Rescue dogs are commonly used during the urban search-and-rescue (USAR) operations for the initial indication on the presence of trapped victims after the collapse of man-made structures. However, dogs are not able to inform the rescue crews whether the trapped victims are alive or not and where exactly they are located. Other complementary tools, such as acoustic- and audio-visual equipment, are prone to inaccuracy, interference and inadequate range of operation. Ultra-wideband (UWB) radar is considered a promising tool for more exact assessment on the range of trapped victims. However, implementation of UWB radar for trapped-victim detection faces challenges such as low signal-tonoise ratio (SNR) conditions, interference from non-stationary clutter, residues due to amplitude instability originating in the equipment as well as narrowband radio interference. There are four commercially available UWB radar technologies and it is not clear which radar technology is the most suited one for the purpose of detecting trapped victims. There is very little available knowledge on the two target features that enable detection of a trapped human body using radar (respiratory- and cardiac motion). In need of further investigation is the choice of the optimal operational frequency band as well as assessment on the amount of attenuation of a few obstacles that represent, to various accuracy, real-life rubble. Chapter 2 introduces the reader with the basic principles of generation, sampling and pre-processing of UWB signals for the four available UWB radar technologies. It investigates the applicability of two time-domain- and the continuous-wave (CW) UWB radar technology for the purpose of trapped-victim detection, both based on the inherent properties as well as by means of an experimental verification study evaluated under as similar measurement conditions as possible. The results of both the theoretical and experimental verification indicate that the CW UWB radar technology is to prefer over the time-domain radar technologies due to generally higher dynamic range, better use of the designated spectrum and higher transmit power as well as that it enables the extraction of two target features, as opposed to only one (respiratory motion). The study is not definitive nor final and should serve as guidelines for further studies and/or system design. Chapter 3 investigates in detail the time-domain and frequency-domain behaviour of the two available target features for various body positions. It shows that the respiratory motion responses are in average 13 dB stronger than the cardiac motion responses. The position such that the chest is turned toward the receive antenna produces strongest respiratory motion responses due to larger chest displacement and reflective area than the other positions. Detectability of respiratory motion responses as function of aspect angle was investigated under line-of-sight conditions for three body positions, four bi-static angles and three antenna-pair polarisations using a single test person. It showed that there is no considerable difference in detectability among the investigated bi-static angles and co-polarised antenna pairs. However, it was concluded that cross-polarised antenna pairs should be avoided in real life as they produce significantly lower detectability values. The attenuation as function of frequency of two types of obstacles (piles of sandstone blocks and a 60-cm concrete wall) was investigated in chapter 4. The results show that the attenuation for both materials is ca 10-15 dB across the frequency range of interest. However, realistic rubble thicknesses and types of rubble can heavily increase the attenuation and thereby lower the probability of detection. Measurements involving a test person resting under 80-cm concrete rubble pile and behind two concrete walls, showed that the centre frequency well below 1 GHz gives rise to highest SNR values. Bandwidths of ca. 400 MHz centred at frequencies below 1 GHz give rise to higher SNR values than larger investigated bandwidths. On the hand, lower bandwidths result in poorer down-range resolution, which is necessary for resolving non-stationary clutter responses or multiple trapped victims. One of the fundamental tasks of this thesis is the development of a respiratory motion detection algorithm. Chapter 5 details a novel and computationally efficient algorithm which is able to improve SNR conditions and better suppress non-stationary clutter compared to an existing algorithm, assessed both experimentally and in a simulated environment. The algorithm further incorporates a threshold which aids in the decision making process by the operator. The performance of three common stationary-clutter suppression methods is investigated on a single measured data set containing respiratory motion and linear amplitude instability (linear trend). It was shown that the linear-trend removal method, which removes any potential linear trend and DC level in the slow-time dimension, is the preferred approach to stationary-clutter suppression. Narrowband interference (NBI) results in increase of noise floor and thereby worsens the probability of detection, when using stroboscopic sampling (such as in impulse radar). Chapter 6 analyses the performance of four developed methods for NBI suppression implemented in stroboscopic samplers. The most suitable method for NBI suppression in stroboscopic samplers is to filter out the NBI in the analogue domain and, after sampling, implements linear interpolation of the missing spectrum in order to avoid ringing of the backscattered waveforms from the victim, is regarded the most suitable method. It shows an improvement factor of 12.9 dB in noise reduction and manages to preserve the signal waveform and energy very well. The thesis is completed by the conclusions and recommendations for future studies.