The Tunka Radio Extension (Tunka-Rex) is an array of 63 antennas located in the Tunka Valley, Siberia. It detects radio pulses in the 30-80 MHz band produced during the air-shower development. As shown by Tunka-Rex, a sparse radio array with about 200 m spacing is able to reconstruct the energy and the depth of the shower maximum with satisfactory precision using simple methods based on parameters of the lateral distribution of amplitudes. The LOFAR experiment has shown that a sophisticated treatment of all individually measured amplitudes of a dense antenna array can make the precision comparable with the resolution of existing optical techniques. We develop these ideas further and present a method based on the treatment of time series of measured signals, i.e. each antenna station provides several points (trace) instead of a single one (amplitude or power). We use the measured shower axis and energy as input for CoREAS simulations: for each measured event we simulate a set of air-showers with proton, helium, nitrogen and iron as primary particle (each primary is simulated about ten times to cover fluctuations in the shower maximum due to the first interaction). Simulated radio pulses are processed with the Tunka-Rex detector response and convoluted with the measured signals. A likelihood fit determines how well the simulated event fits to the measured one. The positions of the shower maxima are defined from the distribution of chi-square values of these fits. When using this improved method instead of the standard one, firstly, the shower maximum of more events can be reconstructed, secondly, the resolution is increased. The performance of the method is demonstrated on the data acquired by the Tunka-Rex detector in 2012-2014.