Sub 1-V Asynchronous Event Driven Analog to Information Converters

TITLE: Sub 1-V Asynchronous Adaptive Event Driven Analog to Information Converters (Completed)

Advisor: Sameer R Sonkusale

Researchers: Michael Trakimas, Ritika Agarwal



In wireless sensor networks there is a need to increase the operational lifetime of each sensor node.  Continued scaling of CMOS technology has enabled energy efficient digital signal processing but with modest returns, because majority of the energy is consumed in acquiring data at the nyquist-rate and its wireless transmission. The traditional acquire-digitize-store-and-transmit paradigm is suboptimal for many sensing applicationsA revolutionary and more holistic approach at both the architectural and circuit level is needed for data acquisition.



The key to power savings in wireless sensor nodes is to “generate less data to begin with”. Emerging ideas on compressed sensing implies that signals that are sparse in some basis set could be acquired at rates lower than the nyquist-rate. Speech, video, EEG, ECG and other naturally occurring signals are all sparse and compressible. We have implemented the first of its kind random sampling based data converter that utilizes the theory of compressed sensing for both spectrum sensing application and in compressive digitization of ECG and EEG signals. Signal reconstruction is carried out at the base station and thus the burden of reconstruction is transferred away from the sensor node. We have also proposed a novel signal adaptive approach for data acquisition based on event driven sampling without the use of any clocks. Such digitizers do not require anti-aliasing filters and their data rate is dependent on the input signal activity.  There is no amplitude quantization error and the information is contained in time between events, rather than amplitude, which has inherent high dynamic range and better resolution in digital CMOS process. We proposed an improved adaptive resolution algorithm where the data rate is further reduced and is not limited by the inherent loop delay in the digitizer. In applications where the goal is feature extraction from analog signals, such as extraction of QRS complex from ECG waveform, we propose an “input-feature correlated asynchronous sampling” paradigm that utilizes activity information from asynchronous samples just discussed to directly extract features of interest from the timing waveform. This is the ultimate realization of a true analog to information converter.




Our random sampling compressed sensing A/D converter was implemented in 90nm CMOS with an event-driven SAR ADC core (IEEE TCAS1 2013). Our adaptive asynchronous ADC generated impressive results for ECG, audio and accelerometer data (IEEE TCAS1 2011). Lastly the work on input feature correlated analog to information converter was applied to the detection of QRS complex in ECG waveform from MIT-BIH Arrhythmia database with more than 99% accuracy (IEEE Transactions of Biomedical Circuits and Systems or TBioCAS, 2011).


Please see publication page