Albrecht, JohannesHasse, Christoph2019-12-172019-12-172019http://hdl.handle.net/2003/38453http://dx.doi.org/10.17877/DE290R-20373Starting in 2021, the LHCb experiment will switch to a fully software-based trigger system. It is an ongoing challenge to ensure that this system will be able to process events at the required rate of 30 MHz. Two alternative approaches to improve and speed up the particle track reconstruction performed in the first stage of LHCb’s software trigger are presented. An alternative method to the Kalman filter track fit inside the VELO reconstruction is presented. This method uses neural networks to perform the fit and additionally provides an estimation of a track’s momentum based on the track’s scattering. The momentum information enables improved uncertainty estimates for a trajectory’s impact parameter, an important quantity to select tracks from secondary vertices. It is shown that the neural network’s uncertainty prediction enables an equally efficient selection of secondary tracks while reducing the amount of falsely selected prompt tracks by 30%. Additionally, an alternative procedure to reconstruct tracks that traverse the entire LHCb detector is presented. These tracks are LHCb’s main physics objects and their reconstruction previously required a third of the overall processing time of the first trigger stage’s reconstruction sequence. A new algorithm design is proposed, which is shown to yield similar reconstruction efficiencies while providing an over six-fold speedup over the current algorithm. A comparison of trigger configurations, which are able to process data at the rate of 30 MHz, is presented. It shows that a trigger based on the new algorithm yields significantly higher selection efficiencies.enCERNLHCbHigh level trigger530TextLHCb <Teilchendetektor>ALICE <Teilchendetektor>