Running Cross-Correlation using Bitstream Processing Tor Sverre Lande, Timothy G. Constandinou, Alison Burdett and Chris Toumazou Abstract: A novel architecture for running cross-correlation and convolution using bitstream processing is proposed. The computationally intensive multiplications inherent in cross-correlation and convolution are replaced by simple logic operations (AND XOR) using bitstream representation. The reduced complexity enables compact and energy efficient silicon solutions suitable for small, portable devices such as wearable heart beat detecting electronics embedded in the actual ECG patch. Introduction: The time-domain operation known as cross-correlation is a computationally intensive algorithm due to the large number of multiplications required. The index I is the shift or offset parameter between two sampled signals x(n) and y(n). For each shift in time all n samples within the correlation window must be multiplied and accumulated, i.e. results summed. In a system operating a running correlation these n multiplications must be performed in parallel (or multiplexed at higher speed). Since multiplication is a power hungry (or slow) operation, power efficient hardware for cross correlation is challenging to make. However, by changing the representation or coding of the signal, hardware efficient equivalents of running cross-correlators exists. Bitstream processing: A popular data conversion technique is known as Δ−Σ modulation using over-sampling to move in-band noise to higher frequencies (noise shaping). As the sampling rate is increased, the precision of the quantiser is relaxed. ) ( ) ( 1 l k y k x r n k l − = ∑ =