Modelo de selección y priorización de proyectos

2.1.1. General aspects of channel modeling

As with any other communications system, it is thechannelthat determines the ul- timate (information-theoretic) performance limits, as well as the practical perfor- mance limits of various transmission schemes and receiver algorithms. For UWB systems, this channel is the ultra-wideband propagation channel. Understanding this channel is thus a vital prerequisite for designing, testing, and comparing UWB systems. Just like UWB communications itself, channel modeling for UWB is a rel- atively new area. And just as the interest for UWB systems has intensified in the last years, so has the importance of modeling the UWB propagation channels. In this chapter, we will give a comprehensive overview of the state of the art in this exciting area.

Quite generally, wireless channel modeling is done for two different purposes. (i) Deterministic channel modeling tries to predict the behavior of a wire- less channel in a specific environment. If a complete description of the geometry, as well as of the electromagnetic properties of the materials, of the surrounding of transmitter and receiver is given,1_{then Maxwell’s} equations can be solved exactly, and the channel impulse response (or an equivalent quantity) can be predicted. This approach, which had long been deemed too complicated, has become popular in the last 15 years. Ray tracing and other high-frequency approximations, as well as the ad- vent of more powerful computers, have made it possible to perform the required computations within reasonable time.

(ii) Stochastic channel models try to model the “typical” or “canonical” properties of a wireless channel, without relating those properties to a specific location. As the most simple example, the probability density

1_{The size of the “surroundings” depends on the environment, as well as on the desired dynamic}

function of the (narrowband) amplitude of the received signal is mod- eled as Rayleigh-distributed within a small area (part of a room). In con- trast, a deterministic model would try to predict the exact amplitude values at each location in that part of a room.

There is an enormous amount of literature fornarrowbandwireless channels; for textbooks and overview articles, see [1–10]. However, those models cannot be easily generalized to UWB channels, due to some important basic differences in the propagation processes and the resulting models. Some of these differences are the following.

(i) Each multipath component (MPC) can lead to delay dispersion by itself, due to the frequency-selective nature of reflection and diffraction coef- ficients. This effect is especially important for systems with largerelative bandwidth.

(ii) The signals are received with excellent delay resolution. Therefore, it of- ten happens that only a few multipath components make up oneresolv- ableMPC, and the amplitude statistics of such a resolvable MPC is not complex Gaussian anymore. Similarly, there is an appreciable probabil- ity that areas of “no energy” can exist, that is, (resolvable) delay intervals during which no significant amount of energy is arriving at the receiver. These issues are mostly important for systems with largeabsoluteband- width.

By definition, a UWB channel is a more general description of nature than a narrowband channel: it is always possible to obtain a narrowband channel from a UWB channel by just filtering the UWB channel. Extrapolation from narrowband to UWB is tempting, as it allows the reuse of existing measurements. However, it is extremely dangerous, and has to be backed up by measurement data. Every aspect of narrowband channel modeling has to be questioned before it can be accepted for UWB modeling!

2.1.2. Regulatory and application-specific aspects

The frequency range of UWB systems is limited by transceiver design issues (tran- sit frequency of semiconductor components, antenna bandwidth, etc.), as well as rulings of the frequency regulators. Channel models should be mindful of those limitations: after all, the main purpose of channel models is the design and testing of UWB systems, and as the systems face practical and legal limitations, there is no need to be overly general in the models either.

Up to now, only the regulator in the USA has issued rulings that allow the use of UWB.2_{The report and order of the FCC from 2002 [11] allows unlicensed} operation of UWB communications devices in indoor environments, mainly in the frequency range between 3.1 and 10.6 GHz. The admissible power spectral density is−41.3 dBm/MHz, and considerably lower outside that band (the exact

2_{Singapore has recently established a “UWB-friendly zone,” and other countries are expected to}

101 100 Frequency (GHz) −75 −70 −65 −60 −55 −50 −45 −40 UWB EIRP emission le ve l (dBm) Indoor limit Part 15 limit

Figure2.1. FCC mask for indoor communications.

mask is shown inFigure 2.1). Communications is also allowed at frequencies be- low 960 MHz. There are certain restrictions in the applications, as well as in peak- to-average ratios; however, those have no impact on the channel modeling. It is important that outdoor devices must not be fixed, or have a connection to power mains-only peer-to-peer communications of portable devices is allowed. These re- strictions limit the frequency range in which the channel model needs to be valid, as well as the possible distance between transmitter and receiver.

The envisioned applications give us guidelines about what propagation envi- ronments are of greatest interest for UWB applications. A main point of emphasis is high-data-rate personal area networks (PANs). A typical application there is the transmission of digital TV signals via a wireless link. It follows that such high data rates, in conjunction with the limits on the power spectral density, can be per- formed only over short distances (on the order of 10 m), and are of interest mostly for residential and office environments. Another application of interest is sensor networks. As the data rates are much lower in this context, longer ranges are possi- ble. Furthermore, sensor networks can be of great interest in factory environments, where “indoor” distances can be larger, and the abundance of metallic reflectors can lead to quite different propagation conditions.

Yet another application for UWB is emergency communications, possibly combined with geolocation, for example, from a building that collapsed during an earthquake, or from an avalanche. Propagation conditions in such rubble or snow environments will obviously differ considerably from “normal” environments; re- quired distance can be up to 100 m. Outdoor UWB applications, as mentioned above, are limited to peer-to-peer communications in the USA; however, since other countries might allow fixed-location transmitters, it is worthwhile to also study the “base-station to mobile-station” scenario. Finally, communications be- tween cars is of interest.

