensor is initially rejected by hardware implementation of resistor capacitor-low pass filter (RC-LPF) circuit. The accuracy of the position feedback is increased by calibrating the DSP’s on-chip analogue-digital converter (ADC) in order to reduce conversion error due to inherent gain and offset errors. To further reduce the influence of the position feedback noise, an average of ten repeated samples based on mean filter is implemented by the DSP in order to reduce the influctuation of the sensor reading. The tracking performance of the actuation system based on two Hall effect sensors on the opposite coil’s poles is investigated under step trajectory input. In an improved actuation system, position feedback is provided by using an AC magnetic field to obtain the capsule position information, decoupling this from the DC actuation field. The noise of the position feedback in the improved system is reduced by replacing the PWM current drive with a linear power amplifier driven from a digital to analogue converter (DAC), hence reducing AC interference. Positioning sensor noise was found to be further reduced by implementing digital filtering based on a coherent detector using the DSP, without increasing response time. The performance of the actuation system using these position sensors is compared based on settling time, overshoot, steady-state error, and control input parameters in order to validate the proposed improvement in the position feedback. The experimental results have shown that the controller based on both sensing strategies satisfactory control of the magnet’s position. However, the response of the system based on AC position sensing has the shortest settling time, smallest overshoot value and steady-state error. In the MIMO system, several linear controllers such as pole placement (PP), Entire Eigenstructure Assignment (EEA), and linear Quadratic regulator (LQR) techniques are designed and their tracking performances are compared. Simulation results have shown that, based on acceptable control inputs, the LQR controller has the fastest response with minimal overshoot value and steady state error. However, the LQR controller based on 2DOF is unable to maintain stable control of the magnet due to the insufficient position feedback from the two coil sensors. Specifically, it is not possible to achieve a stable 2D system since the orientation angle of the magnet is not resolvable. Therefore, the position feedback is improved by obtaining the device position and orientation information from a pair of 3-axis orthogonal coils. A realistic simulation model for the 3DOF LQR controller is designed and implemented to evaluate the developed system. Simulation results have shown that this controller is can achieve the necessary stability. In conclusion, based on the results from the 1D control system, the thesis shows that the DC magnetic field, which is used for capsule movement, can be also used to provide the controller acceptable position feedback. However, the use of AC magnetic field for positioning purpose provides more accurate position information. In order to implement 2DOF control system successfully, two 3-axis orthogonal coil sensors are considered which are used to provide the actuation algorithm with more accurate feedback of position and orientation information.