Abstract
Cross-correlation gives a measure of the similarity between two signals in the time domain and is desirable in pattern matching and recognition, but other solutions are often sought after due to the heavy computational load. The work includes processing of signals in their oversampled delta sigma domain, bitstream processing, and asynchronous techniques. Employing these techniques resulted in a novel signal processing solution.
This master thesis presents a power efficient implementation of a time-domain bitstream cross-correlator suitable for integration in CMOS. The developed generic SKILL script can generate layouts with different dimensions and one version is fabricated in 90 nm CMOS. Cross-correlation can be used for heartbeat detection in an ECG analysis and the implemented chip is tested with real-world Electrocardiography (ECG) beat detection. The power dissipated by the chip is compared to the power dissipated by a microcontroller performing equal computations. The bitstream cross-correlator has an estimated improvement in power dissipation by a factor of 84.