2.1.3. Synopsis of the chapter

A first step in understanding UWB propagation has to be measurements of UWB channels. For this task, we have two different types of measurement techniques at our disposal: frequency-domain measurements, employing a vector network an- alyzer, give the transfer function of the channel; alternatively, the channel can be excited with a short pulse, and a time-domain measurement of the received signal returns the convolution of the pulse with the channel impulse response. In either of those cases, the impact of the antenna on the measurement results is crucial. In fact, it is much more important than in the narrowband case, because the an- tenna characteristics vary significantly with the frequency, and exhibit a different response in different directions.Section 2.2byKunischgives a detailed mathemat- ical model for these effects, and suggests a procedure for the deconvolution of the antenna effects.

As mentioned above, a crucial property of UWB channels is the fact that each multipath component can show delay dispersion by itself. That means that a short pulse that, for example, undergoes only a single diffraction will arrive at the receiver with a larger support (extent in the delay domain) compared to the originally transmitted pulse. The reason for this is that the diffraction coef- ficient is frequency-dependent.Qiudescribes inSection 2.3the mathematics of these processes. Exact and approximate formulations, based on the uniform the- ory of diffraction, allow a modification of the classical Turin model, so that the impulse response in a multipath environment is now the sum ofdistorted, scaled, and delayed pulses. This has important implications for the design of correlation receivers, which are also addressed in this section. From a more experimental point of view,Molisch and Buehrer discuss the impact of frequency-selective reflection coefficients inSection 2.5.5. The results from these sections can be combined with classical ray tracing approaches, resulting in a deterministic channel prediction and modeling method.

For stochastic channel modeling, we have to investigate the path loss and shadowing, as well as the delay and angular dispersion of the channel. These two topics are treated in Sections2.4(byCassioli) and2.5(byMolisch and Buehrer), respectively. While the path loss is a well-known quantity in narrowband chan- nel modeling,Section 2.4shows that it actually has to be treated as frequency- dependent when the relative bandwidth becomes significant. On the other hand, many narrowband measurement campaigns in different frequency bands can be reused for path loss predictions. The delay dispersion properties also show a significant dependence on the bandwidth. While the rms delay spread shows only a weak dependence on the bandwidth, the amplitude statistics and the statis- tics of the arrival times of the multipath properties strongly vary with the band- width, as well as with the center frequency of the considered signal. All the dif- ferent aspects of stochastic channel modeling had an impact on the standardized IEEE 802.15.3a channel model, which is described byPendergrassinSection 2.6. It was used in the downselection process for the standard for the physical layer of high-data-rate UWB communications, and is therefore restricted to modeling

short-range communications in indoor office environments. More general chan- nel models are currently under development within the IEEE 802.15.4a group, but not yet finalized at the time of this writing, and therefore not further treated here.

All of the above considerations are for a channel without human presence— measurement campaigns (as well as the models based on them) usually try to avoid the influence of human bodies on the measurement results as much as possible. This assumption might be realistic, for example, for an antenna mounted on a DVD-player, which establishes a video link. However, it becomes untenable for “body-area network,” where the antenna is placed directly on a body of a user— antenna and body become inseparable. This case requires new measurement tech- niques, and also new channel models.Kovacs, Pedersen, and Eggersdescribe mea- surement setups that are especially suitable for this situation (seeSection 2.7). A main difficulty is the impact of cables between antenna and network analyzer (or oscilloscope); this problem is solved by using RF-over-fiber.

Finally,Section 2.8byRoydeals with the topic of channel estimation. While it is clearly similar to the problem of measuring channels, there are also important conceptual differences. Channel estimation always must be viewed in light of the system it is being performed for, while channel measurement and modeling tries to remain as system-independent as possible. Also, channel estimation has to use a signaling structure that is similar to the signaling (modulation) of the payload data—cost considerations do not allow the building of an additional transceiver with a different structure, just to obtain channel estimates. Due to these reasons, the estimation techniques for OFDM-based UWB systems on one hand and for impulse-radio and direct-sequence CDMA systems on the other hand are quite different. Both are treated inSection 2.8.

The state of the art of UWB channel modeling is by now sufficiently advanced that it merits a review such as the one given in this chapter. The main problems and challenges for UWB channel models have been identified, and generic model structures can be established. However, we have to recognize that UWB channel models are still far less comprehensive than narrowband models—this is no won- der when we compare the duration of UWB channel research with the more than 30 years that have been spent on narrowband channels. A main gap in our knowl- edge is the parameterization of the models in different environments. Up to now, almost all measurements have been performed in indoor office and residential en- vironments. Those measurements give a first indication of occurring parameters, though important questions (angular dispersion, what is the best distribution to describe fading statistics?, how do the results depend on the measurement band- width and center frequency?) need more thorough investigation. Parametrization for other environments, like factory halls, disaster areas, and so forth, are com- pletely unknown, and many measurements will be required to obtain statistically reliable parametrizations. It is thus safe to say that the understanding of UWB channels has made some important advances, but much more work is required before we can claim to have a thorough understanding of this important commu- nications medium.

2.2. Measurement techniques