When Mark Weiser first coined the phrase “ubiquitous computing” around 1988 [1,2], the idea of “information available at our fingertips during a walk” was stilla dream. Today, through relentless pursuit of innovations in wireless communication and energy efficient hardware technology, we are gradually turning this dream into reality. Smart phones and tablets, enabled with adaptive software, have generated lucrative revenues for business and thus have inspired numerous waves of research efforts in mobile computing. In this context, location-based services have attracted considerable attention because of their potential to empower mobile computing with highly personalized and context-aware services [3,4]. However, there still exist considerable challenges, particularly in indoor location determination. As suggested by Mahadev Satyanarayanan [5], this challenge embodies a number of themes that could be central to the evolution of mobile computing. This paper continue sour previous research [6-12] and develops and evaluates a 3Dindoor radio propagation model that will not only help understand radio propagation in complex indoor environment but also contribute a set of genuinely applications that depend on real-time radio propagation signatures for a wide range of radio frequencies.

Indoor location determination typically uses a range of technologies in order to overcome constraints of indoor structures and environments. These technologies include: 1) infrared [13]; 2) sound [14-17]; 3) vision [18-21]; 4) inertial sensors [22,23]; 5) radio frequency (including both WiFi and Zigbee) [6,10,24-33]; and 6) landmarks [34-37]. High-sensitivity GPS sensors have also been evaluated in indoor environments, however, even with the ability to receive signals up to 1000 times weaker than normal GPS, high-sensitivity GPS receiver is still unreliable for indoor environments [38,39]. Among these technologies, however, most location determination systems used only one or a combination of three localization techniques:

1) Trilateration/Triangulation or Area Overlapping [40-51], where distance, angle, or area information from a mobile to known reference positions is used to estimate the mobile’s position.

2) Location Fingerprinting [6,10,16,26,33,52-58], where location dependent information signatures (i.e. fingerprints like signal strength, image, sound, or other unique information) are pre-collected and evaluated against current measurements in order to pinpoint the location of a mobile.

3) Dead Reckoning (DR) [22,23,59-62] that determines, without the aid of external observations or references, the position and/or orientation of a mobile client from the record of traveled courses, distance and drift angles.

Trilateration/triangulation involves less overhead for offline training processes; however, it requires high resolution sensor devices in order to obtain accurate distance or angle measurements between mobile and reference positions. In reality, indoor radio waves from non-lineof-sight paths may result in a stronger signal than radio from a shorter path or even a direct path. This multipath problem could result in an average of 8.5 meters of range estimation errors and is “not able to be solved or even mitigated by simply increasing the bandwidth of the signal transmission” [63]. Consequently, for more than a decade, researchers have been trying to identify alternative approaches that correlate less with multipath radio propagation. Dead reckoning is obvious a good choice, and is gaining more interest lately because of the advance of sensor technology. But it is still difficult to apply this method for low-speed (orpedestrian-based) indoor mobile computing. Therefore, location fingerprinting could be considered to be a promising method, but the key issue is that it requires the construction of a database table to map radio propagation signatures (such as signal strength, images, or sound) with location coordinates. Usually, database construction process is labor intensive and is generally not updatable in real-time applications.

Consequently, research community has been focusing on identifying an effective and precise indoor radio propagation model [6,24,64-69] that would make the database construction an automated process. Among those models, ARIADNE [6] is one of promising radio models that could be used to build dynamic database to map location coordinates with radio signatures (such as signal strength values from reference transmitters) for the changing environment. For example, Figure 1 shows an example in a rectangular space with 40 × 40 meters that uses signal strength values (dynamically estimated by ARIADNE) for location determination (indicated by the vertical axis). In this example, three reference radio transmitters form an equilateral triangle in the space. The figure shows that better localization results could be achieved when the user is located within or near the area of the equilateral triangle. Figure 1(b) gives a corresponding contour representation of the localization performance (in meters) for the space, and it clearly indicates that within and/or around 1.0 meter localization error could be easily achieved at locations that are covered by three reference radio transmitters. Obviously, real-time radio propagation model that correctly estimates radio propagation signatures is a key factor for the performance of location determination.

We believe simple 2-D signal-location maps may not be sufficient for complex3-D environments. As indicated in Figure 2, if a signal-location map is constructed on a horizontal plane at a distance above the floor, then signal strength measurements for mobile clients at different heights (by Δh_i, i = 1, 2) may be misrepresented by values at others locations roughly δ_i (i = 1, 2) away from their true (horizontal) positions. Moreover, for a large floor-plan or a 3-Dbuilding, simple representation of signal-location map with uniform grid distance at both directions in a 2-D plane (or three directions for 3-D buildings) may not be efficient. Therefore, a more advanced 3-D radio model must be exploited in order to provide comprehensive radio signatures, in real time, for various mobile computing applications.

