Double-directional Superresolution Radio Channel Measurements

The paper describes measurements of radio propagation at 2 and 5 GHz in urban environments in Vienna, Austria, and Ilmenau, Germany, and in a rural highway environment near Salzburg, Austria, taken with virtual and physical antenna arrays. Multiple antenna elements at both the base and the mobile stations permit, by estimation with superresolution algorithms, evaluation of directions-of-arrival and directions-of-departure. We aim at a high-precision deterministic full 3-D characterization of the mobile radio channel and succeed in yielding delay, azimuth and elevation of individual multipath components. Comparison with site-maps will allow tracking up to third-order reflections/scattering, thus questioning the applicability of the popular single-bounce models for mobile radio propagation. We introduce a global measure, the multipath component separation, to describe how well an environment lends itself to smart antenna application. The channel-state knowledge at the receiver alone, or both at the receiver and the transmitter, bears great impact on the achievable capacity of the current MIMO (multipleinput multiple-output) systems. Our approach to measurement and evaluation forms a solid basis for the calculation of the MIMO capacity in specific environments. Motivation We have three motives for carrying out double-directional radio channel measurements: First, a highly precise characterization of the mobile radio channel becomes available when this type of measurements is combined with high-resolution parametric methods. We expect answers to questions like: is reflection, diffraction or scattering the dominant propagation mechanism in urban or office environments? What is more important: over-the-roof or around-the-corner propagation? Comparison with maps will show that up to third-order reflections/scattering can be traced, a result that seriously questions the applicability of the popular single-bounce models in mobile radio propagation. The results will benefit the improvement of ray-tracing algorithms and site-specific network planning. Second, knowledge of the double-directional radio channel is a prerequisite to exploit the highly current MIMO (multiple-input multiple-output) systems, proposed to enhance the transmission capacity of radio links by orders of magnitude. Whereas a number of interesting theoretical MIMO channel models have been put forward recently, very few measurements accurately describing these channels have been published. In fading channels, the "capacity" itself becomes a random variable (instantaneous capacity). Determining its distribution and related quantities (outage capacity) might require measurements of many realizations in one and the same environment. We would like to characterize the environment by a single measurement run, and independently of the equipment used for the measurement (including the antenna arrays itself). Third, a more far-reaching goal is site-specific deployment of radio systems, based on an initial double-directional measurement. Just imagine the considerable savings in transmitted power and in system self-interference by transmitting only in the directions where one can be sure that the signal will be actually received, but will not interfere other mobiles in the cell or system. Deterministic or stochastic channel models? Propagation is at the heart of radio communications. It sets the ultimate limits for the transmission speed and throughput of any system built upon radio. Given the obvious attractiveness of mobile radio on one hand, and the complexity of the radio channel on the other, it is clear that good propagation models for electromagnetic waves are needed for both the development and the deployment of the novel MIMO systems envisaged. These systems have multiple antennas at each end of the link, i.e. at the receive and the transmit side. Ideally, independent, orthogonal data transmission channels are set up between the individual antenna elements to maximize transmission capacity. Stochastic channel models are state of the art for system design and for testing of terminals. Their implementation must be simple and allow fast simulation of systems or part of systems (e.g. receiver algorithms). Whatever environments, situations and fields of application of a system have been included in the mix for the stochastic channel model, the choice of this mix will influence the statistics. Using such a stochastic channel model in system optimisation yields a system that is optimum on average. But in a specific environment, the system will not be optimal, unless it exploits the properties of this very specific environment. We will show that, for MIMO systems, we can supply full knowledge of the main propagation paths determined by the environment. We do not denounce stochastic channel models in general. Such models have been proposed by e.g. Pajusco [1] for directional channels, Bölcskei et al. [2] and Bach Andersen [3] for MIMO channels. They rely on such concepts as angular spread as seen from the antenna arrays. Scatterers are assumed numerous and randomly distributed. Our deterministic approach to the description of a specific environment, where double-directional channel measurements have been carried out, is only the basis for further stochastic considerations. One note of caution toward supporters of stochastic channel modelling, however, is in place right now: The popular assumption of wide-sense stationary uncorrelated scattering (WSSUS) [4] will fail most probably in directional channel models [5]. The assumption of an infinite number of scatterers is difficult to justify for each delay bin considered [6], but is usually not questioned any further. The situation is aggravated if we switch to directional stochastic channel models: there should be a large number of scatterers in each direction given by a certain value of azimuth and elevation, for each delay bin. This is unrealistic. With increasing angular resolution, the number of scatterers per delay-angle bin will become smaller and smaller, and eventually a single multipath component will remain. Figure 1 illustrates this situation. It shows a modification of the channel model Vehicular A that ITU recommended for IMT 2000 candidate system evaluation [7], where Pajusco extended it to include a spatial component [1].