GHz Scale Realtime Least Squares Pulse Fitting

Thus far, few-body hadronic parity violation experiments have measured the average current instead resolving individual events. New experiments involving reactions such as n+d → t+γ (NDTGamma) have scattering cross sections 10^3 times smaller than background, which must be filtered out through spectroscopy. However, it is computationally infeasible to fit the energy of each event using standard non-linear least squares techniques at the expected rate of 10^7 events per second. In least squares fitting, the design matrix maps the parameter space into the solution space. By approximating the design matrix using a piecewise polynomial expansion, recursive convolution filters can be used to calculate the fitting parameters at high sample rates (10^9 hz)[1]. The design matrix expansion is optimized using adaptive moment estimation stochastic optimization. We will discuss the implementation of such a system on relatively cheap ∼ $350 commodity hardware.