Embedded Electronic Systems for Electronic Skin Applications

The advances in sensor devices are potentially providing new solutions to many applications including prosthetics and robotics. Endowing upper limb prosthesis with tactile sensors (electronic/sensitive skin) can be used to provide tactile sensory feedback to the amputees. In this regard, the prosthetic device is meant to be equipped with tactile sensing system allowing the user limb to receive tactile feedback about objects and contact surfaces. Thus, embedding tactile sensing system is required for wearable sensors that should cover wide areas of the prosthetics. However, embedding sensing system involves set of challenges in terms of power consumption, data processing, real-time response and design scalability (e-skin may include large number of tactile sensors). The tactile sensing system is constituted of: (i) a tactile sensor array, (ii) an interface electronic circuit, (iii) an embedded processing unit, and (iv) a communication interface to transmit tactile data. The objective of the thesis is to develop an efficient embedded tactile sensing system targeting e-skin application (e.g. prosthetic) by: 1) developing a low power and miniaturized interface electronics circuit, operating in real-time; 2) proposing an efficient algorithm for embedded tactile data processing, affecting the system time latency and power consumption; 3) implementing an efficient communication channel/interface, suitable for large amount of data generated from large number of sensors. Most of the interface electronics for tactile sensing system proposed in the literature are composed of signal conditioning and commercial data acquisition devices (i.e. DAQ). However, these devices are bulky (PC-based) and thus not suitable for portable prosthetics from the size, power consumption and scalability point of view. Regarding the tactile data processing, some works have exploited machine learning methods for extracting meaningful information from tactile data. However, embedding these algorithms poses some challenges because of 1) the high amount of data to be processed significantly affecting the real time functionality, and 2) the complex processing tasks imposing burden in terms of power consumption. On the other hand, the literature shows lack in studies addressing data transfer in tactile sensing system. Thus, dealing with large number of sensors will pose challenges on the communication bandwidth and reliability. Therefore, this thesis exploits three approaches: 1) Developing a low power and miniaturized Interface Electronics (IE), capable of interfacing and acquiring signals from large number of tactile sensors in real-time. We developed a portable IE system based on a low power arm microcontroller and a DDC232 A/D converter, that handles an array of 32 tactile sensors. Upon touch applied to the sensors, the IE acquires and pre-process the sensor signals at low power consumption achieving a battery lifetime of about 22 hours. Then we assessed the functionality of the IE by carrying out Electrical and electromechanical characterization experiments to monitor the response of the interface electronics with PVDF-based piezoelectric sensors. The results of electrical and electromechanical tests validate the correct functionality of the proposed system. In addition, we implemented filtering methods on the IE that reduced the effect of noise in the system. Furthermore, we evaluated our proposed IE by integrating it in tactile sensory feedback system, showing effective deliver of tactile data to the user. The proposed system overcomes similar state of art solutions dealing with higher number of input channels and maintaining real time functionality. 2) Optimizing and implementing a tensorial-based machine learning algorithm for touch modality classification on embedded Zynq System-on-chip (SoC). The algorithm is based on Support Vector Machine classifier to discriminate between three input touch modality classes “brushing”, “rolling” and “sliding”. We introduced an efficient algorithm minimizing the hardware implementation complexity in terms of number of operations and memory storage which directly affect time latency and power consumption. With respect to the original algorithm, the proposed approach – implemented on Zynq SoC – achieved reduction in the number of operations per inference from 545 M-ops to 18 M-ops and the memory storage from 52.2 KB to 1.7 KB. Moreover, the proposed method speeds up the inference time by a factor of 43× at a cost of only 2% loss in accuracy, enabling the algorithm to run on embedded processing unit and to extract tactile information in real-time. 3) Implementing a robust and efficient data transfer channel to transfer aggregated data at high transmission data rate and low power consumption. In this approach, we proposed and demonstrated a tactile sensory feedback system based on an optical communication link for prosthetic applications. The optical link features a low power and wide transmission bandwidth, which makes the feedback system suitable for large number of tactile sensors. The low power transmission is due to the employed UWB-based optical modulation. We implemented a system prototype, consisting of digital transmitter and receiver boards and acquisition circuits to interface 32 piezoelectric sensors. Then we evaluated the system performance by measuring, processing and transmitting data of the 32 piezoelectric sensors at 100 Mbps data rate through the optical link, at 50 pJ/bit communication energy consumption. Experimental results have validated the functionality and demonstrated the real time operation of the proposed sensory feedback system.