More importantly, although research community has already applied both WiFi 802.11 technology (at 2.4 GHz) [6,25,26,31,53,55-57,70] and RFID¹ devices at lower frequency bands (433 MHz - 928 MHz) [36,71-77] in indoor localization research, there is still no clear conclusion on two major issues:

1) Which radio technology works more robust and effectively in indoor environment?

2) Is there a common radio propagation model that works effectively for radio technologies from 433 MHz to 2.4 GHz?

There have been a few studies that compare these wireless techniques. For example, Valenzuela [78] explored signal measurement techniques for radio transmission at 900 MHz and 2 GHz. Ali-Rantala et al. [79] studied indoor radio propagation at both 2.45 GHz and 433 MHz and concluded that 433 MHZ band is a good choice for wireless applications due to its advantages of low power consumption and large fractional coverage. Alvarez et al. [80] implemented the ZigBee system based on received signal strength for 433 MHz, 868 MHz, and 2.4 GHz frequency bands. However, the study used only a free-space model, and the evaluation was conducted in a limited space inside an industrial warehouse.

Additionally, it did not take into consideration radio attenuation effects from walls. To this date, majority research and applications mainly focus on a 2-Denvironment without considering effective radio transmission range and radio attenuation performance through floors (i.e., vertically instead of horizontally), and therefore the application of those systems in 3-D space is still not evaluated.

In this research, we first introduce an interesting experiment during an indoor construction process where radio attenuation was measured at different stages of wall construction. We then study a realistic 3-D radio propagation model to approximate radio propagation in an indoor environment. We also conduct field measurements using both WiFi (at 2.4 GHz) and RFID radios (at 433 MHz) in a two-story academic building and evaluate the effectiveness of both radio technologies in indoor wireless applications. The remainder of the paper is or-

ganized as follows: Section 2 gives measurement results through a full cycle of an indoor construction process, Section 3 introduces the 3-D radio propagation model, Section 4 presents experimental and evaluation results in a two-story building, Section 5 discusses key issues and/or parameters critical to the indoor radio propagation, and Section 6 concludes the paper and outlines future research.

Indoor radio propagation is complex, and to estimate radio signal strength, researchers have tried to identify a “super” attenuation exponent n for the log power model (Equation (1)) based on the distance to the radio transmitter [64,68,69]. Similarly, other researchers proposed that radio attenuation comes from both the distance and (wall) obstructions, where obstruction attenuation is determined by the multiplication of an attenuation factor WAF with the total number of (wall) partitions C between the receiver and the radio transmitter, see Equation (2) [24,66].

where P(d), in dBm, is the power at distance d to the transmitter in meters; P(d₀) is the power at a reference distance d₀, usually set to 1.0 meter.

where C is the total number of obstructions (wall partitions) and WAF is the wall attenuation factor.

In order to understand the intriguing nature of the indoor radio propagation, we conducted signal measurement during a remodeling process of an academic building at the University of South Carolina Beaufort. As indicated in Figures 3 and 4, the remodeling process involved the construction of four offices from an open space. We placed a radio transmitter in one of the rooms, and recorded radio signal strength values at three neighboring rooms (see Figure 4). The experiment took roughly eight months (from March to October, 2011), and we compared signal strength values at four major stages along with progress of the construction. The four stages include: 1) initial aluminum frames for walls (Figure 3(a)); 2) one side dry wall with aluminum frames; 3) one side dry wall with insulation material (Figure 3(b)), and 4) completed dry walls and finished ceiling. We show measurement results in Figure 5.

The top figure of Figure 3 gives a section view (without ceiling) of the 3 D office environment. In the experiment, receiver’s positions in rooms ♯2 and ♯3 are symmetric to the room for the transmitter, and room ♯1 is on the far left, with room ♯2 in the middle, of the transmitter. Figure 5 gives measured (average) signal strength values (y-axis) at three rooms (x-axis). We show measurement uncertainty using standard deviation among all measurements during the data collection process.

Figure 5 shows that: 1) severe attenuation from wall obstructions was clearly observed, signal strength values at room ♯3 are much lower than those from the other two rooms, especially for measurements from both “finished walls” and “one-side dry wall with insulation material”; 2) distance contributes minor attenuation, when only aluminum frame was set up (without drywall on either side), radio signal strength at three rooms are similar; 3) positions with similar wall obstructions and distance to the transmitter have similar radio signal strength values, in the figure, measurements from rooms ♯1 and ♯3 are very similar to each other; and 4) the ceiling structure may potentially increase radio signal strength for locations that are close to the transmitter. Measurements from room ♯2 and ♯3 show that radio signal strength values from the “finished dry walls and finished ceiling” are much higher than other scenarios. On the other hand, because of the larger distance between room ♯1 and the transmitter, potential contribution from the ceiling is not obvious in room ♯1. This may suggest that reflected radio rays from the ceiling may not be able to reach the receiver because of (extra) wall obstructions along the (long) propagation path. Therefore, we believe that simple distance or obstruction based models (Equations (1) and (2)) may not be able to capture the essence of the indoor radio propagation. We consequently use ray tracing method in this research.