The Tunka Radio Extension (Tunka-Rex) is a digital antennaarray (63 antennas distributed over≈1 km2) co-located with the TAIGAobservatory in Eastern Siberia. Tunka-Rex measures radio emission ofair-showers induced by ultra-high energy cosmic rays in the frequencyband of 30-80 MHz. Air-shower signal is a short (tens of nanoseconds)broadband pulse. Using time positions and amplitudes of these pulses,we reconstruct parameters of air showers and primary cosmic rays. Theamplitudes of low-energy event (E<1017eV) cannot be used for suc-cesful reconstruction due to the domination of background. To lower theenergy threshold of the detection and increase the efficiency, we use au-toencoder neural network which removes noise from the measured data.This work describes our approach to denoising raw data and furtherreconstruction of air-shower parameters. We also present results of thelow-energy events reconstruction with autoencoder.