Reducing the processing loss of windowed transforms using linear discriminant analysis

Using window functions in the time-domain to suppress side-lobes in the frequency-domain comes at the cost of a broadened main lobe and a loss in signal-to-noise ratio known as processing loss. This letter proposes a multi-variate detection algorithm, which reduces the processing loss of windowed transforms by up to 65% in comparison to a threshold decision, which translates to an increased detection rate of up to 11%. The comparison of simulation and measurement shows a very close match. Furthermore, an adaption of the ordered statistics constant false-alarm rate algorithm is introduced, which incorporates the performance increase of the new detector.