Pascal Heim

A 1V RF SoC with an 863-to-928MHz 400kb/s radio and a 32b Dual-MAC DSP core for Wireless Sensor and Body Networks

A 150¿A/MHz DSP with two MAC/cycle instructions is integrated with a configurable 863-to-928MHz RF transceiver that yields 3.5mW in continuous reception, 2¿C per channel sampling and 40mW for 10dBm output. The SoC includes voltage converters that allow 1.0-to-1.8V or 2.7-to-3.6V primary voltage supplies. In sleep mode, it consumes 1¿A with a 32kHz crystal-based RTC running.

Quantization effects in digitally behaving circuit implementations of Kohonen networks

Implementing a neural network on a digital or mixed analog and digital chip yields the quantization of the synaptic weights dynamics. This paper addresses this topic in the case of Kohonens self-organizing maps. We first study qualitatively how the quantization affects the convergence and the properties, and deduce from this analysis the way to choose the parameters of the network (adaptation gain and neighborhood). We show that a spatially decreasing neighborhood function is far more preferable than the usually rectangular neighborhood function, because of the weight quantization. Based on these results, an analog nonlinear network, integrated in a standard CMOS technology, and implementing this spatially decreasing neighborhood function is then presented. It can be used in a mixed analog and digital circuit implementation.

High dynamic range vision sensor for automotive applications

A 128 x 128 pixels, 120 dB vision sensor extracting at the pixel level the contrast magnitude and direction of local image features is used to implement a lane tracking system. The contrast representation (relative change of illumination) delivered by the sensor is independent of the illumination level. Together with the high dynamic range of the sensor, it ensures a very stable image feature representation even with high spatial and temporal inhomogeneities of the illumination. Dispatching off chip image feature is done according to the contrast magnitude, prioritizing features with high contrast magnitude. This allows to reduce drastically the amount of data transmitted out of the chip, hence the processing power required for subsequent processing stages. To compensate for the low fill factor (9%) of the sensor, micro-lenses have been deposited which increase the sensitivity by a factor of 5, corresponding to an equivalent of 2000 ASA. An algorithm exploiting the contrast representation output by the vision sensor has been developed to estimate the position of a vehicle relative to the road markings. The algorithm first detects the road markings based on the contrast direction map. Then, it performs quadratic fits on selected kernel of 3 by 3 pixels to achieve sub-pixel accuracy on the estimation of the lane marking positions. The resulting precision on the estimation of the vehicle lateral position is 1 cm. The algorithm performs efficiently under a wide variety of environmental conditions, including night and rainy conditions.